Semigroups in Geometrical Function Theory

Semigroups in Geometrical Function Theory by

David Shoikhet Department of Mathematics, Technion-Israel Institute of Technology, Haifa, Israel

KLUWER ACADEMIC PUBLISHERS DORDRECHT/BOSTON/LONDON

A C. J.P. Catalogue record for this book is available from the Library of Congress.

ISBN 0-7923-7111-9

Published by Kluwer Academic Publishers, P.O. Box 17, 3300 AA Dordrecht, The Netherlands. Sold and distributed in North, Central and South America by Kluwer Academic Publishers, 101 Philip Drive, Norwell, MA 02061, U.S.A. In all other countries, sold and distributed by Kluwer Academic Publishers, P.O. Box 322, 3300 AH Dordrecht, The Netherlands.

Printed on acid~j'ree paper

All Rights Reserved © 200 I Kluwer Academic Publishers No part of the material protected by this copyright notice may be reproduced or utilized in any form or by any means, electronic or mechanical, including photocopying, recording or by any information storage and retrieval system, without written permission from the copyright owner. Printed in the Netherlands.

Contents Preface

vii

Preliminaries 0.1 Notations and notions . . . . . . . . . . . . . 0.2 Holomorphic functions of a complex variable 0.3 Convergence of holomorphic functions . 0.4 Metric spaces and fixed point principles . . 1 The 1.1 1.2 1.3 1.4 1.5

Wolff-Denjoy theory on the unit disk Schwarz-Pick Lemma and automorphisms . Boundary behavior of holomorphic self-mappings Fixed points of holomorphic self-mappings . . . . Fixed point free holomorphic self-mappings of~The Denjoy-Wolff Theorem. . . . . . . . . . . . . . . . . . . . Commuting family of holomorphic mappings of the unit disk.

1 1 4 6 7 9 9 17 25 32 36

2 Hyperbolic geometry on the unit disk and fixed points 2.1 The Poincare metric on~ . . . . . . . . . . . . . . . 2.2 Infinitesimal Poincare metric and geodesics . . . . . . . . 2.3 Compatibility of the Poincare metric with convexity . . . 2.4 Fixed points of p-nonexpansive mappings on the unit disk

39 39 44 46 52

3 Generation theory on the unit disk 3.1 One-parameter continuous semigroup of holomorphic and p-nonexpansive self-mappings . . . . . . . . . . . . . . . . 3.2 Infinitesimal generator of a one-parameter continuous semigroup 3.3 Nonlinear resolvent and the exponential formula . . 3.4 Monotonicity with respect to the hyperbolic metric . . . . . . . 3.5 Flow invariance conditions for holomorphic functions . . . . . . 3.6 The Berkson-Porta parametric representation of semi-complete vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

59

4 Asymptotic behavior of continuous flows 4.1 Stationary points of a flow on ~ . . 4.2 Null points of complete vector fields v

59 62 67 79 83 95

101 101 104

CONTENTS

vi

4.3 4.4 4.5 4.6 4.7

Embedding of discrete time group into a continuous flow . . . . Rates of convergence of a flow with an interior stationary point A rate of convergence in terms of the Poincare metric . . . . . Continuous version of the Julia-Wolff-Caratheodory Theorem . Lower bounds for p-monotone functions . . . . . . . . . .

109 113 120 124 135

5 Dynamical approach to starlike and spirallike functions 5.1 Generators on biholomorphically equivalent domains . . . 5.2 Starlike and spirallike functions . . . . . . . . . . . . . . 5.3 A generalized Visser-Ostrowski condition and fanlike functions 5.4 An invariance property and approximation problems . . 5.5 Hummel's multiplier and parametric representations of starlike functions . . . . . . . . . . . . . . . . . . . . . . 5.6 A conjecture of Robertson and geometrical characteristics of fanlike functions . . . . . . . . . . . . . . . . . . . . . . . . . 5.7 Converse theorems on starlike, spirallike and fanlike functions 5.8 Growth estimates for spirallike, starlike and fanlike functions 5.9 Remarks on Schroeder's equation and the Koenigs embedding property . . . . . . . . . . . . . . . . . . . . . . . . . .

153 154 157 163 166

Bibliography

205

Author and Subject Index

216

List of figures

221

172 176 186 194 198

Preface

Historically, complex analysis and geometrical function theory have been intensively developed from the beginning of the twentieth century. They provide the foundations for broad areas of mathematics. In the last fifty years the theory of holomorphic mappings on complex spaces has been studied by many mathematicians with many applications to nonlinear analysis, functional analysis, differential equations, classical and quantum mechanics. The laws of dynamics are usually presented as equations of motion which are written in the abstract form of a dynamical system: dxjdt+ f(x) = 0, where xis a variable describing the state of the system under study, and f is a vector function of x. The study of such systems when f is a monotone or an accretive (generally nonlinear) operator on the underlying space has been recently the subject of much research by analysts working on quite a variety of interesting topics, including boundary value problems, integral equations and evolution problems (see, for example, [19, 13] and [29]). In a parallel development (and even earlier) the generation theory of oneparameter semigroups of holomorphic mappings in en has been the topic of interest in the theory of Markov stochastic processes and, in particular, in the theory of branching processes (see, for example, [63, 127, 48] and [69]). Later, such semigroups appeared in other fields: one-dimensional complex analysis ([17, 28]), finite-dimensional manifolds [2, 5], Banach spaces geometry [12, 142], control and optimization theory [64], and Krein spaces [147, 148]. At the same time very rapid developments in multi-dimensional complex analysis, functional analysis and a variety of techniques and methods have diverted attention from the roots of one-dimensional dynamic approaches. Furthermore, some interesting results on evolution equations, presented in various papers in journals on nonlinear analysis, abstract analysis, and differential equations, have remained unnoticed by experts in one-dimensional complex variable. One of the first applicable models of the complex dynamical systems on the unit disk arose more than a hundred years ago from studies of the dynamics of stochastic branching processes. In 1874 F. Galton and H.W. Watson [151] in treating the problem of the extinction probability of family names, formulated a mathematical model in terms vii

PREFACE

viii

of the probability generating function: 00

F(z)

=

LPkZk,

lzl::; 1,

k=O

where z is a complex variable, Po, Pl, ... , Pk are nonnegative numbers (probabilities) such that E~o Pk = 1, and its iterations: p(O)(z)

= z,

p(n+l)(z)

= p(n)(F(z)).

The first complete and correct determination of the extinction probability for the Galton-Watson process as the limit points of the iteration sequence was given by J.F. Steffensen in 1930 [139]. Since then the interest in this model has increased because of connections with chemical and nuclear chain reactions, the study of the multi plication of electrons in the electron multiplier, the theory of cosmic radiation, and many other biological and physical problems. Detailed description of classical results concerning branching processes can be found in the books of T.E. Harris [63] and B.A. Sevastyanov [127]. We only note here that while the original Galton-Watson process has been related to discrete time branching process (i.e., it is described by an iteration process of a single probability generating function) the further development involved also the consideration of continuous time branching processes based on one-parameter semigroups of analytical self-mappings of the unit disk. One of the problems in analysis is, given a function F(z) find a function F(z, t), with F(z, 1) = F(z) satisfying the semigroup property F(z,t+s)=F(F(t,z),s),

t,s;:=:o,

where z is a complex variable. Since this formula expresses the characteristic property of iteration when t and s are integers we may consider F(z, t) as a fractional iterate ofF, when t is not an integer. G. Koenigs (1884) [78] showed how this problem may be solved, ifF is analytic self-mapping on the unit disk with an interior fixed point z 0 = F(z 0 ), such that 0 < IF'(zo)l < 1, by considering the convergence of the sequence {F(n)(z)} to zo, as n ~ oo in a neighborhood of the point z0 . These and other problems led to the following general question: Let D be a bounded domain in C and let F : D ~----> D be a holomorphic mapping of D into itself. Does the sequence of iterates {F(n)(z)}~=l converge uniformly on compact subsets of D to some holomorphic mapping h : D ~----> C? In 1926 J. Wolff (see [155]-[157]) and A. Denjoy [31] solved this problem for D = ~, the unit disk in C. Applying Schwarz's and Julia's Lemmas they proved the following remarkable results: Let F : ~ ~----> ~ be a holomorphic mapping of the unit disk of C into itself. lf F is not an automorphism of~' with exactly one interior fixed point, then {F(n)(z)} converges uniformly on compact subsets of~ to a holomorphic mapping h:~t--->~.

Moreover, if F is not the identity then h is a constant.

PREFACE

ix

This result has given a powerful thrust to the development of different aspects of complex dynamical systems on the unit disk, the complex plane, or, more generally, hyperbolic Riemann surfaces. In 1930 G. Julia [72] in publishing of geometrical principles of analysis characterized the dynamics of analytic motions in the unit disk. Over the last 40 yean; these results have been developed in at least three directions. In 1960 V.P. Potapov [110] extended the classical lemma of G. Julia to matrixvalued holomorphic mappings of a complex variable. Next I. Glicksberg [49] and K. Fan [43] established the version of Julia's lemma for function algebras. Also K. Fan [44, 45] extended a result of J. Wolff [155] on iterates of self-mapping of proper contraction operator in the sense of functional calculus. Recently K. Wlodarczyk [152]-[154] and P. Mellon [94] motivated by the research ofV. Potapov and K. Fan generalized these results to holomorphic mappings on J* -algebras. In general, they used operator theoretic methods. Finite-dimensional generalizations are to be found in G.N. Chen [24], Y. Kubota [80], B. MacCluer [91], M. Abate [2, 5], P. Mercer [96], among others. In all cases it appears that some sort of 'finiteness' or 'compactness' is required. E. Vesentini [144, 145], P. Mazet and J.P. Vigue [92, 93] have used spectral analysis for the case when a holomorphic self-mapping of a domain in a Banach space has an interior fixed point. Further observations of the Wolff-Denjoy Theory (even for one-dimensional cases) yield extensions to those mappings which are not necessarily holomorphic but are nonexpansive with respect to a hyperbolic metric on a domain. This approach has been developed by C. Earle and R. Hamilton [34], K. Goebel, T. Sekowski, T. Kuczumov, A. Stachura, S. Reich, I. Shafrir including others (see, for example, [50, 52, 53, 52, 82, 83, 129, 130]) and it is based on the study of the so called approximating curves. These results can also be considered as an implicit analogues of the DenjoyWolff Theorem. It is also remarkable that the asymptotic behavior of the approximating curves is actually nicer than that of the usual iterative process. Moreover, these implicit methods may be useful not only for a self-mapping of a domain, but also for a wider class of mappings which satisfy certain flow invariance conditions. Actually, the study of these methods shows their deep connections with continuous time dynamical systems. In fact, in 1964 F. Forelli [46] established a one-to-one correspondence of the group of linear isometrics of a Hardy space HP, p > 0, p =/= 1, and the group of Mobius transformations on the unit disk. E. Berkson and H. Porta in [17] established a continuous analog of the Denjoy-Wolff Theorem for continuous semigrou ps of holomorphic self-mappings of the unit disk. This approach has been used by several mathematicians to study the asymptotic behavior of solutions of Cauchy problems (see, for example, [16, 2, 5, 28] and [117]). Berkson and Porta [17] also apply their continuous analogue of the DenjoyWolff Theorem to the study of the eigenvalue problem of composition operators in Hardy spaces. Similar approaches were used by A.G. Siskakis [136] to study the Cesaro operator on HP [136], F. Jafari and K. Yale [70], Y. Latuskin and M. Stepin [85] for weighted composition operators and dynamical systems on HP. See

PREFACE

X

also the book of C. C. Cowen and B. MacCluer [28] and references there. These arguments motivate us to give a systematic exposition of the WolffDenjoy theory and its application to dynamical systems. In fact, it does not seem possible to cover the extensive literature concerning this subject in a single book. Nevertheless, I believe that the first step in this direction should be the understanding of dynamic processes on the unit disk of the complex plane. Actually, one can see that multi-dimensional generalizations may often be obtained by the reduction to the one-dimensional case, or by generalizations of attendant notions to higher dimensions. The study of images of domains under holomorphic (or biholomorphic) transformations links some of the most basic questions one can ask about semigroups with nice classical results in geometric function theory. As a part of analytic function theory, research on the geometry of domains in the complex plane is of old origin, dating back to the 19th century. The basic methods of geometrical function theory include the square principles, methods of contour integration, variational methods, extreme metrics, and integral representation theory. Good introductions in these topics can be found in the books [55, 87, 11] and [57]. An exhaustive bibliography on geometrical function theory was compiled by S.D. Bernardi [18] before and until 1981. This book is not intended to be a survey of the theories mentioned above. Our primary focus will be only on the material investing old theorems with new meanings which are related to applications of evolution equations to the geometry of domains in the complex plane. For example, it is a well known result, due to R. Nevanlinna (1921) [102], that if h is holomorphic in [z[ < 1 and satisfies h(O) = 0, h'(O) # 0, then h is univalent and maps the unit disk onto a starshaped domain (with respect to 0) if and only if Re[zh'(z)/h(z)] > 0 everywhere. This result, as well as most of the work on starlike functions on the unit disk, can be obtained from the identity f)

ae arg h(re

iiJ )

= Re

{rei 8 h'(re;8 )} h(reiiJ) .

This idea does not readily extend to a higher-dimensional space. Moreover, such an approach is crucially connected with the initial condition h(O) = 0. Much later Wald [150] characterized of those functions which are starlike with respect to another center (sometimes these functions are called weakly starlike [65, 66]). Observe that although the classes of starlike, spirallike and convex functions were studied very extensively, little was known about functions that are holomorphic on the unit disk t:l. and starlike with respect to a boundary point. In fact, it was only in 1981 that M.S. Robertson [121] introduced two relevant classes of univalent functions and conjectured that they were equal. In 1984 his conjecture was proved by A. Lyzzaik [90]. Finally, in 1990 Silverman and Silvia [134], using a similar method, gave a full description of the class of univalent functions on t:l., the image of which is starshaped with respect to a boundary point. However, the approaches used in their work have a crucially one-dimensional character (because of the Riemann Mapping Theorem and Caratheodory's Theorem on Kernel Convergence). In addition, the conditions given by Robertson and by Silverman and

PREFACE

Xl

Silvia, characterizing starlikeness with respect to a boundary point, essentially differ from Wald's and Nevanlinna's conditions of starlikeness with respect to an interior point. Hence, it is difficult to trace the connections between these two closely related geometric objects. Therefore, even in the one-dimensional case the following problem arises: To find a unified condition of starlikeness (and spimllikeness) with respect to an interior or a boundary point. By 1923 K. Lowner [89] described the problem of continuous deformations of a domain of certain class by using a first order differential equation au at

au k(t)

+u

= az k(t) - u u,

u(O, z) = z, z E ~ = {lzl < 1}, 0 < t < T, lk(t)l = 1, the right hand of which is time dependent and has a moving polar singularity. This equation is central in the theory of parametric extensions of univalent functions and its application (see, for example [11] and references there). In particular, L. de Branges [30] used it to solve the famous Bieberbach conjecture. It seems that the idea to use autonomic dynamical systems was first suggested by Robertson [120] in 1961 and developed by Brickman [20] in 1973, who introduced the concept of Ill-like functions as a generalization of starlike and spirallike functions (with respect to the origin) of a single complex variable. Suffridge, Pfaltzgraff [104, 105, 140, 141, 106] and Gurganus [58] (see also [26] and [109]) developed a similar approach in order to characterize starlike, spirallike (with respect to the origin), convex and close-to-convex mappings in higher-dimensional cases. Since 1970 the list of papers on these subjects has became quite long. Nevertheless, it seems that there is no extension of Wald's, as well as of Silverman and Silvia's, results to higher dimensions. In the last chapter we demonstrate how one can study starlike and spirallike functions by using the behavior of trajectories defined by related dynamical systems and their characterizations. For example, in the description of starlike functions, this approach, roughly speaking, consists of the following observation: If we consider the flows defined by the evolution equations dujdt + h(u) = 0 and dvjdt+v = 0, the condition of starlikeness (with respect to a point in the closure of a domain), translates to the fact that h takes integral curves of the first evolution equation into those of the second one. So the problem is to find a suitable condition to describe those holomorphic functions which generate a converging semigroup of holomorphic self-mappings of a domain. Note that such approach can be useful also for higher dimensions (see, for example, [140, 141, 104, 105, 106, 58, 26, 109, 56, 36] and [41]). My purpose has been to write a tutorial rather than a scientific book. And though many results are previously contained only in journal papers, I have tried to facilitate the understanding of their proofs to students and nonspecialists. Certain chapters of this book were used in my lectures on Geometrical Theory of Complex Variables at Ort Braude College, Karmiel. More advanced sections were presented in a special course for the College teachers and at the Seminars on Functional Analysis and Nonlinear Analysis at the Technion, Haifa.

xii

PREFACE

It seems that the book will be useful to students and postgraduate students in engineering who may use Complex Dynamic Systems. Furthermore, I hope that it will be of interest to mathematicians specializing in complex analysis and differential equations. My joint work, as well as numerous talks, with Simeon Reich, Mark Elin aud Dov Aharonov convinced me that Generation Theory may serve as a showcase for the Classical Geometrical Function Theory. Simeon Reich has given a series of lectures on Complex Analysis and Hyperbolic Geometry in the Complex Plane. His initiative has stimulated me to design a course on the relationships of various topics related to one-dimensional analysis. I would like to thank ORT Braude College, Karmiel and the Technion, Haifa for their support throughout the project. I am very grateful to my colleagues Mark Elin, Giora Enden, Yakov Lutsky, Ludmila Shvartsman for their help. Special thanks go to Mark Elin who examined the manuscript in detail and whose proofreading and astute observations led to significant improvements in Chapter V. I am deeply indebted to my wife Tania for typing the final version.

Preliminaries In this short chapter we compile the series of very basic notions and results, probably familiar to most readers. As we mentioned in the Preface, only a very modest preliminary knowledge is required to read the following material. Nevertheless, certain fundamental topics, such as those related to integral representations, convergence theorems, and fixed point principles, will be used throughout the text, and therefore should be presented at least an auxiliary material.

0.1

Notations and notions

Throughout the book we shall use the following notation: lR- the set of all real numbers (real axis); lR + = { x E lR : x ~ 0} - the set of all nonnegative real numbers (the nonnegative real half-axis);

C - the set of all complex numbers z = x + iy, x, y E JR, i 2 = -1 (i.e., the complex plane). If z = x + iy E C then z = x- iy is its conjugate complex number, Re z := x, Imz := y and [z[ = Jx2 + y2. As usual, if z E Cis represented in the form z = [z[ei 9 , then () = arg z. II+ = {z E C : Re z ~ 0} ~ the right half plane. ~r(a) = {z E C: [z- a[< r} -the open disk of radius r > 0 centered at a E C. ~ = ~r(O) = {z E C: [z[ < 1}- the open unit disk in C. Definitions of a few technical terms will be given here. An open connected subset D ofC is called a domain in C. The symbol aD will denote the boundary of D. In particular, a~= {z E C: [z[ = 1} is the unit circle. When z represents any number of a set D C C, we call z a complex variable. The set D is usually a domain in C. If to each value z in D we assign a second complex variable w, then w said to be a function of the complex variable z on the 1

2

set D: w

= f(z).

The term 'function' signifies a single-valued function unless specified differently. When w = f (z) and w and z are complex variables some information about f may be conveniently illustrated graphically, however, two separate complex planes for the two variables z = x + iy and w = u + iv are required.

f

Figure 0.1: The function w = f(z). The correspondence w = f (z) between points in the two planes is called a mapping or transformation of points (or sets) given by the function f.

-~

-2

Figure 0.2: The translation f(z)

= z +a, a= 4 + 2i.

We use the following geometrical terms in the form: translation: f(z) = z +a, a E C, (Figure 0.2) rotation: f(z) = ei 0 z, BE (0, 2n), (Figure 0.3) contraction: f(z) = kz (Figure 0.4). It is sometimes convenient to consider the mapping as a transformation in a single plane. Sometimes to conform our terminology with the theory of dynamical systems we will refer to the 'vector field' f : D ~----> C, as a vector f(z) whose origin is at z E D (cf., Figure 0.1 and Figure 0.5).

PRELIMINARIES

3

-1

-2

Figure 0.3: The rotation f(z) = ei-rroz, 8 = -2rr/3.

0

-

::2

:L

:L

Figure 0.4: The contraction f(z)

= kz,

k

= 1/3.

f(c;

Figure 0.5: The veclu1 f1dd w = f(z).

4

0.2

Holomorphic functions of a complex variable

Iff is a complex-valued function defined on an open subset D of C we say that f is holomorphic in D if for each zo E D the quotient

f(z) - f(zo) z- zo has a limit inC as z approaches z0 . When this limit exists it is called the derivative off at zo and is denoted as f'(zo). Iff is holomorphic in D and w = f(z) maps D into the set n C C we say that f maps D holomorphically into n and write f : D ~--+ n. The set of all holomorphic functions (mappings) in D with values in a set n c C will be denoted by Hol(D, f!). If such a mapping is one-to-one on a domain D a function f E Hol(D; C) is said to be univalent on D. In particular, Hol(D, C) is the vector space of all holomorphic functions in D. We will simply use Hol(D) to denote the set Hol(D, D) of holomorphic selfmappings of D. IfF and G are in Hol(D), then If»= FoG is also in Hol(D), where o denotes the composition operation on this set, i.e., IP(z) = F(G(z)). A fundamental property of holomorphic functions is the Cauchy Integral

Formula: If f E Hol( D, C), then for each closed contour "' in D and each z E D \ "' _1

27ri

J

f(()d( = (- z

{ f(z), 0,

if z inside "' otherwise.

(0.2.1)

'"'(

This formula leads to the most basic and important results in the theory of holomorphic functions. In particular, iff E Hol(D,C), then f'(z) also belongs to Hol(D,C), hence by induction, f is infinitely differentiable. We adopt the notation ~:t(z), for the n-th derivative off at the point zED. The following integral formula is known as the general Cauchy formula

dn f n! dzn (z) = 27ri

J

f(()d( ((- z)n+l'

n = 0, 1, 2, ... ,

(0.2.2)

'"'(

where "' here is any closed positively oriented contour in D, such that z is inside "f.

In this formula as well in the Cauchy formula (0.2.1) the contour "' can be replaced by the oriented boundary aD of a domain D c C, whenever f admits a continuous extension to an. Furthermore, if f E Hol(D, C) then for each point zo E D there is a disk 6-r(zo) = {z E C : lz- zol < r} in D such that f admits Taylor's power series

PRELIMINARIES

5

extension at zo: 00

f(z)

=

L an(z- zot,

z E D-r(zo),

n=O

where ao = f(zo), That is, the infinite series converges to f. Another very important consequence of the Cauchy integral formula is the so called maximum modulus principle: Iff E Hol(D, C) is not a constant and z 0 E D then every neighborhood U of zo contains points z such that

lf(z)l > 1/(zo)l. In other words, if for f E Hol(D, C) the modulus 1/1 attains a maximum in D, then f is constant. Further, any real function u = u(x, y) of two real variables x and y that has continuous partial derivatives of the second order and that satisfies Laplace's equation {)2u

cPu

8x2

+ 8y2

= 0

throughout a domain Dis called a harmonic function in D. For f E Hol(D,C) the functions u = Ref(x + iy) and v = lmf(x + iy) are known to be harmonic. The maximum modulus principle imply maximum and minimum principles for harmonic functions: If u = u(x, y) is a nonconstant harmonic function in D, then it cannot attain neither maximum nor minimum in D. The function u = Ref plays a crucial role in the study of holomorphic functions on the unit disk D.. In particular, each f E Hol(D., C) can be expressed in terms of its real part: f(z)

1

r

( +_

= ilmf(O) + 21ri 1-r Ref(() (

z d( z ('

(0.2.3)

where"(= {( : 1(1 = r} is a circle of radius 1 > r > lzl. This formula is often referred to as the Cauchy-Schwarz representation formula. The holomorphic functions fonD. whose image lies in the right half-plane II+, play a special role in further discussions. A remarkable result proposed by (0.2.3) and proved by Herglotz and Riesz (see, for example [9, 57, 122]) is an integral representation of positive harmonic functions on D.: ifp E Hol(D.,q with p(O) = 1 and Rep(z) > 0, zED., then p(z) =

1

1 +z(

- - - dm((), IC:I=l 1 - z(

where m is a probability measure on the unit circle.

zED.,

6

0.3

Convergence of holomorphic functions

The vector space Hol(D, 0. Then It" _, h if and only if the following conditions hold: (i) for every wE h(l:!.) ther·e is a neighborhood n such that n lies in h,.(l:!.) for n larye enough; (ii) for w E: 8h(l:!.) there is a sequence w, E 8h,(.6.) such that w,. --+ w as n --+ oo. Note that according to Caratheodory's concept, conditions (i) and (ii) are known for the sequence {h,.(l:!.)} to be convergent with respect to h(O) in the sense of kernel converqence to h( .6.).

PRELIMINARIES

0.4

7

Metric spaces and fixed point principles

A metric space is a nonempty set D endowed with a distance function (or metric) d : D x D -+ R+ = [0, oo) which satisfies the following properties: (i) (symmetry) d(x, y) = d(y, x), for any pair x, y in D; (ii} (reflexivity) d(x, y) = 0 if and only if x = y; (iii) (triangle inequality) d(x, y) ~ d(x, z) + d(y, z) for any triplex, y, z in D.

Note that in differential geometry or higher-dimensional complex analysis, such a function d satisfying (i)-(iii) is called 'distance' (or pseudo-distance, if it satisfies (i) and (iii) only) [47], but we prefer to stay with the notation metric of the classical functional analysis. If d is a metric on a domain D C C it is sometimes convenient to call the pair (D, d) a metric space. If each Cauchy sequence (with respect to this metric) in D converges to an element of D then (D, d) is said to be a complete metric space. If a ED the set B(a,r) ={xED: d(x,a) < r} is called the (open) ball of radius r and (metric) center in a. The function d 1 (z,w) = lz- wl, z,w E
dz(z, w) :=

I1-zw z- w_l·

It can be shown that (~, d2 ) is a complete metric space (see Sections 1.1 and 2.1). The hyperbolic metric on the unit disk 1 d3(z,w) :=-In 2

z-w 1+ I-I

1- zw z -w 1 - -1 1-

1-

zw

is due to Poincare. Although the Poincare metric requires additional computations it has some nice features related to the measure of the lengths in the spirit of Riemann (see Section 2.1). Constructing a metric that is invariant under a class of mappings is one of the basic tools for a geometrical approach in analysis. Let (D, d) be a general complete metric space and let F be a self-mapping of D. We will write in this case F : D f--+ D. If in addition, for each pair x, y in D d(F(x), F(y))

~

d(x, y)

we will say that F is d-nonexpansive (or nonexpansive, with respect to d). The term: 'F is a contraction with respect to d' may also be used.

8 We will see below (Sections 1.1 and 2.1) that each holomorphic self-mapping of the unit disk A is a contraction with respect to metrics d2 and d 3 . The central problem in the theory of d-nonexpansive mappings of an underlined metric space (D, d) is finding their fixed points. A point a E D is said to be a fixed point ofF : D f-4 D if F(a) =a.

A very powerful tool in solving existence problems in many branches of analysis is The Banach Fixed Point Theorem (or Banach's contraction principle) which appeared explicitly in his thesis in 1922 (see, for example, [125]). Let (D, d) be a complete metric space and let F : D f-4 D be a strict contraction of (D, p) in the sense that: d(F(x), F(y))::; qd(x, y),

x,yE D

with some 0 < q < 1. Then F has a unique fixed point in D. Moreover, for each x E D the sequence of iterates {F(nl(x)} converges to this fixed point. Here F< 1 l = F, F< 2 l = F oF, p(n+ll = F o (F
Chapter 1

The Wolff-Denjoy theory on the unit disk There is a long history associated with the problem on iterating holomorphic mappings and their fixed points, the work of H.A. Schwarz (1869), G. Pick (1916), G. Julia (1920), J. Wolff (1926), A. Denjoy (1926) and C. Caratheodory (1929) being among the most important.

1.1

Schwarz-Pick Lemma and automorphisms

Let D be a domain (open connected subset) in the complex plane C, and let Hol(D, C) be the set of all holomorphic functions (mappings) from D into C. Throughout what follows we will denote by .6. the unit disk of the complex plane C, and by Hol(.6.) the set of holomorphic self-mappings of .6.. This set is a semigroup with respect to the composition operation. The subgroup of Hol(.6.) of all holomorphic automorphisms of .6. will be denoted by Aut(.6.). In other words, FE Hol(.6.) is an element of Aut(.6.) if and only if p-I is well defined on .6. and belongs to Hol(.6.). We begin with the classical Schwarz Lemma which plays a crucial role in geometric function theory.

Proposition 1.1.1 (The Schwarz Lemma) Let FE Hol(.6.) be such that F(O)

=

(i) (ii)

0. Then

IF(z)l

~

lzl

for all z E .6.;

IF'(O)I ~ 1.

Moreover, if the equality in (i) holds for at least one z E .6., z =I 0, or IF'(O)I = 1, then F(z) = ei'Pz for some tp E [0, 271'], and thus the equality in (i) holds for· all z E .6.. 9

10

Chapter I Proof. Since F(O) = 0 the function F has the following Taylor representation: 00

F(z) = :~::..:Ukzk.

k=l

Therefore the function

F(z) G(z) = - - - = "'

L akzk-l k=l 00

is also holomorphic in D.. Take any r E (0, 1) and denote D.r = {zED.: lzl < r}. By the maximum modulus principle we obtain: IG(z)l =

IF~z) I :S ~'

z E D.r.

Letting r---. 1- we obtain the inequality: IG(z)l :S 1 for all z E .6., which is equivalent to (i). Note also that G(O) = F'(O), whereby (ii) follows too. If, in addition, IG(z)l = 1 for some z E .6., then once again by the maximum modulus principle, G(z) = ei'P for some rp E [0, 211"], hence F(z) = ei'Pz for all z E .6.. 0 Let us now fix a point a E D. and consider a fractional linear transformation (the so called Mobius transformation) of the unit disk defined by

z+a 1 + az

ma(z) = --_-.

(1.1.1)

It is easy to see that the following inequality holds:

In turn, this inequality implies that lma(z)l < 1, whenever zED. and a is a fixed element of D.. That is, ma maps D. into itself and ma E Hol(D.). Since m;; 1 = m_a also belongs to Hol(D.), we have that ma E Aut(D.). Now let h be an arbitrary element of Aut(D.). We now claim that h is of the form: h(z) = Ama(z) for some unimodular number A. Indeed, if h(O) =bE .6., then m-b(b) = 0, and 9 := m_b o h

satisfies the conditions of the Schwarz Lemma, i.e., 9 E Hol(D.) and g(O) = 0. Hence lg(z)l :S izl for all z E .6..

THE WOLFF-DENJOY THEORY

11

On the other hand g- 1 = h- 1 o mb also belongs to Hol(D.), and g- 1 (0) = 0. Therefore, lg- 1 (w)l ::; lwl, wE .6.. Substituting here w = g(z) we obtain lg(z)l ~ lzl, hence, in fact, lg(z)l = lzl. Consequently g(z) = ei'Pz for some cp E [0,27r], and h( z )

= mb (g ( z )) =

e i

=

+ e -i
·i•n Z

e ,..

Setting a= e-i
and (1.1.3)

with cp = 2 arg(l + ab). Exercise 2. Show that the following relations hold: m;; 1 ( -m-a) - I = -1n-a· Exercise 3. Let I denote the identity mapping in C, i.e., I (z) = z, z E C. For a fixed a E .6. define a Mobius transformation Ma by the formula Ma(z) = -m-a(z). Show that Ma satisfies the so called involution property: Mao Ma =I, i.e.,

M;; 1 =

Ma.

The invariant form of the Schwarz Lemma is the following assertion. Proposition 1.1.3 (The Schwarz-Pick Lemma) Let F E Hol(.6.). Then for each pair z and w in .6. the following inequality holds:

or, explicitly, F(z)- F(w) I < I z- w I I1F(z)F(w) - 1- zw

(the Schwarz-Pick inequality).

Moreover, if equality holds in this relation for at least one pair z -1- w in .6. then FE Aut(D.).

12

Chapter I

Proof. Fix w E ~ and consider the mapping G = m-F(w) oF o mw. It is clear that G E Hol(~) and G(O) = 0. Then by the Schwarz Lemma we have IG(z)l :::; lzl, z E ~- Replacing z by TlLw(z) we obtain the required inequality. If this relation becomes equality for some pair z j w, then IG(m-w(z))l = lm-w(z)l. Again by the Schwarz Lemma G(z) = ei'Pz for some r.p E [0, 21r] and for all z E ~­ Hence G belongs to Aut(~), and so does F = mF(w) o Go m-w· 0 Corollary 1.1.1 Let F E Hol(~, C). Then F only if it satisfies the condition:

maps~

into its closure

~

if and

(1.1.4)

Proof. Indeed, if (1.1.4) holds, then 1- IF(z)l 2 2: 0 and IF(z)l :::; 1. Conversely. Suppose that IF(z)l :::; 1. If IF(z)l = 1 for some z E ~. then F is a constant and (1.1.4) is trivial. Thus we may assume that IF(z)l < 1 for all z E ~- Now, rewriting the Schwarz-Pick inequality in the form

and letting w

--+

z we obtain (1.1.4). 0

Exercise 4. Show that ifF is not a constant and (1.1.4) holds as an equality at one z E ~. then F E Aut(~). Hint. As in the proof of Proposition 1.1.3, consider the mapping G = m-F(w)o F o mw. Show that G'(O) = 1. Corollary 1.1.2 (Lindelof's inequality, [88], p. 11) IfF E satisfies the following estimate: lzl + IF(O)I IF(z)l:::; 1 + lziiF(O)I'

z

E ~-

Proof. Since 1- lmF(O)

F(zW = (1- IF(O)I2)(1- IF(z)l2) 11 + F(O)F(z)l2

we have

F(z)- F(O) 12 I1F(z)F(O)

> 1 _ (1- IF(O)i2)(1- IF(z)i2) (1 -IF(O)IIF(z)l) 2 (IF(z)I-IF(O)I) 2 (1 -IF(O)IIF(z)l) 2 .

Hol(~).

then it

THE WOLFF-DENJOY THEORY

13

On the other hand, it follows by the Schwarz-Pick inequality that F(z)- F(O) I < I1F(z)F(O) -

Jzl

and IF(z)J-JF(O)J 1 -IF(O)JJF(z)J :::; lzi. Solving the last inequality for JF(z)J we obtain our assertion.

0

Corollary 1.1.3 (see L.A. Harris [61]} IfF E Hol(~) then it satisfies the following estimate 1-IF(O)i2 JF(z)- F(O)J:::; Jzl1 -JF(O)Jizi Proof. Consider the mapping G = m-F(O) oF. Solving this equation for F we obtain: F z - m G z - G(z) + F(O) ( ) F(o)( ( )) - 1 +F(O)G(z)

and F(z) _ F(O)

But G(O)

= G(z) + F(O)- F(O)- G(z)JF(O)i2 = G(z) 1 -JF(O)i2 1 + F(O)G(z) 1 + F(O)G(z)

= 0, hence JG(z)J :::; JzJ,

z E ~.

This implies the required estimate.

D

Remark 1.1.1 Observe also that (1.1.4) implies the estimate:

(1.1.5) for each F E Hol(~) and z E ~. Therefore for each r E (0, 1) we obtain the following uniform Lipschitz condition: 1 IF(z)- F(w)J :S 1 _ .,. 2 Jz- wJ,

(1.1.6)

whenever z and w belong to ~r = {z E ~: Jzl < r}. Now iff E bounded, i.e., Jf(z)J ::; M for all z E ~'then F

1

= Mf

Hol(~,q

is

E Hol(~).

Thus we have that each bounded holomorphic function on Lipchitzian with respect to Euclidean distance in C.

~

is locally uniformly

Remark 1.1.2 From the point of view of geometric function theory on the unit disk, the Euclidean distance on the disk is inappropriate. The pseudo-hyperbolic

14

Chapter I

distance between points defined as follows:

z

and w of /:).. is the function d : /:).. x /:).. d(z, w)

-->

JR+ = [0, oo)

= lm-w(z)i.

It is easy to see that, actually, d is a metric on /:).. that induces the usual Euclidean topology. Thus the Schwarz-Pick Lemma states the contraction property of a mapping F E Hol(/:)..) with respect to this metric: d(F(z), F(w)) ~ d(z, w),

z, wE/:)...

Moreover, F is an isometry with respect to d (i.e., d(F(z), F(w)) and only ifF E Aut(~).

= d(z, w))

if

Exercise 5. Show that for each r E (0, 1) lim

inf d(z.w) = 1.

lwi-+I- lzi:Sr

Consider now a few geometrical properties of the action of the group which will be useful in the sequel.

Aut(~)

y

X

c

r Figure 1.1: An orthogonal circle

to()/:)..

and its image under an automorphism.

Let r be a circle orthogonal to a~, the boundary of~. r n ~ f. 0. If cis the center of r, then lei 2: 1, and Jicl 2 - 1 is the radius of r (see Figure 1.1). Thus the circle r satisfies the equation:

r = {z E C:

izi 2

-

2Re(zc)

+ 1 = 0}.

(1.1.7)

THE WOLFF-DENJOY THEORY

15

Consider now the set 'Y = r n ~ which is the arc of r in ~ with end points A and B (see Figure 1.1). We raise the following question: what is the image of 'Y under F E Aut(~)? The next assertion answers this question.

Proposition 1.1.4 lf r is an orthogonal circle to 8~1 and 'Y = r n ~I then for each FE Aut(~) the image F('Y) of'Y is either the intersection of~ with another orthogonal circle to 8~ 1 or it is a diameter of~Proof. IfF is a rotation of~. i.e. F(z) = ei'Pz, r.p E [0,2rr], z E ~.then by (1.1.7) we obtain: F('Y) n ~={wE C: lwl 2

-

where c1 = ei'Pc and the proof is completed. IfF is a Mobius transformation ma, a E

2Re(wcJ.) ~.

+1=

0},

we have for wE ma('Y) n ~:

or ((lwl 2 + 1)- 2Re(wf)) = 0, where ( = (1 + 2 Re(ca)

+ lal 2) and TJ = c + ca 2 + 2a. If ( #- 0 then lwl 2 - 2 Re(wc2)

+1=

(1.1.8) (1.1.8) implies:

0,

where c2 = TJ / (. Hence ma ('Y) n ~ is an intersection of~ with an orthogonal circle to a~. If ( = 1 + 2 Re(ca) + lal 2 = 0, then by (1.1.7) we have that -a E 'Y n ~- Hence 0 = ma( -a) E ma('Y) n ~ and once again it follows from (1.1.8) that Re (wi))

=

0.

This equation represents a straight line through zero, hence ma ('Y) n ~ is a diameter of~Since each F E Aut(~) has the form ei'Pma, r.p E [0, 2rr], z E ~. (see Proposition 1.1.2.) our proof is completed. D This assertion states that each automorphism of~ preserves the class of subsets of ~ which consists of all diameters of ~ and all arcs of the circles which are orthogonal to the boundary of ~We will see below that this result allows us to define the so called geodesic segments on~. whose 'length' is preserved under automorphisms. Another useful image property of automorphisms is the following. Let the mapping ma : ~ r-+ ~be defined as above by formula (1.1.1).

Proposition 1.1.5 The images of concentric disks ~r = {z E ~: lzl < r}, r E (0, 1) under the Mobius transformations ma, a E ~~ are disks in~ centered at 1

c

1- r 2 r21 a 12 . a

= c( r) = 1 -

(1.1.9)

16

Chapter I

------> Figure 1.2: A Mobius transformation of

~r·

with radius d = d(r) =

1-JaJ 2 J J • r. 1- r 2 a 2

(1.1.10)

Proof. Let nr(a) := ma(~r)· Then ~r = m_a(nr(a)). In other words, nr(a) ={wE~: Jm-a(w)J < r}. So, we need to solve the inequality

w-a - I
F(nr(a)) <;;; nr(F(a)),

r E (0, 1),

aE

~.

(1.1.12)

i.e., F maps the disk nr(a) centered at c = c(r), defined by (1.1.9), into the disk nr(F(a)) centered at 1- r 2 cl = 1- r2[F(a)l2 . F(a) with the same radius d = d(r), defined by (1.1.10). However, in fact, the numbers c = c(r) and d = d(r) also depend on the location of the point a E ~. In particular, if a approaches the boundary 8b.. then for a fixed r E (0, 1) the point c also goes to the boundary while the radius r tends to 0. This is a deficiency in the study of the boundary behavior of holomorphic self-mappings of the unit disk. We will consider this problem in the next section.

THE WOLFF-DENJOY THEORY

Exercise 6. For

17

a E .6., the closure of .6., and K ;::: 1 - \a\ 2 , define the set ·{ \1- za\ 2 } D(a, K) = z E .6.: _ \z\ 2 < K . 1

(a) Show that for an interior point a E .6. and r E (0, 1) the sets flr(a) (= ma(.6.r)) and D(a,K) with K = (1 -\a\ 2 )(1- r 2 )- 1 coincide. (b) Show that if a E 8.6. -the boundary of .6., then for each K > 0 the set D(a, K) is a disk in .6., centered at a/(1 + K) E .6. with radius K/(K + 1) < 1.

Remark 1.1.4 To avoid confusion with the symbol flr(a) we point out that it differs from the symbol .6-r(a), which usually denotes the disk centered at a with radius r. At the same time the sets flr(a) are balls with respect to the metric defined by pseudo-hyperbolic distance on .6. (see Remark 1.1.2). Thus an interpretation of the Schwarz-Pick Lemma (or inclusion (1.1.13)) is that a holomorphic mapping of .6. into itself is a contraction in this metric.

1.2

Boundary behavior of holomorphic self-mappings

In this section we want to trace dynamics of the Schwarz-Pick inequality when the center of a pseudo-hyperbolic ball is pushed out to 8.6.. We will present some classical statements which are based on celebrated results of G. Julia in 1920 [71] and C. Caratheodory's contribution in 1929 [22] (see also [23]). For a point ( E .6. and K > 1 - \(\ 2 let us define the sets D ((, K) by the formula

\1-z(\2 } D((, K) = { z E .6.: l _ \z\ 2 < K .

(1.2.1)

It is not difficult to see that for an interior point ( E .6. the set D((, K), K > 1 -\(\ 2 , is exactly the pseudo-hyperbolic ball flr(() centered at the point (with

VI-

radius r = l-k(l 2 (cf., Exercise 6, Section 1.1). Note, however, that D((, K) make sense even if (is a boundary point. In this case computations show that for each K > 0 the set D((, K) is geometrically a with radius K~ 1 < 1, i.e., disk in .6., centered at

rfx(

D((,K) = { z

E.6.: 1: lz-

K(l <

K: 1}.

(1.2.2)

Chapter I

18

This disk is internally tangent to the boundary of D. at the point ( (Figure 1.3). It is called a horocycle in D..

Figure 1.3: A horocycle at the point ( E 8D.. The following assertion establishes the invariance property of horocycles D( (, K) with respect to the family Hol(D.), similarly as was mentioned for domains S1r(() (see Remark 1.1.3).

Proposition 1.2.1 (Julia's Lemma) Let F E Hol(D.) and let ( E 8D. be a boundary point of D.. Suppose that there exists a sequence {zn}::'=l C D. converging to ( as n goes to oo, such that the limits

a= lim 1n--+oo

IF(zn)l

1 - lzn I

(1.2.3)

and T}

=

lim F(zn)

n--+oo

(1.2.4)

exist {finitely). Then, for each z E D. the following inequality holds

11- F(z)ijl 11- z(l 2
2 -'------,-,~~2

In other words, for each K

(1.2.5)

> 0 we have the following inclusion:

F (D((, K)) ~ D(TJ, aK).

Proof. It follows that (1.2.4) and (1.2.5) induce ITJI = 1. For z, w E D. we define the function _ (1 - lwl 2 ) (1 - lzl 2 ) (1.2.6) a ( z, w ) I _ . 1-wz 12

THE WOLFF-DENJOY THEORY

19

It is a simple exercise to show that the Schwarz-Pick inequality (Proposition 1.1.3) can be rewritten in the form

u(z,w):::; u(F(z),F(w)) for all FE

Hol(~)

(1.2. 7)

and z, wE~- In particular, we have:

u(z, Zn) :::;

0"

(F(z), F(zn))

for each z E ~ and all n = 1, 2, .... Writing the latter inequality explicitly we obtain after simple manipulations: 11- F("z.JF(zW (1 -IF(znW)I1- z;;:zl 2 2 (1-IF(z)l ) :::; (1 -lznl 2)(1 -lzl 2) · Letting n go to infinity we obtain (1.2.5). D Exercise 1. Prove formula (1.2.6) directly. Exercise 2. For FE Hol(~) and ( E number of F at () by the formula:

8~,

define the value a((, F) (the Julia

a(( F)= liminf 1 -IF(z)l ' z-+( 1- lzl '

(1.2.8)

where z tends to ( unrestrictedly in ~If a((, F) is finite, it follows by Julia's Lemma, that 11- F(z)iJI 2 < (~" F) 11- z(l 2 1 -IF(z)l2 -a .,, 1 -lzl2 ' where Tf

(1.2.9)

= n-+oo lim F(zn), {z,.} being a sequence along which the lower limit in (1.2.8)

is achieved. Show that if for at least one z E automorphism of ~-

~

we have equality in (1.2.9) then F is an

Remark 1.2.1 Actually inequality (1.2.5) means that for each K > 0, under

conditions (1.2.3) and (1.2.4), F maps the horocycle with radius R = K

K

+1

centered at - K ( into the horocycle with radius r = ;K centered at KTf +1 a +1 a +1 (see formula (1.2.2)). Therefore, ifF is a nonconstant holomorphic mapping, then the number a in (1.2.3) must be positive. Also, the same conclusion can be obtain by using the following result of the Schwarz-Pick Lemma. Exercise 3. Show that for each FE Hol(~) and z E holds: _1---'---IF,. .:. ,(z.. .:. .:.)l > 1- IF(O)I_ 1- lzl - 1 + IF(O)I

~the

following inequality

20

Chapter I

Remark 1.2.2 Observe also, that if the limit a in {1.2.3) exists for a sequence Zn -+ ( E 8/:l, then so does limit rJ E 8/:l in {1.2.4). Indeed, (1.2.5) implies that IF(zn)l -+ 1 as n -+ oo. Therefore, each subsequence of {F(zn)};:"= 1 has a further subsequence which converges to some unimodular point. Choose two such subsequences and take their limit points, say 'r/1 and "12, lrJ1I = lrJ2I = 1. If 'r/1 f. 'r/2, by decreasing K if necessary, we can find two horocycles D( ry 1 , o:K) and D( ry2 , o:K) whose intersection is empty. But from (1.2.5) for each z E D((, K) the element F(z) must lie in both of them. Contradiction. Hence 'r/1 = 'r/2, and the limit in (1.2.4) exists. Moreover, we will show that, in fact, this limit exists (and equals to ry) for each sequence {zn} which converges to ( along so called nontangential directions. More precisely: Definition 1.2.1 For a point ( on the unit circle 8/:l and"' > 1 a nontangentiai approach region at ( is the set r((, K)

= {z E !l: lz- (I < K(1 -lzl)}.

(1.2.10)

The term 'nontangential' refers to the fact that f((, K) lies in a sectorS in !l at point ( which is the region bounded between two straight lines in !l that meet at (and are symmetric about the radius to(, i.e., the boundary curves off((, K) have a corner at (, with an intersection angle less than 1r (see Figure 1.4).

1

Figure 1.4: A nontangential approach region at a boundary point.

Definition 1.2.2 We will say that a function f E Hol (!l, C) has a nontangentiu (or angular) limit L at a point ( E 8/:l if f(z)-+ L as z-+ (, z E f((, K) for eac, "' > 1. We will write in this case L

= L.lim

z-+(

f(z).

THE WOLFF-DENJOY THEORY

21

Exercise 4. A Stolz angle at ( E 8!::l. is the set

S= {z

E !::l.: larg(1- (z)l

< /3, lz- (: < r, f3 E (O,n/2), r E (0,2cosf3)}

(see, Figure 1.5). Show that f E Hol(!::l., C) has a nontangentiallimit L at a point ( if and only if f(z)---+ Las z---+ (,for each Stolz angleS at(.

1

Figure 1.5: A Stolz angle at a boundary point.

A simple consequence of .Tulia's Lemma is the following. Corollary 1.2.1 Let F be a nonconstant holomorphic self-mapping of !::l., and let Zn which converges to (, such that

( E 8!::l.. Suppose that there is a sequence

. . 1 -IF(zn)l hmmf =a< oo. z,.-+( 1- lznl Then: (i)

Q

> 0;

(i·i) the nontangential limit TJ :=

L lim F(z), z-+(

which is a point of the boundary 8!::l. exists; (iii) for each K > 0 the following inclusion holds

F(D((, K))

~

D(ry, aK).

(1.2.11)

Chapter I

22

Proof. It remains to show only that condition (1.2.11) implies assertion (ii). Indeed, given e > 0 choose K > 0 such that D( ry, aK) is contained in the e-disk centered at ry. Further, let S be a sector in .0. with its vertex at (. Then one can find r5 > 0 such that S1 = S n {z E C: lz- (I

< r5} c D((, K).

Hence by (1.2.5) we have

IF(z)- TJI < e for z E S1 (see figure 1.6) .D

1

Figure 1.6: Boundary behavior of a self-mapping of .0.. Finally, we adduce a much stronger assertion, established by Caratheodory (see [22] and [23]).

Proposition 1.2.2 (The Julia-Caratheodory Theorem) Let FE Hol(.0.) an let ( E 8.0.. Then the following are equivalent: (i)

li~1~~1f 1 ~ ~~? =

a

<

x, where the limit is taken as z approaches (

unrestrictedly in .0.; (ii) L_ lim F(z) Z->(

Z-

~ TJ

:= L_F'(() exists for a point 71 E 8.0.;

(iii) L lim F'(z) exists, and L lim F(z) = TJ E 8.0.. z-.(

z-<

THE WOLFF-DENJOY THEORY

23

Moreover, (a) a> 0 in {i); (b) the boundary points TJ in (ii} and (iii} are the same; (c) L lim F'(z) = LF'(() = a(ry. z~(

The value LF'(() is called the angular derivative ofF at ( (see also Section 4.6). There are various proofs of the Julia-Caratheodory Theorem (see for example, the papers [157, 86, 143, 126, 54, 123] and the books [23, 103, 131, 28]. In fact, we will prove below (see Section 4.6) in detail several general assertions which extend Proposition 1.2.2. Nevertheless, in order to demonstrate a direct method and for the completeness we give here a proof which is based on two important classical statements. Proposition 1.2.3 (Lindelof's Principle, [88]) Let ( E 86. and let function f E Hol(6., C) be bounded on each nontangential approach region at(. If for some continuous curve 'Y E 6. ending at ( there exists the limit

L = lim f(z), z_,(

z E "(,

then the angular limit

L lim f(z) = L z_,(

also exists.

Proposition 1.2.4 (see [107], p. 79) Let ( E 86. and let f E Hol(6., C). Suppose that the limit L = L lim f(z) := f(() z_,(

exists (finitely). Then

Llim f(z)-f(() Z- (

z_,(

exists if and only if L lim J'(z) z_,(

exists and both coincide.

For a proof of this proposition see also Section 4.6. Proof of Proposition 1.2.2. First we note that the implication (iii) => (ii) is obvious due to Proposition 1.2.4. In turn, (ii) => (i) because of the inequality 1 -IF(r()l ITJ- F(r()l < ' 1- r I(- r(l

--~~~

which holds for all r E (0, 1).

Chapter I

24

Now if (i) holds, then by Corollary 1.201 there exists

L lim F(z) := TJ E 8!:::.0

(1.2012)

Z-+(

Thus we need to show that (i) and (1.2012) imply (ii)o To this end we observe that for each K > 1 and z E f((, K) we have by Julia's Lemma -2

11- F(z)i/1 2 1 -IF(z)l2

< o:ll-z(l =o:lz-(llz-(1 1 - lzl2

1 - lzl2 1-lzl

< o: lz- (I"' 1- lzl2 = o:Kiz- (lo On the other hand

IF(z)- TJI (1 -IF(z)l)

IF(z)- TJI 1 + IF(z)l

1 -IF(z)l 2 11- F(z)i/1 2 1- IF(z)l2 0

< This implies that

~~~z~ ~ITJI

:::; o:"' (1 + IF(z)l) :::; 2o:K, 0

whenever z E f((, K)o In other words, the functwn f(z) :=

F(z)-TJ

z-(

0

IS

bounded

on each nontangential approach region at (o Therefore, to complete the proof, it is sufficient to show (by using Proposition 1.203) the equality 0

Inn

r--+1-

TJ-F(r() (-

r(

= o:(ryo

or, equally,

11m TJ- F(r() 0

r--+1-

1-

1"

=

O:TJo

(1.2013)

Indeed, since o: is the lower limit in (i) we have lim 117- F(r()l > lim 1 -IF(r()l > O:o r--->1-

1-

1"

-

r-+1-

1-

1"

-

Ou the other hand, setting .:: = r( in (1.205) we obtain

10 1111 r-+1-

117- F(r(W (1

-

1" )"-

<

(}

0 1 -IF(r()l 2 11111 ----'---'---:'..:....:.__ r-11-r2

a

0 1 -IF(r()l (1 + IF(r()l) 2 1- r 2 11m r_,1- 1 + IF(r()l (1 + r) 2 (1- r) 2

a

0 1 -IF(r()l 1- r 2 11111 r-1- 1 + IF(r()l (1- r)2

(1.2014)

THE WOLFF-DENJOY THEORY o

=

25

lim (1-IF(r()l)2 1-r2 1-IF(r()l2 (1-r)2 111 - F(r()J2 1 - r 2 . lIm ~~~~----~ r~1- 1 -IF(r()l 2 (1- r)2 r~1-

<

Q

<

Q

2

. 11- r((l-2 1- r 2 1Im = Q2. 1-r2

r--+1-

(1.2.15)

(1-r)2

Hence, we obtain from (1.2.14) and (1.2.15) lim 111- F(r()l

1- r

r~I-

=

lim 1 -IF(r()l r--+1-

=0

(1.2.16)

1- r

and

111-F(r()l . l 1m 1 -IF(r()l

HI-

. 11- ryF(r()l l Im 1 -IF(r()l

r~1-

l1·m r--+1-

11- ryF(r()l = 1. ( Re 1 -17F(r())

(

1.2.17

)

Thus by (1.2.16) we can write

. 17- F(r() i

(1.2.18)

r--+1-

where

c.p = lim arg(1- ryF(r()). r--+1

But (1.2.17) implies that c.p = 0 and we obtain (1.2.13) from (1.2.18). Thus the proof is completed. D

1.3

Fixed points of holomorphic self-mappings

The Schwarz-Pick Lemma implies that if ( E .1. is an interior fixed point of FE Hol(.1.), i.e., (1.3.1) F(() = (, then F leaves each pseudo-hyperbolic ball f2r(() centered at (invariant. In other words, for each r E (0, 1), (1.3.2) where

.

f2rl()

= {z

I z _, I } = {z E .1.: 111 __ z(l lzl 2

E .1.: 1 _ (z < r

2


}

,

(1.3.3)

Chapter I

26

with K = (1- 1(1 2 ) (1- r 2 ) -l (see Section 1.1, Exercise 6). In turn, this result shows that a holomorphic self-mapping of~ which is no· the identity has at most one interior fixed point in !}. (see Proposition 1.3.4). An additional consequence of the Schwarz-Pick Lemma is that if ( E /}. is 1 fixed point of /}., then (1.3.4 IF'(()I ~ 1. Moreover, the equality in (1.3.2) or (1.3.4) holds if and only ifF is an auto morphism of /}.. These facts are helpful in the study of the asymptotic behavior of the dis crete time semigroup defined by iterates of a holomorphic self-mapping of!}. (se4 Proposition 1.3.2). The situation becomes more complicated if F E Hol(!}.) has no fixed point: inside /}.. If ( E 81}., the boundary of/}., one can define it as a boundary fixed point o F by the relation (1.3.5 lim F(r() = (. r-+1-

However, simple examples show that holomorphic self-mappings of!}. may hav' many fixed points on the circle 81}.. We begin first with the case ofF E Aut(!}.). 1. Fixed points of automorphisms. We already know that ifF E Aut(!}.) then it can be presented in the form:

F(z) =

.

e''~'m-a(z)

. z-a 1- az

= e''~'--_-

for some a E /}. and 'PER Hence, for such F, equation F(z) the quadratic equation:

(1.3.6

= z is equivalent t4 (1.3.7

or

az 2 + (e-i

(1.3.8

The simplest situation is when a = 0. In this case either F is a rotation abou the origin F(z) = ei'~'z, 'P E (0, 211"), or F is the identity F(z) = z, z E /},, Ther respectively, either F has exactly one fixed point ( = 0 in /}. or F has infinite! many fixed points in /}.. If F is not the identity and it is not a rotation about zer then a f- 0, and z = 0 is not a root of (1.3.8). So we may multiply this equatio by -ei'l' jz 2 , resulting in:

Thus z f- 0 is a root of (1.3. 7) if and only if 1j z is a root of ( 1.3. 7). Consequently, ( 1. 3. 7) has at most one solution inside !}. . If we assume the (1.3. 7) has a unimodular solution, then either it is unique or the second solutio has also modulus 1.

27

THE WOLFF-DENJOY THEORY

Exercise 1. Show that if (1 and ( 2 are solutions of equation (1.3.7), then for all z E ~ \ { (2} the following relation holds:

F(z)- F((l) F(z) - F((2) where FE

Aut(~)

1- a(2 z- (1 1 - a(1 z- (2'

(1.3.9)

is defined by (6).

Thus we see that for FE Aut(~) the following three situations arise according to the location of its fixed points: {i) F has exactly one fixed point in ~; {ii) F has exactly one fixed point on a~ and no fixed points (iii} F has two different fixed points on a~. The automorphisms of tively:

~

in~;

are classified according to these situations, respec-

In {i} F is said to be elliptic; In (ii) F is said to be parabolic; In {iii} F is said to be hyperbolic. Applying the Schwarz Lemma it is easy to see that an elliptic automorphism F of ~ has the form: F = mc; o r'P o m_c;, where ( E ~is the solution of (1.3.1) and r'P is a rotation about zero. Therefore, by Proposition 1.1.5, F is a 'rotation' about ( (see Figure 1.7).

F( 3) ( zo) ..--------- ---/_.... F( 2 ) ( zo) ·---,,

'

/

(

\f(zol

\ F(n)(zo)

} 1

\\

0

./ ,. '//

..,..

---,_______________________.... zo

Figure 1.7: Elliptic automorphism.

28

Chapter I

The dynamical behaviors of parabolic and hyperbolic automorphisms are presented on the Figures 1.8 and 1.9.

0

\

1

tF(n)(zo)

Figure 1.8: Hyperbolic automorphism. To explain tho::>e behavior::> we first as::>ume that F, defined by (1.3.6), has two different fixed points, say (1 and (z, (1 =!= (z, on 86.. Set >. := 1 - a(z = (z- a . (2. (1.3.10) 1 - a(1 ( 1 - a (I Since (j = F((j ), j = 1, 2, we have -

1- a(1 = e

i

cp

(j -a --c;-·

Hence

>. = 1 - a(z = 1 -a(!

(2 - a . (1 (I -a (2

=

(2 - a . (2. (I -a (1

Comparing the latter expression with (1.3.10) we conclude that >. = X is a real number. Further, it is clear that >. =!= 1, since a =/= 0 and ( 1 =/= ( 2 . Also >. =/= -1, because otherwise 2 l(t + (21 = lal > 2 = l(tl + 1(21· In other words, we have shown that 1>-1 =/= 1. Exercise 2. Show that

>., defined by (1.3.10), is positive.

29

THE WOLFF-DENJOY THEORY

0

\ F( 3 )(zo),/

\\ ····---...

F( 2) ( zo)

_.//

............ _____________......

Figure 1.9: Parabolic automorphism.

If we now introduce the fractional linear transformation L : C the formula:

t---+

C defined by

then, by using relation (1.3.9) we obtain:

L (F(z))

= >-.L(z),

and consequently:

F(z) = L - 1 (>-.L(z)). This implies by induction that

p(n)(z)=L- 1 (>-.nL(z)),

n=0,1,2, ... ,

where p(n) are iterates ofF: F( 0 l(z) = z, p(n)(z) = F(F(n- 1 l(z)), n = 1, 2, .... Thus, if 1>-.1 < 1 then for each z E ~the sequence {F(n)(z)}:'=o converges to

L- 1(0) =

(1- If

1>-.1 > 1 then this sequence converges to

(2. Since

{p(n)}~=O

is a

normal family on ~. this convergence is uniform on each compact subset of ~. So ifF is a hyperbolic automorphism of~' i.e., F has exactly two (distinct) fixed points on the boundary of~' then the sequence of itemtes { p(n)} ;:"= 0 converges uniformly on compact subsets of~ to one of them. Moreover, one can estimate such a convergence.

Exercise 3. Prove that for each z E

~

the following rate of convergence holds:

30 where ( 1>-1 > 1.

Chapter I

=

(t and

E

= 1>-1 if 1>-1 < 1, or, respectively, ( = (z and

E

= 1>-l- 1 if

Now we consider the case when F is a parabolic automorphism, i.e., F has exactly one fixed point ( on the boundary of Ll. In this case ( is a double root of equation (1.3.6) and we have:

1 - ei"'

2(=-_-. a

Since 1(1 = 1 and lal < 1 it follows that ei'P cannot be -1. Then by direct calculations one can verify the relations:

( F(z)-(

ei"'- 1 ( + -ei'P+1 z-(

---..,.--'--- = - - -

and

( ei"' - 1 = n--p(n)(z)-( ei'P+1

=-:-....,...,......,..---

(

+z-(

for all z ELl and n = 0, 1, 2, .... The latter equality implies that p(n)(z) converges to ( E ail as n tends to infinity. To summarize our considerations we formulate the following assertion.

Proposition 1.3.1 If F E Aut(Ll) is not an elliptic automorphism, then the sequence of iterates { p(n)} is convergent. Moreover, ifF is not the identity, then the limit of this sequence is a unimodular constant, which is a boundary fixed point of F.

:"=o

This result is the first step in the proof of a more general assertion on the asymptotic behavior of holomorphic self-mappings of the unit disk, called the Denjoy-Wolff Theorem. 2. Iterates of holomorphic self-mappings oft. with an interior fixed point.

Proposition 1.3.2 Let F E Hol(t.) have a fixed point ( E Ll. Then: (i) for each r E (0, 1) and n = 0, 1, 2, ... , the following invariance condition holds:

p(n)(flr(()) ~ flr((), where nr(() is defined by (1.3.3); (ii) ifF is not the identity then the point ( E Ll is a unique fixed point ofF in Ll. Moreover, the following are equivalent:

:"=o

(the orbit) converges to ( as n (a) For each z ELl, the sequence { p(nl(z)} goes to infinity. (b) The mapping F: Ll ~--> Ll is not an automorphism of Ll. (c) IF'(()I < 1.

THE WOLFF-DENJOY THEORY

31

Proof. By the induction method condition (i) is an immediate consequence of inclusion (1.3.2). Also, (1.3.2) implies that ( E b. is a unique fixed point of F, if F is not the identity. Indeed, if we assume that F has two different fixed points, say ( 1 and (2, ( 1 =f. ( 2 , then one can choose r 1 and r 2 in (0, 1) such that (2 ~ Drt((I), ( 1 ~ Dr 2 ((2) and Drt((I) nDr 2 ((2 ) = !1 =f. 0. It is clear that !1 is a convex closed subset of b. and F(D) ~ !1. Then it follows by Brouwer's Fixed Point Principle (see Section 0.4) that there exists ( 3 = F((3 ) in n which is obviously different from ( 1 and (2. Repeating these arguments we can find a converging sequence {(n}:'= 1 C b. such that (n = F((n)· By the uniqueness property this implies that F(z) = z, for all z. To prove the second part ofthe assertion we first note that implications (a)=>(b) and (b)=>(c) follow directly from the Schwarz-Pick Lemma. Therefore it is enough to prove the implication (c)=>(a). Since F'(z) is a continuous function on b. there is a disk b.r(() C b. centered at (with radius r > 0 such that

IF'(z)l < 1

(1.3.11)

for all z E b.r((), the closure of b.r((). In turn, (1.3.11) implies that F satisfies the Lipschitz condition (1.3.12) IF(z)- F(w)l ::; q lz- wj, where q = max { IF'(z)J, z E b.r(() }· In addition, we have from (1.3.12) that F maps b.r (() into itself:

IF(z)- (j ::; qjz- (J. So F is a self-mapping of b.r(() which is a strict contraction. It then follows by the Banach Fixed Point Theorem (see Section 0.4) that {F(nl(z)}:'=o converges to ( for all z E b.r((). Using the Vitali theorem (see Section 0.3) we prove our assertion. 0 Combining Proposition 1.3.2. with Brouwer's Fixed Point Principle we obtain the following sufficient condition of existence and uniqueness of an interior fixed point for holomorphic self-mappings of the unit disk.

Corollary 1.3.1 Suppose that FE Hol(b.) maps b. strictly inside, i.e., for some rE(0,1)

IF(z)i::; r for all z E b.. Then F has a unique fixed point ( E b., j(J ::; r, and for each z E b. the orbit { pCnl(z)} :'=o converges to ( as n goes to infinity.

1.4

Fixed point free holomorphic self-mappings of .6.. The Denjoy-Wolff Theorem.

:=

We will say that F is power convergent if the sequenceS= { p(n)} 1 converges uniformly on any subset strictly inside D.. If the limit of this sequence is a constant ( E D. then it is called an attractive point of S. Clearly, if ( is an interior point of D. then it is a unique fixed point of F. In this section we intend to study the dynamics of holomorphic self-mappings of .6. with no fixed points inside. A simple case of this situation occurred in the previous section where we saw that an automorphism of D. with no fixed points has to be either hyperbolic or parabolic, with its fixed point on the boundary of D.. The content of a remarkable result which was essentially obtained simultaneously by J. Wolff and A. Denjoy is that this fact continues to hold for any holomorphic self-mapping of D. with no fixed point inside. In other words, each F E Hol(D.) which has no fixed points in D. is power convergent to its boundary fixed point in the following sense: lim p(nl(z) = ( E 86.. n-+oo

and lim F(r() = (.

r-1-

We have already mentioned that a holomurphic self-mapping of .6. may have many fixed points on the boundary 86.. of D.. So, an additional question is how to recognize which of these is attractive. The key to the answer arrives from Julia's Lemma and the Julia-Caratheodory Theorem where the value of the angular derivative defines such a point (see Proposition 1.4.2 below ). Note also that a consequence of the Schwarz-Pick Lemma (Proposition 1.3.2) tell us about the invariance condition in neighborhoods of an interior fixed point of .6.. For mappings with no fixed points a similar result was established by Wolff [157] where pseudo-hyperbolic disks were replaced by horocycles at a certain boundary point of D..

Proposition 1.4.1 (Wolff's Lemma) Let F E Hol(D.) have no fixed points in D.. Then there is a unique unimodular point ( E 86.., such that for each K > 0 and n = 0, 1, 2, ... , the horocycle

J1- z(J2 } D((,K) = { zED.:
(1.4.1)

Proof. Actually it is sufficient to show that there is a sequence { Zn} ~= 1 converging to (, which satisfies the conditions of Julia's Lemma, i.e., the limit

THE WOLFF-DENJOY THEORY

33

(1.2.8):

a= a((, F):= liminf 1 -IF(z)l z-+( 1 - lzl exists. Moreover we will show that 0 < a ::; 1 and L lim F (z) = (. Then our assertion results as a consequence of Proposition 1.2.2. Take any arbitrary sequence {rn}~=O E (0, 1) increasing to 1, and consider the mappings Fn = rnF. Then by Corollary 1.3.1 Fn has a fixed point (n E l:J... In passing to a subsequence, we may assume that {(n} ~=O converges to a point ( E ~. If ( E ~' then by continuity we have

( = n-+oo lim (n = lim rnF((n) = F(() n-+oo contradicting the assumption that F has no fixed point in Consequently ICI = 1. In addition,

~.

for all n = 1, 2 .... Once again, on passing to a subsequence we conclude that the Julia number

a((, F)= liminf 1 -IF(z)l, z-+(

1 -lzl

(see Section 1.2) exists and is less or equal to 1. It is clear that

1 F((n) = -(n rn converges to ( and by Julia's Lemma we obtain the inequality (1.4.2) which implies (1.4.1). D Exercise 1. Prove the uniqueness of such a point ( E 8~ for which (1.4.2) is satisfied. Hint: Use the following geometrical property of horocycles as in Figure 1.10 and the invariance property (1.4.2).

We will call such a point ( E /J~ which satisfies the Wolff's Lemma a sink point ofF on 8~. Combining Proposition 1.4.1 with Proposition 1.2.2 one obtains a characterization of a fixed point free holomorphic self-mapping of the unit disk ~ which is sometimes called the Julia-Wolff-Caratheodory Theorem.

Chapter I

34

Figure 1.10: Uniqueness of a point (.

Proposition 1.4.2 Let FE Hol(A). Then the following are equivalent. (i) F has no fixed points in A; (ii} there is a unique unimodular point ( E 8A such that a:= L lim F(z)- ( Z- (

Z-+(

exists with 0
and

L lim F' (z) z-+(

=

a ~ 1;

(iv) there is a unique unimodular point ( E 8A such that liminf 1 -IF(z)l =a< 1· z-+( 1 - lzl - '

(v) there is a unique unimodular point ( E 8A such that sup cp<(F(z)) =a< 1, zEtl.

cp(lz)

-

where

Moreover, (a) the boundary points ( and the numbers a in (i)-(v) are the same; (b) the nontangential limits in (ii) and (iii) can be replaced by the radial limits.

THE WOLFF-DENJOY THEORY

35

Remark 1.4.1 Proposition 1.4.2 (as well as previous Julia's Lemma, the JuliaCaratheodory Theorem and Wolff's Lemma), can be therefore considered as a boundary version of the Schwarz-Pick Lemma; the point (is pushed out to 8/::l. in a suitable manner. The Julia number a= o:((, F) defined by condition (iv) is, in fact, the angular derivative ofF at the boundary fixed point(, which is also a sink point of F by condition (v). In addition, if a < 1, then for each n = 0, 1, 2, ... , and all z E !::l. we have by induction (1.4.3) So ( E 8/::l. is an attractive boundary fixed point ofF, i.e., F(n) - the iterates of F converge to ( uniformly on each compact subset of !::l. with the power rate of convergence (1.4.3} in the sense of the non-Euclidean 'distance' cpc;(z) (cf. Corollary 1.3.2}. The question is whether this point is also attractive when a = 1. The affirmative an:swer to thi:s que:stion i:s given in the next a:s:sertion, following Wolff [155, 156, 157] and Denjoy [31].

Proposition 1.4.3 IfF E Hol(!::l.) has no fixed points in !::l., then there is a unique unimodular point ( E 8/::l. which is a sink point ofF, and the iterates F(n) converge uniformly on each compact subset of !::l. to (.

:'=l

Proof. We may assume that F ~ Aut(!::l.). Since { F(n)} is a normal family there is a subsequence { F(n] l} which converges to a holomorphic mapping G : !::l. ~ !::l.. First we show that G must be constant. Indeed, assuming the contrary we have by the maximum modulus principle that G E Hol(!::l.). In passing to a subsequence (if necessary), we may assume that the sequences of integers Pj = nJ+ 1 - nj and qj = pj - 1 tend to infinity, and that the corresponding sequences { F(PJ l} and { F(qJ l} converge to holomorphic mappings, say h and g, respectively. It follows by the continuity of the composition operation, that

hoG= lim (F
j---+00

Since G is not a constant, h is also not a constant and it has more than one fixed point in !::l.. Hence h must be the identity. At the same time

go F = lim p(qJ l oF = lim pCPJ l = h = I J---+00

J---+00

and

Fog= F o lim p(q1 ) = lim pCPJl = h =I. j-HXJ

j---+oo

Thus F = g- 1 is an automorphism. Contradiction. Thus G = ( E !::l. is a constant. Now, by Wolff's Lemma (Proposition 1.4.1) (must be a sink point of F. Then each convergent subsequence {F(nkl} has the same limit ( E 8/::l.. D

Chapter I

36

To summarize, we formulate the following result which is sometimes called the Denjoy-Wolff Theorem.

Proposition 1.4.4 A mapping FE Hol(~) is power convergent on~ if and only if it is not an elliptic automorphism of~- Moreover, the limit of the sequence { p(n)} is a constant ( E ~-

:=O

1.5

Commuting family of holomorphic mappings of the unit disk.

The following result of A.L. Shields [132] is a consequence of the Wolff-Denjoy Theory on the unit disk ~-

Proposition 1.5.1 Let:=: be a commuting family of continuous mappings on ~. such that each F E 3 is an element of Hol( ~) Then there is a common fixed point ( E ~for all FE:=:. Proof. If there is G E :=: which has a unique fixed point ( in FE:=: we have: G (F(()) = F (G(()) = F((),

~.

then for each

i.e., F(() is also a fixed point of G, hence F(() = (. Suppose now that all mappings in :=: have at least two different fixed points. This means that for F E :=: there is a point ( E ~. such that { p(n)(z)} converges to ( for all z E ~- But for each G E 3:

:=o

and we have finished. D Let us consider now a family 3 0 of commuting holomorphic self-mappings of which are not necessarily continuous on ~- If at least one element of 3 0 has a fixed point in ~. then this point is unique, hence it is a common fixed point for all mappings in 3 0 . However, it may be not attractive for all elements of 3o if 3o contains an elliptic automorphism of ~- If all elements of 3 0 have no fixed points, then each of them has a sink point on the boundary, which is attractive for a given mapping from 3o. The question is whether there is a common sink point for all elements of 3 0 ? In general, the answer is 'no', as the following example shows. ~

37

THE WOLFF-DENJOY THEORY

Example 1. Consider the pair F and G of hyperbolic automorphisms of D. defined as follows: F(z)= z-_a, 1- az

G(z) = z +a

1 + iiz '

It is clear that F and G have the same fixed points commuting with G:

a E (I

D.

·

and ( 2 on 8!:!. and F is

FoG= GoF= I. But if one of these points, say (I, is a sink point of both mappings F and G we will have that for a given point w E D. there is a horocycle D((I, K), K > 0, such that w fJ. D( (1, K). But there is an integer n such that p(n) ( w) E D( (I, K) and hence Q(n) (F(n)(w)) E D((I, K). At the same time

that is, a contradiction. D So it does not hold in general that commuting elements of 2o have the same sink points, but it was shown by D.F. Behan [15] that the only exceptional cases involve pairs of hyperbolic automorphisms of D.. We will give his result without proof.

Proposition 1.5.2 IfF E 2 0 is not a hyperbolic automorphism of D. and G E 2o is fixed point free, then F and G have the same sink point on the boundary of D...

Chapter 2

Hyperbolic geometry on the unit disk and fixed points Another look at the Wolff-Denjoy theory is the using of the so called hyperbolic metric of a domain.

2.1

The Poincare metric

on~

Here, we consider some elements of the classical hyperbolic geometry on the unit disk and we will show that the Wolff-Denjoy theory can be extended from Hol(b.) to a wider class of self-mappings of .6.. More detailed representation of the hyperbolic geometry can be found in [55, 47, 52, 32] and [28]. For a class of self-mappings of a domain it is a natural approach to introduce a metric, such that each mapping of this class becomes nonexpansive with respect to such a metric. For the class of holomorphic mappings there are different ways to define such metrics. For example, one may use the invariant metric defined by the pseudohyperbolic distanced(= d(·, ·))on b. (see, section 1.2). However, from the classical Riemann metric theory point of view, it is impossible to measure the 'lengths' of vectors tangent to b. to obtain d (see details below). It turns out that the only way to eliminate this deficiency is to define the so called Poincare hyperbolic metric:

p = tanh- 1 d, which inspite of an additional level of computation provides the needed properties in the construction of a non-Euclidean geometry in the spirit of Riemann. Let z and w be arbitrary points in .6.. Let 1 be a circle such that z, wE/, and 1 is orthogonal to 8.6. (see Figure 2.1). 39

40

Chapter II

Figure 2.1: The points of anharmonic relation. Let z* E -y n 86. and w* E -y n 86. be such that z* is on the same side as z and w* is on the same side as w. The anharmonic relation for these four points is: (z*,z,w,w*)

z- z* w- w*

= -----z- w* w- z*

(2.1.1)

Exercise 1. Take a fractional linear transformation g of IC, g(z)

=

(az

+ b)(cz + d)- 1 ,

ad- be =1- 0,

and denote u

= g(z),

v

= g(w),

u*

= g(z*),

v*

= g(w*).

Show that (z*,z,w,w*)

=

(2.1.2)

(u*,u,v,v*).

Formula (2.1.2) shows that anharmonic relation is invariant under a fractiona: linear transformation. In particular if we take the Mobius transformation

we have '11Lz(z)

where r < 1, cp E R Hence, the line l =

nLz

= 0 and m_z(w) = rei'P

E 6.,

b) is a straight line passing through zero and

Using (2.1.2) we obtain: *

•

ei'P

re'"' - ei'P

(z , z, w, w ) = ---e'-.
1-r 1+r

1 -jrrLz(w)l 1 + lm-z(w)l'

(2.1.3

HYPERBOLIC GEOMETRY

Proposition 2.1.1 Let p: b. x b.

r-;

41

lR be a function with the following properties:

(i) for each FE Aut( b.), and each pair z, w in b. p(F(z), F(w)) = p(z, w); (ii) for each pair s,t E (0, 1), s < t p(O, t) = p(O, s) (iii)

+ p(s, t);

lim p(O, t) = 1. t-+O+

t

Then ( p z,w ) =

1

-2 1og

*) ( * z ,z,w,w

=

1 1 + lm-z(w)l 2 1og 1-lm-z(w)f"

(2.1.4)

Proof. First we note that by (i)

p(z, w) = p(m-z(z), nLz(w)) = p(O, m_z(w))

(2.1.5)

because m_z E Aut(b.). Also, using the rotation automorphism r'f' : r'f'(u) ei'f'u, u E b. with

p(O, u) = p(r'f'(O), r'f'(u)) = p(O,

iuf).

=

(2.1.6)

Comparing with (2.1.5) we obtain (2.1.7)

p(z, w) = p(O, lm-z(w)f). So it remains to show that for each r E (0, 1)

(2.1.8)

p(O, r) = tanh- 1 (r). Indeed, for each 6 > 0, such that r

p(O, r

+ 6 < 1 we have

+ 6)- p(O, r)

by (ii) and (i)

p(r,r+6) p(m-r(r), m_r(r, +6)) p ( O, 1 -

r! -r2) ·

Hence by (iii)

d+ p(O, r) dr

1

d- p(O, 1·) dr

1- r 2

1

-r2"

Similarly, 1

and we have (2.1.8):

p(O,r) =

1 r

0

ds

1 -2 = tanh- (r)

1- s

Chapter II

42

since p(O,O) = 0. It is also clear that the function tanh- 1 (lm-z(w)l) satisfies conditions (i)-(iii). 0 Exercise 2. Prove the following property of the real function 1-; d

=
Proposition 2.1.2 The function p(·, ·) defined by (2.1.4} is a metric

on~.

Proof. It is clear that p(z, w) 2: 0 for all (z, w) E ~ x ~'and since lm-z(w)l = lm-w(z)l we have p(z,w) = p(w,z) and p(z,w) = 0 if and only if z = w. Now we need to show the triangle inequality: p(z,w)

~

p(z,u)

+ p(u,z)

(2.1.9)

for each triple z, w, u in ~. Indeed, let 'Y: [0, 1] ....... ~be a curve joining the points z and w, such that 'Y'(t) is piecewise continuous. We will call such a curve an admissible curve. Consider the function: o(z, w) = inf{L,., 'Y is an admissible curve joining z and w },

where L,. is 'length' ofT

1 1

L,.=

0

b'(t)l 2 dt. 1- l"f(t)l

(2.1.10)

We claim that o(z,w) = p(z,w). Indeed, it is sufficient to show that o(·,·) satisfies the conditions of Proposition 2.1.1. In fact, if F E Hol(~) and 'Y is an admissible curve joining z and w, then F o 'Y is an admissible curve joining F(z) and F(w). It follows by Corollary 1.1.1 that

Hence o(F(z), F(w)) If now FE

Aut(~),

o(z, w).

(2.1.11)

then applying (2.1.11) for p-I we obtain the equality: o(F(z), F(w))

which proves (i).

~

= o(z, w),

HYPERBOLIC GEOMETRY

43

To prove (ii) and (iii) it is sufficient to evaluate 6(0, s ), for 0 0 and s, then so does f3 = ReI· Then we obtain:

L..,

=

{1 h"(t)i dt

lo

1 1

>

< s < 1. If 1 joins

0

1 - ir(t) 12

f3'(t) [/3( t )j2dt = 1-

1 8

0

dr

- =tanh 1-r2

-1

(s),

i.e., 6(0, s) 2:: tanh- 1 (s). On the other hand, for 1(t) = ts, L.., = tanh- 1 (s) we have: 6(0,s) = tanh- 1 (s). Now it follows by properties (b) and (c) of Exercise 2 that function 6 satisfies conditions (ii) and (iii) of Proposition 2.1.1. Thus 6(z, w)

=

p(z, w).

Now it can be seen directly that (2.1.9) is a consequence of the definition of 6(·, ·). 0 As a matter of fact we establish somewhat more in our proof. Proposition 2.1.3 Each mapping F E

Hol(~)

is nonexpansive with respect to

the metric p, i.e., p(F(z), F(w))

~

(2.1.12)

p(z, w).

Moreover, if equality in (2.1.12) holds for some pair z,w E ~.then

FE

Aut(~).

Actually, the conclusion of this proposition is also a direct consequence of the Schwarz-Pick inequality and (2.1.4). The metric, whose existence we have just established, is called the Poincare metric on ~ and will be denoted by p. Proposition 2.1.4 The pair ( ~. p) is a complete unbounded metric space, and p defines on

~

the same topology as the original topology on

~.

Proof. We list several topological properties of the Poincare metric which prove our assertion. (a) limn-. 00 p(O, Zn) = oo if and only if izni Indeed, by (2.1.6) we have:

--+

1-.

and the assertion follows.

(b) Each p-ball B(a,R) = {z E ~: p(a,z) formula (1.2.4), where r = tanhR, i.e.,

< R} is the disk Or(a) defined by

B(a,R) = {z E ~: iz- sal< t· tanhR},

(2.1.13)

Chapter II

44 where s

=

1- (tanh R) 2 2' 1 -(tanh R) 2 lal

=

In particular, if a= 0, then B(O, R) at zero.

1-lal2

(2.1.14) 2" 1- (tanh R) 2 1al {z E .6. : lzl < tanh R} is a disk centered

t=

(c) B(a, R) = ma(B(O, R)), i.e. each p-ball is an image of a disk centered at zero under the corresponding Mobius transformation. Now, it is clear that if Zn E .6. converges to z E .6. in the sense of p-metric, i.e., p(zn, z)-> 0, then lzn- zl ::; lzn- szl + (1- s)lzl -> 0

because s --+ 1-, when R ---> 0 (see (2.1.14)). Conversely, if lzn - zl Zn, z E .6., then there is r E ( 0, 1) such that Z11 E .6.r = {z E .6. : Iz I ::; r}. Hence,

--+

0 for

tanh- 1 lm-z(z.. )l

< t

an

h-IIz.. -zl 1- r2

0 --+

.

Finally, note that each p-ball is bounded away from the boundary of .6., and this implies the completeness of the metric space (.6., p). D

2.2

Infinitesimal Poincare metric and geodesics

Using property (iii) of Proposition 2.1.1 and formula (2.1.7) one can prove a more general assertion than mentioned in part (iii) of Proposition 2.1.1.

Proposition 2.2.1 Let z be any point of .6. and u E
+ u) =

lui

1 -lzl2 (1

+ .s(u)),

(2.2.1)

where e (u)-> 0, as u-> 0.

Exercise 1. Prove formula (2.2.1). Formula (2.2.1) shows that the linear differential element dpz in the Poincare metric is defined by the formula. ldzl dpz = 1 - lzl2.

(2.2.2)

HYPERBOLIC GEOMETRY

45

Definition 2.2.1 The form a:(z,u)

lui

= 1-lzl2'

z E ~'

(2.2.2') on~.

is called the infinitesimal (or differential) Poincare hyperbolic metric

The following properties are an immediate consequence of the definition:

(a) a:(z, u) ;:::: 0, z E ~' u E C. (b) a:(z, tu) = ltl a:(z, u) , t E C. By Corollary 1.1.1 of the Schwarz-Pick Lemma, each FE tion for the infinitesimal Poincare metric.

Proposition 2.2.2 IfF E

Hol(~)

Hol(~)

is a contrac-

then dpF(z) ::; dpz or equivalently,

a: (F(z), dF(z)) ::; a:(z, dz).

(2.2.3)

IfF E Aut(~), then the equality in (2.2.3) holds for all z E ~. Moreover, if the equality in (2.2.3) holds for at least one z E ~~ then FE Aut(~).

This notion allows us, using Riemann integration, to define the 'length' of any admissible curve in ~.

Definition 2.2.2 Let "( : [0, 1] r> ~ be an admissible curve in points z and w in ~. Then the quantity L-y(= L-y(z, w)) =

1

dp-y(t)

"'

=

1 1

0

b'(t)l 1- b(t)l

~

joining two

2 dt

is called the hyperbolic length of 'Y.

We have already used this notion in Section 1 and have seen that the hyperbolic length is greater than or equal to the hyperbolic distance between its end points, i.e., (2.2.4) p(z, w)::; L-y(z, w).

Definition 2.2.3 A curve"( joining points z, w in~ is called a geodesic segment in 6.. if its length is equal to the hyperbolic distance between its end points z and w, i.e., (2.2.5) L-y(z, w) = p(z, w). The following property of geodesics has been discussed, for example, in [47, 52].

Proposition 2.2.3 For each pair z and w in~ there is a unique geodesic segment joining z and w and it is either a direct segment, if z and w lie on a diameter of ~~ or a segment of the circle in C which passes through z and w and is orthogonal to 86.. the boundary of~.

46

Chapter II

Proof. Indeed, for the points 0 and s, 0 < s < 1 the curve ')'1(t) = ts is a unique curve joining 0 and s such that p(O, s) =

1 1 -I dp . n =

1'1

1 b~(t)ldt -1 ( )l 2 =tanh (s), 0 1 1'1 t

i.e., 1'1 (t) is the geodesic segment joining 0 and s. If z and ware arbitrary points in A, then the automorphism g = r'P om_z, where m_z is the Mobius transformation and r'P is the rotation with

(2.2.6)

"Y(t) = g- 1 l1'1(t)l

(2.2. 7)

and we obtain and 1'(1)

= =

g- 1(1'1(1)) = g- 1 (lm-z(w)l) mz(ei'Pim-z(w)l) = (mz o m_z)(w)

= w.

So "Y(t) is an admissible curve joining z and w. In addition, by Proposition 2.2.2

L-y

=

t

[ 1 I"Y'(t)ldt = l(g- 1 )'(1'1(t))1·11'~(t)ldt Jo 1- b(t)l 2 Jo 1- n- 1(1'1(t))l 2 { 1 11'~ (t)l Jo 1 - 11'1 (t)l2 dt = Lw

But p(z, w) = p(O, lm-z(w)l) = L-y, and we hence have relation (2.2.5). The second part of our assertion is a direct consequence of Proposition 1.1.4. D

2.3

Compatibility of the Poincare metric with convexity

In this section we will show that the Poincare metric is compatible with the convex structure of the unit disk A.

HYPERBOLIC GEOMETRY

47

Proposition 2.3.1 The Poincare metric p on the unit disk!:::.. satisfies the following properties: (i} If z, w are different points in!:::.., then for each k E (0, 1): p(kz, kw) < kp(z, w). (ii} If z, w and u are three different points in!:::.., then for each k E (0, 1)

p((1 - k)z + ku, (1- k)w

+ ku)

~

qp(z, w),

where q = (1- k) + klul < 1. (iii} If z, w, u and v are four points in!:::.., then for each k E [0, 1]

p((1- k)z + ku, (1- k)w + kv)

~

ma.x{p(z, w), p(u, v)}.

Proof. (i) If 'Y is the geodesics joining z and w, then k"'f is an admissible curve joining kz and kw. In addition, calculations show: k"'f'(t)

=

k 1 - l"'f(t)i2 . b'(t)i 1- k21"'f(t)12 1- l"'f(t)12

l1' (t) I < k 1 -lr(t)i 2 . Hence,

p(kz, kw)

~

Lk-r

< kL-y

= kp(z, w).

(ii) Consider a holomorphic mapping F on!:::.. defined by:

F(z) = (1- k)z + ku, where u is a fixed element of !:::... It follows by the maximum principle that

IF(z)l ~ q = (1- k) Thus

+ kiul < 1.

1 G = - F E Hol(!:::..). q

Using Proposition 2.1.3 and assertion (i) above we obtain

p(qG(z), qG(w))

p(F(z),F(w)) .- p((1- k)z + ku, (1- k)w + ku) < qp(G(z),G(w)) ~ p(z,w).

(iii) Suppose that

ma.x{p(z,w), p(u,v)} = p(z,w). Choose F and Gin Aut(!:::.) such that F(z) G=m_,..

= 0 and G(u) = 0, say F = m_z

and

48

Chapter II Denote JF(w)J

= r1 and JG(v)J = r2 and take rotations rcp and rcp 2 with 1

(/)1 = - argF(w), (/)2 = - argG(v). Then F1

= rcp

1

= rcp 2 o G belong to Aut(6.)

oF and G1

and we have

Then

p(u,v) and we obtain r2 :S r1. Denoting the mapping

= p(O,r2) < p(F1(z),F1(w)) = p(O,r1), p(G1(u),G1(v))

c- 1 (~~F)

by .P we have .P E Hol(6.) and

.P(z)

= G! 1 (~~ F1(z)) = G! 1(0) = u

.P(w)

= G! 1 (~~ F1(w)) = G! 1(r2) = v.

and

If we consider the mapping (1- k)I + k.P which obviously belongs to Hol(6.), again by Proposition 2.1.3 we obtain: p(((1- k)I + k.P)(z), ((1- k)l

+

k.P)w)

=

p((1- k)z + k.P(z), (1- k)w + k.P(w)) p((1- k)z + ku, (1- k)w + kv) :S p(z, w).

Now we continue to study the function ip: [0, 1] ip(t)

--->

D

JR+ defined as:

= p((1- t)z + tu, (1- t)w +tv)

(2.3.1)

for given points z, w, u, v in 6.. Using the equality 1 -Jm-z(w)J 2 = a(z, w), where

a(z, w) = (1 - JzJ2)(1 -2JwJ2) J1-zwJ (see section 2.1) we can write p(z, w) in the form p(z, w) = tanh- 1(1- a(z, w))!.

(2.3.2)

(2.3.3)

Therefore, p(z,w) :S p(u,v) if and only if a(z,w);:: a(u,v) and a(z,w) = 1 if and only if z = w. Also, defining the function '1/J: [0, 1] f--+ [0, 1] by the formula:

'lj;(t) =a ((1- t)z

+ tu, (1- t)w +tv)

(2.3.4)

HYPERBOLIC GEOMETRY

49

we have by (2.3.3)

cp(t) = tanh- 1 J1-,P(t).

(2.3.5)

It is clear that 1/J is a continuous function on [0, I] and 1/J(O) = a(z, w). We will show that 1/J is right-differentiable at the origin. Indeed, for t #- 0 we have: ·'·(t) ~

=

+ tui 2 )(I -I(I- t)w + t2vl 2 ) = a(t), () J1- [(I- t)z + tu][(I- t)w + tvJJ bt

(I -I(I- t)z

(2.3.6)

where [I- ((1- t) 2 lzl 2 + t 2 1ul 2 + 2 Re(I- t)tzu)]

a(t) x

[1- ((I- t) 2 lwl 2 + t1vl 2 + 2 Re(I- t)twv)]

=

[I -lzl 2

x

[I -lwl 2

=

(1 -lzi 2 )(I -lwl 2 ) ·[I- tc(t)][I- td(t)],

(2.3.7)

b(t)

II- [(I- t)z + tu]· [(I- t)w

(2.3.8)

c(t)

(t- 2)lzl 2 + tlul 2 + 2(I- t) Re zu (I - lzl 2 )

(2.3.9)

(t- 2)lwl 2 + t1vl 2 + 2(I- t) Rewv (1 - lwl 2 )

(2.3.10)

d(t)

=

(t 2 1zl 2

-

2tlzl 2 + t 2 lul 2 + 2 Re(I- t)tzu)]

-

(t 2 1wl 2

-

2tlwl 2 + t 2 1vl 2 + 2 Re(I- t)twv)]

+ tv]l 2

Similarly, we calculate

b(t) = 11- zwl 2 11- te(t)l 2 ,

(2.3.11)

where e(t) = (t- 2)zw + (1- t)(zv + uw) 1-zw

+ tuv.

(2.3.12)

So, by (2.3.6)-(2.3.8) we obtain:

·1/J( t)

(1 -lzl 2 )(1 -lwl 2 ) (1- tc(t))(I- td(t)) 11- zwl 2 II- te(t)l 2 ( ) (1- tc(t))(I- td(t)) · a z, w II- te(t)l 2

This implies

1/J'(O+) = lim

t-+O+

~t (1/J(t)- a(z,w)) =

-a(z,w) · lim

t-+0+

~t p(t),

Chapter II

50 where

1 _ (1- tc(t))(1- td(t)) = 11- te(t)1 2 1 - 2t Re e(t) + t 21e(t)l2 - (1 - tc(t))(1- td(t)) 11- te(t)l 2 1- 2tRee(t) + t 21e(t)l2- (1- t(c(t) + d(t)) + t 2c(t) · d(t)) 11- te(t)l 2 2 2 1- 2t Ree(t) + t 1e(t)1 - 1 + t(c(t) + d(t))- t 2c(t) · d(t) 11- te(t)1 2

p(t)

=

or

1/J'(o+) = -a(z, w) · lim[-2 Ree(t) t ..... o = -a(z, w) · lim [c(t) t ..... o+

+ c(t) + d(t)]

+ d(t)- 2 Re e(t)].

Using (2.3.8), (2.3.9) and (2.3.11) we obtain: lim c(t)

t .....o+

lim d(t)

2(Re zu - lzl 2) 1-lzl2 2(Re wv - lwl 2

t ..... o+

1-lwl2

and . R ( ) R [(zv l1m ee t = e

t ..... o+

+ uw)- 2zw] . 1- zw

Thus we obtain:

1/J'(o+)

= _ 2a(z,

w) [Re zu- lzl 2 1-

lzl 2

+ Rewv- lwl 2

_

Re [(zv

1 -lwl 2

+ uw)- 2zw]l· 1 - zw

(2.3.13) Denoting the expression inside the square brackets of (2.3.13) by J.l we see that if z = w, then J.l = 0. Hence, in this case:

1/J'(o+) = -2J.La(z,w) also equals to zero. If z =1- w, then by using (2.3.5) and (2.3.13) we obtain:

'(o+)'P

-1/J'(o+)

- 2yfl -1/1(0) ·1/1(0)

(2.3.14)

We are now ready to formulate our main result in this section.

Proposition 2.3.2 (cf., [116], Lemma 2.2) For given four points z, w, u and v in D. the following statements are equivalent:

HYPERBOLIC GEOMETRY (a) the function
t

(b) c.p(O)::; c.p(t), (c) c.p'(o+) ~ O; (d) Re [z(z- u) 1-

lzl 2

E [0,

f--->

JR.+,

c.p(t)

51

= p((l- t)z + tu, (1- t)w +tv), is not

1];

+ w(w- v)] < Re z(w- v) + w(z- u). 1-

lwl 2

1 - zw

-

Proof. It is clear that (a) implies (b) which, in turn, implies (c). The equivalence (c) and (d) follows from (2.3.12) and (2.3.13), because f.L ~ 0 if and only if (d) holds. Now, we will show that (c) implies (a). Assume that c.p'(O+) > 0. We claim that max{c.p(ti), c.p(t2)}. Setting Wl

= (1- tl)w

t2)z + t2u,

w2

= (1- t2)w +

= (1- ti)z

z2

= (1 -

we have p(z 1, wi)

+ t 1 v,

+ t1u,

Zl

t2v,

= c.p(tl) and p(z2, w2) = c.p(t2).

At the same time: 1 1 2z1 + 2z2 1 2w1

1

= (1- to)z + t 0 u,

+ 2w2 = (1 -

to)w

+ t 0 v.

Thus we have by property (iii) of Proposition 2.3.1

c.p(to)

p

(~z1 + ~z2 ~wl + ~w2) 2 2 , 2 2

< max{p(z1,wi),p(z2,w2)} max{ c.p(t1 ), c.p(t2)}. Contradiction. Since c.p'(O+) > 0 this implies also that

Chapter II

52

2.4

Fixed points of p-nonexpansive mappings on the unit disk

Definition 2.4.1 A mapping F: pair z, w

~ ~----> ~~

p(F(z), F(w)) lf for all z, w in

is said to be p-nonexpansive if for each ~

p(z, w).

(2.4.1)

p(F(z), F(w)) = p(z, w),

(2.4.2)

~

then F is called a p-isometry

of~-

It follows by Proposition 2.1.3 that each F E Hol(~) is p-nonexpansive and F E Aut(~) is a p-isometry. However, the class of p-nonexpansive mappings is essentially wider than Hol( ~). Indeed, it is not difficult to see that ifF E Hol(~), then F is also p-nonexpansive. Therefore, it follows by the Proposition 2.3.1(iii) that each convex combination of F and F is also a p-nonexpansive. In particular, for FE Hol(~), ReF= ~(F+F) is a p-nonexpansive mapping. Generally, denoting the class of p-nonexpansive mappings by Np(~) we conclude: Np(~) is a closed convex subset of the space of all continuous mappings on~. Np(~) is a semigroup with respect to composition, and Hol(~) is a proper subset of Np(~). At the same time it can be shown that the set of all p-isometries of~. usually denoting by Isom(~) is the joint of Aut(~) and Aut(~), i.e., FE Isom(~) if and only if either FE Aut(~) and FE Aut(~). We will study here fixed points of p-nonexpansive mappings. It will be shown that the Wolff-Denjoy Theory can be extended for the class Np(~). First we note that ifF has a fixed point a in~' then each p-ball B(a, R) = {z E ~' p(z, a)~ R} is F-invariant. This fact contains Proposition 1.2.1 for holomorphic mappings. To study more deep properties of fixed points of p-nonexpansive mappings we need the following observation. IfF E Np(~), and u is a fixed element of~. then the mapping Gt defined by the formula:

Gt(z) = (1- t)u + tF(z),

0

~

t

~

1,

(2.4.3)

also belongs to Np(~). Moreover, it follows by Proposition 2.3.1(ii) that all z, wE ~. ;; =/= w, and t E [0, 1)

p(Gt(z), Gt(w)) < qp(z, w),

(2.4.4)

where q = (1- t)JuJ + t < 1, i.e., G is a strict contraction of(~, p). Therefore by the Banach Fixed Point Theorem Gt has a unique fixed point Zt = Zt(u) in~. and

Zt = lim G~n) n-oo

(2.4.5;

HYPERBOLIC GEOMETRY

53

uniformly on all compact subsets of .6.. In addition, it is clear that if for some sequence {tn}, 0 < tn < 1 there is the limit (2.4.6)

a= t lim Zt , n

n-1-

in .6., then a must be a fixed point of F. Indeed, since a E .6. and F is continuous we have:

F(a)- a

lim [F(ztn) -

t, -+1-

Ztn]

lim [F(zt,)- (1- tn)u- tn(F(ztJ)]

tn~l-

lim (1 - tn)[F(ztJ - u] = 0.

tn -+1-

This implies F(a) =a.

Definition 2.4.2 ([52]) For a fixed element u E .6. the curve Zt = Zt (u) : [0, 1) >---+ .6. defined by formulas (2.4.3} and (2.4.5} is called an approximating curve for F. From a different perspective, fixing t E [0, 1) and varying u E .6. we can consider Zt as a mapping from .6. into itself which will be denoted by It = It(u), that is It : .6. >---+ .6. is a mapping implicitly defined by the equation:

It(u) = (1- t)u + tF(It(u))

(2.4. 7)

It= (1-t)I +tFoit.

(2.4.7')

or, in operator form,

Proposition 2.4.1 IfF E Np(.6.), then for each t E [0, 1) the mapping It is also in Np(.6.) and fortE (0, 1), the mapping It has the same fixed points as F. Proof. Let u and v be different points in .6.. For iterates c;"l, c;o) have:

p(G;n+ll (u), c;n+tl (v))

p((1- t)u + tF(c;nl (u)), (1- t)v

= I we

+ tF(G;"l (v)))

< max{p(u, v), p(c;n)(u), c;")(v))} = p(u, v) by induction. Hence, from (2.4.5) and (2.4.7) we have:

p(It(u),

It(v)) :=; p(u, v),

(2.4.8)

i.e., It E N for all t E [0, 1). Now it is clear that if a E Fix It, i.e. a is a fixed point of It. then by (2.4.7) we obtain:

a= (1- t)a + tF(a),

t E (0, 1).

Assume now that F has a b, which is not a fixed point of It for some t E (0, 1). Then It(b) = (1- t)b + tF(It(b)) #b.

Chapter II

54

At the same time b is a solution of the equation: Zt

= (1 - t)b + tF(zt)·

Since the latter equation has a unique solution for each 0 It(b) must be equal to b. D

< t < 1, it follows that

Remark 2.4.1 Since the set Hol(~) is closed with respect to the topology of uniform convergence on each compact subset on ~, we have by the constructions of Proposition 2.4.1, that ifF belongs to Hol(~) then so does It for each t ;:::: 0. We can also obtain the following consequence of this proposition.

Proposition 2.4.2 IfF is a fixed point free p-nonexpansive mapping for each u E

~

of~'

then

we have:

(2.4.9) Conversely, if (2.4.9) holds for at least one u E ~' then F is fixed point free. In other words, F E N has a fixed point in b. if and only if there are u E b. and a subsequence tn ---+ 1- such that the approximating sequence Ztn = Itn (u) is bounded away from the boundary of~-

Proof. Indeed if for some u E ~ the sequence { Itn ( u)} lies strictly inside ~, then there is a subsequence which converges to a point a E ~- This point must be a fixed point ofF, as we saw above. Now, ifF has a fixed point a E ~ then It(a) = a for all t E (0, 1] and we obtain by (2.4.8) It(a)) ~ p(u, a)

p(It(u),

Thus It(u) is strictly

inside~

< oo,

u E ~-

for all t E (0, 1) contradicting (2.4.9). D

Actually, we can show somewhat more: IfF is fixed point free, then there is a unique point e E 8 ~, such that for each ·u E ~, the net {It (u)} converges to e, as t tends to 1-. The following assertion contains an extension of Wolff's Lemma.

Proposition 2.4.3 ((52], p. 131) Let F be a p-nonexpansive mapping of the unit disk ~' and suppose that F has no fixed point in point e E 8~ such that: (a} each horocycle

_..

D(e,l\) =

~-

Then there is a unique

{ z E ~: l1-zel22 < K, 1-lzl

internally tangent to 8~ ate is F -invariant, i.e., F (D(e, K)) ~ D(e, K). (b) For each t E (0, 1) It(D(e, K)) ~ D(e, K).

HYPERBOLIC GEOMETRY

55

(c) For each u E 6. the approximating curve zt(u) = lt(u) converges toe, as t tends to 1-, i.e., lim Zt(u) =e. t--+ 1-

Proof. We already know that for each u E 6.: lim llt(u)l

t--+ 1-

=

1.

(2.4.10)

Take firstly u = 0 and consider a net Zt = lt(O). Since { zt}tE(0, 1) is precompact there is a sequence Zn = ltn (0) which converges to a point e E 6.. It follows by (2.4.10) that lei = 1. Since F is p-nonexpansive we have: (2.4.11) for each

z E

6.. Recall also that (2.4.12)

and using (2.4.11) and (2.4.12) we obtain:

II- F(.ZJF(z)i2

<

1 -IF(z)l2

<

1 -IF(znJI2 1 -lznl 2

11- z-;.zl 2 1- lzl 2 1- t,lznl 2 11- z-;.zl 2 1-lznl 2 . 1-lzl 2 11- z-;,zl2 1 -lzl2 .

(2.4.13)

In addition (2.4.12) implies that: lim F(z,.) = e,

n--+oo

and letting n-+ oo in (2.4.13) we obtain: 2 11- eF(z)l __:._....:....;.2 1 -IF(z)l

.:....__

:::;

11- ezl22• 1 -lzl

This means that

F (D(e, K))

~

D(e, K)

for all K > 0. So e is indeed a sink point of F. Its uniqueness can be proved exactly as in Proposition 1.4.2. Now, since D(e, K) is convex it follows that for each t E [0, 1) and u E D(e, K) the mapping Gt, defined by the formula

Gt(z) = (1- t)u + tF(z) maps D(e, K) into itself, and so does lt = limn--+oo G~n).

56

Chapter II

Thus for each t E (0, 1) the horocycle D(e, K) is Jt-invariant. Hence for a given z E b. one can find K > 0 such that z E D(e, K) and

But together with (2.4.10) this implies that Jt(z) converges toe as t tends to 1-. Finally, if F is contractive then it follows by the general theory of compact mappings on metric spaces (see, for example, [84]), that p(n) converges to a point a E b.. Arguments similar to that above show that a must be equal to e E 8b., and we have completed the proof. 0

Remark 2.4.2 Note that the situation is simple when F E Isom(b.) and it is fixed point free. In this case either F or F is an automorphism which is power convergent to a sink point ofF on the boundary of b.. The case of p(n) iterations when F is a fixed point free p-nonexpansive mapping neither holomorphic nor antiholomorphic, seems to be rather complicated in the framework of the pure one-dimensional metric theory. The full answer to this question was given by R. Sine [135] in 1989: ifF is a fixed point free p-nonexpansive mapping, then it is power convergent. The result for contractive F (i.e., p(F(z), F(w)) < p(z, w)) has established

earlier by K. Goebel and S. Reich (see [52], p.138). Now we turn to the situation when FE Np(b.) has a fixed point in b.. Generally, such a situation differs form the holomorphic case. Namely, we mean that there are p-nonexpansive mappings which have more than one interior fixed point. The following simple example F(z) = !(z + z) shows that in fact F may have infinite number of fixed points. Actually, in this example F is a p-nonexpansive retraction (F o F = F) of the unit disk onto the open interval (-1, 1). In general, the iterates of a p-nonexpansive mapping with interior fixed points may not be convergent. A rather complete discussion on their behavior can also be found in [135]. Thus the following question arises: does the approximating curve Jt(u) for a fixed u E b. converge to a fixed point ofF, when t tends to 1- ? The following assertion states that the answer to the above question is affirmative.

Proposition 2.4.4 ([52], Theorem 28.3, p. 134) Let F be a p-nonexpansive mapping on b. with the nonernpty fi:red point set. Then there exists the limit

lim Jt

t~I-

which is a retraction onto this set.

This assertion together with assertions (b) and (c) of Proposition 2.4.3 can be considered as an implicit continuous analog of the Wolff-Denjoy Theory for p-nonexpansive mappings.

HYPERBOLIC GEOMETRY

57

Since lt for each t E (0, 1) has the same fixed point set as F, one may ask what happens with the discrete iterations Jt(n) for a fixed t E {0, 1). The following assertion is a special case of Theorems 30.5 and 30.8 in [52]. For the holomorphic case, see, also [51].

Proposition 2.4.5 Let F be a p-nonexpansive mapping on the unit disk .6.. Then for each t E (0, 1) the mapping lt defined by (2.4.3) and (2.4.5) is power convergent. In particular, ifF has no fixed point in .6., then the iterates Jt(n)(z), n = 1, 2 ... , converge to the sink pointe E 8.6. ofF as n--> oo for each t E (0, 1) and each z E .6.. This assertion is quite simple if F is holomorphic. Indeed, it follows by algorithm (2.4.5) that for each t E (0, 1) the mapping lt belongs to Hol(-6.). Thus to prove our assertion we need only to show that for every F E Hol (.6.) and t E ( 0, 1). the mapping lt cannot be an elliptic automorphism. In fact, if a E .6. is the fixed point of F then it follows by the chain rule that (Jt)'(a)

= (1- t) + tF'(a)(Jt)'(a),

i.e.,

1-t ( J)'() t a = 1- tF'(a) If F'(a) = 1 then F, hence Jr, is the identity for each t E (0, 1) and we have completed the proof. If F'(a) =/:- 1, then it follows by the Schwarz-Pick Lemma (Proposition 1.1.3(ii)) that IF' (a) I ~ 1 and we have: I(J:(a)l < 1.

This means that lt is not an automorphism of .6.. D For a complete exposition of the Wolff-Denjoy Theory under the hyperbolic geometry approach it would be appropriate to mention the results due to P. R. Mercer [95, 97, 98], R. Sine [135] and P. Yang [160] for the one-dimensional case, M. Abate [1]-[5], G.-N.Chen [24], Y. Kubota [80] and P. R. Mercer [96] for the finitedimensional case, T. Kuczumov and A. Stachura [82], [83], P. Mazet and J.P. Vigue [92], [93], S. Reich [111]-[113], S. Reich and I. Shafrir [114], I. Shafrir [129, 130] for the Hilbert ball and its product, and P. Mellon [94] and K. Wlodarczyk [152]-[153] for J*-algebras (see also a survey of S. Reich and D. Shoikhet [118]). P. Yang in [160] suggested a hyperbolic metric characterization of the horocycles D(e, K) in .6. as follows:

D(e, K) = {z

E

.6.: lim [p{z, w)- p(O, w)] < z-e

!2 log K}.

(2.4.14)

Even for holornorphic case this characterization is a very useful tool in the study of the boundary behavior of self-mappings of .6.. By using (2.4.14) and a strengthened Schwarz-Pick inequality, P. R. Mercer [98] has recently obtained a sharper description of Julia's Lemma which makes the Denjoy-Wolff Theorem more transparent. We state his result here.

58

Cbapter II

Proposition 2.4.6 Let F E Hol(~) be not a constant. Suppose that there exists a unimodular pointe E {)~ such that liminf 1 -IF(z)l =a< oo. z-+e 1 - lzl

Then there is a point"' on the boundary 8~, such that for any z E aD(e, K), z e, and wE D(e, K), w f. e, the following inclusion holds: 1+a F(z) E D(ry, - 2-aK),

where a= b+c b= IF'(w)l(1-lwl2)' 1+bc' 1-IF(w)l2

1

(2.4.15)

c=l 1-wz z-w I·

Note that it follows by Corollary 1.1.1 that b::::; 1. Moreover, the Schwarz-Pick Lemma implies that b (hence a) is equal to 1 if and only ifF is an automorphism of ~- If this is not the case (i.e., F is not an automorphism of ~), then a in (2.2.15) is strictly less than 1, whence this inclusion is stronger that the original one in Julia's Lemma.

Chapter 3

Generation theory on the unit disk For physical, chemical, and biological applications it is sometimes preferable to study a great variety of iterative processes, including the processes of continuous time. In spite of their simplicity these processes have been applicable in many fields, involving mathematical areas such as geometry, theory of stochastic branching processes, operator theory on Hardy spaces, and optimizations methods. A problem that has interested mathematicians since the time of Abel is how to define n-th iterate of function when n is not integer.

3.1

One-parameter continuous semigroup of holomorphic and p-nonexpansive self-mappings

Let A be a topological Abelian (additive) semigroup with zero and let there exist a natural ordering of A, i.e., 7 2 t if and only if there is s E A such that 7 = t + s. A mappingS: A~---> Hol(~) (respectively, Np(~)) which preserves the additive structure of A with respect to composition operation on Hol(~) (respectively, Np(~), i.e.,

(i) S(t + s) = S(t) o S(s), whenever s, t and s + t belong to A; (ii) S(O) = I, the identity embedding of~. will be referred to its holomorphic (respectively, p-nonexpansive) action on~If A= NU{O} = {0, 1, 2, ... } then S = {Fo, Fr, F2, ... , Fn, .. .}, Fn E Hol(~),

59

Chapter III

60

(respectively, Np(tl.)) is called a one-parameter discrete semigroup. Actually such a semigroup consists of iterates of a holomorphic (respectively, p-nonexpansive) self-mapping F = F1, because of conditions (i) and (ii), i.e., Fo = I, Fn = Fin), n=1,2, .... If A is an interval of IR, containing zero and actionS: A 1-+ Hol(tl.) (Np(tl.)) is continuous with respect to the topology of pointwise convergence on tl., we will say that Sis a one-parameter continuous semigroup of holomorphic (p-nonexpansive) self-mappings Ft, tEA. In other words, a one-parameter continuous semigroup of holomorphic selfmappings of tl. is a family S = {Ft}tEA C Hol(tl.) (Np(tl.)) such that (i)' Ft+s(z) = Ft(F8 (z)), z E tl., t, sand t +sEA, (ii)' Fo(z) = z for all z E tl. and

limFt(z) = F8 (z), t--+s

(3.1.1)

whenever t, s E A. Mainly we will concentrate on two cases: (a) A= [O,T), T > 0; (b) A= (-T,T), T > 0. If T = oo we will say that S = {Ft}t~o is a flow on tl.. In the case (b) conditions (i)' and (ii)' imply that Ft oF-t = Fo = I, hence Ft E Aut(tl.) (Isom(tl.)) and S is actually a group which will be called a oneparameter group of holomorphic automorphisms (p-isometries) of tl..

Exercise 1. Prove the following: if A = [0, T) and at least one element Ft of S belongs to Aut(tl.), then so does each element of S, and Scan be extended to a one-parameter group:

S = {Ft}tE(-T,T) c

Aut(tl.).

Example 1. Let a be a complex number such that Rea ;::: 0 and let t E [0, T), T > 0. Define Ft E Hol(tl.) as follows:

Ft(z) =e-at· z,

z

E tl..

(3.1.2)

It is clear that

Fo(z)

=

z,

z E tl.

and z E

tl.,

whenever 0 ~ s, t < T and t + s < T. Since T > 0 is arbitrary we have that such a semigroup can be continuously extended to a flow on tl.. As a matter of fact, we will show below that each one-parameter continuous semigroup of holomorphic self-mappings of tl. can be continuously extended to a holomorphic flow on tl..

61

GENERATION THEORY

Note also that the semigroup (flow) considered in this example is a linear action on ~- In fact, each one-parameter semigroup which is a linear action on ~ has the form (3.1.2) with some a E C, Rea 2:0.

c

Example 2. Define the family {Ft}

Hol(~),

t E IR, by the formula:

z + tanh t . ztanht + 1

Ft =

It is easy to see that for each t E IR, Ft E Aut(~), and that conditions (i) and (ii) are satisfied. Hence, Ft is a flow of automorphisms on ~Exercise 2. Show that each of the automorphisms in Example 2 is hyperbolic. Example 3. Set D ( ) _

I't

Since Ft(z)

z -

Wt

1 + it . Z + t 1-tt 1+ 1_itz 0

0

t +rt

1 +it

= Atma,(z), where At=--. has modulus 1 and at= -1--.

1-rt for all t E IR, we have that Ft belongs to Aut(~).

E ~

Exercise 3. Let S = { Ft} tEIR be defined as in Example 3. Show that the following statements hold: (i) S = {Ft} tEIR is a flow of automorphisms of~; (ii) for each t =F 0 the mapping Ft in Example 3 is a parabolic automorphism. Exercise 4. Consider the family S = { Ft} defined as follows:

z

Ft(z)

= e t - z (e t -

1)'

t > 0.

Show that (i) for each t 2: 0, IFt(z)J < 1, i.e., Ft E Hol(~); (ii) the family S = {Ft} is a one-parameter continuous semigroup on ~ and every element of S is not an automorphism of ~; (iii) for each Ft, t > 0, only elements {0, 1} are solutions of the equation Ft(z) = z;

(iv) find limt_, 00 Ft(z),

z E

~-

Exercise 5. ([115]) Consider the family S = {Ft}, where

(i) Show that Ft E Hol(~), and S (ii) Find lim Ft(z).

= {Ft} is a one-parameter semigroup;

t--+oo

Remark 3.1.1 The semigroups in Examples 2 and 3 consist of automorphisms of b. and consequently are one-parameter groups, while the semigroups in Exercises

62

Chapter III

4 and 5 cannot be extended to subgroups of Aut(~). In fact, Ft in Exercise 4 is defined also for t < 0, but it does not map the unit disk ~ into itself. Therefore fort> 0 the mapping Ft is biholomorphic on~, but Ft(~) is a proper subset of ~- This fact holds in general.

Proposition 3.1.1 Let the family S = {Ft}tE[O,T), T > 0 be a one-parameter continuous semigroup on ~' S C Hol(~). Then each element Ft E S is a local biholomorphism on~Proof. Let U be any convex open subset strictly inside~- Since Sis a normal family on ~ the net {Ft} converges to the identity uniformly on U, as t goes to zero. Therefore for fJ =dist(U, a~) and £ E (0, 1 - fl) one can find to E (0, T) such that jz- Ft(z)l < £ for all z E U and t E [0, t 0 ). It follows by the Cauchy inequality that 11- (Ft)'(z)l

£

< 1_

{J

< 1,

hence (Ft)'(z) =f. 0 for all z E U. Consequently, for each pair z and w in U we have by the Lagrange formula:

IFt(z)- Ft(w)l = I(Ft)'(z

+ B(w- z))l·lz- wj,

0 :::; (} :::; 1, equals to zero if and only if z = w. So, Ft is injective (hence biholomorphic) on U for all t E (0, T) small enough. Now it follows by continuity that if Ft 0 is locally injective on ~ then so is Ft for all t close enough to t 0 . At the same time, if {Ft} is a net of locally injective mappings which converges to Fs when t goes to s E (0, T), then Fs must be either also locally injective or zero. But the latter is impossible. Hence the set of t E (0, T) such that Ft is locally injective is a nonempty open and closed subset of the interval [0, T). Therefore it must be [0, T). D

3.2

Infinitesimal generator of a one-parameter continuous semigroup

As we mentioned above each one-parameter continuous semigroup of holomorphic self-mappings of ~ defined on an interval [0, T) can be extended to the flow on ~, i.e., we may always assume that T equals to oo. Moreover, only the right continuity at zero of a semigroup implies continuity (right and left) on all of JR+ = [0, oo). These facts can be shown by different approaches, but here we

GENERATION THEORY

63

will establish them by using a very strong property of a continuous one-parameter semigroup of holomorphic self-mappings of 6.. to be differentiable with respect to a parameter at each point t on the interval of definition. This nice result for the one-dimensional case is due to E. Berkson and H. Porta [17]. We will give here another proof which can easily be extended to a higher dimension (see, for example, [115, 118], cf., also [5]). Proposition 3.2.1 Let S = {Ft, t E [0, T)} be a one-parameter semigroup of holomorphic self-mappings of 6.., such that for each z E 6..:

=

lim Ft(z)

t-->0+

(3.2.1)

z.

Then for each z E 6.. there exists the limit:

. z-Ft(z) - f ( ) l Im (3.2.2) z' t which is a holomorphic mapping of 6.. into C. Moreover, the convergence in (3.2.2} is uniform on each subset strictly inside 6... In other words, if a semigroup S = {Ft} , t E [0, T), is right continuous at t-->O+

zero, it is also right-differentiable at zero. Proof. Step 1. Let ¢ E Hol(b..) and { ¢Ck)} be its iteration family. Let Ar denote the disk centered at zero with radius r. Suppose that there are positive r1 < r2 < 1 and 0 < J.l < r2 - r 1 and an integer p ~ 1 such that:

\z- ¢Ckl(z)l <

(3.2.3)

J.l

for all k = 1, 2, ... ,p and z E Ar 2 • Thus for all z E Ar 1 the following inequality holds \ z- ¢CPl(z)- p(z- ¢(z))\ :::;

J.l

r2- r l - J.l

(p- 1) ·[z- ¢(z)[.

(3.2.4)

Indeed, let z E b..r 1 and w E b..r 2 be such that [z - w[ :::; J.l· Then the disk Ar 2-r 1-IL(z) centered at z with radius r2- r1- J.l lies in Ar 2. Hence it follows from (3.2.3) and the Cauchy inequalities that: 1- (¢(k))'(z)\ :::;

1

J.l

r2- r1- J.l

(3.2.5)

.

Therefore, for z E b..r 1 and w E Ar 2 such that [z- w[ :::; J.l we have by (3.2.5):

[z- ¢Ckl(z)- (w- ¢Ckl(w)[ < Setting w

J.l

r2- r l - J.l

[z- w[.

(3.2.6)

= ¢(z) and using the triangle inequality we obtain:

[z- ¢CPl(z)- p(z- ¢(z))[

=

~~ [¢
<

L

\¢
k=l

<

J.l r2- r1-

J.l

(p- 1)[¢(z)- z[,

Chapter III

64

and we are done. Step 2. Let a semigroup S = {Ft}, t E [0, T) satisfy condition (1). For s > 0 define the mapping 1 fs = -(1- Fs) E Hol(6., C). s We claim that the net {fs}s>O is uniformly bounded on each disk 6.n with 0 ::; r < 1. Indeed, set r1 =rand choose r2 E (rt, 1). Take any f-L > 0, such that f-L < r2-r1 and

f-L

< ~-

r1- f-L 2 By (3.2.1) one can find a E (0, T) such that for all r2-

T

E (0, a) and all z E 6.r 2

[z- Fr(z)[ < f-L·

(3.2. 7)

Setting p = [a Is] we see that for all s E (0, a 12), the following relations hold: p ~ 2 and ps ~ a 12. Also ks ::; a for all k = 1, 2, ... , p. Hence by (3.2. 7) and the semigroup property for all s E (0, a 12) we have:

[z- F;k)(z)[ < f-L, whenever z E 6.r 2 and k = 1, 2, ... ,p(= p(s)). From Step 1 we obtain for all z E 6-r:

p [z- Fs(z)[-[z- F;Pl(z)[

<

[p(z- F8 (z))- (z- F;Pl(z)[

<

2 p[z- F (z)[

or

[z- F8 (z)[ <

~

1

8

jz- F;Pl(z)j,

(3.2.8)

whenever s E (0, a 12) and p = [a Is]. So, by (3.2.7) and (3.2.8) we obtain: 2

[fs(z)[ ::; -[z- Fsp(z)[ ps

2/L

4/L

ps

a

< - ::; - < oo,

as claimed. Step 3. Now we will show that the net {!8 }, fs = lls(I -F5 ), s E (0, T), converges to a holomorphic mapping f on 6., when s goes too+. Set n = n(s) = [11 s 2 ] and consider the sequence {ft;n}. Since by Step 2 this sequence is uniformly bounded for all large n one can find a subsequence {h;nk} which converges to a mapping f E Hol(6., C) uniformly on each compact subset of 6.. That is for each r E (0, 1) and all z E 6.r and for a given e > 0 we can choose k large enough such that: (3.2.9) lft;nk(z)- f(z)l < €. In addition, for such nk and s E (0, 1) we have:

GENERATION THEORY

65

Observe, that since, n = [1/s 2 ], we havens---+ oo and [ns]/ns---+ 1 ass---+ o+. Moreover, for a given net s ---+ o+ we can find (if necessary) a sequence Sk ---+ o+ such that sk/ s ::::: 1. Then, setting nk = [1/ s~], we have:

!.__ - s < snk s~

s/ Sk

Since,

s

=

[2_] < + 1 · s. s~ sk 8 2

1 we obtain:

;:::

-

s

s~

1 >-----> - sk

00

as

Sk---->

0+.

Therefore nks ___.. oo as s ___.. o+ and [nks] /nks ___.. 1 as s ----> o+. Thus we can find 8 > 0 such that 1- [nks] /nks < e and F[nks]/nk (z) C ~r 2 C ~. whenever s E (0, 8) and z E ~r, r < r 2 < 1. Then we obtain by step 1 and the semigroup property:

~s IFbd (z)- Fs(z)l

~S IFrnk•l (z)- Fru o F _bd (z)l 8

"k

"'k

nk

<

~M~z-Fs_r:~·J(z)l

<

4J-L (s _ [nks]),

as

nk

n

(3.2.10)

where M =sup I(Fs)'(z)l. Finally we obtain by Step 1:

~I'

-

F:!:"'))(z)- n,s (z- F,', ('l)

<

I

~ lz- F
nk

nk

S

nk

< ::S ([nks] + l[nks- ns]l) lz- F .J.... (z)l nk < (e:[nks] +I [nks] -11) nklz- F.J....(z)l nks

nks

"'

8J-L

< -e:.

(3.2.11)

a

Thus for z E

~r

and s E (0, 8) we obtain from (3.2.10) and (3.2.11):

lfs(z)- f(z)l

<

if ..1k (z)- f(z)l + ifs(z)- f ..1k (z)l

<

16J-Le: a

and we have completed the proof. D

66

Chapter III

Further we will show that even for the wider class of semigroups of the pnonexpansive mappings the property of the right-differentiability at zero, implies the continuous extension of the semigroup to a flow in Np(D.) and the differentiability of this flow at each point t on (0, oo ).

Definition 3.2.1 LetS= {Ft} C Np(D.), t E [0, T), T > 0, be a one-parameter semigroup of p-nonexpansive self-mappings of 6.. If there exists the pointwise limit f : 6. ~ C defined by . I - Ft f = 11 (3.2.12) m--,

t-.o+ t we will say that S is generated by f and f is called the infinitesimal generator of the semigroup S = {Ft} tE [o, T) .

If this will not imply a confusion we will often simply say 'generator of a semigroup', omitting the word 'infinitesimal'. Thus Proposition 3.2.1 states the existence and holomorphity of generators for every one-parameter continuous (even only right-continuous at o+) semigroup of holomorphic self-mapping of 6.. Proposition 3.2.2 LetS= { Ft} tE[O,T) be a one-parameter semigroup of p-nonexpan1 self-mappings on 6., which is generated by f : 6. ~ C and suppose that its generator f, defined by (3.2.12), is a continuous function on 6.. Then the complex valued function u(·, ·) defined on [0, T) x 6. by u(t, z) = Ft(z) is the solution of the following Cauchy problem:

au~; z) + f(u(t, z)) =

0,

t E [O,T),

{ u(O,z)

= z,

z E 6..

Proof. Fix s E [0, T) and set F.(z) = w. We have: a+u(t, z) at t=s

. Fh(F.(z))- F.(z) . Fs+h(z)- F.(z) 11m = 11m h-o+ h h_,o+ h lim h->O+

Fh(w~- w

= - f(w) = - f(u(s, z)).

This means that the right-hand partial derivative of u(t, z) = Ft(z) in t = s E [O,T) exists and is exactly equal to -f(u(s,z)). Now for fixed s E (O,T) we denote Wh = Fs-h(z) for hE [O,e) with € small enough. Since z is an interior point of 6. we can find € small enough such that the set {wit} hE [o,c:) lies in a compact subset !:1 of 6.. Then we obtain by the uniform continuity of f on n a-u(t, z) I at t=s

=

. Fs-h(z)- Fs(z) - 1lm h-o+ h -

-

. wh- F(w,.) 1lm

h-o+ h lim fh(w,.) h->O+

= - f(w),

GENERATION THEORY

67

where w

= lim

h--+0+

Wh

= Fs(z).

So u(t, z) is also left-differentiable on the interval (0, T) and therefore satisfies the Cauchy problem (*). 0

Remark 3.2.1 As a matter of fact, it is known (see, for example, [159]) that if a function has a right-hand derivative which is continuous, then it also has a left-hand derivative and they coincide. Corollary 3.2.1 If S = {Ft}, t E [0, T) is a one-parameter semigroup of holomorphic self-mappings, which is right-continuous at zero, then it is continuous at each point of [0, T) and, moreover, it is differentiable on this interval. Remark 3.2.2 Also we note that since the generator f of a semigroup of holomorphic mappings is holomorphic, it is locally Lipschitzian. Therefore the Cauchy problem (*) has a unique local solution in a neighborhood oft= 0 for each initial value z E ~. This fact and the analytic continuation principle can be successfully used to prove the extension of the semigroup to a flow of holomorphic mappings on ~ (see [115]). However, to prove this for the class of p-nonexpansive mappings we will use another approach which is based on the so called resolvent method. We will discuss it in the next section. At the end of this section we present a property which is specific for semigroups of holomorphic self-mappings. Exercise 1. Prove that if u( ·, ·) is a jointly continuous function on [0, T) x ~ which satisfies (*) with f E Hol(~, C), then it also satisfies the linear partial differential equation: 8u(t,z) au(t,z)f(-)=0 at + az "' . Exercise 2. Find the (infinitesimal) generators in Example 1-5 of Section 3.1.

3.3

Non linear resolvent and the exponential formula

In this section we consider the notion of the nonlinear resolvent of a continuous (or holomorphic) function and establish a result in the spirit of the Hille-Yosida theory

Chapter III

68

(see, for example, [159]). More precisely, we will show that the global solvability of the Cauchy problem (*) (see Section 3.2) is equivalent to the solvability of a functional equation (see [115, 116, 118]). To make the ideas more transparent we first explain them for the simplest linear case. We already mentioned that the only linear (in the complex sense) semigroup S on Cis of the form: S = {FtL>o, Ft(z) = exp( -ta)z, a E C, with the infinitesimal generator f : C f-+ C defined bY . z-Ft(z) f(z) := hm = az.

t

t-+O+

Conversely, for a given number a E C two following equivalent definitions of the exponential function u(t,z) =exp(-ta)z usually are employed in the classical analysis. The first one is based on the solution of the linear Cauchy problem:

{

Du(t,z) ( ) at +au t,z = 0 (3.3.1) u(O,z) = z E ~'

a E C, while the second one uses the exponential formula:

u(t, z) = lim n~oo

(1

+.!.a) -n z. n

(3.3.2)

On the other hand, if for given z E C, a E C, and r 2:: 0, we solve the linear equation (3.3.3) w+raw = z, we have that its solution

w = lr (z) = (1

+ ra)- 1 z

can be considered as a linear mapping defined on C. Then the exponential formula (3.3.2) can be rewritten in the form u(t, z) = lim Jt(/n\z), n-+oo

n

(3.3.4)

where J$nl denotes the n-fold iterate of the mapping lr, r 2:: 0. In addition, it is clear that S = { exp( -ta)} is a p-nonexpansive action on the unit disk ~ if and only if Rea 2:: 0. This is, in fact, equivalent to the property that for each r 2:: 0 the mapping lr : C f-+ C is a self-mapping of~' hence is p-nonexpansive.

69

GENERATION THEORY Now, if we consider, for example, a (nonlinear) analytic function defined by

f : .6.

~-----+

C

f(z)=z-z 2 , we see that the solution of the Cauchy problem

()u~; z) + f (u(t, z)) = 0, { u(O, z) = z .6.,

(3.3.5)

E

is a well defined semigroup (even a flow) of holomorphic (hence p-nonexpansive) mappings of .6.: u(t, z)(= Ft(z)) =

z t

( t

e-ze-1

)

(see Exercise 4 in Section 3 .1). But in this case the similar formula (3.3.2) makes no sense with respect to (3.3.5). At the same time, it turns out, that formula (3.3.4) continues to hold with Jr : .6. ~-----+ .6., r 2: 0 defined as the solution w of the equation w

+ r(w- w2 )

= z

(3.3.6)

(compare with equation (3.3.3)).

Exercise 1. Show directly that for each z E .6. and r > 0 the quadratic equation (3.3.6) has a unique solution w = Jr(z) E .6. which holomorphically depends on z E .6.. Show that lim Jt(/n) ( z) = Ft ( z). n--+oo

n

In general such an approach, which we call the resolvent method, is very useful in the study of generated p-nonexpansive semigroups and their properties. In addition, by using this method one can give different characterizations of the class of functions which generate a flow of p-nonexpansive or holomorphic mappings, or in other words, to give sufficient and necessary conditions for the global solvability of the Cauchy problem. To continue we need the following definition.

Definition 3.3.1 Let f : .6. ~-----+ C be a continuous function. We will say that f satisfies the range condition (RC) if for each r > 0 the nonlinear resolvent Jr =(I+ rf)- 1 is well defined on .6. and belongs to Np(.6.). In other words, the equation

f satisfies the range condition (RC) if for each r > 0 and;; w+rf(w)=z

has a unique solution w = Jr(z) in .6., such that

whenever

w 1 , w2

belong to .6..

E .6.

(3.3.7)

Chapter III

70

Proposition 3.3.1 A continuous {holomorphic) function f on~ is an infinitesimal generator of a one-parameter semigroup S = {Ft} C Np(~) {respectively, Hol(~)) defined on [0, T), T > 0 if and only if it satisfies the range condition (RC). Moreover, the semigroup S = {Ft} is unique and can be continuously extended to a flow of p-nonexpansive (holomorphic) self-mappings defined on JR+ = [0, oo) by the following exponential formula

z E ~' t > 0.

Ft(z) = n-+oo lim J'1n)(z) t n

(3.3.8)

Since the proof of this assertion is rather long we will give it step by step using several lemmata which will also be needed independently in the sequel. Lemma 3.3.1 Let p be the Poincare metric on~ and let {Gt}, t E [0, T), T > 0 be a family of p-nonexpansive mappings of~' i.e.,

p(Gt(z), Gt(w)) :S p(z,w) for all z, wE~' t E [0, T). Suppose that for each z E

(3.3.9) ~

there exists the limit:

. 1 f(z) = hm -(z- Gt(z)) t-.o+ t

(3.3.10)

and assume that f is continuous on each compact subset r > 0 and each w E ~ the equation w

has a unique solution z

Proof.

of~-

Then for each

+ r f(w) = z

= Jr (w) and Jr : ~

f-->

(3.3.11)

~

is also p-nonexpansive.

Given t E (0, T) denote: 1

(3.3.12)

ft =-(I- Gt) t and consider the equation: w

+ r ft(w)

= z,

z E ~'

7'

> 0.

(3.3.13)

This equation can be rewritten in the form: r r+t

w = --Gt(w)

t

+ --z.

(3.3.14)

r+t

For fixed z E ~ the mapping defined by the right-hand side of (3.3.14) is a strict contraction with respect to the metric p (because of (3.3.9) and Proposition 2.3.1(ii)). Therefore, for each z E ~ and r ~ 0, this equation has a unique solution Wt = lrt(z) E ~- In addition, this solution can be obtained by the iteration method:

Wn+l(= Wn+l(z))

=

r r+t

--Gt(wn)

t

+ --z, r+t

(3.3.15)

GENERATION THEORY

71

where z0 is an arbitrary element in b.. Setting wo(z) = z we have by induction and Proposition 2.3.1(iii) that p(wn+l (z1), Wn+l (z2))

< max {p( Gt(Wn (z1) ), Gt(Wn (z2)) ), p(z1, z2)} < max{p(wn(zl),wn(z2)),p(zl,z2)}

~

p(z1,z2).

It means that all wn(·) are p-nonexpansive and so is Jr,t(·) = limn-+oo wn(·). :\ote, in passing, that if Gt E Hol(b.) then Jr,t : b.~---+ b. defined as the solution of !13.14) is also holomorphic on b.. ~ow we want to show that for some r > 0 and z E b. the net {Jr,t(z)}tE(O,T) converges to Jr(z), as t tends too+ and its limit is a solution of equation (3.3.11). To do this we first claim that this net lies strictly inside b.. The latter is equivalent to the inequality:

p(z, Jr,t(z)) ~ M (= M(z)) < oo, as t---+ o+. Indeed, since

r Jr t(z) = -Gt(Jr t(z)) ' r+t '

(3.3.16)

t

+z, r+t

we have by Proposition 3.3.1(ii):

p(Jr,t(z) , _r_z + _t_z) r+t r+t

(-r+ _t_lzl) p(Gt(Jr t(z)), z) r+t r+t ' (-r+ _t_lzl) [p(Gt(Jr t(z), Gt(z)) + p(Gt(z), z))] r+t r+t ' (-r+ _t_lzl) [p(Jr t(z), z) + p(Gt(z), z)]. r+t r+t '

< < <

This inequality implies that

p(Jrt(z),z) ~ ( r+tI I) ( - r ' t1- z r+t

t zl) + -r+t l

p(Gt(z),z).

Since fortE (0, T) small enough the element Gt(z) is close to z we have that rhere exists a positive number M 1 < oo such that

.

1

hmsup -p(Gt(z), z) t-+O+

t

< limsup M 1 IGt(z)- zl t-+O+ t M1l!(z)l.

Consequently, we have the estimate:

.

r M1lf(z)l =: M, 1- 1Z 1

hmsup p(Jr,t(z), z) ~ t-+O+

~rhich

implies (3.3.16). 0

72

Chapter III Thus this lemma proves the necessary assertion of Proposition 3.3.1 if we set

Gt = Ft, t E [0, T), T > 0. To accomplish this matter we note that it follows by the uniqueness of the local solution of the Cauchy problem (see also remark at the end of this section) that if f is holomorphic in D., then so is Ft for each t E [0, T). Then our constructions in Lemma 3.3.1 show that the resolvent Jr : D. ___. D. is a holomorphic mapping for each r > 0. This fact can be shown also by using the local Implicit Function Theorem (see, for example, [115]). Let as above p be the hyperbolic Poincare metric on D. and B(a, R) = {w E D.: p(a, w) < R}, a ED., R > 0. Lemma 3.3.2 Let f be a continuous function which satisfies the range condition (RC). Then for each a E D. and R > 0 there are 7 = 7(a, R), 0 < 7 < 1 and L = L( a, R) < oo such that p(J~k)(z), z):::; rkL

for all r E (0, 7) and k = 0, 1, 2, .... Proof. Since each p-ball is bounded away from the boundary of D. for given a ED. and R > 0 we can find 0 < s < 1 such that B(a, R) C D.s ={zED.: lzl < s}. Denote M = max{lf(w)l, wE B(a, R)} and set 7 = d/M, where d =dist{ aB(a, R), aD..} > o. Then for each r E (0, 7) and wE B(a, R) we have

z=w+rf(w)ED.

(3.3.17)

and w = Jr(z) E B(a, R). Hence, for such 1· and all wE B(a, R) we obtain by (3.3.17) p(Jr(w), w)

p(Jr(w), Jr(z)):::; p(z, w)

< tanh- Ilz-wl rM 1 _ 82 = arctanh 1 _ 82 .

(3.3.18)

Further, it follows by the Lagrange mean value theorem that for each t E [0, to], to < 1, -1 1 tanh t < t · - - . 1- t6 Then setting

7M to = 1 -

s2

d 111(1- s 2 ) = 1 - s2 and L = (1 - s2)2 - d2

we obtain by using (3.3.18): p(Jr(w), w) :::; 1·L

for all wE B(a, R) andrE [0, 7).

GENERATION THEORY

73

Now using the triangle inequality we have

(

k

p(J$kl(z), z) ~I:> J$il(z), J$1-l}(z)) ~ kp(Jr(z), z). j=l

Hence p(J~k)(z), z) ~ rkL

and the Lemma is proved. D Lemma 3.3.3 Let f be a continuous function in ~ which satisfies the range condition (RC). Then for each a E ~. R > 0, and c > 0, there is f.l = f.J.(a, R, c) > 0 such that for all r E [0, f.l) and each p = 0, 1, 2, ... the following inequalities hold

f(z)-

z-

fPl(z)

~c

rfp

r

(3.3.19)

and

(3.3.20) whenever z E B(a, R).

Proof. Since f is continuous in ~. for each a E ~. R > 0 and c > 0 one can find b > 0 such that !f(z)- f(w)! < c, whenever z and w belong to B(a, R) and p(z, w)
p(z, J r I p (z))

~

r k kL-p < -b, p

z E B(a, R).

Hence, for all k = 1, ... , p we have

lf(z)-

f(J~~~(z)l < c, z E

B(a, R).

In addition, it follows by the definition of the resolvent that for all w E r

Jrjp(w)-

W

=p

f (Jrjp(w)).

Now using the triangle inequality we estimate

f(z)-

z-

J(P) (z) rfp

r

~~I~ f(z)r~ p

J(k-l}(z) ~p

+ J(k) (z)l ~p

k=l

<

~ ~ ~~f(z) + Jr;p(J~~;l}(z))- J~~~(z)l ~ t ~ lf(z)- f(J~~~(z))l ~ c, k=l

~

74

Chapter III

whenever z E B(a, R). So, (3.3.19) is proved. In turn (3.3.19) implies (3.3.20):

llr(z)-

J;j~(z)l <

lz-

J;j~(z)- rf(z)l + irf(z)- z + lr(z)l z -J(p)

< r f(z)-

r

rfp

+ r lf(z)- f(Jr(z))l:::;

2rc,

and we have completed the proof. 0

Lemma 3.3.4 (The resolvent identity) Let f be a continuous function in ~ which satisfies the range condition (RC). Then for 0 :::; s :::; t the following resolvent identity holds:

lt(z) = 18 Proof. Since for each

z E ~

w= by the convexity

of~.

(f z+ (1- f) lt(z)), and t

;::=:

z

E

~.

0 the element lt (z) E

f z + ( Dlt (z) 1-

E

~

we have

~

(3.3.21)

It follows by the definition of the resolvent that

z- lt(z) = tf(Jt(z))

(3.3.22)

ls(w) + sf(Js(w)) = w.

(3.3.23)

and On the other hand (3.3.21) and (3.3.22) imply

w = lt(z)

s

+ -(zlt(z)) = lt(z) + sf(Jt(z)). t

Since equation (3.3.23) has a unique solution we have the equality

lt(z) = ls(w), which is equivalent to the resolvent identity. 0 Now we are able to complete the proof of Proposition 3.3.1. As we mentioned above the necessity of the assertion of this Proposition follows from Lemma 3.3.1. To prove the sufficiency we first show that for all t ;::=: 0 and z E ~ the limit in (3.3.8) exists. In fact, it is enough to prove that for each t 2: 0 and z E ~ the sequence {JD'~ z)} is a Cauchy sequence in the Poincare metric on ~. Indeed, fix any t 2: 0 a E ~and consider the sequence {zt}~ 1 C ~defined as follows (k(l)) ( (3.3.24) Zl = Jt/l a), k(l) :::; l.

(

For an arbitrary s E (0, 1) close to 1, one can find l 0 such that elements W!

t =a+ yf(a) E ~s = {lzl < s < 1},

GENERATION THEORY

75

whenever l;::: l. Also, for such l we have lt;t(wt) =a. Then as in Lemma 3.3.2 we obtain that there is d = d(s) > 0 such that for alll;::: l 0

p ( Jt<;?))(a),

a)

< k(l)p (Jt;t(a), a) = k(l)p (Jt;t(a), lt;t(wt)) < k(l)p(a, wt)

~

d k(l)lwt- al · 1 _ 82

k(l) d d -z-tif(a)l· 1- s2 ~ t1f(a)l1- s2. Setting now

R=max{p(a,zt), l-1,2, ... ,lo,tlf(a)l~} 1-s we obtain that the sequence { zt}~ 1 defined by (3.3.24) is contained in B(a, R). Using this fact we will show now that the sequence {Jtjl(a)} is a Cauchy sequence in Poincare metric on .0... Taking any c: > 0 and z E B(a, R), define 11 > 0 as in Lemma 3.3.3. Then we have by this lemma that there are n 0 > 0, m 0 > 0 and L > 0 such that for all n > no and m > mo (m)

)

(n)

)

p ( lt;n(z), Jtfnm(z) and P ( lt;m(z), Jtfnm(z)

<

2Lc:t

~

2Lc:t

< ~'

whenever z E B(a, R). Using these inequalities we obtain after several manipulations with the triangle inequality that

as required. Since a is an arbitrary element of .0.. we have that the limit in (3.3.8) exists and the mapping Ft : .0.. r-t .0.. defined by this formula is a p-nonexpansive mapping on

.0... Now we have to establish the continuity of {Ft(z)}t>o, for each z E .0... By using (3.3.8) it is enough to prove continuity of the resolvent lr(z) for r sufficiently small. To this end let us choose 0 < t < r < T for any a E .0.. and R > 0 as in Lemma 3.3.1. Then we have for such r, that p(lr(a), a)
p (It(~ a+ r

p (lt(a), lr(a))

< as r - t ---. 0.

~ t Ir(a)), lt(a))

p(~a+ r~tlr(a),a) ~p(a,a-(r-t)f(lr(a))---.0

Chapter III

76

The semigroup property of S = {Ft}t~o can be proved in a standard way (see, for example, [29]), using (3.3.8) and passing from rational t ;::: 0, s ;::: 0 to real numbers by the continuity of {Ft}t~o. Now using (3.3.19) (see Lemma 3.3.3) and again (3.3.8) one can easily show that for each z E fl lim z- Ft(z) = f(z). t--+0+

t

Thus f: fl....-. Cis the infinitesimal generator of the flowS= {Ft}t~o, defined by (3.3.8). Finally, the uniqueness of this flow follows by Proposition 3.2.2, and the proof is complete. 0 Remark 3.3.1 By 9Np(fl) (respectively, 9Hol(fl)) we will denote the set of all continuous (respectively, holomorphic) functions on fl which are generators of one-parameter semigroups (flows) of p-nonexpansive (respectively, holomorphic) self-mappings of fl. A direct consequence of the above Proposition and Lemma 3.3.1 is that these sets are real cones, i.e., iff and g belong to 9Np(fl) (respectively, 9Hol(fl)) then so does the function h = af + {Jg

for each pair of nonnegative numbers a and {3. Indeed, let {Ft}t~o, {Gt}t~o be the flows generated by Then if we define the family { Ht}t~o by

f and g respectively.

we have that

h(z) = af(z)

+ (Jg(z) =

lim _z_-_H_t....:...(z--'-) t--+0+

t

for each z E fl, and we are done. This fact can be also established by using representation theory of generators or the so called flow invariance conditions (see following sections). Moreover, we will see below that the sets 9Np(fl) and g Hol(fl) are closed with respect to the open compact topology on fl. 0 Exercise 1. Prove directly that if f n E g Hol( b.) is a convergent sequence on each compact subset of fl, then its limit function f also belongs to g Hol(fl). Hint: Use Proposition 3.2.2 and the properties of the solution of the Cauchy problem (*) (see section 3.2). At the end of this section we will give another important consequence of Proposition 3.3.1, which will be used in the sequel. Corollary 3.3.1 Let F : fl.....-. fl be a p-nonexpansive (respectively, holomorphic) self-mapping of fl. Then f =I-F (i.e., f(z) = z- F(z)) belongs to 9Np(fl) (respectively, g Hol(fl)).

GENERATION THEORY

Proof. We have to show that for each r

~

77

0 and z E D. the equation

(3.3.25)

w+r(w-F(w)) =z has a unique solution w = lr(z) in D.. Indeed, setting t = rj(r + 1) we can rewrite (3.3.25) in the form

w = tF(w)

+ (1- t)z.

(3.3.26)

It was shown in Section 2.4 that for each t E [0, 1) equation (3.3.26) has a unique solution w = Gt(z) and that for each t E [0, 1) the mapping Gt : D. ~---+ D. is a p-nonexpansive self-mapping of D. (see Proposition 2.4.1). In addition, if FE Hol(D.), then so is Gt (Remark 2.4.1). Thus the function f =I- F) satisfies the range condition and its resolvent lr : D. ~---+ D., r ~ 0 is defined by (3.3.27) This completes the proof. 0 Note, by the way, that the family of resolvents {Jr(z)}r~o at the point zED. f = I - F is, in fact, the rescaling approximating curve of F at this point. Thus the question whether the mapping I-F belongs to gNp(D..) whenever F E Np(D.) has been answered in the affirmative. In its turn, this fact implies another useful resolvent identity formula for the general case when f E gNp(D.).

for

Lemma 3.3.5 ([116]) Let f be a continuous function in D. which satisfies the range condition (RC), i.e., for each s ~ 0 the resolvent ] 8 = (I+ sf)- 1 is well defined on D. and belongs to Np(D.). Then for each pair s, t ~ 0 the mapping Gt := (I+ t(I- J.))- 1 is also well defined on D. and belongs to Np(D.) and the following identity holds: (3.3.28)

Proof. The existence of the resolvent Gt :=(I+ t(I- J.))- 1 follows directly by Proposition 3.3.1 and its Corollary 3.3.1. The same assertions imply that Gt E Np(D.). By definition this mapping satisfies the identity: (I+ t(I- J.))(Gt(z)) = z,

zED..

(3.3.29)

Reminding that I - ] 8 =sf (J.) we obtain

Gt(z)

+ stf (J,(Gt(z))

= z, zED.

for all zED.. At the same time rewriting (3.3.29) in the form:

(3.3.30)

78

Chapter III

we have by (3.3.30) the following identity

ls (Gt(z)) for all z E

~.

+ (1 + t)sf (J. (Gt(z)))

= z

(3.3.31)

Since the equation zE~,

w+(1+t)sf(w)=z, has a unique solution w = (I+ (1 (3.3.31) holds, (3.3.28) results. 0

+ t)sf)- 1 (z)

:= J( 1+t)s(z), provided that

Here follows the basic assertion which demonstrates the resolvent method.

Proposition 3.3.2 Let f be a continuous function in ~ which satisfies the range condition (RC) and let lr =(I+ rf)- 1 E Np(~), r ~ 0, be its resolvent. Then: (i) for each r > 0 the sets Fix( lr) and Null (f) in ~ coincide; (ii) for each z E ~ the net {Jr(z)} r2:0 is convergent as r ~ oo. Moreover, (a) if W =Null(!)

in~

is not empty, then the limit mapping F = lim lr is T--->00

a p-nonexpansive retraction on W; (b) if W =Null(!) in~ is empty, then the limit mapping F = limr_, 00 lr is a unimodular constant ( which is the sink point for each lr, r > 0. In other words, for each K > 0 the horocycle

D((, K) = internally tangent to

8~

{ z E ~:
11- z(l 2 < K }

:= 1 -lzl 2

at ( is lr-invariant, i.e.,

lr(D((,K))

~

D((,K).

Proof. Assertion (i) follows directly by the definition of the resolvent and the uniqueness of the solution of equation (3.3.11). Thus for each r > 0 we have W := Null(!) = Fix (lr)· Consider now the mapping Gt := (I+ s(I- J.)- 1 ) that was introduced in the previous lemma. By the construction of this mapping we have, in turn, that for each pairs, t > 0 W =Fix (J.) =Fix (Gt)· Moreover, it follows by Propositions 2.4.3 and 2.4.4 that for each z E ~ there exists the limit F(z) = lim Gt(z) E ~. More precisely, if W =Null(!) in ~ is not empty, then t->oo

by Proposition 2.4.4 F is a p-nonexpansive retraction on W. Otherwise, F is a unimodular constant ( which is the sink point for J•. But in both cases equation (3.3.29) implies: J. (Gt(z))- Gt(z) ~ 0 as t ~ oo for all z E ~. Setting now r = (1 + t)s and letting t to oo we obtain by Lemma 3.3.5 that F = lim lr. Finally, condition (b) follows from the resolvent identity T--->00

(Lemma 3.3.4) and we are done. 0

GENERATION THEORY

3.4

79

Monotonicity with respect to the hyperbolic metric

Our main goal in the following considerations is to find analytical characterizations and parametric representations of the classes IJNp(fl) and g Hol(fl), respectively. It turns out, that a good tool in these investigations is the notion of monotonicity with respect to the hyperbolic Poincare metric on the unit disk. To motivate the definition below, we recall that a mapping f : IR 2 ~----> IR2 is said to be monotone with respect to the Euclidean norm of IR2 if for each x, y in a domain of definition f (x- y, f(x)- f(y)) 2:: 0,

(3.4.1)

where by(·,·) we denote the inner scalar product in IR2 . Since each complex-valued function f :
(3.4.2)

for each pair z, win a domain of definition of f. At the same time it is easy to see that (3.4.2) is equivalent to the following condition: iz+rf(z)-(w+rf(w))l 2::

lz-wl

(3.4.3)

for all r 2:: 0. The latter condition is a key to define the notion of monotonicity with respect to the Poincare hyperbolic metric on fl (see [116]).

Definition 3.4.1 Let f be a complex valued function on fl, and let p be the Poincare metric on fl. The function f is called p-monotone {monotone with respect to the metric p} if for each pair z, w E fl the following condition holds: p(z+rf(z),w+rf(w)) 2:: p(z,w) for all r 2:: 0 such that z

+ r f(z)

and w

+ r f(w)

(3.4.4)

belong to fl.

Proposition 3.4.1 Let f : fl ~----> C be a continuous function in fl. Then f is p-monotone if and only if it satisfies the range condition (RC). Proof. Let f satisfy the range condition (RC), i.e., for all r ;::: 0 the nonlinear resolvent lr = (I+ r f)- 1 is a well defined p-nonexpansive self-mapping of fl: p((I +rf)- 1 (u),(I +rf)- 1 (v)) :s; p(u,v)

(3.4.5)

for all u, v E fl. Take now any pair z and w in fl and let r ;::: 0 be such that z + rf(z) := u and w + rf(w) := v belong to fl. Then by definition z = (I

+ r f) - 1 ( u)

and w

=

(I

+ r f)- 1 ( v).

80

Chapter III

It is clear now that (3.4.5) implies (3.4.4), i.e., f is p-monotone. Conversely. For a pair z, wE~ denote u = z + f(z) and v = w for r ~ 0 sufficiently small (3.4.4) implies: p ((1- r)z

+ ru, (1- r)w + rv)

~

+ f(w).

p(z, w).

Then

(3.4.6)

Denoting the left hand of (3.4.6) by ¢(r) we can rewrite it as ¢(r) ~ ¢(0). Now it follows by Proposition 2.3.2 that the latter inequality is equivalent to the condition: Re [ (z-u)z

1 - lzl 2

R z(w-v)+w(z-u) + (w-v)w] > e . 2

1 - lwl

Substituting u = f(z) + z and v = f(w) characterization of a p-monotone function: Re [ f(z)z

1-

Substituting now z

lzl 2

1 - zw

-

+ w into this inequality we obtain a

] R zf(w) + wf(z) + 1f(w)w > e . 2 - lwl 1 - zw

(3.4. 7)

= 0 into (3.4.7) we obtain the condition: Ref(w)w ~ Ref(O)w(1-lwl 2 ).

(3.4.8)

for all wE~Now we will show that condition (3.4.8) implies the solvability of the equation: w+rf(w)=z

(3.4.9)

for each r 2:: 0 and z E ll. To this end we will establish a more general assertion which we will call the numerical range lower bound (cf., [60] and [62]). Lemma 3.4.1 Let a : [0, 1] a(O) ::; 0 and the equation

f--->

lR be a continuous function on [0, 1] such that s+m(s)=t

has a unique solutions= s(t) E [0, 1) for each t E [0, 1) and r ~ 0. Suppose that f : ~ f---> C be a continuous function on ~ which satisfies the following condition: Ref(w)w ~ a(lwl) lwl, wE~Then for each z E such that

~

and r 2:: 0 equation (9} has a unique solution w = w(z)

lw(z)l ::; s(t), whenever Iz I ::; t < 1.

(3.4.10)

(3.4.11)

GENERATION THEORY

81

Proof. Fix t E (0, 1) and z E .6., izl :S t < 1, and consider the equation: s+m(s)=izj.

It follows from our assumption that this equation has a unique solution s 0 = so (izl). Then setting -y(s) = s+ra(s) -lzi we have -y(O) < 0 and ')'(so)= 0. Note that')' must be monotone on [0, 1]. Hence, for an arbitrary 0 < o< 1- so we can find e > 0 such that -y(so + o) 2': e. Taking now wE .6. such that lwl =so+ owe have by (3.4.10) for those w:

Re(w+rf(w)-z)w > lwl 2 +m(lwl)lwl-lwllzl lwi'Y (iwi) 2': lwle > 0. Then it follows by the Bohl-Poincare theorem (see, for example, [79]), that equation (3.4.9) has a unique solution w = w(z) such that lwl < s 0 + o. Since o is an arbitrary sufficiently small number, we must have lwl :S so :S s(t). D To complete the proof of Proposition 3.4.1 we first note that the function

a(s) = -lf(O)I (1- s2 ) satisfies the conditions of the above Lemma. Therefore inequality (3.4.8) and this Lemma imply that for each r 2': 0 the resolvent Jr, defined by Jr(z) = w(z)- the solution of (3.4.9) is a single-valued self-mapping of .6.. It remains to show that this mapping is p-nonexpansive on .6.. Indeed, for a pair u and v in .6., setting z = Jr(u) and w = Jr(v) we obtain z + rf(z) = u and w + rf(w) = v. Since f is p-monotone we obtain finally

p(u, v)

= p (z + r f(z), w + r f(w)) 2': p(z, w) = p (Jr(u), Jr(v)).

The Lemma is proved. D We already mentioned in the proof of Proposition 3.4.1 that, in fact, condition (3.4.7) is equivalent to the property of a continuous function f on .6. to be pmonotone. Thus combining these assertions with Proposition 3.3.1 and Lemma 3.3.1 we can formulate a summary assertion for this chapter which characterizes the property of a continuous function on D. to be in class QNp(.6.).

Proposition 3.4.2 ([116]) Let f be a complex-valued continuous function on .6.. Then the following conditions are equivalent:

(i) f E QNp(.6.), i.e., it is a generator of a continuous fiow S = {Ft}t:=::o of p-nonexpansive self-mappings of .6.; (ii) there is a family {Gt} 0 :S t :S T, of p-nonexpansive self-mappings of D. such that f(z) = lim z- Gt(z) t--->0+ t for each z E .6.;

82

Chapter III (iii} the Cauchy problem

{

ou(t,z) at

+ !( u (t, z ))-0 - ,

u(O, z) = z E

~

has a unique solution u(t, z) E ~ for all t ~ 0 and z E ~. such that for each t ~ 0 u(t, ·) E Np(~); (iv) f satisfies the range condition (RC}, i.e., for each r ~ 0 and z E ~ the equation

w+zf(w)=z has a unique solution w = Jr(z), such that Jr E (v) f is a p-monotone function on ~; (vi) f satisfies the condition:

Re [ f(z)z 1 - lzl 2 for each pair z, w in

Np(~);

+ f(w)w ] > Re zf(w) + wl[Z) 1 -lwl 2 1 - zw

~;

Remark 3.4.1 Condition (vi) plays a crucial role in our further considerations. Inequalities of such a type (which characterize the classes of generators of flows) are called flow in variance conditions. For the class of holomorphic mappings a simpler inequality (3.4.8) is also a flow invariance condition, since it characterizes the class of holomorphic generators. Indeed, as we saw in the proof of Proposition 3.4.1 (or in Lemma 3.4.1) this condition is sufficient for the existence of the (nonlinear) resolvent (I +r f)- 1 which maps~ into itself. In addition, it follows by the Implicit Function Theorem that this mapping is holomorphic (hence, p-nonexpansive) in ~. so f belongs to g Hol(~). The necessity of this condition follows directly from condition (vi). Thus for f E Hol(~, C) inequalities (3.4.7) and (3.4.8) are equivalent.

We will see below that these conditions can be considered as forms of the Schwarz-Pick inequalities for the classes QNp(~) and g Hol(~) of generators of flows of p-nonexpansive and holomorphic mappings, respectively. A geometric nature of these conditions for holomorphic functions will be explained in the next section. Here we note that an immediate consequence of these flow invariance conditions is the following (cf., Remark 3.3.1): Corollary 3.4.1 The sets 9Np(~) and g Hol(~) are closed (with respect to the topology of uniform convergence on compact subsets of~) real cones. Exercise 1. Show that the set QHol(~) n (-QHol(~)) is precisely the set of all generators of one-parameter groups of automorphisms of~- Hence, this set is a real vector space. Exercise 2. Describe the set

9Np(~)

n ( -QNp(~)).

GENERATION THEORY

3.5

83

Flow invariance conditions for holomorphic functions

In this section we study several flow invariance conditions for the class of holomorphic functions. We will use these conditions to obtain parametric representation of functions of the class Q Hol(~), to study their dynamic transformations and the asymptotic behavior of the flows generated by them. In the first step we give a simpler explanation of the necessity of the flow invariance condition (3.4.8) for a function f E Q Hol(~) and later we study it in greater detail. In addition, we will see below that this condition can be improved by a more qualified condition (see Proposition 3.5.3) which has some additional applications. Also, note that in the case of holomorphic functions one uses a special terminology which comes from the theory of bounded symmetric domains (see, for example, [142, 12, 32]).

Definition 3.5.1 A function f E Hol(~, q is said to be a semi-complete vector field on ~ if the Cauchy problem (*):

{

au~~ z) + f

(u(t, z)) = 0,

u(O,z) = z E ~' has a unique solution u(t, z) E

~for

all z E

~

and all nonnegative t, i.e., t E

JR+ = [0, oo). If the Cauchy problem (*) has a solution u(t, z) defined for all real t, i.e., t E lR = (-oo, oo), then f is said to be complete (or integrated) (see, for example, {32, 142}). Thus f is semi-complete if and only if it is an infinitesimal generator of a one-parameter semigroup; f is complete if and only if it is a generator of a oneparameter group. Exercise 1. Show that the mapping semi-complete vector field on ~-

f :~

f-+

C, defined as

Exercise 2. Show that f : ~ f-+ C, defined by f(z) semi-complete vector field. Find the flow generated by f. Exercise 3. Show that for each a complete vector field.

E

C the mapping

=

f (z)

=

z- 1 +

f : f(z)

z - z 2 is a

J1="Z

is a

= a- az 2 is a

We begin with a characterization of all complete vector fields. Note again that if f is a complete vector field, then it generates a one-parameter subgroup S = {Ft} tElR of the group Aut(~) of all automorphisms of~- Consequently, each

84

Chapter III

Ft E S is a fractional linear Mobius transformation. Thus Ft has a holomorphic continuation on a neighborhood of b. and so does f, because of the equality:

f

=

lim I - Ft. t-.o+

t

(3.5.1)

The family of complete vector fields on b. will be denoted by aut(b..). As we mentioned above, this family is a real vector space, which can be described as follows. Since IFt(z)l::; 1, for z E b. we have by (3.5.1) that:

Ref(z)z 2:0 for all z E 8/::i. At the same time the function fore we also obtain that

f is also a complete vector field on

-Ref(z)z 2:0 for all z E 8!1.

(3.5.2) b.. There-

(3.5.3)

Comparing (3.5.2) and (3.5.3) we obtain the necessary boundary condition for

f

E aut(b..):

Re f(z)z = 0 for all

z E 8!1

(3.5.4)

(see Figure 3.1)

····················...

Figure 3.1: Boundary condition for f E aut(b..). Actually, this condition is also sufficient for f E Hol( b., C) to be complete. Indeed, suppose that f E Hol(b.., C) and satisfies (3.5.4). Rewriting f in the Taylor series form: we have for

z E

8/::i:

Ref(z)z = Reg(z) = 0,

GENERATION THEORY where g(z) = a1

85

+ (ao + a2)z + a3z 2 + ... E Hol(-6., C).

It follows by the maximum principle for harmonic functions that:

Hence f(z) is actually polynomial of the second order at most: f(z) = ao

+ a1z + a2z 2

(3.5.5)

with (3.5.6)

Now it can be shown by direct computation that the Cauchy problem:

{ au~; z) + f

(u(t, z)) = 0,

u(O,z) = z E b.,

with f satisfying (3.5.5) and (3.5.6) has a unique solution u(t, ·) E Hol(b.) for all t E R Moreover, for fixed t E lR the mapping Ft = u(t, ·) is a Mobius transformation of the unit disk. So, we have proved the following result (see, for example, [16, 8]). Proposition 3.5.1 (Boundary group invariance condition) A mapping f E Hol(-6., C) is a complete vector field on b. {i.e., f E aut( b.)) if and only if it has a continuous extension to b. and Ref(z)z = 0

for all z E 8.6..

This is equivalent to the statement that f is a polynomial of the second order at most: f(z) = ao + a1z + a2z 2 with coefficients ao, a 1, a2 which satisfy the conditions:

Corollary 3.5.1 The family aut(.6.) is a real vector space of entire functions. Moreover, this space has the following decomposition: aut(b.) =auto( b.) El7 P2, where

{! E aut(.6.) : f(O) = 0} {! E Hol(b., C) : f(z) = bz, Reb= 0}

auto (.6.)

is the subspace of linear functions, and

{! E aut(b.), f'(O) = 0}

P2 =

{!

E Hol(b., C): f(z) = i i - az 2 , a E

is the subspace of so called 'transvections '.

C}

Chapter III

86

Now we will turn to the general case when f E Hol(b., C) is a semi-complete vector field on the unit disk. To explain the nature of the condition (3.4.8) we adduce first some heuristic grasps. Assume temporarily that f E QHol(b.) is holomorphic in the neighborhood of b.. In this case we will write just f E Hol(b., C). Then, again we have the following boundary flow invariance condition: Re f (z) · z 2:: 0 for all

z E ab.

(see Figure 3.2).

Figure 3.2: Boundary flow invariance condition.

It implies that

Re(f(z)- f(O))z 2: - Re f(O)z,

z E aD..

Dividing the left hand side of this inequality by lzl 2 Re ( f(z)

~

f(O)) 2::- Ref(O)z,

= 1 we obtain:

z E aD..

Now again it follows by the maximum principle for harmonic functions that the latter inequality holds also for z E b.. Multiplying it by lzl 2 =f. 0, z E b., we obtain: (3.5.7) Ref(z) · z 2:: Ref(O) · z · (1 -lzl 2 ), zED.. However, there are holomorphic mappings on the unit disk which are semicomplete, but have no holomorphic extension to D., the closure of b.. Consider, for example, f(z) = z- 1 + v/1 - z. Nevertheless, we will see that even f E Hol(D., C) does not extend continuously to D., condition (3.5.7) is necessary and sufficient for f to be in QHol(D.). The

87

GENERATION THEORY

necessity can be shown also by the following simple considerations which are useful, however, to obtain a parametric representation of the class g Hol(~). As we already know this class is a real cone. Therefore, if we present f E g Hol(~) in the form:

f(z) = g(z) + h(z),

(3.5.8)

where

g(z) = f(O)- f(O)z 2 is a transvection (i.e., g E P2 , see Corollary 3.5.1), we have

+ (-g(z))

h(z) = f(z)

E QHol(~)

(3.5.9)

and (3.5.10) h(O) = 0. Now, conditions (3.5.9) and (3.5.10) imply that there is a semigroup Sh = {Ht}t~o of holomorphic self-mappings Ht of~. such that Ht(O) = 0, for all t ~ 0. Hence by the Schwarz Lemma we have:

[Ht(z)[ ::::; [z[,

for all

Since

h(z) = lim

t-->0+

z E ~-

~(z- Ht(z)) t

we obtain:

Reh(z)z

~ 0

for all z E

~-

(3.5.11)

In addition, note that

Reg(z)z = Re f(O)z · (1 -[z[ 2 ).

(3.5.12)

Then by (3.5.11), (3.5.12), and (3.5.8) we obtain (3.5.7) and the necessity of this condition is proved. The sufficiency of condition (3.5.7) for f E Hol(~, q to be a semi-complete vector field was established in Lemma 3.4.1 (see, also, Remark 3.4.1). However, for the case of holomorphic functions one can make this lemma more precise. Lemma 3.5.1 Let a E g Hol(~) be such that a ([z[) E Hol(~, q satisfy the following condition:

Re f(z)z

~a

([z[) [z[,

zE

~,

z E

~.

and let f E

~-

Then: (i) f is a semi-complete vector field on ~; (ii) if Sf = { Ft}, t ~ 0, is the semigroup generated by f, then for all t and x E ~

fFt(x)[ ::::; ,Bt([x[), where f3t is the solution of the Cauchy problem:

{

d,Bt(s) + a(B (s)) = 0 dt

,Bo(s)

.

=

s,

t

s E [0, 1).

,

~

0

Chapter III

88

Proof. It follows by Proposition 3.3.1 that the function a : b. f-+
+ rf)-1 (z)l ~(I+ ro:)-1

(lzl).

Using again Proposition 3.3.1 and the exponential formula we obtain our assertion. 0 Now, observing that the function:

= -lf(O)I (1- z 2 )

a(z)

satisfies all the conditions of Lemma 3.5.1 we obtain the following result:

Proposition 3.5.2 (see [7]) Let f E Hol(b.,
~

lzl lzl

+ 1- e-2lf(O)It(l -lzl) + 1 + e-2lf(O)It(1 -lzl).

(3.5.13)

Remark 3.5.1 Note that if eq·uality in (3.5. 7} holds for at least one value z0 E b. then it holds for all z E b., and f ( z) is actually a complete vector field. Indeed, by (3.5.8) we have: f = g + h, where g E aut( b.), and h E QHol(b.), with h(O) Therefore if we present h in the form

= 0.

h(z) = z · p(z), we obtain that p

E

Hol(b.,
z E b..

(3.5.14)

Thus (3.5.7) is equivalent to the following equation

f(z) = f(O)

+ zp(z)- f(O)z 2 .

If now for some zo E b. we have equality in (3.5.11) we also have the equality Rep( z0 ) = 0. It then follows by the maximum princi pie that p( z) = p = const, hence f(z) has the form (3.4.5)

f(z)

=

f(O)

+ ipz- f(O)z 2 ,

GENERATION THEORY

89

i.e., it is a complete vector field. As a matter of fact (3.5.14) enables us to find a more qualified estimate which also characterizes semi-complete vector fields. To do this we need the following assertion.

Lemma 3.5.2 (Harnack inequality) If p E Hol(.6., C) maps .6. into the right half-plane, i.e., Rep(z) 2: 0, z E .6., then it satisfies the strong estimate:

1 -lzl Rep(0)-1- 1 1+z

::;

1 + lzl Rep(z)::; Rep(0)-1- 1 . 1-z

(3.5.15)

Proof. If Rep(O) = 0, then as above Rep(z) = 0 for all z E .6. and (3.5.15) is obvious. If p E Hol(.6., C) satisfies the strong inequality in (3.5.14), then the function PI: 1

.

PI(z) = Rep(O) [p(z)- dmp(O)] belongs to the so called class of Caratheodory: Rep1(z) > 0, z E .6.

and PI(O) = 1.

Since the fractional linear transformation:

G(w) = w -1 w+ 1

(3.5.16)

maps the right half-plane {w E C : Re w > 0} into the unit disk (see Exercise 4 bellow), we have that the mapping F:

F(z) = G(pi(z)) is a holomorphic self-mapping of .6. and F(O) = 0. Then the Schwarz Lemma implies that for each 0 ::; r < 1 and z E .6. : lzl ::; r the following inequality holds: PI(z)- 11 < r. IPI(z)+1Now it is easy to see that the circle Iw- 1 1 = r is symmetric with respect to w+ 1 1 +-r an d 1 - -r (see F'1gure 3.3) . · an d mtersects · · at t he pomts · t he rea1 axis It 1-r 1+r Therefore we obtain the following inequality for each function p 1 of the class of Caratheodory: 1 - lzl 1 + lzl - < R ep 1 (z) < 1 + lzl - 1- lzl.

Chapter III

90

1

0 1-r 1+r

l±!:

l±r~

1-r

1-r

Figure 3.3: Values of functions of Caratheodory's class. Since

1

Rep1 (z) = Rep(O) Rep(z) we obtain (3.5.15). D Exercise 4. Prove that for all w such that Re w inequality holds:

~

0 (w E II+) the following

1:~~~~1. Exercise 5. Prove the equivalence of the following assertions:

(a) a function p E Hol(A, q has values (for all z E A) in a compact subset of the open right-half plane {w E C : Re w > 0}; 1- w(z) ( ) , where w E Hol(A) is such that lw(z)l ~ c < 1 for all (b) p(z) = 1 +w z

z EA. Exercise 6. Show that under the conditions of Exercise 5 the function p satisfies the strong Harnack inequality:

c:z: ~ Rep(z) ~ Rep(O) 11+cz - c:z:,

Rep(O) 1 + 1-cz

0

~ c < 1.

Finally, observe that from the representation off E Q Hol(A):

f(z) = f(O)- f(O)z 2 + z · p(z), we deduce that: p(O)

= f'(O).

Thus we have proved the following assertion.

Rep(z) ~ 0

(3.5.17)

GENERATION THEORY

91

Proposition 3.5.3 ([6]) A function f E Hol(~, q is a semi-complete vector field on ~ if and only if Ref' (0) 2 0 and the following inequality holds:

7

2 Ref'(O) lzl (1 \zl) + Ref(O)z(1 -lzl 2) 2 Ref(z)z 1- z

2 Re f(O)z(1 - lzl 2) + Re f'(O)

lzl:(~ lzllzl).

(3.5.18)

Moreover, the equality in (3.5.16} holds if and only ifRef'(O) = 0. Corollary 3.5.2 ([7]) A mapping f E Q Hol(~) belongs to

aut(~)

if and only if

Rej'(O) = 0.

Corollary 3.5.3 ([7]) Iff E Q Hol(~) is given with f(O) = f'(O) = 0, then =.0.

f

This assertion can be considered as a tangential version of the Schwarz Lemma. Indeed, ifF E Hol(~) is a holomorphic self-mapping, then f = I-F is a semicomplete vector field. Therefore, if F(O) = 0 and F'(O) = 1, then f(O) = f'(O) = 0. Therefore, by the Corollary 3.4.2 we obtain that F(z) = z. This is the statement of the second part of the Schwarz Lemma. Recently M. Abate [5] established another condition, which characterizes a semi-complete vector field f by using the estimate for its derivative f'. To establish his condition we first prove the following characterization of the class P := {p E Hol(~, IC): Rep(z) 2 0, z E ~}.

Lemma 3.5.3 ([7]) Let p E

Hol(~,

q.

Then condition (3.5.14}:

Rep(z) 2 0, holds for all z E ~. if and only if there is a positive function 'lj;: [0, 1) that: Re (zp'(z) + 'lj;(lzl)p(z)) 2 0 for all z E

1--4

JR+ such (3.5.19)

~.

Proof. Let p E Hol(~, q satisfy (3.5.14). Define F = (p- 1)(p + 1)- 1 which maps ~ into itself, F E Hol(~). Applying the Schwarz-Pick Lemma to F we obtain the inequality:

l(

p-1)'1 2lp'l lp+11 2 -IP-11 2 p+1 =l1+pl2:::; lp+112(1-lzl2)'

which implies: I '( )I p z

:::;

2 Rep(z) 1- lzl2 .

Chapter III

92 Consequently,

Re( -zp'(z)) :::; lzp'(z)l :::;

2lzl Rep(z) 1 + lzl 2 :::; Rep(z). 1 - lzl2 1 - lzl2

1 + t2 Setting here 1/J(t) = - -2 we obtain (3.5.19). 1-t Conversely. Suppose that (3.5.19) holds with a positive Setting z = Tei 6 we have:

1/J: [0, 1) f-+ JR+.

zp'( z ) = r ap ar and (3.5.19) becomes: ( Re

(r:~) +1/J(r)Rep(z)) ~ 0,

z = rei 8 E !:J..

(3.5.20)

Assume that there exists z 0 = r 0 ei!Jo in !:J. such that Rep(zo) < 0. But (3.5.20) implies that Re p(O) ~ 0, hence there is r 1 E [0, r 0 ) such that Rep(r 1 ei80 ) = 0 due to continuity. Then one can find r2 E ( r 1 , ro) such that Rep(r2ei 80 ) < 0 and Re

:~ heiBo)

< 0.

But these inequalities contradict (3.5.20). Thus it follows that Rep(z) everywhere and we are done. D

~

0

Now it is easy to verify that condition (3.5.19) with 1/J(r) = 1 + r: is equivalent 1-r to the condition: Re [2f(z)z + f'(z)(1 -lzl 2)] ~ 0. Thus we can summarize the assertions of this section in the following result. Proposition 3.5.4 Let f E Hol(!:J., q. Then the following are equivalent:

(i) f E QHol(!:J.,C), i.e., f is a semi-complete vector field on l:J.; (ii) Ref(z)z ~ Ref(O)z(1-lzl 2); (iii) Ref'(O) ~0 and Re [f(O)z(1 -

lzl2)

+ f'(O)Izl21 + lzll 1-lzl

> Re f(z)z > Re [f(O)z(l -lz1 2) + f'(O)Izl 21 - lzl] . 1 + lzl '

(iv) Re [2f(z)z + f'(z)(1 -lzl 2)] ~ 0;

93

GENERATION THEORY (v) f(z) = f(O)- f(O)z 2

+ z · p(z),

with p E Hol(~, C), Rep(z) ;::: 0.

Moreover, if for some z 0 E ~ the equality in one of the conditions {ii) or (iv) holds, then it holds for all of these conditions and for all z E ~. In this case the function p(z) in (v) is constant and f is actually a complete vector field.

Remark 3.5.3 The class offunctions of the form: h(z) = z·p(z), where Rep(z);::: 0, z E A usually referred to as class N. The class M consists of all elements of N which are not linear functions, i.e.,

M ={hE

Hol(~,

C): h(z) = z · p(z), Rep(z) > 0, z E

~}.

Note that N = auto(~) E9 M. Thus condition (v) of the above proposition means that the class of the semi-complete vector fields on ~ admits the following decompositions: (3.5.21) QHol(~) = P2 +N =aut(~) E9 M. (see Corollary 3.5.1). In addition, it is well known (see, for example, [57] and [122]) that for each holomorphic function p on ~ with values in the closed right half-plane II+ (i.e., Rep(z) ;::: 0) there exists a positive increasing finite function /-lp on the unit circle 8~, such that 1 + z( . (3.5.22) p(z) = ---df..lp(() + zb 1- z(

J

a.c::..

with some real b. This formula is called the Riesz-Herglotz representation of functions in P = {p E Hol( ~.C) : Rep( z) ;::: 0}. It establishes a linear one-to-one correspondence between the set of all positive measures on 8~ and P. We will call the function /-lp : 8~ 1--4 lR the measure characteristic function for pEP.

Thus by (3.5.21) and (3.5.22) we have the following integral parametric representation for f E Q Hol(~) : f(z) =a- az 2

+ izb +

J

a.c::..

1 + z( z---df..l((), 1- z(

(3.5.23)

where a E C, b E lR and J.l is a positive function on 8~. Another parametric representation of the class g Hol( ~) which is determined by the location of null points off E QHol(~) is due to E. Berkson and H. Porta [17]. We will give it in the next chapter.

Exercise 7. LethE 0, z E

Hol(~,q

belong to class N: h(z) = z · p(z), Rep(z);:::

~.

(a) Show directly that the Cauchy problem: {

8 u~; z) + h (u(t, z)) u(O,z)

=

z E

~.

= 0,

(3.5.24)

94

Chapter III

has a unique solution u(t, z) E 6. for all t ~ 0, and u(t, 0) (b) Show that lim u(t, z) = 0

= 0, t

~

0.

t-->oo

if and only if hEM, i.e., Rep(z) > 0, z E 6.. (c) Show that if hEM with h'(O) = 1, then the limit lim etu(t, z) := F(z)

(3.5.25)

t-->oo

exists for all z E 6. and F E Hol(6., q. Hint: Define the function Pl E M by p 1 ( z) = 1/p( z) and show that the Cauchy problem (3.5.24) is equivalent to the following integral equation

1

d(

u(t,z)

In [etu(t,z)] = lnzwhich

impl~s

t~~ etu(t, z)

(Pl(() -1) ( '

z

t

d(

= zexp(Jo (p 1 ( ( ) - 1) ().

(3.5.26)

(d) Show that F E Hol(6., q defined by (3.5.25) (or (3.5.26)) satisfies the following inequality:

zF'(z)] Re [ F(z)

(3.5.27)

> 0.

(It is well known due to R. Nevanlinna [102] that the latter inequality characterizes all univalent F functions on 6., normalized by F(O) = 0, F'(O)-# 0, whose image F(6.) is starlike with respect to zero. See Chapter 5.)

f, defined by:

Exercise 8. Show that the function

h(z) = z belongs toM, and h'(O)

= 1.

1 + zeiiJ

.9 ,

1- ze'

() E

[0, 27r]

Find explicitly the function F(z) in (3.5.25). 00

Exercise 9. Show that if the function f defined as: f(z) is a semi-complete vector field on 6. and Re a 1

= 0,

then a0

= ao+a1z+ z::::akzk

= ak = 0 for

k=3

all k

~

3.

95

GENERATION THEORY

3.6

The Berkson-Porta parametric representation of semi-complete vector fields

An important consequence of Proposition 3.4.1 and Corollary 3.4.1 is the following representation of semi-complete vector field, which is originally due to E. Berkson and H. Porta [17] (see, also [6]).

Proposition 3.6.1 A mapping f E Hol(D., C) is a semi-complete vector field on D. if and only if there is a point T E D. and a function p E Hol( D., C) with Rep( z) ~ 0 everywhere such that: (3.6.1) f(z) = (z- r)(l- zr)p(z). Moreover, such a representation is unique and

D., or the boundary sink point of the resolvent lr

T

:=

is either a null point off in , r > 0.

(I+ r f) - l

Proof. Firstly, let f be a semi-complete vector field on D. with a null-point TED.. Then it follows by formula (3.4.7) that the following inequality holds: Ref(w)w ~ (1

f(w)r

-lwl 2 ) Re -_ 1-WT

(3.6.2)

or,

~W l2 - ~) 1- WT ~ 0.

Ref(w) ( 1-

(3.6.3)

We calculate

w

w - lwl 2 7 -

7 + rlwl 2 lwl 2 )(1 - wr) w-r (1 - lwl 2 )(1 - wr) lw-rl 1 (1 -lwl 2 ) . (w- r)(1- wr) ·

7

-....,...-=- - 1 -lwl 2 1- WT

(1 -

(3.6.4)

Now by identifying w E D. with z E D. we obtain from (3.6.3) an'd (3.6.4):

Re (Z -

f(z) T) ( 1 -

>0

(3.6.5)

ZT) -

for all z E D.. Denoting:

p(z = f(z) ) (z-r)(1-z7)

(3.6.6)

we have (3.6.1). Now suppose that f E QHol(D.) has no null point in D.. Then it follows by Proposition 3.3.2 that in this case there is a unique boundary point T, such that for each w E D., the net { Zr (w)} r>O defined as the solution of the equation

Zr(w)

+ r f(zr(w))

=

w

(3.6.7)

Chapter III

96

converges toT, as r----+ oo. (Indeed, for each r 2: 0 the value zr(w) is just the value of the resolvent Jr = (I + r f) - l at the point w E .6). Fixe > 0 and consider the mapping fe E Hol(.6, C) defined as fe(z) = e · z + f(z). It is clear that fe converges to f as e goes to zero. Since QHol(.6) is a real cone, it follows that f" E Q Hol(.6) for each e 2: 0. In addition, the equation

fe(z)

= 0

is a particular case of (3.6.7) with r = 1/e and w = 0. Hence fe has a unique null point T,;; E .6 and the net {T,;; }00 converges to T as e tends to zero. Since fe satisfies the inequality: Re

fe(z) >0 (1-ZT,;;)(z-T,;;)- '

letting e tend to zero we have the same (inequality (3.6.5)) for f(z), which in turn implies representation (3.6.1). Conversely. Suppose that f E Hol(~) admits representation (3.6.1) with T E .6 and Rep(z) 2: 0 everywhere. If Rep(z) = 0 for some z E .6, then by the maximum principle it follows that p(z) = im for some mER In this case:

f(z)

(z- r)(1- zf)im = (z- T - z 2 f -imT- z 2 fim + (1 + ITI 2 )imz.

Denoting -imT :=a,

(1

+ ITI 2 )im := b we obtain:

f(z) =a- az 2 + bz, i.e.,

+ ziTI 2 )

Reb= 0,

f is a complete vector field (see Proposition 3.5.1). Therefore, we have to consider only the case when Rep(z) > 0. Let us present f E Hol(.6) in the form: f(z) =a- az 2

+ z · q(z),

(3.6.8)

where a= f(O). Comparing (3.6.1) with (3.6.8) we have f(O)

q(z) = (1- f'z

+ ITI 2 )q(z)-

q(z)- q(O) z

=

-rq(O) and

T- zTq(O).

(3.6.9)

To proceed we need the following lemma which will be also useful in the sequel.

Lemma 3.6.1 (cf., [6]) Let T be in .6 and let p and q be those holomorphic functions in .6 which satisfy equation (3.6.9). Then Rep(z) > 0 if and only if Req(z)>O. Moreover, the values of p lie strictly inside IT+ = {Re w > 0} if and only if the values of q lie strictly inside IT+.

97

GENERATION THEORY

Proof. First we note that assuming one of the functions p or q to be holomorphic in ~ we have that the second one is holomorphic on ~ too. Observe also that it is enough to prove our assertion under the above stronger assumption. Indeed, for the general case one can use an approximation argument: given p (or q, respectively) with Rep(z) > e: ~ 0, set Pn(z) = p(rnz) for r E (0, 1), rn--> 1-. So we assume that both these functions are holomorphic on ~­ Then substituting z = eiiJ, () E JR., in (3.6.9) we calculate

+ ITI 2 - fei 0 )p(ei0 ) - Te-i 0 p(ei0 )} (1 + ITI 2 - 2 Re fei 0 ) Rep( ei 0 )

Re{(1

Req(ei0 ) =

11- fei 0 12 Rep(ei 0 ).

Since by our assumptions, T E ~. and the functions Rep and Re q are harmonic, we see that Rep(z) > e: ~ 0 if and only if Req(z) ~ 8 > 0. Moreover, if e: is positive, then 8 can be chosen positive too, and conversely. D Returning to the proof of the Proposition 3.6.1, we see that if T E ~ then > 0, hence it is semi-complete (see Remark 3.5.2). If Tin (3.6.1) belongs to{)~ we just apply again the following approximation argument. We choose any sequence Tn E ~ such that Tn --> T and set fn(z) = (z- Tn)(1- zfn)p(z). It is obvious that {fn}~=l converges to f uniformly on each compact subset of~- Since we already know that each fn belongs to Q Hol(~), we have that so does f, and we have completed the proof. D f defined by (3.6.1) admits representation (3.6.8) with Req(z)

Remark 3.6.1 Thus this proposition implies that a semi-complete vector field f on ~ has at most one null point in ~- If such a point exists it must be T in representation (3.6.1). For T E ~ we will denote by Q Hol(~, T) the class of functions f E Q Hol(~) with f(T) = 0. Thus, Q Hol(~, T)

= {f

E Hol(~,
If T in this representation is a boundary point of~. then f is null point free. Moreover, it follows by the exponential formula that T is, in fact, the sink point of the semigroup { Ft} t;:::o generated by f, i.e., for each K > 0 the horocycle

D(T,K) = inter·nally tangent to{)!:;,. at

zfl 2 } { z E ~: 'Pr(z) := 111 -lzl 2 < K T,

is Ft-invar'iant, i.e.,

Ft(D(r,K))

~

D(r,K).

Also, it can be shown (see Section 4.6) that T E in the following sense: lim f (rr) = 0. r----.1-

{)~

is the limit null point off

Chapter III

98

However, it happens that f E g Hol(A) may have more than one null point on the boundary of A (consider, for example, f(z) = 1 - z 2 ). So the question is which one of them is the sink point of the semigroup {Ft}t>o generated by f. Another question relates to the asymptotic behavior of such a-semigroup. In the next chapter we intend to answer these questions as much as to find the best rates of the exponential convergence. At the end of the section we will add some preparatory material concerning this matter. To clarify our further reasoning we first summarize briefly different characterizations of semi-complete vector fields.

Summary Let f E Hol(A, C). The following are equivalent: {i} f E QHol(A), i.e., f is a semi-complete vector field; is a well defined holomorphic (ii} for each r > 0 the mapping Jr = (I+ r self-mapping of A; (iii} f is p-monotone with respect to the Poincare hyperbolic metric on A, i.e., for each pair z, w E /::). p(z+rf(z),w+rf(w)) ~p(z,w),

n-l

whenever z + r f ( z) and w + r f (w) belong to A for some positive r; (iv) f admits the following parametric representation f(z) =a-

az 2 + zq(z)

for some a E C and q E Hol(A,C) with Req(z) ~ 0, z E A; (v) f admits the Berkson-Porta parametric representation f(z) = (z- r)(l- zf)p(z) for some TEA andp E Hol(A,C) with Rep(z) ~ 0, z EA. In addition, different flow invariance conditions given in terms inequalities are presented in sections 3.4 and 3.5. The simplest one can be formulated as follows (vi) there exists a number m E lR (in fact, m :::; 0) such that Ref(z)z ~ m(l-lzl 2 ),

z EA.

In the study of asymptotic behavior of the semigroup generated by f E g Hol(A) with an interior null point, a few stronger conditions than (i)-(vi) will be relevant.

Definition 3.6.1 (cf., [39]) A function f : A f-+ C is said to be strongly pmonotone if for some c: > 0 and for each pair z, w in A there exists J = J(z, w), such that p(z + rf(z), w + r f(w)) ~ (1 + rc:)p(z, w), whenever

0:::; r < J.

Of course, a strongly p-monotone holomorphic function in A is semi-complete. Moreover, such a function must have a unique null point in A. Indeed, by definition

99

GENERATION THEORY we have that (at least for r :2: 0 small enough) the resolvent lr E contraction with respect to the hyperbolic metric pin ~:

Hol(~)

1 p(Jr(z), lr(w)) :::; --p(z, w). 1 +rc

is a strict

(3.6.10)

Thus lr has a unique fixed point T in ~ because of the Banach Fixed Point Principle. This point is a null point of f. Further, putting w = T in (3.6.10) and differentiating it with respect to r at the point r = o+ we obtain

or Rep(z) :2: c(1

-ITI 2 )- 1 > 0,

where pis the factor in the Berkson-Porta representation (see (3.6.1)). Now Lemma 3.6.2 enables us to conclude that q(z) in representation (3.6.8) has a real part strictly separated from zero, i.e., Req(z) > c 1 for some c 1 > 0 and all z E ~- In turn, the same formula (3.6.8) implies

for all z close enough to a~, the boundary of~­ Thus we have proved the following assertion. Proposition 3.6.2 Let f E

Hol(~,

q

be strongly p-monotone

in~-

Then:

(i) f admits the representation f(z)

= (z- T)(1- zr)p(z)

with T E ~ and Rep(z) > E for some E > 0; (ii) there exist positive numbers 8 and TJ such that Ref(z)z :2: TJ for all z in the annulas {1- 8 <

>0

lzl < 1}.

Again from Lemma 3.6.1 it can easily be seen that (ii) implies (i). Thus these conditions are equivalent. As a matter of fact, we will see below (Section 4.5) that (ii) implies the strong p-monotonicity of a function f E Hol(~, q.

Chapter 4

Asymptotic behavior of continuous flows In this chapter we want to trace a connection of the iterating theory of functions in one complex variable and the asymptotic behavior of solutions of ordinary differential equations governed by evolution problems. Therefore our terminology is related to both these topics.

4.1

Stationary points of a flow on

~

Quoting M. Abate [2], note that E. Vesentini seems to be the first person who suggested an analog of the Denjoy-Wolff Theorem for continuous time semigroups. In fact, in 1938 J. Wolff [158] himself initiated the consideration of dynamical systems determined by holomorphic functions. However, the first general continuous version of the Wolff-Denjoy Theory was given by E. Berkson and H. Porta [17] in their study of the eigenvalue problem for composition operators on Hardy spaces.

Definition 4.1.1 A point ( E {Ft}t>D C Hol(~), if

~

is said to be a stationary point of a flow S

lim Ft(r()

r-1-

= (

=

(4.1.1)

for all t > 0. In other words, ( E ~ is a stationary point of S if it is a common fixed point of all Ft E S. Note that the family S = {Ft}t~o is commuting, that is, FtoF5 = F.oFt = Ft+s for all t, s ~ 0. Hence, it follows by the Shield theorem [132] that if each Ft had 101

102

Chapter IV

been continuously extended to {)jj. the boundary of /j., then the stationary point set of S would not be empty. As a matter of fact, it is enough to require the existence of an interior fixed point only for one t > 0 to ensure the existence of such a point for the whole semigroup. Indeed, if for at least one t > 0 the mapping F1 E S has an interior fixed point ( E jj. then it is a unique fixed point for F1 , and for each s 2: 0 we have: i.e., (is also a fixed point of Fs E S, s 2: 0. Henceforth this fixed point is a unique stationary point of S.

Exercise 1. Show that if ( E jj. is a stationary point for a semigroup S = {Ft}t>O• F1 E HoJ(Ij.), t > 0, then (Ft)'(a) = e-at is a contraction linear semigroup, i.e., Rea 2: 0. Hint. Use the chain rule and the Schwarz-Pick lemma. Naturally, the strategy now is to study the convergence of a semigroup to its stationary point. The foregoing result is the first step in the study of the asymptotic behavior of a flow in /j.,

Proposition 4.1.1 ([81]) Let S = {Ft}t>O C Hol(lj.) be a flow on jj., Then this net converges uniformly on compact subsets of jj. to a holomorphic mapping FE Hol(fj., C) if and only if for at least one to the sequence {Ft 0 n};;c'= 0 converges uniformly on compact subsets of jj., Moreover, if F 10 is not the identity then F is a constant with the modulus less or equals to 1. Proof. The necessity is obvious. To prove the sufficiency we assume that Ft is not the identity for t > 0, otherwise the assertion is trivial. Then the limit:

. IIm

n--+oo

r.

rton

1'1m p(n) = n-+oo t0

is a constant mapping, say ( (=((to)) E 1.3.2 that

/j.,

If ( E

jj.

then it follows by Corollary

Consequently, the chain rule and the semigroup property imply that for all

t > 0: (4.1.2)

(see also Exercise 1). Hence (is an attractive fixed point for each Ft. t

. 11m

p(n)(~)- r· t "' - ..,,

zE

> 0, i.e., (4.1.3)

/j.,

11---+00

If (

(=

lim F1~"))

11--+00

l

E f)jj.,

then F10 has no fixed point inside

/j.,

In this situation,

as we mentioned above, each F1. t > 0, must be fixed point free on /j., Then the Denjoy-Wolff Theorem implies that for each integer m > 0 the sequence of iterates

{ Ft~fm} ~=O converges uniformly on compact subset of

~

to a point (m E

~.

ASYMPTOTIC BEHAVIOR

103

But (m = lim p(nm)(z) = lim Ft(n)(z) = (,

n-+oo tofm

n-+oo

0

and it follows that (m = ( does not depend on m. So, in both cases (either ( E ~or ( E 8~) we have the following equality lim Ft(n/) (z) = (

n~oo

(4.1.4)

om

for all z E ~~ and each mEN. Now we will show that (4.1.4) implies that for each z E ~the net Ft converges to (, as t tends to infinity. Indeed, for a given £ > 0 and z E ~ one can choose 8 > 0 such that JFt(z)-Ft(w)l < e/2 for all t > 0, whenever wE~ and lw-zi < o. For such take m E N so large that jF8 (z) - zi < whenever s E [0, t 0 /m]. Finally, for such m and t > 0, large enough settings= [tm/to] we have by (4.1.4):

owe

o

IFtonfm(z)Noting that s such t > 0:

=

Cl = JFt~n)m(z)- cJ

< ~·

t- ton/m E [0, t 0 /m] and setting w = F 8 (z), we obtain for

IFt(z)-

(I

IFton/m(Fs(z))-

cl

<

IFtonfm(Fh(z))- Ftonfm(z)l

+

IFtonfm(z)-

cl : :; ~ + ~ =

c:,

and we have completed the prof. D The result which we have established implies immediately a continuous analog of the Denjoy-Wolff Theorem.

Proposition 4.1.2 ([81]) LetS = {Ft}t>o c Hol(~) be a flow on ~. If for at least one to the mapping Fto is not the identity and is not an elliptic automorphism of~. then the net {Ft}t~o converges to a constant ( E ~ as t ----> oo uniformly on each compact subset of~. Remark 4.1.1 Since every continuous semigroup S = {Ft}t~o of holomorphic self-mappings Ft of~ is differentiable (by parameter t ;::: 0), it is natural to describe 1

its asymptotic behavior in terms of the generator f = lim - (I- Ft). This becomes t-+0 t more desirable when such a semigroup is not given explicitly, but is defined as the solution of the Cauchy problem:

{

8u(t, z) at +f(u(t,z))=O, (4.1.5)

u(O,z) = z We set here Ft(z) = u(t, z).

E ~.

Chapter IV

104

Note also that iff is holomorphic in a neighborhood of the point ( E D. then it follows by the uniqueness of the solution of the Cauchy problem that f(() = 0 if and only if ( is a stationary point of S = {Ft}t~O· In particular, an interior null point of a semi-complete vector field is a stationary point of the generated semigroup. However, this fact is no longer true for a boundary null point.The following example shows that even a semi-complete vector field f has a continuous extension to D.; it may have two null points in D. (one of them on aD.), while the semigroup generated by f has a unique stationary point in D. (which is the interior null point of f). Example 1. Set f(z) = z - 1 + vfl=Z. It is clear that f(O) = f(1) = 0. At the same time, solving the Cauchy problem (4.1.5) one can find the solution explicitly:

u(t,z) = 1- [1-e-t/ 2 +e-t1 2

Jf=Zr

Setting Ft = u(t, ·), it is easy to verify that Ft E Hol(D..), hence complete. But for all t > 0: lim Ft(r·)

r--+1-

= 1- [1- e-t/ 2 ] 2 <

f is semi-

1

and therefore ( = 1 is not a stationary point of S = {Ft}t>O· Nevertheless, as we will see below (see Section 4.6), if

f has no null point in

D. then it must have a boundary null point on aD. which is an asymptotic limit of the semigroup generated by

4.2

f.

Null points of complete vector fields

In this section we deal with a one-parameter group of automorphisms of D... Since the generator of a one-parameter group is a complete \'ector field, it is a polynomial at most of the second order, aucl hence holomorphic in C. Now the assertion follows: Proposition 4.2.1 The stationar-y point set of a one-parameter groupS = {Ft} tEIR' Ft E Aut(D..), has either one or two points in D.. which are exactly the null points of the function f: f(z) =a+ ibz- az 2 , (4.2.1)

ASYMPTOTIC BEHAVIOR where a= _ aFt(O) at

I

and

then z 1

= 1/ z0

+ ibz -

~ a2 Ft (z) I i

Exercise 1. Show directly that if iiz 2

b=

t=o

z 0 =1-

a = 0,

105

ataz

.

t=O,z=O

0 is a solution of the equation b E IR,

a E C,

(4.2.2)

is also a solution of this equation.

Thus equation (4.2.2) has at least one solution in b.. If one of the solutions of abo. the second one (if it exists) lies on abo. also. Moreover, given a complete vector field f one can characterize the group of automorphisms generated by f.

(4.2.2) lies on

Proposition 4.2.2 Let f E aut(b..) be a complete vector field on b. and let S = {Ft}, t E IR be the group of automorphisms generated by f. The following assertions hold: 1) If 21 f (0) I < I!' (0) I then S has a unique stationary point zo in b. which is actually in b. and S consist of elliptic automorphisms of b.. In this case Ft(z) does not converge to zo for each z E b., z =1- z 0 . 2} If 2lf(O)I = lf'(O)I then S has a unique stationary point zo in b. which actually lies on abo. and S consists of parabolic automorphisms of b.. In this case for all z E b.: lim Ft(z) =ZoE abo.. t-oo

3) lf2lf(O)I < lf'(O)I then S has exactly two different null points z1 and z2 in b.. Both of them lie on abo., and S consists of hyperbolic automorphisms of b.. Consider in more details the group of hyperbolic automorphisms. In this case its generator f has the form: f(z) =a+ ibz- az 2 ,

(4.2.3)

2lal > lbl.

(4.2.4)

with a E b., b E IR, such that Using (4.2.4), by direct calculations one obtains that the null points off are

(4.2.5) It is clear that z1 =1- z2 and lz1l = lz2l = 1. Let now S = {Ft}tEIR be the flow generated by f. Since for a fixed to E IR+ the points z1 and z2 are fixed points of Ft 0 , therefore one of them is a sink point for Ft 0 and thus this point is the limit of the net { Ft} when t goes to infinity. It is easy to understand that the second null point is the sink point for the mapping Ft~ 1 = F-to· Therefore it is the limit of the net {Ft} when t---> -oo.

Chapter IV

106 Write Ft in the form:

Ft(z) = ei'Pt z- at ' 1 - atz

(4.2.6)

where at E b., 'Pt E JR., t E R It is easily seen that for each t E JR. and z E b.:

Ft(z)- z1 Ft(z)- Z2

1- atZ2 Z - Zl ·-1- atZl Z - Z2

This equation may be written in the form

L(Ft(z)) = AtL(z),

(4.2.7)

where L is the fractional linear transformation defined by

L(z)= z-z1 Z - Z2

(4.2.8)

and (4.2.9) At the same time, since

Zi,

i = 1, 2 are fixed points of Ft we have by (4.2.6):

It follows that (4.2.10) In addition note, that (4.2.5) implies that

(It is also clear because of the equality zizi = 1, i = 1, 2). Comparing (4.2.9) and (4.2.10) we obtain that:

so the result that At is real is established for all t E R Furthermore, it follows by the group property and (4.2. 7) that

i.e., So At = ekt for some k E R If k > 0 we have again by (4.2.7):

Ft(z) = L - 1 (ekt L(z)),

ASYMPTOTIC BEHAVIOR

107

or (4.2.11) Hence lim Ft(z) t--+-00

= z 1 while

lim Ft(z)

t--+-00

= z2

for all z E

~.

In case of k < 0 we obtain similarly: lim Ft(z) =

z1

t--+-00

and lim Ft(z)

t--+-00

= z2.

Therefore we need to recognize k. Since u(t, z)

= Ft(z)

satisfies the equation

au(t,z) at +f(u(t,z))=O,

with the initial data: u(O,z)

= z,

we obtain by solving this equation: Ft(z)-zl _ e -ii(z,-z 2 }·t ·Z-Zl _:........:.... ___ --. Ft ( z) - Z2 z - Z2

Comparing this formula with (4.2.7) we obtain that k must be -a(z 1 By using (4.2.5) we calculate:

k

-

z2).

=

So k is negative in our case, and we have the second situation. Now we are ready to formulate our result.

Proposition 4.2.3 Let f(z) = az 2 + ib- a such that bE lR and 2lal > lbl. Then f generates a flow of fractional linear transformations:

which converges to z

1 -

-

2a

J 4lal2 -

and to z -

2-

J

2a 4lal2 - b2

b2 - ib'

+ ib'

as t

as t

-->

-->

oo,

-oo.

Chapter IV

108

In addition,

where

k

= -J4Ial2- b2

and

Exercise 2. Let

f

E aut(.6.). Show that the fractional linear transformation

G, defined by the formula:

where c =

j)~j

z-c G(z) = -1- zc'

has the same fixed points as the null points of f.

Exercise 3. Show that if f E aut(.6.) generates a one-parameter group of parabolic automorphisms of .6. then f has the form:

f(z) = T'(z- T)('Fz- 1), where T E 8.6. and T' E IC, ReT' generated by f and show that

=

lim Ft(z)

t--+oo

0. Find explicitly the group S

= {Ft} tEIR

= t___.-oo lim Ft(z) = T.

Exercise 4. Show that iff E aut(.6.) has a null point Tin .6., i.e., f generates a one-parameter group S = {Ft} tEIR of elliptic automorphisms, then Ref' (T) = 0. Show that the group S = {Ft} tEIR can be presented in the form:

where and

Rea= 0.

ASYMPTOTIC BEHAVIOR

4.3

109

Embedding of discrete time group into a continuous flow

A classical problem of analysis is: given a self-mapping F(z) (FE Hol(A)). Find a semigroup {Ft(z)}t>o C Hol(A) with F 1 (z) = F(z). In this case we say that F satisfies the embedding property on A. In general, however, this problem cannot be solved globally even if F E Hol(A) is a fractional linear transformation (see Example 4.3.1).

Example 1 Consider the fractional linear transformation defined as follows:

F(z)= ~ 1- bz

l

with a= exp(i3.1) and b = ~· Since lal+lbl = 1 we have that F is a self-mapping of the unit disk A. Therefore its iterations F(n) map A into itself. Moreover, F(n) converge to zero uniformly on compact subsets of A as n goes to infinity. At the same time the semigroup

atz

Ft

:=

b

1 - -1-a · (1 - at)z

is a continuous extension of the discrete time semigroup {F{n}} because of the equality F = F1. However, for some t > 0 (which is not integer) Ft does not map A into itself. If, for example, we choose z = 0.99 + 0.1i E A then by calculation we obtain that

IF~ (z) I > 1 (see Fig.

4.1).

Nevertheless, by using some geometric methods we will give in Section 5.9 a complete characterization of those fractional linear transformations of the unit disk into inself that satisfy the above embedding property. In this section we show how to solve this problem, when FE Aut(A). Suppose first that F is a hyperbolic automorphism of A, i.e., F has exactly two distinct fixed points ZI and Z2 on the boundary aA. As we saw above (see section 1.1. 3), in this case if F = r 'P • ma, then

F(z) = L- 1 (>-.F(z)), where

A=

~

-

QZ 2 QZI

(4.3.1)

E JR+ \ {0, 1}

and L is defined by

L(z)=z-zl. Z - Z2

(4.3.2)

Chapter IV

110

Figure 4.1: Fractional iterations of the self-mapping F(z). In addition, if z 1 is the sink point ofF, then 0 < A < 1. Denote k

= log A,

a

= __k_, Z2 -

Since lz1l Indeed,

= lz2l = 1, and z1

a k -=--· c

z1-z2

Z2 - Z1 z1·z2k

b=

=I= z2

=

~ z 1 + z 2 , and c = Zl · z 2 t Zz -

Z]

Z2

Z2 -

it is easy to verify that

Z2 - Z1 (z2-zl)z1z2

=

a=

k.

Z1

-c, and bE R

Z2 - Z1 =-1 z1iz2l 2 -z2izll 2 ·

In addition, Reib

i.e., bE R Therefore the mapping

f:

f(z)=az 2 +ibz+c

is a complete vector field. It follows by the Vietta formulas that z1 and z2 are the roots of the equation f(z) = 0. Hence as above (see section 4.2) the group of automorphisms generated by f has a form:

where Lis defined by (4.4.2).

ASYMPTOTIC BEHAVIOR

111

For t = 1 we have:

because of (4.4.1). This gives us the desirable embedding. As a matter of fact, we have proved somewhat more:

Proposition 4.3.1 Let z 1 =/:- z 2 are two arbitrary points on the boundary 8!:1 of the unit disk !:1. Then for each k E JR, k =/:- 0, there is a one-parameter group S = {Ft}, t E JR, such that the following assertions hold: i = 1, 2. (a) Ft(zi) = zi, dFt(z) k kz1+z2 2 (b) - - lt=o= -(az + ibz- a) where a=-.- - , b = --:---. dt Z2- Z1 t Z2 - Zl (c) If k < 0 then z 1 is a sink point of {Ft}t>o; if k > 0 then z2 is a sink point of {Ft}t>O· (d) In particular, if z 1 and F(z)

are fixed points of a hyperbolic automorphism

z2

= ei'f'

z +a 1 + O:z'

a E

!:1

c.p E JR,

'

then for

we have the equality:

F1(z)

= F(z),

z E !:1.

Now we turn to a parabolic automorphism F E Aut(/:1). In this case F has a unique fixed point z0 E 8!:1 which is a sink point of F. Moreover, as we already know F satisfies the equation: z0 -=-:---:-'---

F(z)-zo

ei'f' - 1

z0

e''f'+1

z-zo

= -.-- + - -

(4.3.3)

for some c.p E JR, c.p =/:- 1r + 2n7r (see section 1.3.1). It is clear that in such a situation we must solve the following Cauchy problem:

8u(t,z) {

at

( )2 _ +a z- zo - 0 ' (4.3.4)

u(O, z) = z E !:1, with suitable a E C in order to find a one-parameter group {Ft(·)}, Ft such that: F1(z) = F(z). Solving (4.3.4) we have that Ft satisfies the equation: 1 Ft(z)-zo

-=-:--:---

1 z-zo

= at + - - .

= u(t, ·), (4.3.5)

Chapter N

112

Comparing the latter equality with (4.3.3) and (4.3.5) we obtain:

ei"'- 1 1 a=---·ei'P + 1 zo ·

(4.3.6)

Now we only need to check that the mapping f defined by:

f(z) = a(z- zo) 2

(4.3.7)

is a complete vector field. In fact, substitute (4.3.6) in (4.3.7) we obtain:

f(z) =a z 2 + ibz + c, where 2

c = az0

ei"' - 1 = -. -- · zo = -a-

e'"' + 1

and -2 ei"'-1 41mei"' b=-·-.--=E!R.. i e•'P + 1 jei'P + 1j 2

Thus we have proved the following assertion:

Proposition 4.3.2 Let zo and T =f. -1 be two unimodular points. Then the mapping f defined by (4.3. 7} with T-1 a= --zo r+1 is a generator of the one-parameter groupS= {FthelR E Aut(Ll), such that each Ft is a parabolic automorphism of Ll, and the point zoE 8Ll is a sink point of S. If, in particular, FE Aut(Ll), is a parabolic automorphism of Ll: F(z)=ei"'z+a

1 +za

such that F(zo) = z0 , then forT= eicp we have: F 1 (z) = F(z). Finally, we consider the simplest situation where F E Aut(Ll) is an elliptic automorphism of Ll, i.e., F has a unique fixed point z0 E Ll. In this case F can be presented as: (4.3.8) where

z

mzo(z) is a Mobius transformation of Ll, i.e.,

r"'(z) = ei"'z, Setting in this case

+ zo

= 1 + zoz

mz 0 (0) = z 0 , and r"' is a rotation around zero,

cp E JR.,

k = 0, 1, 2, ....

ASYMPTOTIC BEHAVIOR

113

and (4.3.9) we have that S = {Ft}, t E lR is a one-parameter group of elliptic automorphisms of~ such that Ft(zo) = zo for all t E lR and Ft(z) = F(z), z E ~-

Exercise 1. Prove that group (4.3.9) is generated by the mapping f(z) =

-ic.p I l 2 (z- zo)(1- z 0 z). 1- zo

Exercise 2. Prove directly that the mapping plete vector field.

f: (4.3.10)

f defined by (4.3.10) is a com-

Finally, we formulate the result:

Proposition 4.3.3 Let zo E ~ and c.p E R, c.p = 21rk, k = 0, 1, 2 .... Then the mapping f defined by (4.3.10) generates the one-parameter groupS= {Ft}, t E lR of elliptic automorphisms defined by formula (4.3.9). In particular, ifF E Aut(~) is an elliptic automorphism, defined by (4.3.8), then Ft(z) = F(z), z E ~-

4.4

Rates of convergence of a flow with an interior stationary point

Let f E Hol(~, C) be a semi-complete vector field on ~. and let us assume that Null f n ~ =f:. 0. Actually, it means that if f is not zero identically, then f has exactly one null point in ~ (otherwise the semigroup S = { Ft}t~o generated by f should have more than one stationary point in ~. and a contradiction results). A :standard que::;tion of dynamical ::;y::;tems aualysis is: given a vector field J, describe the asymptotic behavior of the flow defined by the evolution equation dujdt + f(u) = 0, u(O) = z, in a neighborhood of its singular point T (J(T) = 0). In this section we intend to answer this question for a semi-complete vector field, as much as to find global rates of convergence of the generated flow to its interior stationary point. For our purpose we need the following definition.

9 Hol(~) is said to be a strongly semicomplete vector field if it has a unique null point T in ~ which is a uniformly attractive stationary point for the flow S = { Ft}t~o generated by f, that is, the net {Ft}t~o converges toT uniformly on each compact subset of~ as t--+ oo.

Definition 4.4.1 ([40]) A function f E

Chapter IV

114

We begin with establishing a simple assertion which is a (nonlinear) analog of the Lyapunov stability theorem.

Proposition 4.4.1 Let f E Hol(D., C) be a semi-complete vector field with f(r) = 0 for some T E D.. Then: (i) Ref'(r) ~ 0; (ii) Re f'(r) > 0 if and only iff is strongly semi-complete.

Proof. Let S = {Ft}t~o be the flow generated by f. Since Ft(T) = T for all t ~ 0, it follows by the chain rule, that (Ft)'(r) = At satisfies the semigroup property At+s =At· A., Ao = 1, hence At= ekt, with k = f'(r) E C. In addition, it follows by Corollary 1.1.1 that

But this inequality holds if and only if Re k = Re !'( r) ~ 0 and assertion (i) follows. Furthermore, Re k = Ref' (T) > 0 if and only if for all t 2: 0

and then assertion (ii) is a consequence of Proposition 4.1.1. and Corollary 3.1.1. D

Exercise 1. Show that if Ref'(r) = 0 then the semigroup S = {Ft}t~o is actually a group of elliptic automorphisms. Thus, iff E

9 Hol(D.) is not a complete vector field of the type f(z) = (z- r)(l- 'fz)a

with TED. and Rea= 0, then the semigroup S = {Ft}t~o generated by f does not consist of elliptic automorphisms and converges uniformly on each compact subset of D. to a point in D.. If this point belongs to D., then it is a unique uniformly attractive stationary point of the semigroup, and the question of finding a rate of convergence arises.

Remark 4.4.1 We already known that for a holomorphic function fonD. the property of being semi-complete is equivalent to the property of being monotone with respect to the Poincare metric p on the unit disk. Also, it can easily be seen that iff is strongly p-monotone then it is strongly semi-complete. For this case one can obtain a rate of convergence in terms of the hyperbolic metric on D. (see Section 4.5). The converse, however, does not hold in general: a strongly semi-complete vector field is not necessarily strongly p-monotone. Therefore in this section we will establish some rates of convergence in terms of the Euclidean and pseudo-hyperbolic distance on D.. First we consider the case when f(O) = 0, i.e., zero is a stationary point of the semigroup S = {Ft}t>D·

ASYMPTOTIC BEHAVIOR

115

Proposition 4.4.2 (see (58] and (109]) Let f E g Hol(~) be a strongly semicomplete vector field with f(O) =·O and .A= Ref'(O) > 0, and letS= {Ft}t~o be the semigroup generated by f. Then there exists c E [0, 1] such that for all z E ~ and t :2: 0 the following estimates hold: (a) IFt(z)l

~ lzl·exp( -.A~~~:::t).-

IFt(z)l < lzl (b) (1- c1Ft(z)l)2 - exp( -.At) (1- clzl)2.

Proof. The procedure that is to follow is based on general flow invariance conditions established in sections 3.4 and 3.5. We start with the consideration of an auxiliary Cauchy problem: 8u(t,z) {

8t u(O,z)

+p

= z,

. ( )·1-cu(t,z)_ 0 u t, z ( ) - ' 1 + u t, z

zE

(4.4.1)

~-

1-z Noting that the function cx(z) = p · z - - satisfies the condition 1+z

for all z E ~. we conclude that ex E g Hol(~) and the Cauchy problem above has a unique solution u(t, z) E ~ for all z E ~ and t :2: 0. In addition, u(t, 0) = 0 for all t :2: 0, because of cx(O) = 0. Then by the Schwarz Lemma we obtain that lu(t, z)l ~ lzl,

z E ~-

(4.4.2)

= ps 1 -

s, defined on the interval [0, 1] 1+s is real we have by the uniqueness property that the differential equation:

Furthermore, since the function cx(s)

(4.4.3)

with the initial data:

v(O)

=s

(4.4.4)

has a unique solution v(t, s) for all t :2: 0 and s E [0, 1) which coincides with u(t, s) restricted on the real interval [0, 1). Consequently, by (4.4.2) we obtain:

v(t,s) < s for all t :2: 0 and s E [0, 1). Also, it follows from (4.4.3) that: dv 1- v = -p--dt. v l+v

-

(4.4.5)

Chapter IV

116 Hence, by (4.4.4):

v = s exp ( / -p

~ ~ ~ dt) .

Finally, inequality (4.4.5) implies the estimate: (4.4.6) Recall that by Proposition 3.4.5 inequality:

f E QHol(.6.) with f(O) 1-lzl

Ref(z)z 2:: Ref'(O)Izl 2 1 + lzl = a(lzl) ·lzl,

0 satisfies the

z E .6..

Then by Lemma 3.4.1 and (4.4.6) we have: IFt(z)l

~ u(t, lzl) ~ lzl· exp (- Ref'(O) ~ ~

:::t).

This proves (a). To prove (b) we return to (4.4.3) and rewrite it in the form: 1 +v 1 - - · -dv = -pdt

1- v v

(we substitute a(v) = vp 1 + v into this equation). Integrating the latter equality 1-v in [0, t] we obtain: v(s,t) _ e-pt s (1 - v( s, t) )2 (1 - s )2,

[ ) s E 0, 1 .

Again, settings= lzl and using the relation

IF(t, lzl)l ~ v(s, t), we obtain (b). D Note that estimate (a) with c = 1 is due to K. Gurganus, whilst estimate (b) was established by T. Poreda. Now we are prepared to consider a more general case, when f E 9 Hol(.6.), with f( r) = 0 for some T E .6., which is not necessarily zero.

Proposition 4.4.3 Let .6. equipped with the pseudo-hyperbolic distance

d(z,w)

w-z I , = I--_1-wz

and let Sf = {Ft }t~o be the semigroup generated by f E E .6. the following conditions are equivalent:

T

9 Hol( .6.).

Then for some

ASYMPTOTIC BEHAVIOR ~

{i) f(r) = 0 with Ref'(r) {ii} for some c E [0, 1]:

d (Ft(z), r)

p

~

117

0;

~ d(z, r) exp ( -p ~: ~~~:: ~~ t)

for all t ~ 0 and z E !::.. In addition, iff is bounded and strongly p-monotone then the number c in {ii} can be chosen as strictly less than 1.

Proof. Define the Mobius transformation Mr : !::. T-Z

Mr(z) = -1 _ , -TZ

z

E

t---4

!::. by:

!::..

(4.4.7)

It is clear that (4.4.8) and it follows that Mr is an involution, i.e., (4.4.9) Consider the family

{Gt}t~o

defined as follows: {4.4.10)

It is obvious that {Gt} is also a semigroup of holomorphic self-mappings of !::., and for all t ~ 0 we obtain {4.4.11) Gt(O) = 0, t ~ 0, since Ft(r) = r, for all t ~ 0. Let g be the generator of the semigroup {Gt}t>o , i.e.,

g(z) =

-dGt(z) dt

(4.4.12)

lt=O+ •

By direct calculations we obtain:

d

dMr

dFt

dt (Mr(Ft(Mr(z)))) = ----;J:;- (Ft(Mr(z))} · dt(Mr(z))

=

d~r (Ft(Mr(z))) · (- f(Ft(Mr(z))),

z

E

!::..

Hence by (4.4.12)

g(z) =

d~r (Mr(z)) · f(Mr(z)),

z

E

!::..

(4.4.13)

Note that the latter formula expresses the direct connection between generators

f and g. Also, (4.4.8) and (4.4.13) imply: g(O)

= 0,

(4.4.14}

Chapter IV

118

which is equivalent to (4.4.11). In turn, differentiating (4.4.13) we calculate 1

9 (z) =

d 2Mr ( dMr ) dz 2 (Mr z)) · dz(z · J(Mr(z))

dMr df dMr ( +dz(Mr(z)) · dz (Mr(z)) · d z z).

(4.4.15)

Substituting here z = 0 and using that f(r) = 0 we obtain: g'(O) = dMr (r). df (r). dMr (O). dz dz dz

(4.4.16)

Since dMr d ( T- z ) lrl 2 - 1 d z = dz 1- TZ = (1- Tz)2

relation (4.4.16) becomes: g'(O) =

It 1/ -1 f'(r) · (lrl 2 -

1)

= f'(r).

(4.4.17)

So by Proposition 4.4.2 (a) we obtain:

IGt(z)l

~ lzl exp (- Reg'(O) · ~ ~ :~:

·

t),

and using (4.4.10) and (4.4.17) we obtain from the latter inequality:

I(Mr o Ft

o M r) ( z) I

For arbitrary w, setting I(Mr o Ft)(w)l ~

~ Iz Iexp (- Re f' (T) · ~ ~ :~:

z =

·

t) .

(4.4.18)

Mr(w) in (4.4.18) we obtain:

IMr(w)l exp

(

-Ref

1

1 -IMr(w)l ) (r) · 1 + IMr(w)l · t ·

(4.4.19)

-I r-~: I t) '

(4.4.20)

Rewriting (4.4.19) in the form:

I ( - Re !'(

Ft(w) < T -w exp 1-Ft(w)r- 1--rw T-

T

)· 1 1

+1r---r:l

·

we obtain (ii). Conversely, assuming that (ii) holds we first observe that f(r) should be zero. Of course, this is an immediate consequence of (ii) if p > 0, hence Ref'(r) > 0. Indeed, in this case lim Ft (z) = r, t-oo

so, Tis the stationary point of St = {Ft}t~o, whence, a null point of f. But even if Re/'(0) = 0, condition (ii) means that each pseudo-hyperbolic ball f2r(7) (see Section 1.3) is F 1-invariant for all t ~ 0, and the same conclusion follows.

ASYMPTOTIC BEHAVIOR

t

=

119

To arrive at the second condition of (i) we just need to differentiate (ii) at o+ to obtain:

Letting z -+ T we obtain immediately that Re f' (T) ~ p as desired. Finely, note that our last assertion follows from the strong Harnack inequality (see Section 3.5 and Proposition 3.6.2). D Remark 4.4.2 Since IFt(w)l < 1 one may obtain from (4.4.20) (setting c = 1) the following estimate

IFt(w)- rl

~

1 + lrl ( , 11 - rwl -lr- wl ) jw- rl1 -lrllwl exp -Ref (r). 11- rwl + lr- wl t .

(4.4.21)

This formula gives an estimated rate of convergence of the semigroup { Ft}t>O to its stationary point r in the usual Euclidean metric in C. As a matter of fact, this estimate can be slightly improved by using the following considerations. Return to formula (4.4.19). Denoting Ft(w) = J and Mr(J) = u we have Mr(u) = J and Mr(O) = T. Now, it follows by Corollary 1.1.3 of the Schwartz Lemma (see section 1.1.1), that:

or

IFt(w)- rl

< lr-wl - 11-'fwj

(4.4.22)

It is clear that (4.4.22) implies (4.4.21 ). Remark 4.4.3 Note that the above proposition may be considered as an analogy of the Schwarz-Pick inequality. Moreover, for the continuous case this inequality has a more precise form because of the exponential factor 1- d(z, r) ) exp ( -p1+d(z,r)t ·

However, the property of this factor depending also on z E ~ is sometimes inconvenient for applications. In some situations it would be preferable to find a metric on ~ in which such a factor becomes uniform. In addition, as in the discrete time case, condition (ii) in our proposition cannot be useful in the study

Chapter IV

120

of the boundary asymptotic behavior of flows with no stationary point inside ~. Therefore we will treat the last question separately in Section 4.6 by using a continuous extension of the Julia-Wolff-Caratheodory Theorem. However, it turns out that one can define a non-Euclidean distance on ~ (which, in fact, is not a metric on ~' but induces the original topology of ~) such that the rates of convergence for interior and boundary points have some unified form. This makes it possible to study the dynamic behavior of evolution equations when their stationary points approach the boundary. Moreover, we will show in Section 4.6 that in terms of such a distance, the question about a uniform rate of convergence of the exponential type may be solvable too. Finally, we note that a deficiency of the rates of convergence given by formulas (4.4.20) and (4.4.21) is the presence there the value of the derivative f' at the null point T of a generator f. In fact, if f has a complicated form even the existence of its null point is unclear a priori. Therefore it would be nice to establish a rate of convergence by using another condition which will guarantee also the existence of an interior null point of a semi-complete vector field. The next section is devoted to this matter.

Remark 4.4.4 From the geometrical function theory point of view there is a need to point out a lower bound estimate of the flow behavior. Exactly as in Proposition 4.4.2, by using the left side inequality of (3.5.18) and Lemma 3.5.1 one again obtains: if f(O) = 0 and Ref'(O) =A 2:0, then for some c E [0, 1] (a') IFt(z)l 2: lzl exp( -A 1 + c:z: t); 1- c z

,

IFt(z)l

lzl

(b) (1-c1Ft(z)l) 2 2:exp(-At) (1+clzl) 2 ' Moreover, iff is bounded strongly p-monotone then c can be chosen strictly less then 1.

4.5

A rate of convergence in terms of the Poincare metric

In this section we will give several sufficient conditions for f E Hol(~, q to be strongly p-monotone, hence strongly semi-complete on the open unit disk ~ and obtain rates of convergence for the semigroups generated by such functions in terms of the Poincare hyperbolic metric p on~.

ASYMPTOTIC BEHAVIOR

Proposition 4.5.1 ([40]} Let

f

121

E Hol{A, C) satisfy the condition:

Ref(z)z

a(Jzi)izl,

;::=:

z E

A

(4.5.1)

for some real continuous function a( l) on the interval [0, 1], such that:

a{1) =a> 0.

{4.5.2)

Then f is strongly p-monotone, hence a strongly semi-complete vector field on T E A which is uniformly attractive for the semigroup S = { Ft}, t ;::=: 0, generated by f. Moreover, for each pair z, w E A

A. Thus f has a unique null point

at

~ exp { -

p( F1 ( z), F1 ( w))

2} p( z, w).

(4.5.3)

}p(z, w).

{4.5.4)

In particular,

p(F1(z), T)

~

exp { -

Proof. Consider the mapping g. : A

g8 (z) = z

C defined as follows:

f-+

+ sf(z)- w,

at

2

where

w E A, s

Let Ar be the disk centered at zero with radius r z E oAr= {z E C: lzl = r} we have by {4.5.1):

Reg.(z)z

izl 2 r (r

0

;::=:

~

0.

r < 1. For all

+ s Re f(z)z- Rewz ;::=: r 2 + sa(r)r- 1·Jwl + sa(r) -lwl).

(4.5.5)

Since a{1) > 0 it follows that for s > 0 small enough the equation:

cp 8 (r) = r

+ sa(r)

= 1

has a solution r 8 E {0, 1). In fact, cp 8 (0) = sa(O) ~ 1 for 0 < s < 1/a{O) and cp 8 {1) = 1 + sa{1) follows by (4.5.5) that for such fixed n and z E OArn,

Reg.(z) · i

;::=:

(4.5.6)

;::=:

1. It

0.

The latter inequality implies that the equation:

g.(z) = z

+ sf(z)- w = 0

has a unique solution z = J.(w) :=(I+ sf)- 1 (w) EAr. for each wE A. In other words, the resolvent mapping J8 maps A into Ar,· Thi~ mean~ that ] 8 has an interior fixed point T E A which is also a unique null point of f in A. Furthermore, we consider the function h(z) = h8 (z) defined as follows:

hs(z) = Js(z)

a( rs)

+ s -2-

{Js{z)- Js(w)),

Chapter IV

122

where w is an element We obtain:

of~.

and r 8 is the solution of (4.5.6).

zE ~

~-

Since h 8 (w) = J.(w) E ~ it follows by the maximum principle that hs maps into itself. Therefore by Corollary 1.1.1 lh~(w)l

<

1- lhs(w)l 2 for all z E But

1

1-

-

lwl 2

~­

[Js(z) ]' . h,8 (z) = ( 1 + S -o:(rn)) 2Hence

j[J.(w)]'j < 1-

1Js(w)l

2 -

1 1 + s a:(r.)

1 1-

2

lwl 2 .

In other words, the mapping J. is a strict contraction with respect to the

ldwl

.

.

infinitesimal Poincare metric dpw = 1 -lwl 2 , w E ~. defined m Sectwn 2.2. Integrating the above inequality we obtain the same conclusion for the hyperbolic metric p: 1 (4.5.7) p (J8 (z), J 8 (w)) ~ ~ p(z, w). 1 + s 0: ; · Since o:(r) is bounded on the interval [0, 1] it follows by (4.5.6) that r 8 -4 1 as s -4 0 and o:( r s) -4 o: > 0 because of its continuity. Therefore there exist some positive 8 and f such that for all s E (0, 8) we have o:(rs) ?: f > 0. Then for such s we obtain the inequality

p (J.(z), J 8 (w))

~

1 - - p(z, w), 1 +Sf

which means that f is strongly p-monotone. Now settings= t/n in (4.5.7) and using the exponential formula:

Ft(z) = lim J(n)(z) n--->oo tfn we obtain by induction: lim

The proposition is proved.

0

1

k . o(;n))

h--->oo (

1+

exp {-

o:t 2 } p(z, w).

n

p(z,w)

ASYMPTOTIC BEHAVIOR Example 1. Let

123

f E Hol(~, q be defined by 1- cz f(z) =a- az 2 + bz - - , 1 + cz

where a, b E
> 0, and 0 :::; c < 1. a(s)

=

If we introduce the function

-lal(l- s 2 )

1 - cs

+ (Reb)s--, 1 + cs

then we obtain

Reg(z)z;;::: a(izl)izl and

1-c a(l) =Reb--> 0. l+c Hence f(z) is a strongly semi-complete vector field on~Remark 4.5.1 Note that iff E Hol(~, q is known to be a semi-complete vector field on~' then condition (4.5.2) can be replaced by a slightly more general condition, namely, a(l) > 0 for some l E (0, 1], which still ensures f to be strongly semi-complete. The above arguments can be employed in the disk ~~- This note leads us to the following simple sufficient condition yielding the existence of an interior null point of a semi-complete vector field and its attractiveness. Corollary 4.5.1 (cf., EM-RS-SD-2000c) Let

f E QHol(~) be such that:

Re !' (0) > 41/(0) I· Then f has a unique null point verges to r as t goes to infinity.

T

(4.5.8)

E ~ and the semigroup Sf = { Ft}t2':D con-

Proof. Consider the function: a(r) = -a(1- r 2 )

1-r

+ br--, 1+r

where a= l/(0)1, b = Ref'(O). It follows by Proposition 3.5.3 that

Re f(z)z;;::: a(izi)izl,

zE

~-

Then we have under condition (4.5.8), that a(O) = -a < 0, a'(l) < 0. Hence, there islE (0, 1), such that a(l) > 0. 0

(4.5.9) a(1) = 0, and

Remark 4.5.2 Note that conditions (4.5.8) and (4.5.9) imply that for some ~r with r E (1- 8, r] is invariant for the semigroup {Ft} generated by f. However these conditions are not sufficient to ensure the validity of the strong p-monotonicity of f.

8 > 0 each disk

Chapter IV

124

Remark 4.5.3 Observe also that the above Proposition 4.5.1, Lemma 3.6.1, and Proposition 3.6.2 imply that the property off E Hol(~, C) being strongly p-monotone is equivalent to the strong flow invariance condition

Ref(z)z

~ Tf

>0

for all z in the annulus {1- 8 < lzl < 1} for some 8 > 0. In terms of the Poincare metric this property is equivalent to the global uniform exponential convergence (see formula (4.5.3)) of the semigroup {Ft}t~o generated by f on all of ~, whilst the property of f to be strongly semi-complete is equivalent to local uniform exponential convergence on each compact subset of~ (see Proposition 4.4.3).

4.6

Continuous version of the Julia-Wolff-Caratheodory Theorem

Let now f E Hol(~) be a null point free mapping. We already know that in this case the semigroup Sf = {Ft}t>o generated by f is convergent (see Section 4.3). Moreover, there is a point T E 8 ~ (a sink point of Sf), such that for each z E ~, lim Ft(z) = T. However, the study of rates of convergence in this case is more t--+00

complicated than in the case when f has a null point in ~- Indeed, as we saw above, in the latter case the asymptotic behavior of the semigroup generated by f is completely determined by the value of f'(r). Namely, {Ft}t>O is convergent if and only if Ref'(r) > 0. Therefore it is natural that the rates of convergence obtained in Section 4.4 are connected with this value. Iff E QHol(~) is null point free, i.e., Sf= {Ft}t~O has a sink point T E 8~, one cannot use the same approach, since !' (T) is not defined in general. In addition, the above approach of using a Mobius transformation is also unfeasible in this situation. Therefore we need another method to study the asymptotic behavior of the semigroup generated by a null point free holomorphic function. A complete characterization of convergence of a semigroup to its sink point can be done by using the so called nontangential derivative of its generator in the spirit of the Julia-Wolff-Caratheodory Theorem (Proposition 1.4.2). Here we establish a continuous version of this result (Proposition 4.6.2) which will provide another proof of the classical one. Note in passing that if F E Hol(~) is a holomorphic self-mapping of~, then f =I-F is semi-complete. We first prove some auxiliary assertions.

ASYMPTOTIC BEHAVIOR

Hol(~, C)

Lemma 4.6.1 Let f E

(finitely) then: (i) L lim f(z) z-.e

(ii)

and let e E

125

8~.

If (3 := L lim f(z) exists z-e

z- e

= 0;

the angular limit L lim

z-.e

f' (z) also exists and equals to (3.

Proof. (i) is trivial. To prove (ii) let us assume that e the form f(z) = (3(z -1) + h(z).

= 1 and present f in (4.6.1)

Then we have

Llimh(z)=O Z-->1

and L lim h(z) = 0. Z-+1 Z -

(4.6.2)

1

We wish to show that

L lim h'(z) = 0.

(4.6.3)

z-+1

To this end take two sectors SandS in~ both with vertex at e = 1, so that S C S. For z E S we denote by f(z) the circle with its center at z, such that f(z) is tangent to the boundary of S. By(} we denote the angle between segments [0, 1] and [z, 1] and let 2Bs and 2Bs

be the angles of the sectors S and Then . sm(Bs- Bs)

S, ~

respectively, at vertex e = 1 (see Fig. 4.2).

. sm(B_s;- B)

=-

r(z)

(4.6.4)

--, 1

1-z

1

where r(z) is the radius of the circle f(z). Now observe that when z converges to 1 in the sector S, all points w E 8f(z) converge to 1 in the sector S. Therefore using (4.6.2) we obtain that for each e > 0 there is a point z close to 1 such that

lh(w)l < e, lw -11

wE f(z).

It then follows by the Cauchy formula and (4.6.4) that

h'(z)

=

_1

2rri

J

h(z) dw < .!...._ (w- z)2 - 2rr

r(z)

<

.!...._ max lw 2rr wEr(z)

-11

J

J

lw- 11 ldwl lw- zl2

r(z)

ldwl = _e_ max lw lw- zl 2 r(z) wEr(z)

r(z)

rtz) (r(z) + 11- zl) = e (1 + llr~)zl) <

e(

1 +sin(B 1-Bs))· 8

-11

126

Chapter N

1

Figure 4.2: The circle f(z) and the sectors SandS. The latter inequality concludes the proof of the Lemma. 0 The value (3 := L.lim f(z) is called the angular derivative off ate E 8tl. z->e

z- e

We say that a function the limits

f

E Hol(tl, C) has the radial derivative at e E 8tl , if

a:= lim f(re) T--+

1-

and

. f(re)-a l lm r---->1-

(r-l)e

exist. We will denote it by j f'(e). It is clear that iff has the angular derivative at a point e E 8tl then it has a radial derivative at this point and

i f'(e) = L.f'(e). We will show that for a null point free semi-complete vector field a somewhat converse statement is also true at the sink point of the flow generated by f. First we establish the following assertion. Lemma 4.6.2 Let f be a semi-complete vector field on tl. Suppose that for a

ASYMPTOTIC BEHAVIOR

127

point e E 8b. there exists the radial limit derivative (3

-

-

. l liD

r-+1-

f(re)

(r -1)e,

such that

Re (3 2: 0.

(4.6.5)

Then f has no null point in b.. Moreover, the point e must be the sink point of the semigroup generated by f.

Proof. Without loss of generality let as assume that e the Berkson-Porta representation

= 1,

and write

f(z) = (z- r)(1- zf)p(z),

f (z) by (4.6.6)

where Rep( z) 2: 0 everywhere, and T is a point of b.. We wish to show that T in (4.6.6) is equal to 1. First we assume that Re(3>0 and conversely suppose that Re(J

= Re

T

-=f. 1. Then we have by (4.6.6) and (4.6. 7)

/(r))

( lim r-+I-

(4.6.7)

Re [lim (r- r)(1- rt) p(r)] r-+Ir- 1

r- 1

[1- 7[ 2 Re [lim Rep(r)] > 0. r-+I- r - 1 On the other hand for all r E (0, 1) p(r) 1 Re - - = - - Rep(r):::; 0, r-1

r-1

which is a contradiction. To complete our proof for the general case we consider the mapping fe : b. f-> C defined by (4.6.8) fe(z) = f(z) + c:(z 2 - 1) with c: > 0. Since c:( z 2 complete and

-

1) is a complete vector field, the vector field /" is semiRe [T /~(1)]

= Re(3 + 2c: > o.

(4.6.9)

Therefore by the previous step, for each c: > 0 the mapping fe has no null points in b.. Now, if Tin (4.6.6) belongs to b., that is, f has an interior null point, then by the Rouchet theorem it follows that one can choose a small enough E: such that !e has an interior null point T 0 close to T. Once again a contradiction. Moreover, (4.6.9) and the step proved above show that for each E: > 0 the point 1 must be a sink point off". Thus there are functions Pe E Hol(b., C) with Re Pe 2: 0, z E b., such that

Chapter IV

128

Comparing the latter formula with (4.6.8) we obtain

( ) _ (z- 7)(1 - z'F)p(z) Pc: z - (1 - z )2

z +1 +e1 - z.

Since Re Pc: ;:::: 0 for all e > 0 and Pc: (z) converges to Po ( z)

( ) _ (z- 7)(1- z'F)p(z) paz-(1-z)2 , of~

uniformly on each compact subset

as e goes to 0, we obtain that

Repo;:::: 0. But

-(1- z) 2 po(z)

= (z- 7)(1- z'F)p(z) = f(z),

hence contradicting the uniqueness of the Berkson-Porta representation. Thus po(z) = p(z) and 7 = 1. D

Proposition 4.6.1 If for a point e E

i

8~

lim f'(z) = (3

z-e

the radial limits

and

i

lim f(z) = 0

z-e

exist with Re f3 ;:::: 0, then

0::; (3 and e =

7

=

Lf'(e)

is a sink point of the flow generated by f.

Proof. Indeed, since lf'(re)l < M < oo we obtain

J 1

lf(re)l =

f'(te) dt ::; M(1- r).

r

Hence lim inf ( f(re~ exists. Let e r-->1r- 1e

= 1 and u(z) = Ref(z).

Then we obtain

J 1

u(r) = -

u'(t)dt = u'(ry)(r- 1),

r

where 17 E [r, 1). Hence lim ( u(r)) = Re(J;:::: 0. D r- 1

r-->1-

To establish the converse assertion to Lemma 4.6.2 we will use the RieszHerglotz integral representation of functions of the class P = {p E Hol(~. C) : Rep(z);:::: O}(see Section 0.2):

p(z) =

J+

1 z( 1- z(

- - - dJ-Lp(()

at::.

.

+ tlmp(O),

(4.6.10)

ASYMPTOTIC BEHAVIOR

129

where f..Lp : 8/:l. ~---+ lR (the measure characteristic function for p E P) is a positive increasing finite function f..Lp on the unit circle 8/:l., such that

J

df..Lp(() = Rep(O).

at:. Lemma 4.6.3 Let f E Hol(b., C) be a semi-complete vector field on b. with no null point in b.. Then there is a point T E 8/:l., such that f has an angular derivative at T. Moreover, if T is a sink point of the flow generated by f then Lf' (T) exists and it is a positive real number which is equal to 2t-Lp( T), where

pz)= f(z) ( (z- T)(1 - ZT) Proof. Of course, it is enough to prove only the second assertion of the Lemma. Let us again present f by the Berkson-Porta formula

j(z) = (z- T)(1- zr)p(z), where Rep(z);::: 0,

-(f(z) --) = z- T

z E b.. Then by (4.6.10) one can write

J at:.

1 + z( (1 - zr) - - - df..Lp( () - (1 - zr)i Im T f(O). 1- z(

(4.6.11)

Let Zn E b. be a sequence of points which converges to T nontangentially. Then if we write Zn in the form Z11 = T(Xn + iyn), we can find 0 < K < oo, such that

IYnl

~

(4.6.12)

K(l- Xn)

(see Fig. 4.3). Further, consider the sequence of functions g11 : 8/:l.

~---+

C defined by

gn := gn(() = (1- TZn)(1 + Zn(). 1- Z 11 (

To estimate 9n on the unit circle we calculate

or Zn( (1 - TZn

This implies that

+ 9n) = gn -

(1 - Tz,.).

( 4.6.13)

Chapter IV

130

Figure 4.3: The nontangential convergence to or

Consequently, _ I

9n

(1- fzn)(l + lznl 2) 12 1 -lznl2

Finally,

I_

(1- fzn)(l

9n

+ lznl 2)

1 -lznl2

1--

21 Zn

111- fznl

1 -lznl2'

and we obtain

lwn(()l < 4 ll- fznl = 4 J(l- Xn)2 + 1 - lznl 2 (1 - Xn) 2 y~ Taking (4.6.12) into account we obtain

Since for n large enough 1 + :rn

1

-IYniK > 2'

y;.

T.

ASYMPTOTIC BEHAVIOR

131

we have for such n

l9n(()l ~ 8V1 + K 2 • In addition, it follows by (4.6.13) that

9n(()

--->

0,

(

#- T.

Then by the Lebesgue Theorem (see, for example, [125]) and (4.6.11) we obtain

. l lm Zn~T

f(zn)

-Zn- T

J(

1 - fzn)( 1 + Zn() dJ-Lp(()- i Im T f(O) · lim (11- Zn( Zn-+T

Znf)

OA

2J-Lp(r), and the Lemma is proved. D Our next goal is to establish relations between angular derivatives of semicomplete vector fields and the asymptotic behavior of their generated semigroups. Lemma 4.6.4 Let F E Hol(D., D.) be a holomorphic self-mapping of D. with no null point in D., and letT E fJD. be its sink point. If Zn converges toT nontangentially, then so does the sequence F(zn)· Proof. 1.4.2) that

It follows by the Julia-Wolff-Caratheodory Theorem (Proposition

0 ~a= L.F'(r) ~ 1 and 'Pr (F(z)) ~ O:'Pr ( z ),

(4.6.14)

where

If z E r(r, k) = {z E D. : 1.2.1) then (4.6.14) implies

lz- rl < k(1- lzl)}

IF(z)-ri <

'---'--c':=-:--:-:-

1 -IF(z)l

-

iz-ri

for some k

iz-ri IF(z)- rl

0 : - - . .,...::::-.,.---,-----'----,-

1 -lzl

1+

> 1 (see Definition

IF(z)l

1 + lzl

(4.6.15)

In turn, The Julia-Caratheodory Theorem states that L.lim F(z)-r =a>O. z,-+T

Z-T

Therefore the third factor in (4.6.15) is bounded. Thus we can find k 1 such that F(z) E f(r, k!) and we are done. D

~

k,

Finally we are able to formulate our main assertion of this section which is a dynamical analog of the Julia-Wolff-Caratheodory Theorem.

132

Chapter IV

Proposition 4.6.2 (see [42]) Let f E QHol(D.,C) be a semi-complete vector field, and let S = {Ft}t~o be the semigroup generated by f. The following are equivalent: (i) f has no null point in D.; (ii) f admits the representation f(z)

=

-'f(z- r) 2 p(z)

for some T E aD. and Rep(z) ~ 0 everywhere; (iii) there is a point T E aD., such that

i J'(r) = f3 exists and Re f3 ~ 0; (iv) there is a point

T

E aD., such that

Lj'(r) = f3 exists andRe f3 ~ 0; (v) there is a point

T

E aD., such that the following limits exist

L lim j'(z) = f3

and

Z~T

with Ref3 ~ 0; (vi) there are a point

T

L lim f(z) = 0 Z~T

E aD. and a real positive number"(, such that

IFt(z)- r[ 2 < e -t-r [z- r[ 2 . 1 -1Ft(z)[ 2 1 -[z[ 2

.:....__...:...,..:,=-:-...,...:..,.

Moreover, (a) the points rEaD. in (ii)-(vi) and the numbers f3 in (iii)-(v) are the same; (b) f3 is, in fact, a nonnegative real number which is the maximum of all "f ~ 0 which satisfy (vi).

Proof. Equivalence of (i)-(iv) has been proved in Lemmata 4.6.2-4.6.3, while equivalence (iv) and (v) follows from Lemma 4.6.1 and Proposition 4.6.1. Thus it is enough to show that (v)==>(vi) and (vi) implies one of the conditions (i)-(v). Suppose that (v) holds. We already know that f has no null point in D. and e = T is the sink point of its generated semigroup. In addition, if lr E Hol(D.), r ~ 0, is a resolvent off: lr(z) +rf(Jr(z)) = z, zED., (4.6.16) then T is also the sink point for Jr, r > 0. It follows by Lemma 4.6.4 that if z converges nontangentially tor, then so does lr(z) for each r ~ 0. Hence (4.6.17) L lim f'(Jr(z)) = (3. Z-->T

On the other hand, it follows by Julia's Lemma that for each r > 0 there is a number O:r, 0 < O:r :::; 1, such that L lim J;(z) z-->r

= O:r

(4.6.18)

ASYMPTOTIC BEHAVIOR

and

llr(z)- rl 2 1 -llr(z)l 2

133

iz- rl 2 ~ O!r 1 -lzl 2 ·

(4.6.19)

At the same time, differentiating (4.6.16) we obtain

or

J;(z) = 1 + r/(Jr(z)) Using (4.6.17) and (4.6.18) we obtain (4.6.20) Substituting (4.6.20) into (4.6.19) and applying the exponential formula we obtain

J~~ 'Pr(Jt(/l(z)) <

lim (

n-+oo

1t 1 + r;:/3

r 'Pr(z)

= exp { -t,B}cpr(z),

(4.6.21)

where (4.6.22) Thus the implication (v)=>(vi) is proved. Obviously (vi)=>(i), because in this case the semigroup {Ft} has no stationary point in b.. Then it remains to prove assertion (b). In other words, we wish to show that if (vi) holds with some"' 2: 0 then"'~ /3. Indeed, consider the real valued function

'!f;(t, z) = 'Pr(Ft(z)), where 'Pr is defined in (4.6.22). Since by (vi) 'lj;( t, z)

(4.6.23) ~

e-'"Yt'lj;(O, z) we have

. 'lj;(t,z)-'lj;(O,z) l liD ----'-----'-----'-----'t--+0+

<

. 1liD

t-o+

t 'lj;(O, z)(e-'"Yt- 1) __ ·'·(O ) '"Y'I' ,z. t

(4.6.24)

On the other hand, differentiating (4.6.23) directly and using

{)~;z) we obtain

~~

lt=o+=

lt=o+=-f(z)

-2'1j;(O,z) · Ref(z)z*,

(4.6.25)

Chapter IV

134 where z

*

=

z 1 -lzl 2

- - f- . 1- zf

Comparing (4.6.24) with (4.6.25) we obtain

'"'(

~

2Ref(z)z*.

Let us again suppose for simplicity that hand side of (4.6.26) in the form:

T

= 1.

(4.6.26) Then we may rewrite the right

1) = 2Re (1-f(z)(z-1) z)(l-Jzl2)"

* ( z 2Ref(z)z = 2Ref(z) 1-jzj2- 1- z Setting here z = r and letting r

'"'( < 2 lim Re -

r->1-

~

1- we obtain by (4.6.26):

f(r) = lim __jJ!J__ = {3. (r- 1)(r + 1) r-+1- (r- 1)

The proof is completed. 0 This result will play a crucial role in our study of the spirallike and starlike functions with respect to a boundary point (see Chapter 5). At the end of this section we consider some examples and a consequence of the above proposition which is an extension of the Julia-Wolff-Caratheodory Theorem.

Example 1. Consider a semi-complete vector field follows zn + 1 f(z) = (1- z) 2 · - - . zn -1

f

E Hol(~, C) defined as

Computations show that f'(1) = 2/n > 0, while Ref'(zk) < 0, where z;: = -1, k = 1, 2, ... , n. Hence if {Ft}t~o is the semigroup generated by f, then the point T = 1 is an attractive point of this semigroup and the following rate of exponential convergence holds

for all z E ~ and t :::: 0. As we mentioned above, ifF is a self-mapping of~ then the function f(z) = z- F(z) defines a semi-complete vector field on~- As a matter of fact, this fact holds even if F E Hol(~, C) is not necessarily a self-mapping of ~, but satisfies the following one-sided estimate: lim F(rz)z

r--+ 1-

~

1 for all z E

8~.

(4.6.27)

Thus we have the following version of the Julia-Wolff-Caratheodory Theorem:

ASYMPTOTIC BEHAVIOR

135

Corollary 4.6.1 Let F E Hol(~, ~) satisfy (4.6.27) and z E following statements are equivalent:

a~.

Then the

{i) F has no fixed point in~; {ii) for some w E a~ there exists the angular limit L lim Z-+W

F(z)- w Z- W

:=a

with Rea=:::; 1; {iii) F admits the representation F(z) = z + (z- w) 2 wp(z) for some w E a~ andp E Hol(~,C) with Rep(z);:::: 0. Moreover, a in (ii) is actually a real number and the boundary points w in {ii) and (iii) are the same.

4. 7

Lower bounds for p-monotone functions

We have seen in previous sections that the asymptotic behavior of semigroups generated by holomorphic functions can be described in terms of their derivatives. If f E g Hol(~) and T E Null(!) then T is (globally) attractive if and only f'(r), the derivative off at T, lies strictly in the right half-plane. Moreover, f E QHol(~) has no null point in~ if and only if for some TEa~ the angular derivative L/'(r) = {3 exists (finitely) with {3 ~ 0. In addition, if S = {Ft}t;;::o is the flow generated by

f then

and the point TEa~ is a (globally) attractive sink point of 8 (even if {3 = 0). However, if Lf'(r) = {3 = 0 the latter formula does not help to establish a rate of convergence of the semigroup to its sink point. Note also that if f E QNp(~) is not holomorphic the characteristics of the derivatives are not relevant. Actually, we will show that for f E QHol(~) the number {3 = Lf'(r) is equal to (4.7.1) inf 2Ref(z)z*, zELl.

136

Chapter IV

where z

*

=

Z

1 -lzl 2

- -T 1- ZT

It turns out that even iff E (}Np(Ll) is not holomorphic, expression (4.7.1) can serve as a characterization of the asymptotic behavior of flows of p-nonexpansive mappings in Ll both in the cases of an interior stationary point and a boundary sink point. In this section we will mostly follow the material of [39]. For a fixed T E Ll, the closure of Ll, and an arbitrary z E Ll, we define a non-Euclidean 'distance' between z to T by the formula:

II- zrl2 Il 1- z 2

d,.(z) =

(4.7.2)

(1- a(z, r)),

where

(see Section 2.3).

Exercise 1. Show that the sets

s > 0,

E(r, s) = { z ELl: d,.(z) < s }, have the following geometric interpretation: (a) If TELl, then these sets are exactly the p-balls

E(r,s) centered at (b) If T

r E E

= B(r,r) = { z ELl: p(z,r) <

Ll and of radius

r

= tanh- 1

J

r}

I l2 .

8

s+ 1- r 8Ll, the boundary of Ll, then these sets

E(r,s)

= D(r,s)

=

{ z ELl:

d,.(z) =

IIzrl 2 < s } , l-lzl 2

s

> 0,

are horocycles in Ll which are internally tangent to the unit circle 8Ll at

T.

Now for fixed TELl and z E 8E(r,s) = { z ELl: d,.(z) = s }, s > 0, z =/:-

T,

consider the nonzero vector *

1

z = 1- a(z, r)

(

1-

1

lzl 2

1

z- 1 - zf

7

) ·

(4. 7.3)

Exercise 2 ([6]). Show that z* is a so called support functional of the smooth convex set E(r, s) at the point z E Ll, d,.(z) = s, i.e., for all w E E(r, s) the following inequality holds:

Re(wz*)

~

Re(zz*).

ASYMPTOTIC BEHAVIOR

137

In order to classify the asymptotic behavior of a flow generated by f E QNp(D.) for a point r E D. we consider two real nonnegative functions on (0, oo ):

wt>(s) :=

inf

2Ref(z)z*,

s > 0,

(4.7.4)

inf

2Ref(z)z*,

s > 0,

(4.7.5)

dT (z)~s

and

w"(s) :=

dT(z)=s

where z* is defined by (4.7.3). If TED. is a stationary (or sink) point for the flow generated by f E QNp(D.), then it follows that Re f(z )z* 2: 0 by p-monotonicity off (see Section 3.4). Hence

w"(s) 2: wt>(s) 2: 0 and wb(s) is clearly decreasing on (O,oo). Let M (0, oo) denote the class of all positive functions w on (0, oo) such that ~ is Riemann integrable on each closed interval [a, b] c (0, oo) and

J

ds

.. IS d1vergent.

w(s)s

o+

Note that for each wE M(O,oo) the function d7

O(s) :=

n defined by

(z}

J

d)..

(4.7.6)

w(>-.)>-.

s

is a strictly decreasing positive function on (O,dr(z)] and maps this interval onto [O,oo). We denote its inverse function by V: [O,oo) ~----> (O,dr(z)].

Definition 4.7.1 We will call a function w E M(O, oo) an appropriate lower bound for f E QNp(D.) if

w(s) ~ wU(s) =

inf dT(z)=s

2 Ref(z)z*,

s > 0.

Exercise 3. Show that if wt> defined by (4.7.4) is not zero then it is an appropriate lower bound for f E QNp(D.). Exercise 4. Let p E Hol(b.,C) be such that Rep(z) 2:0, zED. and L.lim(1- z)p(z) = z-+1

f3 2:0.

Show that iff E Q Hol(D.) is defined by the Berkson-Porta formula (see Section 3.6)

f(z) then for the sink point r

= 1 the

=

-(1- z) 2p(z),

function Wt> is constant which is equal to (3.

Chapter IV

138

Proposition 4.7.1 Let f E QNp(!::J.) be continuous and letS = {Ft}t~o be the flow generated by f. Given a point T E !::J. and a function w E ..A-1(0, oo), the following conditions are equivalent: {i) the function w is an appropriate lower bound for f; {ii} for any differentiable function W on [0, oo) such that V(t) ~ W(t), V(O) W(O) and V'(O) = W'(O), dr(Ft(z)) ~ W(t),

=

z E !::J., t;::: 0,

v

where = n-l and n is defined by (4. 7.6). In particular, dr ( Ft ( z)) ~ V (t); hence T is a globally attractive stationary point for S.

Proof. Consider the function

w: JR+ x !::J.

t--+

JR+ defined by

= dr(Ft(z)).

w(t, z)

(4.7.7)

By direct calculations we have

~~

(4.7.8)

lt=o+= -2\li(O,z) Ref(z)z*.

We first assume that condition (ii) holds. Since w(O, z) = dr(z) = W(O), we obtain by (4. 7.8) and (ii) that

d

aw

-at it=O+;::: - dt [W(t)Jt=O+

2\li(O, z) Ref(z)z*

d

1

- dt [V(t)Jt=O+ =- fl' (dr(z)) dr(z)w (dr(z)).

Varying z E 8E(T,s) = {z E !::J.: dr(z) = s} we see that the latter inequality immediately implies (i). Conversely, let condition (i) hold. It follows by (4.7.7) and the semigroup property that for all z E !::J. and s, t ;::: 0, w(s

+ t, z)

= w (s,

Hence by (4. 7.8) and the continuity we deduce from (i) and (4.7.8) that:

Ft(z)).

off, W is differentiable at each t ;::: 0 and

aw(t,z) tt at ~ -w(t, z)w (w(t, z)) ~ -w(t, z)w (w(t, z)).

Separating variables we obtain dw w(w)w = f2 (drFt(z));::: t.

ASYMPTOTIC BEHAVIOR

139

This is equivalent to condition (ii). Our assertion is proved. D As we mentioned above, if wb(s) := inf :S s 2 Re f(z)z* is positive then it may dT(z)

be used as an appropriate lower bound for

Example 1. Let

f

E QNp(l:i) (see Example 1).

f : 6. ~----+
=

f(z)

1 + zn

-(1- z) -1 n, -z

where n is a positive integer. Since f is holomorphic on 6. it follows by the Berkson-Porta representation (see Section 3.6) that f generates a flowS= { Ft} t>O of holomorphic self-mappings of 6. and T = 1 is a sink point of S. If we now set • z 1 z = --2 1- lzl 1- z then we obtain

-

Ref(z)z* =

l1-zl 2 Il 1- z 2

1+zn 1+zn Re - - = d 1 (z) Re -1 > 0. 1- zn - zn

In addition, it can be shown (see Exercise 4 and Proposition 4.7.5 below) that w(s) = wb(s) =

inf

dl(z)~s

In this case

j

2 2 Ref(z)z* =constant=-. n

d1(z)

!l(s) =

n

2

d>..

n

= exp

(-~t)

T = -2

s ln d1(z)

s

and

V(t) =

n- 1

(t)

d1(z).

Thus we have an exponential rate of convergence of the flow S to the boundary point T = 1: 2 2 } 11- zl 2 -'IFt(z)----'--'-:--:::::-:--:-'11-:::-2 < exp { --t . 1 -1Ft(z)l n 1 -lzl 2 )rote also that although f has n + 1 null points {ak : k = 1, 2, ... , n + 1} on the unit circle, only a 1 = 1 is an attractive point of S = {Ft}t~o· The reason is that Ref'(al) > 0, while Ref'(ak) < 0, k = 2,3, ... ,n+ 1 (see Figure 4.4 for n = 3).

Remark 4.7.1. However, examples show (see Example 2 below) that sometimes wb may be zero identically, while wU itself belongs to the class M(O, oo ). Moreover,

we will see below that for a semigroup of holomorphic mappings with a boundary sink point T the function is always a constant which coincides with the angular

w,

Chapter IV

140

y \ \ \ \ \ \ ' ' ' ' ''\1/ '\\\/!/~~' ___ , ,

\\\J\,, ___ \ i } \ \

\ \

1 \ ,...,. ....,

It

I/~

(1/ / / /

I(

lh/~'\ /~/~,,\

i

/--~' '\ ~~'''\\

1/ll\,,'\\\

\\\\'\'\ \l!

Ill\\\~\\~

\ f.l/~·-,, '''''"\~ l '\\ ;,~-~ ' ' ' ' ' ' " I ,,,,/ -----~' I / .---.--" _.._..._..._..__/ / /, \ ' X /////// ~\

I'

///1//l

f{{1~~#~J}

\\\11 11//

,,...._ V / / / / /

-..::::~~rt'~///1

//I~,.._~/.?/

I

I l \ v-..__..//

Figure 4.4: The asymptotic behavior of the flow generated by 1 +z 3 f(z) = -(1 - z) 2 - - . 1- z 3 derivative off at r. Also observe that the same estimate as in Example 1 can be obtained by using Proposition 4.6.2. Nevertheless, even for holomorphic mappings Proposition 4.7.1 becomes an effective tool when the angular derivative L.f'(r) = 0.

Example 2. Let f : .6.

~

C be defined by

f(z) = -(1 - z)2 1 + czn 1- czn with lei< 1. Once again, if we define z* as in Example 1 we have

11- zl 2 1 + czn 1 -lei Rej(z)z* = 1 I 12 Re 1 2: d1(z)-1- 1 > 0. - z - czn 1+ c In this case L.f'(1)

= wb(s) = 0 for all

s E (0, oo) and we cannot use it as an

1-lcl

appropriate lower bound. However, we can define w(s) =as, where a= - -1- 1, 1+ c and we find d 1 (z)

n(s) =

~

a

j

s

dA =

>. 2

~ (~ a

s

1_).

__

d1(z)

Thus we obtain by Proposition 4.7.1 the following rate of nonexponential con-

ASYMPTOTIC BEHAVIOR

141

vergence:

11- Ft(z)i2 1 11- zl 2 2 1-IFt(z)l ::; 1+atdl(z)1-lzl 2 " The reason is that f E g Hol( D.) has no null point inside D. and the boundary sink point r = 1 is equal to zero.

wb ( s)

defined at

Next we consider the case when f E QNp(D.) is not holomorphic on .6.. In this situation the convergence of the flow generated by f may be of nonexponential type even f has an interior null point.

Example 3. Let z

= x + iy E .6.. Define f : D. ~--+ C by f(z) = x7f3

Since Re f(z)z

+ iy7f3.

= xl0/3 + yl0/3

~ 0,

f is p-monotone and the origin is the unique null point of f. Then, setting T = 0, we have lzl2 do(z) = 1-lzl2 and

2 1/3 82/3( 1 + s)l/3.

Since w~(s) E M(O,oo) we can set w(s)

= w~(s) and we have

do(z)

n(s)

=

I

I

do(z)

1 w(>.)).. = 21/3 d)..

s

d)..

)..5/3()..

+ 1)1/3

0

s

Inverting this function we obtain the estimate

do(z)

do(Ft(z))::; V(t) = 24/3

[ - - tdo(z) 213 3

3; 2

+ (do(z) + 1) 2/

The latter inequality is equivalent to the estimate

IFt(z)l ::;

lzl [

(2lzl)4/3

3

. ] 3/4

t

+1

] 3

-

d0 (z)

Chapter IV

142

Note that one can calculate Ft directly by solving the Cauchy problem and obtain

Thus for x

= y we obtain IFt(z)J =

JzJ . (2JzJ)4/3 ]3/4 [ 3 t +1

So the rate of (nonexponential) convergence we have obtained is sharp.

Remark 4. 7.2. We saw above (see Sections 4.4 and 4.5) that a similar phenomenon is impossible for holomorphic mappings: Namely, if a flow of holomorphic self-mappings converges locally uniformly to an interior stationary point then the convergence must be of exponential type. These examples and Proposition 4.7.1 above motivate the following definitions.

Definition 4.7.2 LetS = {Ft}t>o be a flow with a stationary (or sink) point E b.. We will say that the asymptotic behavior of S at T is of order not less than a> 0 if there is a function wE M(O, oo) such that

T

liminf {wl(/s)} > 0 s-->0+

and dr(Ft(z))

~

(1 +

S

(4.7.9)

"'

~w~dr(z)))"' dr(z)

(4.7.10)

for all z E b. and t 2: 0.

Definition 4. 7.3 We will say that the asymptotic behavior of S at T is of wexponential type if there is a decreasing function w E M(O, oo) such that dr(Ft(z)) ~ exp ( -tw(dr(z))) dr(z)

(4.7.11)

for all z E b. and t 2: 0. In particular, if w can be chosen to be a positive constant a then we will say that S has a global uniform rate of exponential convergence: dr(Ft(z)) ~ exp( -ta)dr(z).

(4.7.12)

The following assertion is a consequence of Proposition 4.7.1.

Proposition 4.7.2 LetS= {Fth::=:o be a flow generated by f E QNp(b.) with a null (or sink) point T E b.. Then the asymptotic behavior of S at T is of order not less than a > 0 if and only if there exists an appropriate lower bound w E M (0, oo) for f such that

w(s) . . sl/a zs decreaszng on (0, oo).

ASYMPTOTIC BEHAVIOR

143

Proof. We first observe that condition (4.7.10) with some w E M(O, oo) satisfying (4.7.9) is equivalent to the same condition with a function WI E M(O, oo) which satisfies both (4.7.9) and(**). Indeed, for a given wE M(O,oo) define a function 1-L : (0, oo) ~---+ (0, oo) by . {w(l) ~-t(s) = mf [1/a: l E (O,s] } ,

s > 0.

It is clear that ~-t(s) is decreasing. Setting now wi(s) = si/a · ~-t(s) we clearly see that WI satisfies (4.7.9) and that wi(s):::; w(s). Hence

J o+

is divergent and

WI

ds wi(s)s

E M(O, oo). Then the inequality

proves our claim. Thus we can assume for the rest of the proof that w satisfies (**). It remains to be shown that w is an appropriate lower bound for f. Indeed, defining n: (O,dr(z)]~-+ [O,oo) by (1.7) and using(**) we have

O(s)

Inverting this expression we obtain V(t) :=

n- I (t):::;

(

1 t ) dr(z) := W(t). 1 + ;:;w(dr(z))

It is clear that the function W(t) satisfies all the conditions of Proposition 4.7.1. This completes the proof of Proposition 4.7.2. D Corollary 4.7.1 LetS= {Ft}t~o be afiow generated by f E QNp(!l) with a null (or sink) point T E .6.. Then: (i) The asymptotic behavior of S at

T

is of w-exponential type if and only if

inf{w"(l): l E (O,sJ} > 0,

s > 0.

(4.7.13)

Chapter IV

144

(ii} The flow S has a global uniform rate of exponential convergence if and only if

(4.7.14) for some a> 0.

Indeed, in both cases (i) and (ii) there is one function wE M(O, oo) such that the asymptotic behavior of S at T is of order not less than a for all positive a. In case (i), w can be chosen to be w(s) := inf{ wU(l) : IE (0, s]}

> 0,

s

> 0,

while in case (ii) w can be chosen to be a constant a. The following example shows that for a semigroup of p-nonexpansive (but not holomorphic!) mappings an asymptotic behavior of w-exponential type does not imply, in general, a global uniform rate of exponential convergence.

Example 4. Define a continuous mapping f : D. f(x

~---+

C by the following formula:

+ iy) = x(1- x) 2 + iy(1- y) 2 .

Since Re f(z )z 2': 0 for all z = x + iy E D., it follows that f is a generator of a semigroup S = {Ft}t>O of p-nonexpansive mappings such that each disk

D.r = {z E C:

lzl < r < 1} ls Ft-invariant.

Setting

T

= 0 and z* = izi 2 ( 1 ~ izi 2 ),

we have

It is easy to see that lim wU(s) = 1 while wtt(s) S-->0+

example, y = 0 and x =

J

8

s+1

--+

--+

0 as s

---->

oo (take, for

1).

Remark 4.7.3 For holomorphic mappings, however, condition (4.7.14) holds for some a > 0 whenever condition (4.7.13) holds for a decreasing positive function w. In other words, for holomorphic flows any convergence of w-exponential type implies global uniform exponential convergence. To explain this phenomenon in terms of lower bounds we observe that for a holomorphic generator with an interior null point the function (4.7.15) is bounded from below by a positive number, while for a boundary sink point the function w17 is just a constant.

ASYMPTOTIC BEHAVIOR

145

In both cases (interior stationary point or boundary sink point) the asymptotic behavior of a flow generated by f E g Hol(A) is completely determined by the value wU(O) := liminf wU(s) which is related to the value of derivative off at its null s~o+

point (for the interior case) or the angular derivative (for the boundary case).

Proposition 4.7.3 Let f E QHol(A) and let {Ft}t~o be the flow generated by f. If for some point T E A there is a decreasing function w : (0, oo) ~----+ (0, oo) such that dr(Ft(z)) :S e-tw(dT(z)) dr(z), z E A, t ~ 0, (4.7.16)

then there exists a number p, > 0 such that (4.7.17)

Moreover, {i) ifr E A, then p, can be chosen asp,= wb(0)/4, but p, cannot be larger than wb(O) { = lim wl> ( s )); S-+0+

(ii) if T E oA, then the maximal p, for which (4. 7.17) holds is exactly wi>(O), that is 0 < p,:::; w•(O). The proof of this Proposition is based on the following two lemmata.

Lemma 4.7.1 Let f E QHol(A) with f(O) = 0, and let wl> and wU be defined by (4.1.4) and (4.1.5), respectively. Then:

{i) wU(O) = wi>(O) = 2 Ref'(O) := 2v; (ii) v/2:::; wl>(s):::; 2v. Proof. First we show that (4.7.18) where

v Since in our case

T

= Re f'(O).

= 0, we have Ref(z)z*

1

= lzl 2 ( 1 -lzl 2 ) Ref(z)z.

Now fixing ( E oA, we set z = r(, where r E (0, 1). Then we obtain -

Re f(z)z*

= Re - -1 , -1 f(r() · -(. 1- r- r

Therefore II

1

1

-

w•(s):::; 2 Re - - - f(r() · (, 1- r2 r

where

Letting s (hence, r) tend to zero we obtain

s

r 2 = lzl 2 = - - . s+1

Chapter IV

146

On the other hand, it follows by Harnack inequality that for all z E b.

This implies that 2 lzl2(1-lzl2) Ref(z)z

2 Ref(z)z*

>

2vlzl 2 1 - lzl lzl2(1 - lzl2) . 1 + lzl

2v (1 + lzl)2.

Hence w 0(s)

>

2v

inf

inf

d.,.(z)~s (1 + lzl) 2

2v

lzl 2 ~

.~1

2v (1 + lzl)2

( 2: v /2).

(4.7.19)

(1+ff+;)2

Letting s tend too+ in (4.7.19) we see that w,(O) 2: 2v. Comparing the latter inequality with (4.7.18) and (4.7.15) we obtain (i). On the other hand, substituting now in (4.7.19) v = w 0 (0)/2, we obtain assertion (ii) and we are done. 0 0,

Suppose now that QHol(b.) has a null point different from zero, say, f(r) = E b., T =/= 0. Note that if the automorphism Mr of b. is given by

T

T-Z

Mr(z)= -1 _, -ZT then the following equality holds: (4.7.20)

Now let us consider the flow

{Gt}t~o C

Hol(b.) defined by (4.7.21)

and let g E QHol(b.) be its generator, i.e.,

g(z) = - :tGt(z)lt=O+ = [(Mr)'(z)r 1 f(Mr(z)). Then Gt(O)

= 0 for

all t 2: 0 and g(O)

Lemma 4. 7.2 The functions wU ( s) and mations (4. 7.21} and (4. 7.22}.

(4.7.22)

= 0. W 0 ( s)

are invariant under the transfor-

ASYMPTOTIC BEHAVIOR

147

Proof. We have seen already in (4.7.7) and (4.7.8) that

~

vt

[dr(Ft(z))]

=

-2dr(z) Re f(z)z*

t=O+

= -2

dr(z) Re [/(z) ( z 2 - ~)]. 1 - lzl 1- zf" 1 - O"(z, r)

(4.7.23)

On the other hand, by (4.7.20) and (4.7.21) we have

(4.7.24) Since

do(w)

w

(

-~ 2 Re g(w) 1 -lwl2

)

dr(z) ( W ) (1 -lrl2)1wl2 2 Re g(w) 1 -lwl2 ' we obtain from (4. 7.24) and (4. 7.23) the required equality 1

[

1- O"(z, r) Re f(z)

(

z

f

1- lzl 2 - 1- zf

)]

w 2] · = lwl1 2 Re [ g(w) 1 -lwl

The Lemma is proved. D Let the flow {Gt ( z) h>o C Hol( 6.) and its generator g E Q Hol( 6.) be defined by (4.7.21) and (4.7.22). By Lemmata 4.7.2 and 4.7.1 we have

. mf do(w)=s

w,(O) v Reg(w)w* 2: - - = -. 4

2

Then by Corollary 4.7.1(ii) we have

do(Gt(w)):::; do(w)exp(v/2) Finally, setting w

=

for all wE 6..

Mr(z) and using (4.7.20) we conclude that (4.7.25)

This enable us to point out the following rates of convergence.

Proposition 4. 7.4 Let {Ft} t>o be a flow generated by f E Q Hol( 6.) and let f(r) = 0, T E 6.. Then the folTowing estimates are equivalent:

148

Chapter IV

(i)

z E ~' t 2: 0;

(ii)

z E ~' t 2: 0,

(iii)

where the numbers J..L in (i) and (ii) can be chosen to be one and the same such that 0 ~ v /2 ~ J..L ~ 2v and v in (iii) is defined by v

= ~wi>(O) = ~wU(o) = Re/'(T).

(4.7.26)

Proof. First we note that inequalities (ii) and (iii) are equivalent to the following ones: (ii*) (iii*)

IGt(w)l ::::; lwl· exp ( -J..L 1 IGt(w)l

~wl 2

t),

~ lwl· exp ( -v~ ~ ::Jt),

t 2: 0,

t 2:0,

where w = Mr(z) E ~ and the flow {Gt}t;::::o is defined by (4.7.21). First let us suppose that estimate (i) holds. By using (4.7.20) for this flow we have ~

do (Gt(w))

do(w) exp ( -tJ..L).

Rewriting the latter inequality in the form

we obtain by direct calculations

IGt(w)l 2 ~lwl 2 l w l2 + ( 1 -

1

1w 12

)

(

exp -tJ..L

)

~lwl 2 ·exp(-tJ..L(1-Iwl 2 )).

which coincides with (ii*). Now we will assume that inequality (ii) (and hence (ii*)) holds. Differentiating both sides of this inequality with respect to t at t = o+ we obtain 1

-~ Reg(w)w ~ -lwiJ..L

1 -lwl 2 . 2

(4.7.27)

This implies that w~(s) ~ J..L. Thus the function w(s) = J..L is an appropriate lower bound and the implication (ii) => (i) follows by Proposition 4.7.2. Let us suppose now again that inequality (ii) (hence, (ii*) and (4.7.27)) holds with some number J..L > 0. Setting in (4.7.27) w = r(, ( E 8~, r E (0, 1) and

ASYMPTOTIC BEHAVIOR

149

letting r to zero (cf., the proof of Lemma 4.7.1) we obtain Reg'(O) 2: 11/2 > 0. A direct calculation shows that

g'(o)

= [(M,.)'(o)r 1 J'(r)

(M.,.)' (O)

= J'(r).

and so v > 0. Therefore, again by Harnack inequality, we have

1-lwl 1+ lwl -

1-lwl 1 + lwl

> Reg'(O)-- > vlwl 2- - .

Reg(w)w

-

On the other hand, 8ln IGt(w)l

8t

1 [ -] = -IGt(w)i2 Re g(Gt(w))Gt(w) .

Also it follows by the Schwarz Lemma that 8ln IGt(w)l

IGt(w)l :s; lwl.

Thus we have

1 -lwl

----'-::--'--'----'-'-<-v--

8t

1+ lwl'

-

Integrating this inequality we obtain the following estimate

which implies (iii*). Finally, if condition (iii*) holds then differentiating it with respect to t at t = o+ we obtain 1/

Reg(w)w 2: (1

1/

+ lwl)2 2: 4 > 0.

Thus w~(O) > 0 and the result follows. D Let us turn now to the case when f E g Hol( l:..) has no null point~ in l:... As a matter of fact, if S = {Ft}t>o converges to a boundary sink point r E 86. with a rate of convergence of exponential type,

d.,.(Ft(z)) :S exp( -tw(d.,.(z))) · d.,.(z), where wE M(O, oo) is a decreasing function, then this estimate can be improved as follows: where w(O) := lim w(s). s-.o+

In other words, we claim that if the inequality

holds for a decreasing

w

then the stronger inequality

150

Chapter IV

also holds. In particular, this property holds for the function w = wb. This implies, in turn, that wb is actually constant: wb(s) = w 11 (0) = f3 for all s E (0, oo) and is equal to the angular derivative off at the point r E 8b.. Moreover, this number f3 gives the best rate of exponential convergence of S = {Ft}t~O· Proposition 4. 7.5 Let the function f E Hol(b., C) be a generator of a semigroup S = {Ft}t~o of holomorphic self-mappings of b.. Suppose that f has no null-point in b. and that T E 8b. is the boundary sink point for S. Then the following are equivalent: (i) the asymptotic behavior of S at T is of w-exponential type; (ii} there is a positive number 'Y such that dr(Ft(z)) :S: dr(z) exp ( -t"(), z E b. and t ~ 0. Moreover, the maximal"( which satisfies condition (ii} is

f3

= L.j'(r)

= 2 zEtl. inf Ref(z)z*.

Proof. Let condition (i) hold, i.e., for some decreasing function wE M(O, oo), dr(Ft(z)) :S: dr(z) exp(-tw(dr(z)))

or explicitly,

II- Ft(z)fl 2 11- zfl 2 1- 1Ft(z)l2 :::; 1- lzl2 exp ( -2tw(dr(z))). This is equivalent to the inequality 2 2 11- Ft(z)fl ( (d ( ))) __ 12 :::; 1 -1Ft(z)l I l exp -2tw r z . 2 11 - ._7 1- Z

(4.7.28)

Once again it follows by the Julia-Caratheodory Theorem that for fixed t

o(Ft)

:=

0,

l. . f 1 - 1Ft (z) 1 nnm z~r 1- IZ I

/[r.]'( ) ·- . 1- Ft(z)f rt T . - 11111 _ ,

L

z~T

1-

ZT

where that last limit is taken along an nontangential approach region at Let us denote w(O) = lim w(dr(sr)) . .'0-l-

Thus setting z = sr in (4.7.28) and letting s tend to 1- we obtain

or

o(Ft) :::; exp ( -t"f), where we set 'Y = 2w(O).

~

(4.7.29) T.

STARLIKE AND SPIRALLIKE FUNCTIONS

151

Now by using Julia's Lemma we obtain the implication (i) => (ii). The converse implication can be shown by differentiating the inequality in (ii) at t = o+. Namely, we obtain Ref(z)z* 2': ~ > 0. So one can set w(s) = 'Y/2 and the asymptotic behavior of Sat Tis seen to be of exponential type. In addition, it follows by Proposition 4.6.2 that 'Y:::; {3 := L.j'(T). Finally, we observe that lim Re f(sr)(sr)*

s--->1-

lim Re f (sr) ( r--->1-

sf' 1- S 2

-

lim Ref(sr)T' _1_ = s-1 1+s

s--->1-

and the assertion is proved. D

1 -T S )

!3/ 2,

Chapter 5

Dynamical approach to starlike and spirallike functions This chapter is devoted to showing some relationships between semigroups and the geometry of domains in the complex plane. Mostly we will study those univalent (one-to-one correspondence) functions on the unit disk whose images are starshaped or spiralshaped domains. Several important aspects, however, had to be omitted, e.g. convex and close-to-convex functions (see, for example, [57, 55]), and other different classes of univalent functions. We have selected the forthcoming material according to the guiding principle that the demonstrated methods may be generalized to higher dimensions. For example, the celebrated Koebe One Quarter Theorem states that the image of a univalent function h on ~ normalized by the condition h(O) = 0 and h'(O) = 1 contains a disk of radius This theorem is no longer true at higher dimensions. Nevertheless, the dynamical approach analogues of the Koebe theorem have been recently established and used for subclasses of starlike (or spirallike) functions (see, for example [141, 109, 26, 56, 14]).

t·

Our second objective is to study the dynamics of starshaped (or spiralshaped) domains when the origin is pushed out to the boundary. For example, a domain which is starshaped with respect to a point may fail to be starshaped with respect to another point. We will study inter alia some unified conditions which describe starlike (or spirallike) functions which are independent of the location of their null points.

153

Chapter V

154

5.1

Generators on biholomorphically equivalent domains

Although the studies in the previous chapters were carried out on the unit disk, one can translate them to any simply connected domain (which differs from C) of the plane as the consequence of the Riemann Mapping Theorem. First we recall some notions and definitions in classical function theory. Definition 5.1.1 Let D be a domain in C. A function h E Hol(D, C) is said to be univalent on D if for each pair of distinct points

h(zl)

z1

and

z2

in D we have

=f. h(z2).

The set of all univalent functions in a domain D C C will be denoted by Univ(D). For h E Univ(D) one can define the inverse mapping h- 1 : n ~--+ D, where n = h(D). The content of the Open Mapping Theorem (see, for example, [122]) is that n = h(D) is also a domain (open connected subset) in C. In addition, h- 1 E Hol(O, D). In other words, h is one-to-one and h- 1 is also holomorphic on h(D). In this case f is also called a (globally) biholomorphic mapping on D. A mapping hE Hol(D, C) is said to be locally biholomorphic on D if for each point z E D there is a neighborhood V CD, of this point such that hE Univ(V). It is well known that h E Hol(D, C) is locally biholomorphic on D if and only if h'(z) =f. 0 everywhere (see, for example, [122, 128]). Two domains D and n in C are called biholomorphically (or conformally) equivalent if there exists hE Univ(D) such that 0 = f(D). The fundamental Riemann Mapping Theorem states that every simply connected domain n in c (but not c itself) is biholomorphically equivalent to the open unit disk 6. in C. Moreover, for each a E 0 there is a unique h E Univ(6.) with 0 = f(6.) such that h(O) =a and h'(O) > 0.

For the special case when D = 6. is the open unit disk in C, the subset of Univ(D) normalized by the conditions h(O)

= 0 and

h'(O)

=1

will be denoted by S(6.). This notation conforms to the one used in the classical geometric function theory. In this case we simply write S (= S(6.)) ={hE Univ(6.): h(O) = 0 and h'(O) = 1}.

In other words, the class S C Hol(6., C) consists of all the mappings h E Univ(6.) such that h has the following Taylor series at the origin: 00

h(z) = z + Lakzk. k=2

STARLIKE AND SPIRALLIKE FUNCTIONS

155

Thus, referring to some geometrical properties of a simply connected domain 0 containing the origin, if we are permitted to shrink or expand it we can find a function hE S (= S(~)) for which the domain fi = h(~) is similar to fl. (Of course, we can translate a domain, if necessary, so that the origin would be its interior point). However, in this way we may sometimes loose some features of the dynamical transformation of a domain n if its geometrical characteristics are related to a certain given fixed point in C. In particular, it happens if such a point lies on the boundary of n. The following simple assertion is the key to our further considerations.

Proposition 5.1.1 (Main Lemma) Let D and 0 be two domains in C, such that 0 = h(D) for some biholomorphic (univalent mapping) h. Then there is a linear invertible operator T on the space Hol(O, q onto the space Hol(D, C), which takes the set QHol(O) C Hol(O,C) onto the set QHol(D) (i.e., QHol(D) = T(QHol(O)). Moreover, such an operatorT: QHol(O) ............ GHol(D) can be given by the formula: (5.1.1) T(
where

t}t::::o be the semigroup of holomorphic self-mappings on 0 generated by
f(z)

=

lim z- Ft(z)

t_,o+

t

= - dFt(z) lt=O .

dt

(5.1.3)

Then substituting formula (5.1.2) into (5.1.3) and using the chain rule we obtain:

f(z)

- [(h- 1 )'(4->t(h(z)). d
=

(h- 1 )'(h(z)) ·
t=O

i.e., f = T

Chapter V

156

Remark 5.1.1. It follows by formula (5.1.2) that if

!(a)= [h'(a)r 1
o.

In addition, note that by direct calculations we obtain:

f'(a) =
(5.1.5)

Therefore a is an attractive fixed point of the semigroup { Ft} = S 1 if and only if b is an attractive fixed point of the semigroup {¢t} = S"'.

Remark 5.1.3. By using Proposition 5.1.1 and Remark 5.1.2 we are able to easily obtain a description of the class g Hol(D., T) of all semi-complete vector fields vanished in D., at a point T E D., which is a particular case of the representation that according to E. Berkson and H. Porta (see Section 4.6). Indeed, let us set D =

n =D.

and h = M-r E aut(D.) for some TED., where T-Z

M-r(z) = --_. 1- ZT

(5.1.6)

Iff E QHol(D.,r) is a semi-complete vector field, such that f(T) = 0 for some r E D., then

f(z) = [h'(z)r 1 .
=

[h'(z)r 1 . h(z). p(M-r(z))

(1- zf)2 .

ITI 2 -

Now, if we denote q(z) =

1

1-

IT l2 p(M-r(z))

1

T-

z_ . p(M-r(z)).

1- ZT

we obtain the desired representa-

tion:

f(z) = (z- 7)(1- zf)q(z),

(5.1. 7)

where Req(z) 2:: 0 everywhere. As a matter of fact, we know that representation (5.1.7) holds also for a null point free f E QHol(D.). In this case just T E 8D.. Therefore we can try to use this representation to characterize different biholomorphic mappings on D. and the geometric structure of their images.

STARLIKE AND SPIRALLIKE FUNCTIONS

5.2

157

Starlike and spirallike functions

Definition 5.2.1 A set n inC is called starshaped (with respect to the origin) if given any w E n, the point tw belongs to n for every t with 0 < t :::; 1. That is, if n contains w then it also contains the entire line segment joining w to the origin. If D is a domain in C then a function h E Univ(D) is said to be a starlike function on D if the image n = h(D) is a starshaped set (with respect to the origin).

n.

In this definition the origin is in If, in particular, the origin belongs to n then we will say that h is a starlike function with respect to an interior point. In this case the function h has a null point r in D. The image of the starlike function z

f(z) = (1-

0.9z) 0 ·7 (1

on the unit disk

~

+

0.9iz) 0 · 5 (1-

(0.7 + 0.4i)z) 0 ·6 (1

+ (0.56 + 0.7i)z) 0 ·3

is illustrated in Figure 5.1.

Figure 5.1: A starshaped domain (0 En).

If the origin belongs to the boundary an of n we say that h is starlike with respect to a boundary point (or fanlike on D). In this case there is a boundary point r E 8D and a sequence {zn}~=I ED, converging tor such that lim h(zn)

n-+oo

= 0.

The domain n in Figure 5.2 is the image of a fanlike (starlike with respect to a boundary point) on the unit disk ~ function

f(z) =

(1 + z) 2 (1- 0.9z)D.5(1 + 0.9iz) 0 · 7 (1- (0.7 + 0.5i)z) 0 ·6 (1

+ (0.6 + 0.67i)z)D· 3 ·

Chapter V

158

Figure 5.2: A starshaped domain (0 E 80). The set of all univalent functions on D which are starlike on D will be denoted by Star(D). If, in addition, there is 7 E D such that

h(7) = 0, then we will write hE Star(D, 7). Of course, in this case such a point 7 is unique because Star(D, 7) C Star( D) C Univ(D).

If h E Star(D) has no null point in D (i.e., h is starlike with respect to a boundary point) we will write that hE Fan( D). Finally, keeping the classical notations, for the special case when D = D. is the unit disk we will just write S* for Star( D., 0) n S. That is S* ={hE Star(D.): h(O) = 0 and h'(O) = 1}. The concept of univalent starlike functions was first introduced by Alexander [10] in 1915. In 1921 Nevanlinna [102] conducted a more detailed study of this class. In particular, the following characterization of the class Star( D., 0) is due to him.

Proposition 5.2.1 Let h be a univalent holomorphic mapping on the unit disk D. such that (5.2.1) h(O) = 0.

Then h is a starlike function on D. if and only if Re [

z~~~~)]

> 0,

zED..

(5.2.2)

STARLIKE AND SPIRALLIKE FUNCTIONS

159

As we will see in the sequel, if h E Hol(b., C) is locally biholomorphic, i.e., h'(z) =f. 0 everywhere, and satisfies (5.2.2) then it is necessarily univalent. Furthermore, condition (5.2.2) leads to the study of other interesting subclasses of S. In particular, in 1936 Robertson [119] had introduced the class S*(.A) of starlike functions of order .A:

S*(.A) = {hE S* : Re [

z~~~~) ~ ~A> 0,

z E b.}.

(5.2.3)

In 1978 Wald [150] characterized starlike functions with respect to another center. Using our notions, his result can be reformulated in the following way. Proposition 5.2.2 Suppose that h E Hol(b., C) is either of the form h(z) = z + 2:::;:'= 2 akzk or of the form h(z) = 2::::::'= 1 bkzk with h'(7) = 1 for some 7 E b.. Then the function h(z)- h(7) belongs to Star( b., 7) if and only if Req(z) > 0, where h(z) q(z) = h'(z) (z -7)(1- zr)'

z E b.\ {7 },

and

We will see below that Proposition 5.2.2 (as well as Proposition 5.2.1) can be easily obtained by using a different approach in more general settings. Different applications of this assertion are presented in Wald's thesis [150] (see also [57]). Definition 5.2.2 A set n in C is said to be spiralshaped (with respect to the origin) if there is a number f-L E C with Re 11 > 0 such that for each w E n and t ~ 0 the point e-tp.w also belongs to n. (see Figure 5.3} If D is a domain in C, then a univalent function h E Hol(D, C) is said to be spirallike on D if the closure of its imagen = h(D) is a spiralshaped set. Once again, if the origin belongs to n then we will say that h is a spirallike function on D with respect to an interior point. Otherwise (if the origin belongs to the boundary of n) we say that his spirallike with respect to a boundary point.

an

The set of all biholomorphic mappings on D which are spirallike on D will be denoted by Spiral(D). If, in addition, there is 7 E D such that h(7) = 0,

then we will write h E Spiral(D, 7). Note also that if in Definition 5.2.2 the number f-L is a real positive number, then n is actually starshaped, i.e., Star(D) c Spiral( D). Consequently, Star(D, 7) C Spiral(D, 7). If h E Spiral(D) has no null point in D (i.e., h is spirallike with respect to a boundary point) we will write that h E Snail(D). (We will see below that the class Snail(D) is quite narrow: in particular, each h E Univ(b.) which has a biholomorphic extension to b. is, in fact, in Fan(b.), i.e., h(b.) is starshaped).

Chapter V

160

Figure 5.3: The spiralshaped domain (0 E of!): e-t1-1w E f! for each w E fl. It seems that the first occurrence of the class Spiral(6., 0) = { h E Spiral(6.) : h(O) = 0} appeared when condition (5.2.2) was modified analytically by inserting the factor ei 8: Re [eiO zh'(z)] > 0 (5.2.4) h(z) (see Montel [101] and Spacek [137]). Actually, this definition is compatible with our Definition 5.2.2 (see also, Remark 5.2.2 below) Proposition 5.2.3 LethE Hol(6., C) have the form (X)

h(z)

= z + Lakzk

(5.2.5)

k=2

(i.e., h(O) = 0 and h'(O) = 1). Then h E Spiral(6., 0) if and only if condition (5.2.4) holds for some() E (1rj2, 7r/2). Of course, Proposition 5.2.1 follows from Proposition 5.2.3 if we set there()= 0. To prove the latter proposition as well as Proposition 5.2.2 we will need another more general assertion which is a direct consequence of the Main Lemma (Proposition 5 .1.1). Proposition 5.2.4 Let h E Hol(D, C) be a spirallike (respectively, starlike) function on D. Then there exists a generator f E g Hoi( D) of a continuous flow on D such that h satisfies the following differential equation:

J.Lh(z)

= h'(z)f (z),

zED,

(5.2.6)

for some J.L E C (respectively, J.L E IR) with Re J.L > 0. Proof. Let fl = h(D) be a spiralshaped (respectively, starshaped) domain. Then for some J.L E C (respectively, J.L E IR) with Re J.L > 0 the mapping

STARLIKE AND SPIRALLIKE FUNCTIONS

161

g = J.d, (g(z) = J-Lz), belongs to QHol(f2). Indeed, in this case the integral curve defined by the Cauchy problem {

av(t,w) av(t,w) at + g(v(t, w)) = at v(O,w) = w, wEn,

+ J-LV(t, w)

= 0,

(5.2.7)

has the form {v(t, w) = e-J.Ltw, t ~ 0, wE f2} and belongs to f2 by Definition 5.2.2 (respectively, Definition 5.2.1). Then Proposition 5.1.1 implies that f E Hol(D, C) defined by: (5.2.8) f(z) = (Tg)(z) = J-L [h'(z)r 1 h(z) is a generator of a flow on D, and we are done. D To establish a converse assertion we may omit the requirement on h to be univalent.

Proposition 5.2.5 Suppose that h E Hol(D) satisfies equation (5.2.6}, with J-L E C, (respectively, J-L E IR) ReJ-L > 0, and f E QHol(D). Then the set n = h(D)

is spiralshaped (respectively, starshaped}. Moreover, if D is bounded and h has a null point TED with h'(T) =1- 0, then h is univalent, hence spirallike (respectively, starlike) with respect to an interior point. Proof. Fix wE f2 and find zED such that h(z) = w. Since f E QHol(D) generates a continuous flow s, = {Ft(z)}t>o on D, setting v(t, z) = h (Ft(z)), we have that v(t, z) E n for all t ~ 0. That is for a fixed z E D, the family {v(t,z), t ~ 0} determines a continuous curve inn, such that v(O,z) = w. In addition, it follows by the chain rule that

av~~ z) = h' (Ft(z)) a~~z) = -h' (Ft(z)) f

(Ft(z)).

On account of formula (5.2.6) we obtain that

av(t,z) at = -J-Lh(Ft(z)) = -J-Lv(t, z).

(5.2.9)

Integrating the latter relation with the initial data

v(O,z) = w

(5.2.10)

we obtain

v(t, z) = e- 11 tw E fl

for all t ~ 0.

Since w E f2 was arbitrary the first assertion of our proposition follows. Assume now, in addition, that h has a null point T E D and h'(T) =1- 0. This implies that there is a neighborhood U C D of the point T such that h is univalent on U and V = h(U) C n is a neighborhood of the origin. We want to show that, in fact, h is univalent on the whole of D. Indeed, assuming the contrary, suppose that for some w E n there are two distinct points Zl and Z2 in D such that h(zl) = h(z 2 ) = w.

Chapter V

162

Observe also that, as a result of equation (5.2.6), the conditions h(r) = 0 and h'(r) # 0 imply that f(r) = 0 and f'(r) = J.L with ReJ.L > 0. Since Dis bounded the family {Ft (·)} t>o is a normal family in D. Then exactly as in Proposition 4.4.1 one can conclUde that T E D is an attractive stationary point of the flow Sf = {Ft(·)}t>o generated by f. In particular, we have that Ft(zi) and Ft(z2) converge toT as t----> 00. Note that Ft(ZI) and Ft(Z2) are different for all t ~ 0. As above, now define v(t, z) = h (Ft(z)), t ~ 0, z E D and choose to > 0 such that Ft(z 1 ) and Ft(z2) belong to U for all t ~to. Then for such t the curves v(t, z;), i = 1, 2, lie in V = h(U) C n. But v(O, z 1 ) = v(O, z2) = w and we have that v(t, z;), i = 1, 2, are the same as the solutions of the differential equation (5.2.9) with the same initial data (S.2.10). Consequently Ft(zi) = Ft(z2) for all t ~to. That is a contradiction. D Sometimes we will say that h E Univ(D, C) is J.L-spirallike if it satisfies equation (5.2.6) with J.L E C, Re J.L > 0, and f E Q Hol(D). Remark 5.2.1. Thus h E Spiral(D) is spirallike with respect to an interior point (that is h E Spiral(D, r) for some T E D) if and only if the generator f in (5.2.6) vanishes at T E D. Moreover, iff is defined, then J.L = f'(r), and Tis an attractive stationary point of the flow S 1 = {Ft (·)} t>O generated by f. In fact, it can be shown (see Section 5.7) that for each f E Q-Hol(D), normalized by the conditions f(r) = 0 andRe f'(r) > 0 there is a unique solution h E Spiral(D, r) of the equation (5.2.6) with J.L = f'(r) normalized by h'(r) = 1. In addition, if J.L is a purely real number, then h defined by (5.2.6) is a starlike function on D. Similarly, we can say that h E Snail(D) if and only iff has no null point in D. In this case the flow S 1 = { Ft (·)} t>o generated by f for each z E D converges to a boundary point T E aD. Since for the special (but most important) case when D = b. is the unit disk we have a complete description of the class Q Hol(b.), the proved propositions imply several corollaries. In particular, applying the Berkson-Porta parametric representation of the class Q Hol(b.) we obtain the following: Corollary 5.2.1 Let h : b. ~---> C be a univalent holomorphic function on b.. Then h(b.) is spiralshaped if and only if the following equation is fulfilled:

J.Lh(z) = h'(z)(z- r)(1- zf)p(z), where T E b., J.L E C with ReJ.L z E b..

>

z E b.,

(5.2.11)

0, andp E Hol(b.,C) with Rep(z) ~ 0 for all

Remark 5.2.2 Thus, if T E b., then h E Spiral( b., r) (i.e., is spirallike with respect to an interior point); if T E 86, then h E Snail(b.) (i.e., spirallike with respect to a boundary point). Separating these two cases we conclude: A locally biholomorphic function h on ~ belongs to Spiral(~, r) if and only if there exist T E b. and J.L E C with Re J.L > 0 such that

R h'(z)(z- r)(l- zf) 0 e J.Lh(z) > '

A

z Eo.

STARLIKE AND SPIRALLIKE FUNCTIONS

163

Indeed, equation (5.2.11) can be rewritten in the form

Re h'(z)(z- 7)(1 - zf) 11h(z)

= Re _1_ > 0 p(z) -

If 7 E b., then differentiating (5.2.11) at

z =

z E b.. '

(5.2.11')

7 we obtain p(7)

which means that inequality (5.2.11') is actually strict. If we normalize 11 by the condition 1111 = 1, we obtain that T and p are uniquely determined by h. Of course, the converse assertion is also true. Moreover, if 7 = 0 then setting 0 = - arg 11 we obtain a description of the set Spiral( b., 0) which coincides with the classical description (Proposition 5.2.3). Letting 11 = 1 and 7 = 0 in (5.2.11') we arrive immediately at Nevanlinna's condition (Proposition 5.2.1) and Wald's condition if 7 E b., 7 ::f. 0 (Proposition 5.2.2). Similarly, we conclude from (5.2.11): A univalent function h on b. belongs to Snail( b.) (respectively, Fan( b.)) if and only if for some T E {)b. and 11 E
We will see in the next section that the boundary behavior (at the point 7) of the quotient Q(z)

= h'(z~~:)- 7 )

(the so called Visser-Ostrowski quotient) in the

latter inequality characterizes those functions in Snail(b.) which, in fact, belong to Fan(b.).

5.3

A generalized Visser-Ostrowski condition and fanlike functions

In this section some relations between classes Snail( b.) and Fan( b.) (of spirallike and starlike functions with respect to a boundary point) will be studied. First we make the following observation. Suppose that h E Snail(b.), that is h E Univ(b.) satisfies equation (5.2.6):

J-Lh(z) = h'(z)f(z), for some J-L E 0, and f E

z E b.,

g Hol(b.)

with no null point in b..

164

Chapter V

We know that in this case there is a unique boundary point T E at:::. which is the sink point of the semigroup generated by f. Assume, temporarily, that one of the following conditions holds: . h" (z) (1) I h'(z) I:::; M < oo, z E !:::.., or (ii) lim h(z) = 0 and lim h'(z) = a, a =f. 0, oo, as >. approaches T nontanz--+r z--+r gentially. The latter condition is known to define h to be conformal at the boundary point T E af:::.. (see, for example, [107]). Without loss of generality one can set T = 1. We recall, that in our situation as a result of the Berkson-Porta formula f E 9Hol(l:::..) has the form:

f(z)

= -q(z)(1- z) 2 ,

(5.3.1)

with some q E Hol(t:::.., q such that Req 2: 0 everywhere. In addition, we know by the established continuous version of the Julia-WolffCaratheodory Theorem (Proposition 4.6.1) that the angular derivative L lim f(z) z-+1 Z - 1

of f exists at 1 and equals to the angular limit

= f3 2: 0.

L lim J'(z) Z-+1

(5.3.2)

We want to show that in both cases (i) and (ii) the number f-l in equation (5.2.6) must be equal to {3, that is, f-l is actually a real number. Indeed, assuming first that (i) holds and differentiating equation (5.2.6), we obtain h" (z)

1

f-l- f (z) Since lim f(r) r-+1-

=

(5.3.3)

h'(z) f(z).

= 0 we obtain f-l = f3 by (5.3.2).

If now condition (ii) holds, then one can write, by using equalities (5.2.6) and (5.3.1), that f-lh(z) = -h'(z)(1- z) 2 q(z) with Req(z) 2: 0, z E !:::... Then we have f..l-1-h(z) = h'(z)(1- z)q(z) = h'(z) f(z).

z-1

Since

h(z)

L lim--= L lim h'(z) =a z-+1Z-1

we see again that

J-L

(5.3.4)

z-1

Z-->1

=f. O,oo,

= f3 is real, hence h is, actually, a fanlike function.

0

Of course, these simple considerations are related to the question of how can one join conditions (i) and (ii). For example, one can replace condition (ii) by the condition that h be isogonal at T, that is h and arg h' have finite angular limits at T. It is clear that if h is

STARLIKE AND SPIRALLIKE FUNCTIONS

165

conformal at r then it is isogonal at T. In general the converse does not hold. At the same time some geometrical arguments indicate that there is no properly spirallike function with respect to a boundary point which is isogonal at it. Another (much weaker than isogonality) condition which is sufficient to the above mentioned property is the Visser-Ostrowski condition:

L lim h'(z)(z- r) z-+T h(z)- h(r)

= 1.

(5.3.5)

It is known (see [107]) that this condition is equivalent to the following: . h"(z) Lhm(z-r)-h() =0. Z--+T I Z

(5.3.6)

Obviously condition (i) implies (5.3.6). Thus the Visser-Ostrowski condition is weaker than both (i) and (ii). We will establish now that a generalized VisserOstrowski condition (see condition (*) below) is necessary and sufficient for h E Snail(~) to be in Fan(~). Let us consider the Visser-Ostrowski quotient:

Q(z)

= h'(z)(z- r).

(5.3.7)

h(z)- h(r)

Proposition 5.3.1 ([38]) Let h be a univalent function on a spiralshaped set with lim h(rr) = 0, T E 8~.

~

such that

r-+1-

Then

h(~)

h(~)

is

(5.3.8)

is, in fact, starshaped if and only if the following condition holds:

(*)

the angular limit

L lim Q(z) := v Z-+T

exists finitely and is a positive real number.

Proof. Assuming again that the generator fin equation (5.2.6) is presented by the Berkson-Porta formula (5.3.1), we could use a similar argument as in (5.3.4) if we show that the point T E 8~ in (5.3.7), (5.3.8) and (*) is equal to 1. Indeed, if this is not the case we have by (5.2.6) and (5.3.1) that 11-h(z) h'(z)(z- r)

(1- z) 2q(z) (z- r)

(5.3.9)

Hence by (*) the limit TJ := L lim Z-+T

[-(1-

z) 2 q(z)]

z-

T

=!!. v

(5.3.10)

exists (finitely). On the other hand, setting z = rr, r E (0, 1), and letting r approach 1- we obtain TJ

= (1- r)(r -1)f

. q(rr) hm - 1- r - 1

r-+

= 11- rl 2

. q(rr) hm - r- 1

r-+ 1-

166

Chapter V

and Re77

=

11- rl 2

lim Re q(rr) ::; 0. r- 1

r-+1-

But the latter equality is impossible because of (5.3.10) andRe J1. > 0. Contradiction. So T = 1. Once again, since J.Lh(z) =

(;~~) h'(z)(z- 1)

we obtain J1. = L lim ( f(z)) Q(z) = (3L lim Q(z), z-+1 z - 1 z-+1

and our assertion follows. 0 Thus the class of properly spirallike (i.e., not starlike) functions with respect to a boundary point contains neither conformal nor isogonal mappings at this boundary point.

Remark.5.3.1. We will show below (see Section 5.6) that the number v = L lim Q(z) yields an important geometrical characteristic of hE Fan(.D.), namely, Z-+'T

the angle (} = 1rv is the smallest one such that h(.D.) lies in the wedge of the angle 0. Thus, in fact, for h E Fan(.D.) with lim h(r) = 0 we will show that r-+l-

L lim Q(z)::; 2. Z-+'T To do this we need an approximation process which is based on Hummel's representation of the class Star( .Do, ·) of starlike functions with respect to interior points. We will give this representation in Section 5.5.

5.4

An invariance property and approximation problems

The following question naturally arises in approximation function theory: given h E Hol(.D., C) such that n = h(.D.) is spiralshaped (respectively, starshaped) and a sequence nn c n spiralshaped (starshaped) domains such that nn = n, find the sequence hn such that lim hn(z) = h(z) for each z E .Do, and hn(.D.) = !1n. n->oo Theoretically, under certain conditions this problem can be solved because of the Riemann Mapping Theorem and Caratheodory's Kernel Convergence Theorem.

u

STARLIKE AND SPIRALLIKE FUNCTIONS

167

However, in general, to find such a sequence implicitly has not seemed plausible, even the sequence nn is well described. At the same time one can define, in a sense, a dual approximation problem which is related to some invariance property of spirallike (respectively, starlike) functions. Namely, if h E Spiral(D.) (respectively, Star(D.)), find a 'nice' sequence of domains Dn in D. such that UDn = D. and h continue to be spirallike (respectively, starlike) on each Dn. In view of Propositions 5.2.4 and 5.2.5 an answer to the above question is provided by the following observation. If h E Spiral(D.) (respectively, Star(D.)), then for some J1. E C (respectively, J1. E IR) with ReJ-L > 0 the function f(z) = J-Lh'(z)[h(z)]- 1 is a semi-complete vector field on D.. Then these propositions yield that h continues to be spirallike (respectively, starlike) on a domain D C D. if and only if this domain is invariant for the semigroup S = {Ft}t~O generated by f. Thus, due to Proposition 5.2.4 we can formulate the following assertion: Proposition 5.4.1 Let h E Spiral(D.) (respectively, h E Star(D.)). Then there is a unique point TED. such that for all K > 1- ITI 2 the sets h(D(T, K)), where D(T, K) =

zfl 2 } {zED.: 111 -lzl 2 < K

,

are spiralshaped (respectively, starshaped).

We recall that forTE D. the sets D(T,K) ={zED.:

Z-T

1----1 < 1- TZ

r}, K =

(1 -ITI 2 )(1- r 2 )- 1 (see Section 1.1, Exercise 6) are pseudo-hyperbolic balls in D. while for T E 8D. these sets are horocycles internally tangent to 8D. at T. In particular, forT= 0 we have that hE Spiral( D., 0) (respectively, Star( D., 0)) is spirallike (respectively, starlike) on each disk D.r = {lzl < r, 0 < r ~ 1} concentric with D.. (Note that for the functions of the class

S* ={hE Star(D.): h(O) = 0 and h'(O) = 1} this result was obtained independently by Takahashi and Seidel [18] as an extension to Nevanlinna's theorem (see Proposition 5.2.1). A simple proof of this fact with the use of the Schwarz Lemma can be found in [33]). One may expect that even when T i- 0 and h E Spiral( D., T) (respectively, h E Star( D.)) on the unit disk, then for at least r close enough to 1 it continues to be spirallike (starlike) on the disks D.r. However, examples show that in general this conjecture has been answered in the negative.

Example 1. Consider the function h( z) = h 0 ( M 1 ( z)), z E D., where 2

z

ho(z) = (1- z)2

168

Chapter V

is the so called Koebe function and

l-z M1;2(z) = -2- 11- 2 z is a Mobius involution transformation of the unit disk taking the origin to T = ~. It is easy to see that the Koebe function belongs to Star(b.,O) (evenS*) and it follows that hE Star(b., T), because of the relations h(b.) = h0 (b.), h(T) = 0. We claim that there is a sequence {zn} C b., lznl----+ 1 such that

R [ zh'(zn)] O e h(zn) < · Indeed, calculating

Re [zh'(z)] h(z)

=

~ Re z(1- z)(1 + z)(z- !)(1- !z) 4 11 + zl 2 1z- !1 2 11- ~zl 2

and setting z = x + iy we obtain (after several technical manipulations) that the numerator N(z) of the right hand side of the latter equation has the form

Now it can easily he seen that for any sequence {zn} E b., lznl ----+ 1 such that D = { z = x + iy : (x-! )(x 2 + y 2 - 2x) > 0} the expression N(z) is negative (see Figure 5.4). Thus the claimed assertion is proved.

Zn

E

Figure 5.4: The set D.

STARLIKE AND SPIRALLIKE FUNCTIONS

169

Nevertheless, it has been shown in [66] that if T is close enough to zero, then the answer to the above question (concerning starlike functions) is affirmative.

Proposition 5.4.2 Let h E Star(~, T) with ITI < 2- J3. Then there exists c: > 0 such that for each 1 - c: < r ~ 1 the function h belongs to Star( ~r). By using the results in Sections 3.6 and 4.5 we can formulate some sufficient conditions for h E Spiral(~, T) (respectively, Star(~, T)) to be spirallike (respectively, starlike) on each ~r for r close enough to 1, which do not depend on the location of T E ~We will say that a function h E Hoi(~, C) is strongly spirallike if for some f.Lh(z)z f.L E 0 the function f ( z) := h' ( z) is strongly p-monotone on ~.

Proposition 5.4.3 Let h E Hol(~, q be a strongly spirallike function. Then there is T E ~ and e > 0 such that for each 1- e < r ~ 1, hE Spiral(~r) and h(T)

= 0.

In particular, by using a criterion of strong p-monotonicity we have the following:

Corollary 5.4.1 Let a be a real continuous function on [0, 1] such that a(1) > 0 and let h E Hol(~, C) be such that for some f.L E 0 the following condition holds f.Lh(z)z Re h'(z) ;::: a(lzl)lzl, Then there exists T E

~

and c:

hE

> 0 such that for each 1 - c: < r

Spiral(~r)

Exercise 1. Let f(z) = az 2

-

z E ~-

and h(T)

~

1

= 0.

a+ bz for some positive a and b.

(i) Show that if equation (5.2.6) is solvable for some f.L E 0, then f.L is, in fact, a positive real number, that is, the solution h of (5.2.6) is, actually, a starlike function on ~- Find f.L. (ii) Show that there exists c: > 0 such that for each r E (1-e, 1], h E Star( ~r ). (iii) Set a = 2 and b = 3. Show that e in (ii) can be chosen arbitrarily in the interval [0, ~).

1- cz

= az 2 - a+ b z - 1 +cz Reb> 0 and lei < 1.

Exercise 2. Let f(z)

with some complex numbers a, b

and c such that (i) Show that there are /-L E 0 such that equation (5.2.6) has a solution hE Spiral(~r, T) for some T E ~and each r E (1- e, 1]. (ii) Find relations between a, b and c such that f.L is real, i.e., h is starlike. (iii) Setting a= 0 and b > 0, show that h defined by (5.2.6) with an appropriate f.L

> 0 is starlike of order a (see Section 5.2) with a

1-lcl . 1 + 1c 1

= b - --

Chapter V

170

Finally, note that if hE Hol(A, C) is known to be in Spiral( A), then sometimes information on values of h and its first and second derivatives at only one point may be useful for solving the dual approximation problem on disks Ar concentric with A. Indeed, using Corollary 4.5.1 and Remark 4.5.2 we can easily arrive at the following sufficient condition: Proposition 5.4.4 Let h E Spiral(A) (respectively, Star(A)), satisfy equation (5.3.6) with some f. L E C (respectively, f. L E IR) with Ref..L > 0 and f E Hol(A,C). Then . h" (O)h(O) (z) Re f. L [h'(0)]2 < Re f..L,"

. . . h(O) [ h" (O)h(O)] (n) if 41 h'(O) I< 1- [h'(0)]2 then for some

E

> 0 and each r E (1- E, 1]

the function hE Spiral(Ar) (respectively, Star(Ar)). Remark 5.4.1. Of course, the problem above is not relevant for the class Fan(A). Indeed, it is clear that if h E Star(A) has no null points in A then there is no disk Ar concentric with A such that h( Ar) is starshaped. In this case another approximation problem arises: given h E Fan(A); find a sequence hn E Star( A, ·) of starlike functions with interior null points such that hn converge to h uniformly on compact subsets of A as n---+ oo. It looks as if the following procedure should work in solving this problem. If, for example, hE Fan(A) is isogonal at its boundary null point, say T = 1, then it satisfies the equation f..Lh(z) = h'(z)f (z), z E A, (5.4.1) for some

f

E

9 Hol(A): f(z) = -q(z)(1- z) 2 ,

with Req(z) > 0, z E A, and f. L = L.f'(1) > 0. Define an approximation sequence fn to f by using the Berkson-Porta representation formula fn(z) = (z- Tn)(1 fnz)q(z), where {rn} C A is any sequence which converges to 1 as n tends to oo. If, in addition, we can choose this sequence such that the numbers f~ (T n) are real, then we may try to solve the equations

f..Lnhn(z)

= h~(z)fn(z),

z E A,

(5.4.2)

with f..Ln = f'(rn) in order to define a sequence of univalent functions hn which are starlike with respect to interior points (h(rn) = 0). However, in this way there is a risk that hn may not converge to h. Indeed, putting it otherwise, we obtain that the numbers f..Ln = f~(rn) = (1 -lrnl 2 )q(rn) must converge to

f. L = L.lim f'(z) = L. lim f(z) = L.lim(1- z)q(z). z-+1

On the other hand, if T n achieves

Z--+ l

T

Z -

1

Z--+ 1

= 1 along the real axis we have

STARLIKE AND SPIRALLIKE FUNCTIONS

171

That is a contradiction. Also it is easy to construct a counter example of the latter relation.

Example 2. Hoi(~,

q

Consider a semi-complete (even complete) vector field defined as follows:

f(z)

= z2 -

f

E

1.

There are two boundary null points z = 1 and z = -1 of f. Since f' (1) = 2 > 0 it follows by Proposition 4.4.1 (a continuous version of the Julia-WolffCaratheodory Theorem) that z = 1 is a sink point of the semigroup generated by f. Therefore, one can present f by the Berkson-Porta formula:

f(z)

=

21 -(1- z) -

+z

1-z

with q(z) = 1 + z. Obviously, q admits real values if and only if z

1-z

So, we have to choose a sequence {Tn} E ~. Tn procedure to be real. Further, if we define

fn(z) we obtain fn(Tn)

= 0,

=

--->

E

~

are real.

1, in the above approximation

1+z (z- Tn)(1- Tnz)--, 1-z

and fn(z)---> f(z)

in~.

while

As a matter of fact, it can be shown that if hn are solutions of (5.4.2) normalized by the condition hn(O) = 1, then they converge to a fanlike function h which is a power of h defined by (5.2.1). However, in general, we do not know whether the numbers JLn are real (i.e., whether the functions hn are starlike). Thus a procedure of using the Berkson-Porta multiplier (z- Tn)(1- Tnz) has been shown to be inappropriate for the approximation of starlike functions with respect to a boundary point by starlike functions with respect to interior points. Nevertheless, it turns out that a modification of this multiplier in the spirit of J.A. Hummel makes such a procedure very effective. Hummel's multiplier has been used by A. Lyzzaik to prove a conjecture of M.S. Robertson on a description of starlike functions with respect to a boundary point, the images of which lie in a half plane. The next section is devoted to this approach.

Chapter V

172

5.5

Hummel's multiplier and parametric representations of star like functions

Hummel's multiplier is a (meromorphic) function of the form

H ( ) _ (z- r)(1 - zf) T Z ' z

z E

A,

where T is a given complex number with JrJ :::; 1. For the case of JrJ < 1, J.A. Hummel showed in [65] and [67] that this function plays a special role in the study of starlike functions. In fact, it turns out that by the multiplication operation this function translates the set Star( A, 0) onto the set Star( A). Moreover, if T E A then Star( A, 0) is translated onto Star( A, r). More precisely: Proposition 5.5.1 Let h E Hol(A, q, h(O) = 0, and g E Hol(A, q be two functions related by the formula

g(z) = HT(z) · h(z).

(5.5.1)

Then g(A) is starshaped if and only h(A) is starshaped.

In this section we will treat only the case when T E A. In this situation it is more convenient (for some symmetry) to consider the meromorphic function WT on A defined by

W ( )= 1 H ( )= 1 (z- r)(1 - zf) Tz 1 - JrJ2 T z 1 - JrJ2 z '

z EA.

(5.5.2)

Of course, it is sufficient to prove Proposition 5.5.1 replacing HT in (5.5.2) by WT. Moreover, in this case the inverse translation is just the same multiplier composed with a Mobius involution transformation.

Proposition 5.5.2 LethE Hol(A, q and g E Hol(A,
where

Moreover, the inverse

T-Z

MT(z) = 1- zf. To prove this assertion we need some properties of the function WT defined by (5.5.2).

STARLIKE AND SPIRALLIKE FUNCTIONS

173

Lemma 5.5.1 For the function Wr the following properties hold:

{i) for all z E 86.: lzl = 1, Wr(z) is a positive real number; {ii} for all z E 86. : lzl = 1,

R zll!~(z) = O· e Wr(z) ' {iii} if Mr denotes the Mobius transformation T-Z

Mr(z) = -1 -_, -ZT then

Wr(z) · Wr(Mr(z)) = 1. Proof. (i) If we set z = ei'P, 0

~

cp ~ 27r we obtain:

(ei'P - r)(1 - ei'Pf) (1 - lrl2)ei'P 11 - ei'Pfl2 1 -lrl2 > 0.

(1 - ei'Pr)(1 - ei'Pf) 1-lrl2

(ii) By direct calculations we have: 1 [(1 - 2::) (1 1 -lrl 2 z

w~(z)

=

1 1 -lrl2

zr)]

[~(1-zf)-(1-2::)r] z2 z

1 [r lrl 2 _ lrl 2] 1 ( r -) 1 - lrl2 z2 - ----;- - r + ----;- = 1 - lrl2 z2 - r .

Again setting z = ei'P we obtain: Re zlll~(z)

Wr(z)

=

Re

e2 i'P (~-f) r- fe 2i'P . . =Re 2 (e''P- r)(1- e''Pr) ei'P 11- e-i<prl 1

- - - .-----,. 2 •

11- e-•'Prl

(iii) Substitute Mr(z)

Wr(Mr(z))

. . Re( re-•'P - fe''P)

= 0.

= r- z_ to Wr instead of z. We obtain: 1- ZT

=

(1- zf) . ( ~ -

1-lrl 2 (r- z- r

r) (1- ~-f) r- z

+

zlrl 2 )(1-

zf

+ zf-

(1 -lrl 2)(r- z)(1- zf) -z(1 - lrl 2) 1

(r- z)(1- zf)

Wr(z)"

lrl 2)

Cbapter V

174 The lemma is proved. D

Following J.A. Hummel [65, 67] we will say that a meromorphic function g on satisfies the condition (A) if for each c > 0 there is p E (0, 1) such that for all z in the annulus p < izi < 1 the following condition holds:

~

zg'(z) Re g(z) > Star(~,

Lemma 5.5.2 Let g E (A).

(5.5.4)

-£.

T) for some T E

~-

Then g satisfies condition

Proof. Since g(~) is starshaped and the origin is an interior point of the function g satisfies the equation:

g(~),

g(z) = g'(z)(z- T)(1- z'f)p(z) with some T E ~.such that g( T) = 0 and p E Then by Lemma 5.5.1(iii) we obtain: ( 1 -ITI

2

).

z · g'(z)

g(z)

Hol(~,

q

with Rep> 0 everywhere.

1

(5.5.5)

= lllr(z) · p(z)

Denoting w = Mr(z) we have in turn from (5.5.4):

z · g'(z) g(z)

(5.5.6)

= Wr(w) · q(w),

1

= (1 -ITI 2 )p(Mr(w)) (recall, that Mr is an involution). Req(w) > 0 for all wE~- Now for each wE~ the right

where q(w)

Hence (5.5.5) can be presented as:

Ill ( ) . ( ) = (w- T)(1- w'f) q(w) = f(w) r w qw (1 - jTj2)w w ,

hand side of

(5.5.7)

(w- T)(1- w'f) q(w) (1 _ ITI 2 ) defines a semi-complete vector field by the Berkson-Porta representation. At the same time, it follows by Proposition 3.4.4 that f satisfies the condition: where f(z)

=

Ref(w) · w;::: Ref(O)w(1-lwl 2 ). Therefore for each c > 0 one can find PI E (0, 1) such that:

f(w) 1 Re - - = - Ref(w)w > w lwl2 -

-£

,

(5.5.8)

whenever PI < lwl < 1. Note now, that if w runs in the annulus PI < lwl < 1, then z = Mr(w) runs in the set A=~\ B(T, r) where B(T, r) C ~is a hyperbolic ball centered at T. But this ball is also strictly inside ~. hence there is p E (0, 1) such that the annulus p < izi < 1 lies in A (see Figure 5.5).

STARLIKE AND SPIRALLIKE FUNCTIONS

175

A=!'!..\ B(.,-,r) Figure 5.5: Condition (A). Thus for such z : p < The lemma is proved. D

lzl <

1 we obtain from (5.5.5)-(5.5.7) relation (5.5.3).

Proof of Proposition 5.5.2. Let h E Hol(.6., C), and g E Hol(.6., q be related by (5.5.2). Assume that g E Star(.6., r). It is clear that T must be a unique null point of g. Note, that Wr satisfies condition (A) due to Lemma 5.5.1(ii). By differentiating (5.5.2) we obtain the equality:

z. h'(z) h(z)

zw'(z) z g'(z) g(z) - IJ!(z) ' 0

which implies that h also satisfies condition (A). At the same time, since h E Hol(.6., C) we must have h(O)

=

0 and h'(O) = - g(O) =F 0. Therefore it follows by T

the minimum principle for harmonic functions that actually

R zh'(z)

eh(;f~o

(5.5.9)

for all z E .6.. Thus hE Star(.6., 0). The converse assertion can be easily proved by replacing the roles of h and g. Indeed, let now h E Hol(.6., C), h(O) = 0 be starlike, and g E Hol(.6., q be defined by (5.5.2) with some T E .6.. Note that a function 91 E Hol(.6., q defined by: 91(z) = h(Mr(z)) is also starlike because: 91(.6.) = h(.6.). Again, using Lemma 5.5.1(iii) we obtain: 91 (z) g(Mr(z)) = h(Mr(z)) · Wr(Mr(z)) = Wr(z)

Chapter V

176 or

91 ( Z) = Ill T

(

Z) " h 1( Z),

where we denote h1(z) = 9(Mr(z)). Since, h 1(0) = 0 and 91 E Star(~. r), we have that h 1 is also starlike, as proved above. But again 9(z) = h 1(Mr(z)), and we are done. D

Remark 5.5.1. Generally speaking we have proved the following: If 91 and 92 are two functions related by the formula: (5.5.10) then also: (5.5.11) In addition 91 E Star(~. 0) if and only if 92 E Star(~, T). To prove Proposition 5.5.1 for the case when T E 8~ we need a more detailed treatment of fanlike functions. We will do this separately in the next section.

5.6

A conjecture of Robertson and geometrical characteristics of fanlike functions

The first note related somehow to an application of the class Fan(~) of starlike functions with respect to a boundary point to a boundary point seems to be due to E. Egervary [35] in connection with Cesaro sums

~n 'f.)n- k + 1)zk

of the

k=1

+ z 2 + z 3 + ... + zn + .... In particular, it turned out that the 1- ( 2 ) ~(n- k + 1)zk belong to Fan(~) with hn(1) = 0. n n+1 ~

geometric series z functions hn =

k=1

Strangely, although the classes Star(~, r) (of starlike functions with respect to interior points) have been studied by many mathematicians over a long period of time it seems that very few papers have been published up to 1981 on starlike functions with respect to a boundary point. A breakthrough in this matter is due to M.S. Robertson [121], who suggested the inequality 1+z} R { 2 zh'(z) (5.6.1) e ~+ 1-z >0, z E ~.

STARLIKE AND SPIRALLIKE FUNCTIONS

177

as a characterization criterion for those univalent holomorphic h : ~ ~ C with h(O) = 1 such that h(~) is starlike with respect to the boundary point h(1) := lim h(r) = 0 and lies in the right half-plane. This characterization was partially

r--+1-

proved by Robertson himself under the additional assumption that h admits a holomorphic extension to a neighborhood of the closed unit disk. Furthermore, he established that this class is closely related to the class of close-to-convex functions. In particular, if h satisfying (5.6.1) is not a constant with h(O) = 1, then g(z) = log h( z), log h(O) = 0, is close-to-convex with 2 h'(z)]

Re [ (1- z) h(z)

< 0.

(5.6.2)

Note, in passing, that because of Proposition 5.2.2 inequality (5.6.2) is exactly a characterization of functions in the class Fan(~) normalized by the condition h(1) := lim h(r) = 0. r--+1-

Different applications of these results to convex functions were also exhibited in Robertson's work (see also [57]). Observe, that the full proof of Robertson's conjecture was given by A. Lyzzaik [90], whilst a generalization of these results was later established by H. Silverman and E.M. Silvia [134]. In view of Caratheodory's Theorem of kernel convergence, a univalent function which is starlike with respect to a boundary point can be approximated by functions which are starlike with respect to interior points (see Figure 5.6 below). This approximation process can be considered dynamically as an evolution of the null points of these functions from the interior towards a boundary point. As mentioned above, this evolution is somehow connected to the evolution of semi-complete vector fields corresponding to starlike functions and the asymptotic behavior of one-parameter semigroups. So a natural question is how to trace these dynamics analytically in terms of inequalities (5.2.11'), (5.6.1) and (5.6.2). In this chapter we will mostly follow the works [134] and [37]. We will show that condition (5.6.2) is equivalent to a generalized form of Robertson's condition (5.6.1). In addition, we will relate these conditions to some geometric considerations in the spirit of Silverman and Silvia [134]. We begin with the following observation. Let g E Star(~, 0) and let h E Hol(~, C) be given by h(z) = (1- z)2g(z). (5.6.3)

z

Consider the functions

hn(z) = HrJz) ( -g(z)), where

Hr..(z) = (z- Tn)( 1 - z'fn"),

Tn E ~'

z

are Hummel's multipliers. Since -g also belongs to Star(~, 0) we have by Proposition 5.5.1 that hn E Star(~, Tn)· Letting Tn E ~ tend to 1 we obtain that hn

Chapter V

178

uniformly approximates h on each compact subset of D., so h(D.) is expected to be starshaped. In addition it follows by Hurwitz's theorem [68] that h is univalent. Thus one may conjecture that condition (5.6.3) is also a characterization of those univalent holomorphic h: D. f-+ C such that h(D.) is starshaped with h(1) := lim h(r) = 0, and hereby should be related to conditions (5.6.1) and (5.6.2). r-1z ) 2 h(z) Indeed, if (5.6.1) holds then setting (because of (5.6.3)) g(z) = (

1-z ·

we obtain Re zg'(z)

g(z)

Re{zh'(z) + l+z} h(z) 1- z

1R { zh'(z) 1+z} 1R 1+z 0 2 e 2 ~+1-z +2 e1-z>' hence g E Star( D., 0). This sketches another proof that an h E Hol(D., q satisfying (5.6.1) ought to belong to Fan(D.). At the same time, if we keep in mind that (5.6.1) characterizes those h E Fan( D.) with h(O) = 1, such that h(D.) lies in the right half-plane, we reason that a stronger conjecture, namely that (5.6.1) is equivalent to (5.6.3), should be refuted. So it might be worthwhile to replace (5.6.3) by a more qualified condition as well as to replace (5.6.1) by a generalized inequality, which are both related to the same geometrical location of the image h(D.). Following [134] for A E [0, 1) we define the class G>. of nonvanishing holomorphic functions h : .6. f-+ C with h(O) = 1 which satisfy the condition:

1 zh'(z) Re [ 1 _ A ~

+

1+z] 1_ z

> 0 for all z

E D..

(5.6.4)

First we will study some properties of the classes G >.,

Lemma 5.6.1 For each A E [0, 1) the set

is dense in G >. in the topology of uniform convergence on compact subsets of D..

Proof. Since the function

p(:.:) = _1_ zh'(z) 1-A h(z)

+ 1+z 1-z

belongs to the class of Caratheodory: {p E Hol( D., C), p( 0) = 1, Rep( z) > 0, z E D.}, it follows from the Riesz-Herglotz representation of this class that:

p(z) =

1.

1 + z(

- - - dm((),

1(1=1 1- z(

z E .6.,

STARLIKE AND SPIRALLIKE FUNCTIONS

179

where m is a probability measure on the unit circle. Approximating the left hand side of the latter equation by the integral sums: n

L D.mj = 1,

(j E aD.,

j=l

and solving the initial value problem _1_ zh~(z) _ 1- >. hn(z) -

8

( ) z '

hn(O) = 1,

where

s z)(

~

-~

(1 + z(j - 1 + z) D.m·- 2 ~ (j -1 D.m· 1 1-z(j 1-z ~(1-z( )(1-z) 1 '

1

we obtain our assertion with >.1

= 2(1 - >.)D.m1 ,

j

= 1, ... , n. 0

To continue we recall that a holomorphic function 9 : D. starlike of order >. E [0, 1) if 9(0) = 0, 9'(0) = 1 and Re (

~---+

C is said to be

z:~~~))

> >. for all

zED. (see Section 5.2). The set of such functions will be denoted by S*(>.).

The foregoing simple but important fact is due to Silverman and Silvia [134]. Lemma 5.6.2 Let>. E [0, 1) and let 9 and h be holomorphic functions in D. related by the equation 9(z) := z(1- z) 2J..- 2h(z). (5.6.5)

Then 9 E S*(>.) if and only if hE G;... Proof. This fact follows immediately by the equation

z9'(z) ->.=( 1 ->.){-1_zh'(z) + 1+z} 9(z) 1->. h(z) 1-z and the relations g(O) = 0, g'(O) 2mm

= h(O).

(5.6.6)

0

Proof. Assuming that h E G ;.. 2 we obtain by Lemma 5.6.2 that g defined by (5.6.5) with >. = >.2, belongs to S*(.A 2). Let us now denote

9(z) g(z) = (1- z)2(J..2-J..J) · We have by (5.6.5) g(z) = g(z) := z(1- z) 2J.. 2- 2h(z) = z(1- z) 2J.. 1 - 2h(z).

(5.6.7)

180

Chapter V

Thus, again in view of Lemma 5.6.2 we have to show that on account of (5.6.7) we calculate

g E 8*(>, 1 ).

Indeed,

R ( zg'(z)) e

g(z)

g(O) = 0, and we have completed the proof. 0

Exercise 1. Show that Re _z_ < ~' z E z-1-2

~­

Exercise 2. Prove that if 0:::; A1, ..\2 < 1, then hE 0>. 1 if and only if

A natural question related to the last lemma is: a given h of the class G.\ for some A E [0, 1) find the maximal ,.\ = ).. *, such that h also belongs to G>.·. In other words, G >.• should be the minimal class which contains h. It turns out, that this question is closely connected to another one: if h is a starlike function with respect to a boundary point, how does one determine the minimal angle () such that h(~) lies in the wedge of this angle. In [37] it was discussed how to resolve the above questions.

Proposition 5.6.1 Let h: ~ ~----+ C be holomorphic and let A E [0, 1). If h is not a constant and h(O) = 1, then the following conditions are equivalent. . [ 1 z h' ( 1+ {z) Re 1 _A h(z) + 1 _ z > 0 for all z E ~-

z)

z]

{ii} There exists a starlike function g : ~ h(z) =

(1- z) 2 z

{iii} The function h belongs to

2 >.

Fan(~)

~----+

g(z)

C of order ).. such that ,

z E ~-

with h(1) := lim h(r) = 0 and

h(~)

r--->1-

lies in a wedge of the angle 211"(1- >.). {iv) The function his a univalent function on~ such that f(z) := h(z)/ h'(z), z E ~, is a semi-complete vector field with

L.lim Re f(z) > - 1z---> 1

Z -

1 - 2 - 2).. '

where the limit in the left hand side of the latter inequality exists finitely. Moreover, the equality in (iv) can be reached if and only if)..=)..*, where G>.• is the minimal class of G.\ which includes h, and if and only if the wedge of the angle 211"(1- )..*) is the smallest one which contains h(~).

STARLIKE AND SPIRALLIKE FUNCTIONS

181

Proof. The equivalence of conditions (i) and (ii) is the content of Lemma 5.6.2. Our next steps are slightly simpler than in [37]. First we claim that condition (ii) and Lemma 5.6.3 imply that hE G>.. \ {1} must be univalent. Indeed, setting in this lemma .>. 1 = 0, .>. 2 = .>. ~ 0 we obtain that h admits a similar representation as (5.6.3):

h(z) = (1- z)2g(z)

(5.6.3')

z

with some

gE

S*. Then, as mentioned above, the approximation process:

) h n( Z ) _- (Tn- z)(1- z'Tn")-( gz, z

implies the desired claim. Next we want to show that h belongs to this fact will be proved once we show that Re ((1- z) 2

Fan(~).

~((:n ~ 0,

To this end let 0 < r < 1, and define hr : ~

t-+

In view of Proposition 5.2.2

z E

~-

(5.6.8)


1- z )2(1->..) hr(z) := h(rz) ( - , 1- rz

z E ~-

If we use the corresponding function g E S*(.X) we can write equivalently that

hr(z)=

(g(~z)) (~)

(1-z) 2 (l->..)_

This last representation of hr shows that it belongs to G>... Its definition shows that hr ----> h as r ----> 1- and that hr is continuous on the closed disk ~- Therefore the claimed inequality will follow if we inspect it for hr and for z = eicp E 8~. Indeed, for such z we have Re

((1=

z) 2

h~(z))

hr(z)

Re [(1-z) 2 z

(zh~(z) + (1 --\) 1+z) + (1-z) 2 ( 1 --\) 1+z] hr(z)

1-z

= Re [ ( z - 2 + z) ( zh~~) + (1 - .>.)

~ ~ :) + (1 -

zh' (z) 1 + z] = 2(coscp- 1) Re [ hrr(z) + (1- .X) 1 _ z

1-z

z

.>.) (z - z)]

~ 0,

as claimed. The fact established in (5.6.8) and the Berkson-Porta formula mean, actually, that f : ~ t-+
Chapter V

182

is a semi-complete vector field. In addition, the same formula implies (because of the uniqueness property) that T = 1 must be the sink point of the semigroup S = { Ft = h- 1 (e-th(z))} t;::>:o generated by f (see Section 5.2). To proceed with the proof of the implication (ii):::}(iii) we show now that lim h(r) = 0. In fact, we intend to prove a more general condition:

r--+1-

L lim h(z) = 0.

(5.6.9)

z--+1

= Ft(zo), t 2: 0.

Fix any element zo in .6. and define Zt

We have Zt ____, 1 and

h(zt) = e-th(z) ____, 0, as t ____, oo.

(5.6.10)

Thus what we need to show is that h is bounded in any nontangential approach region f(l, k) = {z E .6.: II- zl ~ k(l- lzl), k > 1}. Returning to the representation (5.6.3') the latter fact is easily seen with the help of Koebe distortion theorem:

whenever z E f(l, k). Now by the Lindeli:if's Principle with (5.6.10) valid (5.6.9) results. To accomplish (iii) we need to show that h(D.) lies in a wedge of the angle 2rr(l -A). By Lemma 5.6.1 we may assume that his of the form

h(z) =

IT ( 1- z )

where IC1 1 = 1. (J

n

=f 1 and 2:.:: AJ J=1

>.J'

1 - z(j

J= 1

1-z

= 2(1- A). Each function Wj(z) := - 1 - Z(j

maps the open unit disk .6. onto a half-plane. In other words, Re (eif3Jwj(z)) > 0 for some f3J. n

Denoting

2:.:: Ajf3J by {3, we have for each

z E

.6.

j=l

largei,'3

n

n

j=l

j=1

IJ w~ 1 I= II: Aj (argeif

31 wJ)

I

n

<

2: AJ (~) = rr(l- A). j=I

Hence g( .6.) is contained in a wedge of the angle 2rr( 1 - A) as claimed.

STARLIKE AND SPIRALLIKE FUNCTIONS

183

Now following the idea suggested in [90] we will show that (iii)=?(ii). Let

ho(z) = h(z)6. Then h0 (0) = 1, ho(1) = 0, ho is univalent and ho(~) is starshaped with respect to ho{1)

Dn =

= 0. Set

ho(~) U { z E C : lzl < ~},

n = 1, 2, ... ,

and for each n let hn : ~ 1-+ Dn be the conformal mapping of that hn(O) = 1 and argh~(O) = argh~(O) (see Figure 5.6).

~

onto Dn such

Figure 5.6: An approximation of a starshaped domain with respect to a boundary point. By Caratheodory's Kernel Theorem we know that lim hn = ho,

n-+oo

uniformly on each compact subset of~. Since each hn(~) is starshaped there are starlike functions 9n with 9n(O) = 0 and numbers Tn, lrnl < 1, such that

hn(z) (see Section 5.5). ~ote that 1 = hn(O)

=

=

9n(z) _ - - ( z - Tn)(1- Tnz), z

-Tng~(O)

z E ~.

and that

h~(O) = ~ :~~~~ + g~{0)(1 + lrnl

2

)

for all n. If the sequence

{g~(O)}

a contradiction, because

h~(O)--+ h~(O) and ~g~(O)/g~(O)I:::; 4.

had been unbounded then we would have reached

184

Chapter V

Thus {g~(O)} is bounded and we can extract a convergent subsequence of {gn}.It is clear that we can assume that the corresponding subsequence of {Tn} converges to a point T E D.. Denoting the limit function of the convergent subsequence of {gn} by g0 , we see that

go(z) _ ho(z) = - - (z- r)(1- rz), z Letting z approach 1 we conclude that

= 1.

T

zED..

Hence

9o(z)) (1- z) 2 ho(z) = ( - -zand

h(z) = ( - 9o;z)) 1->. (1- z)2-2A, where go : D.

f--+

C is starlike with go(O)

g(z) := z (- go;z))

= 0. 1

Since the function

->. =

z(1- z) 2>.- 2 h(z)

is starlike of order >. we obtain (ii), as claimed. (iii)=>(iv). Let the smallest wedge in which h(D.) lies be of an angle 27r(1-.A*). Then >. * ;::: >., Re [-1- zh'(z) 1- ,\* h(z)

+ 1 + z] > 0 1- z

'

z E ~,

and this inequality no longer holds when >. * is replaced with any number ,\ * < >. < 1. By the Riesz-Herglotz representation theorem we can write

_1_zh'(z) 1-,\1 h(z)

+

1+z = 1-z

J 1+z~dm((), 1-z(

1<1=1 where m is a probability measure on the unit circle. After some calculations we obtain

h(z) = (1- z) 2 (I->.·) exp (-2(1- >.*)

j

log(1- z()dm(()) .

1<1=1 Again we note that (5.6.9) no longer holds when .A* is replaced with any number .A* < .A < 1. Let 8 denote the Dirac measure at ( = 1 E aD.. Decomposing m relative to 8, we can write m = (1- a)v + a8, where 0 ~a ~ 1, and v and 8 are mutually singular probability measures. It follows that

h(z) = (1- z) 2 {l->.) exp (-2(1- .A)

j 1<)=1

log(1- z()dv(()) ,

STARLIKE AND SPIRALLIKE FUNCTIONS

185

where>.= 1- (1- ).*)(1- a). If a > 0 we reach a contradiction because >. > >. *. Thus a = 0 and m = v. Let g = hjh'. Then f is semi-complete by Proposition 5.2.4. Using (5.6.8) or (5.6.9) we see that z-1

f(z)

=

2(1- >.*)

J

1-(

---dv((), 1- z(

z E

Do.

1(1=1 Let { z,} be any sequence in f(1, k) ={zED.: [1- z[

~

k (1- [z[), k > 1}.

which tends to 1. Consider the functions hn : 86.

hn(() :=

1-( 1- Zn(

,

f-+

C, n = 1, 2, ... , defined by

( E 86..

Since the function hn maps the unit circle 86. onto the circle where Cn = (1- z11 )/(1- [zn[ 2 ), n = 1, 2, ... , we obtain that

[~

-

Cn [

= [en [,

[hn(()[ ~ 2[cn[ ~ 2k. Using (5.6.10) and applying Lebesgue's Bounded Convergence Theorem we now obtain z-1 L.lim - !Z( ) Z-+1

T

2(1- >.*) lim { hn(()dv(() n-+OO }1(1=1 2(1- >.*) ~ 2(1- >.).

(5.6.11)

In other words, condition (d) holds. Finally, we show the implication (iv):::;.(iii). Note that by Proposition 4.6.2, = 1 is the sink point of the semigroup generated by f and L.lim f'(z) is, in fact, z-+1

a real number. Therefore L. lim f(z) > 1 z-+1 Z - 1 2(1 -A).

h(z)

Moreover, f(z) = h'(z) = -(1- z) 2 p(z), where p: .6.

f-+

C is holomorphic with

Rep( z) 2 0 for all z E .6.. Again applying Proposition 5.2.4 and repeating the arguments as in the proof of (5.6.9) we obtain that h is starlike with respect to a boundary point with lim h(r) = 0. Let the smallest wedge in which h(D.) lies be r-+1

of an angle 27r(1 - >.*), where ).* E [0, 1). As we saw in the proof of implication (iii):::;.(iv), it follows that L.lim f(z) = 1 z-+1 Z - 1 2(1 -A*)

(5.6.12)

Chapter V

186

Comparing the latter equality with (5.6.11) we see that>. :::; >.*. Thus h(A) lies in a wedge of an angle 211"(1 - >.), as claimed. This concludes the proof of our assertion. D

Remark 5.6.1. Thus given h E Fan(A) with lim h(r) = 0, formula (5.6.12) r--+1

infers that the value of smallest angle () such that h(A) lies in its wedge is multiplied by the angular limit of the Visser-Ostrowski quotient:

() _

/ 1.

-7l"Llm z->1

7r

(z- 1)h'(z) h() . Z

Corollary 5.6.1 If hE Fan(A) with lim h(r) = 0 satisfies the Visser-Ostrowski r--+1 condition: L. lim (z- 1)h'(z) = 1 z--+1 h(z) (in particular, if h is conformal, or, more generally, isogonal at 1}, then the smallest wedge which contains h(A) is precisely the right half-plane

II+

= {z

E C : Re z :;::: 0} .

Remark 5.6.2 Since G;.. C Go, it follows by the above proposition that h E Fan( A) with lim h(r) = 0 if and only if it satisfies the equation r---+1-

h(z) = -(1- z)2 g(z), z

where g E Star( A, 0). This proves Proposition 5.5.1 (Hummel's representation formula) for the case T = 1. Geometrically this fact should be understood as h(A) lies in the wedge of angle 271".

5. 7

Converse theorems on starlike, spirallike and fanlike functions

In this section we consider inter alia the following converse problem: given f E g Hol(A) find conditions such that the equation h(z) = h'(z)f(z)

(5.7.1)

STARLIKE AND SPIRALLIKE FUNCTIONS

187

has a global solution in b. which is univalent (consequently, starlike) in b.. If f(r) = 0 for some T E b. and we are looking for the solution of (5.7.1) satisfying the initial conditions: h(r) = 0 and h'(r)

i

(5.7.2)

0,

then the necessary restriction is that f' (T) = 1. More generally, if f'(r) = J.L -f. 0 then instead of equation (5.7.1) we must consider equation: (5. 7.3) J.Lh(z) = h'(z) · f(z) with initial conditions (5.7.2), which determines a spirallike function h. Proposition 5. 7.1 Let f E Q Hol(b.) be such that:

f(O) = 0

(5. 7.4)

j'(O) = f.L·

(5.7.5)

and Suppose that hE Hol(b.) is a solution of {5. 7.3}. Then: h(z) = h'(O) lim e~-'t · Ft(z),

(5.7.6)

t->CXJ

where {Ft}t>O is the semigroup, generated by f. Proof. Let {Ft}t~o = Sf be a semigroup generated by (5.7.3) that

or

f. Then it follows by

J.LGt(z) =- 8G~t(z)'

where Gt(z)

= h(Ft(z)), t

~ 0,

(5.7.7)

z E b. and

Go(z) = h(z).

(5.7.8)

Solving (5.7.7) with initial data (5.7.8) we obtain:

Gt(z) =

e-~-'

t · h(z)

or

h(Ft(z)) = e-1-' t · h(z).

(5.7.9)

Consequently, to prove our assertion, it is sufficient to show that lim e~-'t [h(Ft(z))- h'(O)Ft(z)] = 0 t->CXJ

for all z E b..

(5.7.10)

Chapter V

188

Note, that by Taylor's formula we have that for each w E

1 dnh n! dzn (0) wn = h(w).

00

h(w)- h'(O)w =

~

L n=2

If M

= sup h(w) then the Schwarz Lemma (see Section 1.1) implies lwl
for all wE~Setting w = Ft(z) we obtain

\h(Ft(z))\ ::=; M 1Ft(z)l 2 .

(5.7.11)

In addition, we know that

IFt(z)l <ex - Re t lzl (1 -1Ft(z)l) 2 p( IL) (1 -lzl) 2 (see Proposition 4.1.1). Hence (5.7.11) implies 1

Mlzl 2

\

h(Ft(z)) ::=; ( 1 -lzl) 4 · exp(-2Re!-Lt),

since (1- 1Ft(z)l) 2 ::=; 1. Finally, since Re IL > 0 we have

\eJLth(Ft(z))\

= exp(Re!-Lt) \h(Ft(z))\ Mlzl2

::=; ( 1 _ lzl) 4

as t

-+

·

exp(- Re ~Lt)

-+

0

oo and the result follows. 0

Thus if equation (5.7.3) has a solution hE

Hol(~)

satisfying the condition (5.7.12)

h'(O) = k-::/:- 0, then there is no other solution satisfying the same condition. Conversely, if the limit

h(z) := k lim eJLt Ft(z) t--+00

exists it is easy to see that h(z) satisfies (5.7.3) with (5.7.12). Indeed, in this case

h'(z)f(z)

. t 8Ft(z) t k hm e JL · -~- · f(z) = k lim e JL f(Ft(z)) t--+oo

u

z

t->oo

k lim etJL [ !-LFt(z) t---oo

l

1 dnJ +'"~ """' --(O)(Ft(zt) n! dzn 00

n=2

.

STARLIKE AND SPIRALLIKE FUNCTIONS

189

Again, as above, it can be shown that

.

oo1dnf

n

hm " " - -d (O)(Ft(z) ) t-+oo L.....J n.1 zn

= 0.

n=2

Hence, we obtain:

h'(z)f(z)

= t-tk t-+oo lim eti-L Ft(z) = t-th(z)

and we have completed the proof. 0 Thus to solve the problem (5.7.3)-(5.7.12) we only need to show that the limit in (5.7.6) exjsts. To this end we define for all t 2: 0 and z E /::i.

u(t, z) = eti-L Ft(z) Then for fixed

z E

L\,

au(t,z) = t-tei-Lt Ft(z)- e~-Ltf(Ft(z)) 8t = t-tu(t,z) -ei-Ltf(e-1-Ltu(t,z)) = ei-Lt (t-te-1-Lt u(t, z)- f(e-1-Ltu(t, z))) and u(O, z) = z. Therefore the function u(t, z): IR+

{

8u(t, z) at

--+

C satisfies the equation:

+ eti-Lg(e-1-Ltu(t, z)) =

0, (5.7.13)

u(O, z) = z, where 00

g(z) = f(z)

-j.LZ

=

L

1 dn J n! dzn (0) zn.

n=2

In turn, (5.7.13) implies: t2

u(tb z)- u(tz, z)

=-

J

eti-Lg(e-1-Ltu(t, z)) dt

tt

for each pair t1, t2 E JR+ and z E L\. It follows by (5.7.14) that g'(O) = 0. Hence, for each z E L\: Jg(z)J ~ M1 Jzl 2 , where M1 = supzE.o.Jg(z)J. Therefore

Jg(Ft(z))l ~ MlJFt(zW

< M e-2Re~J.t 1

JzJ 2 (1-JzJ)4"

(5.7.14)

Chapter V

190

Using (5.7.6) we obtain for each z E

~:

as t 1 and t 2 tend to oo. Therefore the limit lim u(t,z) =lim et~'Ft(z)

t -·HX>

t-+- CXJ

exists. D Thus we have proved the following assertion. Proposition 5. 7.2 Let f E Q Hol(~) be a semi-complete vector field which satisfies the conditions: f(O) = 0, and f'(O) = f-l· Then the equation

f-lh(z) = h'(z) · f(z) has a unique solution hE Hol(~) which satisfies the conditions: h(O) = 0, h'(O) = k "# 0. This solution is a univalent spirallike function and it has the form:

where { Ft} is the semigroup, generated by f. In particular, when f-l is a real number h( z) is a starlike function. Turning to a more general situation, assume now that

f(z) = (z- r)(1- zf)q(z), with Req(z) > 0 everywhere. We already know that for each an interior null point T representation (5.7.15) holds. In this case:

(5.7.15)

f

E QHol(~) with

(5.7.16)

If we define

g(z)

=

(Mr)'(Mr(z))f(Mr(z)),

we obtain g(O) = 0 and f-l = g'(O) = (1 Therefore the equation

with the initial conditions h(O) = 0,

-ITI 2 )q(r)

h~ (0)

= f'(r) and g E Q Hol(~).

"# 0 has a

unique solution:

h1(z) = h~(O) ·lim e~'tGt(z), t~oo

where {Gt(z)}t>O = S9 is the semigroup generated by g. Define now (5.7.17)

from which we obtain h(r) = h 1 (0) = 0 and h'(z) =

h~(Mr(z))

· (Mr)'(z).

STARLIKE AND SPIRALLIKE FUNCTIONS

191

In particular, (5.7.18)

In addition,

fl-hl(Mr(z)) = fl- · h~(Mr(z))g(Mr(z)) fl-hl(Mr(z)) · (Mr)'(z)f(z) = fl-h'(z)f(z).

11-h(z)

(5.7.19)

Note now that if {Ft}t>o = St is the semigroup generated by f, then Gt(z) = Mr(Ft(Mr(z))) is the semigroup generated by g (recall, that Mr is an involution), and we have two symmetric relations:

(i) Mr(Gt(z)) = Ft(Mr(z)); (ii) Mr(Ft(z)) = Gt(Mr(z)). Since, h(z) = h 1 (Mr(z)) and h 1 (z) h~(O)

h(z)

= h~(O) · t-->CXJ lim e~-'tGt(z),

we obtain:

· lim e11 tGt(Mr(z)) t-->CXJ

h'(r)(irl 2 -1) · lim e11 t Mr(Ft(z)) t-->CXJ

t r-Ft(z) -----="":-'-:1 - r Ft ( z) h'(r) lim e11 t(Ft(z)- r). 2

h'(r)(irl - 1) · lim e11 t-->CXJ

t-->CXJ

This formula gives us the solution of (5.7.3) under the conditions h(r) 0, h'(r) =1- 0. D So, we arrive at the following conclusion:

Proposition 5.7.3 Let f E

QHol(~)

have the form {5. 7.15)

f(z) = (z- r)(1- zf)q(z) with Re q(z) > 0, 1' E

~

and q(O) =1- 0. Then the equation 11-h(z) = h'(z) · f(z)

with 11- = (1 -lrl 2 )q(O) has a unique solution h(z) satisfying the conditions: h(r) = 0,

h'(r) = k =1- 0.

This solution is a univalent spirallike function on the formula:

~

which can be defined by

If, in particular, q( 1') is real then h is a starlike function on

~.

Now we consider the case when f E g Hol(~) has no null point in ~. It looks like this case is simpler, because equation (5.7.3) has no singularity in~' and one

Chapter V

192

can define a local solution of (5.7.3) for each p, E C under a corresponding initial condition, say (5.7.20) h(O) = 1, and the boundary condition (5.7.21)

lim h(r7) = 0,

r-+1-

where 7 E 8!:l. is the sink point of the semigroup generated by f (see Sections 5.2 and 5.3). But, on the other hand, it is not clear why such a solution has an extension to all of !:l. to be a univalent (spirallike) function (with respect to a boundary point). Moreover, if we wish this solution to satisfy the Visser-Ostrowski condition at the point 7 E 8!:::. we must require for the number p, in (5.7.3) to be real.

Proposition 5. 7.4 Let f be a semi-complete vector field in !:l. with no null point inside, and let 7 E 8!:l. be the sink point of the semigroup generated by f. If (3 = L lim f' (z) then for a real p, > 0 the problem Z-+1"

J1h(z) = h'(z)f(z), h(O) = 1

(5.7.22)

has a unique univalent solution in !:l. if and only if

(5. 7.23)

J1, ::; 2(3.

This solution is a fanlike function (i.e., a starlike function with respect to a boundary point), the image of which lies in the wedge of angle () = 1rp,j {3.

Proof. Without loss of generality we can assume that 7 = 1. If h E Univ(!:l.) satisfies (5.7.22) with some real p, > 0, then its image h(!:l.) must be starshaped by Proposition 5.2.4. Furthermore, if we define fi(z)

f3I

:= L lim Z-+1"

=..!:.{t f(z)

we have

f~ (z) = ~jJ, > 0

and h satisfies the equation

h(z) = h'(z)fi(z) with h(O) = 1. Then it follows by Proposition 5.6.1 that there is >. E [0, 1) such that h belongs to the class G:.. and (31 := L lim f~(z):::: ~:::: -12 . This implies Z-+1" 2- 2/\ (5. 7.23). In addition the angular limit of the Visser-Ostrowski quotient is

v = L lim h'(z)(z- 1) = _!_ = !!.. z-+1h(z) (31 (3'

7f·

and the smallest wedge which contains h(!:l.) is of the angle Conversely. To solve (5.7.22) we first consider the problem:

(Jh(z)

= h'(z)f(z), h(O) =

1,

(5.7.24)

STARLIKE AND SPIRALLIKE FUNCTIONS where

193

f3 = L lim f'(z). z~l-

Consider the functions:

fn(z)

1 = -z + f(z), n

Firstly, it is known that fn E

n

g Hol(t..)

= 1, 2, ....

for all n

(5. 7.25)

= 1, 2, ... ,

since the class

g Hol(t..) is a real cone. Secondly, for each n = 1, 2, ... the equation fn(z) = 0

(5.7.26)

has a unique solution Tn E t. such that Tn ----> 1-, as n----> oo. Indeed, equation (5.7.26) is equivalent to the following one

z+nf(z)=O, which defines the values of the resolvent ln at the point zero, i.e., Tn = ln(O).

If we denote Jln = .!_ + J'(rn) we obtain that Jln----> f3 as n----> oo and f~(rn) = n Jln· Therefore, by the above Proposition 5.7.3, for each n = 1, 2, ... the equation (5.7.27) has a univalent solution h 11 (z) determined by h~,(rn) 1=- 0. Since Tn 1=- 0 for all n = 1, 2, ... and hn is univalent we have h 11 (0) 1=- 0 for all n = 1, 2, .... Therefore, we can define the functions:

which also satisfy equation (5.7.27), with h~(rn) = hn1(o) In addition, h,.(O) = 1 n = 1,2, ....

·

h~(rn)­ (5. 7.28)

Now for each r E [0, 1) we can find N > 0 such that for all n > N, lrnl > r, that is fn do not vanish on the disk lzl ~ r. Therefore for such z (i.e., lzl ~ r) we can write by using (5.7.27) and (5.7.28):

Furthermore, siuce f(O) 1=- 0 there is a neighborhood U of the point z = 0 in which (5.7.24) has a unique solution:

-h(z)

= exp

{JL • Jor f(z) dz } .

(5.7.29)

Since Jln ----> f3 and fn(z) ----> f(z) for all z E t.., we have that hn(z) converges to h(z) in this neighborhood. It then follows by the Vitali theorem that hn(z)

194

Chapter V

converges to h(z) on all of the disk JzJ < r. Since r is arbitrary we obtain that ~(z) is well defined on all of~. By the Hurwitz theorem (see, for example, [55]) h is univalent on ~. Now again by Proposition 5.6.1 we have that h is a fanlike function the image of which lies in the wedge of angle 1r. Therefore the function h defined as -

h(z) = [ h(z)

] p.f f3

is a univalent function on~ whenever (5.7.23) holds. On account of (5.7.24) it is easy to see that h satisfies (5.7.22). D

Remark 5.7.1 In the proof of the above proposition we have used an approximation process for the generator f (see formula 5.7.25).:_ In turn, this p_:ocess induces an approximation sequence of univalent functions hn converging to h with interior null points defined by (5.7.27). However, in general we can claim only that these functions hn are spirallike, but not necessarily starlike. In fact, we do not know whether the numbers /-Ln in (5.7.27) are real.

5.8

Growth estimates for spirallike, starlike and fanlike functions

The famous Koebe distortion theorem asserts: if hE S ={hE

Univ(~):

(1

h(O) = 0, h'(O) = 1} then

JzJ < Jh(z)l < + Jzl) 2 -

(1

JzJ -lzJ) 2

for all

z E

~.

Equality holds for the Koebe function 00

hi<(z) = (1

~ z)2 = L

nzn.

n=l

Usually the proof of this fact is based on the known bound for the second coefficient in Taylor's expansion of hE S, namely Ja2l < 2. Note that the Bieber bach conjecture, namely that Jan I :S n for each h E S, an =

~! ~:~ (0),

n = 1, 2, ... , was proved by L. de Branges [20] in his remarkable work S*(~) it has been proved earlier by R. Nevanlinna [103].

in 1985, whilst for hE

STARLIKE AND SPIRALLIKE FUNCTIONS

195

Although analogous of the Koebe distortion theorem fail for biholomorphic mappings in the polidisks (or balls) of dimensions greater than 1, it is still relevant for many special cases, in particular, starlike and spirallike mappings. Again as in previous sections one can use the relationships between these classes and semigroups to obtain corresponding bounds for these classes (see, for example, [109, 26]. Moreover, even in the one-dimensional case, by using some characteristics of semi-complete vector fields one can improve the estimates for some subclasses of Spiral(~). 1. We already know that if hE Hol(~, q is a spirallike function on~' with

h(r)

=

0,

h'(r) =a=/= 0,

T

E ~'

then h satisfies the equation

p,h(z) = h'(z)f(z)

(5.8.1)

for some semi-complete vector field f E Hol(~, q with f(r) = 0, f'(r) = p, E C. In addition, if {Ft} = Sf is the semigroup generated by f, then h can be defined by the formula

h(z) =a lim eJ.Lt · (Ft(z)- r) t---+oo

(5.8.2)

(see section 5.8). First let us suppose that T = 0. It follows by Proposition 4.4.2 and Remark 4.4.4 that Ft satisfies the following estimate:

with some c E [0, 1]. Moreover, iff is a bounded p-monotone function in ~then c can be chosen strictly less then 1. This inequality and (5.8.2) immediately imply the following estimates for h: (5.8.3) For the general case, when T =/= 0, lv!,:

T

E ~one can use the Mobius transformation T-Z

M,(z) = --_, 1- ZT

defining the spirallike mapping h = h o /1.1, (note that the image h(~) = h(~)). Since h(O) ~ h(r) = 0 and h'(O) = h'(r)(irl 2 - 1) we obtain by (5.8.3) and replacing h by h,

lh'(r)j(1- lrl 2 ) ( 1 J:llzl) 2 < ih(M,(z))i

< lh'(r)l(1- lrl 2 ) ( 1 ~:lizl) 2

(5.8.4)

196

Chapter V

or (replacing z by M.,.(z))

lh'(r)l(1-lrl 2) d(r,z) · (1 + cd(r, z)) 2

~

lh(z)l (5.8.5)

where d(r, z) = IM.,.(z)l is the pseudo-hyperbolic distance on !:l. Thus we have proved the following assertion. Proposition 5.8.1 Let h E Hol(!:l, C) be a spirallike {starlike) function on !:l, with h(r) = O,r E !:l. Then h satisfies estimates {5.8.5). Moreover, if h is strongly spirallike, then c can be chosen to be strictly less than 1. Remark 5.8.1. To obtain a growth estimate for h E Spiral(!:l, r) in a form which does not contain a Mobius transformation we recall that the sets

D.,.(K)

= { z E !:l: cp.,.(z) =

l1-zrl 2 2} 1 -lzl 2 < K, K > 1- lrl

are Ft-invariant for all t 2:: 0, and h belongs to Spiral(D.,.(K), r) for each K > 1 - lrl 2. Therefore it may be convenient to rewrite the right hand side of (5.8.5) in terms of the function cp.,.. By simple calculations we obtain

< (1- lrl 2)(1 + ciM.,.(z)l) 2 (1- ciM.,.(z)l) 2

=

(1 -IM.,.(z)l2)2 (1 -lrl 2)(1 + ciM.,.(z)l) 2 . 11 _ -~4 (1- lzl2)2(1 - lrl2)2 zr 2 [ ( )]2 (1 + ciM.,.(z)l) cp.,. z 1 - lrl2 .

Now (5.8.5) implies (5.8.6)

2. If h is a starlike function on !:l, normalized by the conditions h(r) = 0,

h(O) =a E C,

(5.8.7)

then one can find an estimate of growth by using Hummel's representation formula (see Section 5.5): h(z) = H.,.(z)g(z), where

H ( ) _ (z- r)(l - zr) .,. z ' z

z E !:l,

STARLIKE AND SPIRALLIKE FUNCTIONS

197

and g E Star(~, 0). Indeed, it follows by the right hand inequality in (5.8.3) that

lh(z)l

lz- rill- zfll ( )I< lz- rill- zfl ·I '(O)I lzl g z (1 - lzl) 2 g =

<;?r(z) 1 + lzl 1-

lz- rl . lg'(O)I. lzl 11 - zrl

Similarly, by using the left hand inequality in (5.8.3) we obtain 1 - lzl lz - rl 1 lh(z)l;::: <;?r(z) 1 + lzl . ll- zrl ·lg (0)1.

Taking into account that

g'(O)

= -ra,

we obtain the following estimate: Proposition 5.8.2 Let h E Hol(~, q be a starlike junction on conditions {5.8. 7). Then the following estimate of growth holds:

lh(O)IIrl d(r, z) · <;?r(z)

~

normalized by

~ ~ ::: ~ lh(z)l ~

1 + lzl lh(O)IIrl d(r, z) · <;?r(z) -1z 1

.

(5.8.8)

1

3. Note that the latter arguments are relevant also when r is a boundary of Moreover, in this case (5.8.8) can be written in a more precise form. Indeed, using Proposition 5.6.l(ii) we have that if his a fanlike function on~, normalized by the conditions: h(l) = 0, h(O) = 1, ~.

then for some >. E [0, 1) it satisfies the equation

h(z) =

(1

-

)2-2>. z g(z), z

where g E S*(>.) (starlike of order>.), i.e.,

, R zg'(z) e g(z) > "· Using the latter inequality it is easy to see that

198

Chapter V

where g 1 belongs to S* (cf., Section 5.6). Then again by (5.8.3) we obtain

[

l1-zl]2-2>. [l1-zl]2-2>. + lzl) :::; lh(z)l:::; (1 -lzl)

(1

(5.8.9)

In particular, we obtain that his bounded in each nontangential approach region at T = 1. 0 Thus on account of Proposition 5.6.1 (see also Remark 5.6.1) and the obvious inequality 1-lzl :::; 11- zl :::; 1 + lzl, we obtain from (5.8.9) the following distortion theorem. Proposition 5.8.3 (see (134]) Let h be a fanlike function on .6. normalized by the conditions h(1) = 0, h(O) = 1,

and let the image of h lie in the wedge of angle Then the following estimates hold :

1r11,

0<

11 :::;

2.

-lzl] v:::; lh(z)l:::; [1 + lzl] v [1 1 + lzl 1 -lzl 1_

z] v

These estimates are sharp for the function [ - l+z

5.9

Remarks on Schroeder's equation and the Koenigs embedding property

The so called Schroeder's (functional) equation:

h(
= )..h

(5.9.1)

has been studied since the late nineteenth century (see, for example, [28] and references there). Here

c
h = h(
(5.9.2)

STARLIKE AND SPIRALLIKE FUNCTIONS

199

on the space Hol(b., q (or its relevant subspaces, in particular, Hardy spaces HP). In this section we shall discuss some relations of Schroeder's equation with spirallike and starlike mappings on b.. We assume that cp has an interior fixed point a E b.. Of course, we will exclude the trivial cases when cp is a constant or the identity. In addition, we will assume that cp is not an elliptic automorphism of !:!.. In other words, our condition for cp is that a E b. is an attractive fixed point of it, or equivalently (5.9.3) lcp'(a)l < 1. We will see below that in this case, in fact,

cp'(a) # 0

(5.9.4)

is a necessary condition for the global solvability of (5.9.1). First we note that if Schroeder's equation has a nontrivial solution (i.e., h is not constant), then >. is neither 0 nor 1. Indeed, if>. = 0 then h(cp(z)) = 0 for z E !:!. and h = 0 by the uniqueness theorem. If >. = 1 then h satisfied the equations h(cp(n))

= h,

n

= 1, 2, ...

,

and we have

h(z)

=

lim h(cp(n)(z))

n->oo

= h(a)

because of assumption (5.9.2). Thus we have to assume that >. is neither 0 nor 1. Note that if>.# 1 then h(a) = h(cp(a)) = >.h(a), and we have

h(a)

=0

(5.9.5)

as a solution h of Schroeder's equation (5.9.1). Hence h must have the form 00

h(z) =

L an(z- a)n,

k 2: 1.

(5.9.6)

n=k

For simplicity we put a= 0. Then (5.9.5) and (5.9.1) imply

Since the left hand side of this equality is a constant we obtain, letting z go to zero, that >. = [cp'(O)]k. (5.9. 7) Since >. # 0 we obtain (5.9.3). D Thus we have proved the following assertion.

200

Chapter V

Proposition 5.9.1 Let 'P E Hol(~) have an attractive fixed point a E ~ and suppose that Schroeder's equation (5.9.1} has a nontrivial solution hE Hol(~, C) for some A E C. Then

(i) h(a) = 0; (ii) A= ['P'(O)]k =f. 0 for some positive k; (iii) if h is locally univalent, then A = 1p1 (a) =f. 0. In 1884 Koenigs proved a remarkable result on the solvability of Schroeder's equation.

Proposition 5.9.2 (see [78]) Let 'P E Hol(~) have an attractive fixed point a E ~. such that 'P'(a) :=).=f. 0. Then there is a function hE

Hol(~,

h(1p(z))

C) such that

= Ah(z)

for all z E ~. In addition, if 'P is univalent then so is h. The idea in the original proof of Koenigs's theorem is based on the convergence of the sequence

h (z) = 'P(n)(z) = n A"

z

+ L......, ~ a(n) zk k ,

n = 1, 2, ... ,

k=2

which evidently satisfies the recursion equation (5.9.8) Its limit function his called the Koenigs function and it is normalized by the condition h'(a) = 1.

If such a function can be found one can present (5.9.9) whenever the right hand side of this equality is well defined. Hence

'P(n)(z) = h- 1 (Anh(z)).

(5.9.10)

The latter expression then serves as a definition of fractional iterations of 'P when n is not an integer (cf., section 4.4) and large enough. Of course, if 'P can be embedded in a globally defined continuous semigroup of holomorphic mappings it must be univalent, and so should h. Although the discrete semigroup of iterates of 'P cannot be embedded, in general, in a continuous semigroup of holomorphic self-mappings of ~. depending on what one requires the answer may be yes in some suitable cases (see for example, G. Srekeres [138], J. Hadamard [59], T. Harris [63], C.C. Cowen [27]). The simplest case (though very useful) is, of course, when 'Pis a fractional linear transformation and so is h. We consider now, in some sense, the dual problem.

STARLIKE AND SPIRALLIKE FUNCTIONS

201

Definition 5.9.1 We will say that a mapping cp E Hol(.6.) satisfies the Koenigs embedding property (K.e.p.) if its iterations cp(n) : .6. ~ .6. can be embedded in a continuous semigroup {Ft}t?:O of holomorphic self-mappings of .6., i.e., F 1 = cp. It turns out that the answer to the question of what are the conditions for a univalent self-mapping cp E Hol(.6.) to satisfy the (K.e.p.) is related to some geometrical properties of the solution of Schroeder's equation.

Proposition 5.9.3 Let cp be a univalent self-mapping of .6., and let cp(r) = T for some T E .6. with 0 < lcp'(r)l < 1. Then cp satisfies the Koenigs embedding property if and only if its Koenigs function is J.L-spirallike, with J.L = -log cp1(T). Proof. Sufficiency. Suppose that equation (5.9.10) has a univalent J.L-spirallike solution with J.L = -log A. Then for each z E .6. and t;::: 0 element e-tJ.Lh(z) belongs to h(.6.). If we define

we obtain by (5.9.10) that

Necessity. Let cp satisfy the Koenigs embedding property, i.e., there is a semigroup {Ft}t?:O C Hol(.6.) such that

F1(z) = cp(z). Denote by

f = - ~ lt=O+ the generator of {Ft} and consider the equation J.Lh(z) = f(z) h'(z),

(5.9.11)

where J.L = - log A. Since J.L = f'(r) =f. 0 it follows by Proposition 5.7.2 that equation (5.9.11) has a univalent J.L-spirallike solution h which satisfies the equality

Setting here t

=

1 we obtain (5.9.10). D

If cp has no fixed point in .6. then it follows by the Julia-Cratbeodory Theorem that there is a unique point T E 8.6. such that L lim cp(z) = T and Z--+T

0 < L lim cp'(z) ~ 1. Z--+T

Similarly, as in Proposition 5.9.3, by using Proposition 5.7.3 one proves the following assertion:

Chapter V

202

Proposition 5.9.4 Let

on~

(5.9.12)

such that L lim h(z) Z--+T

= 0.

Remark 5.9.1 In fact, Proposition 5.7.3 and the Julia-Wolff-Caratheodory Theorem imply that if

0. Z->T

Another direct consequence of the above propositions is a result originally established by A. Siskakis (see [136]).

Corollary 5.9.1 Let Ft be a continuous semigroup of holomorphic self-mappings of~. and suppose that there is a point r E ~' such that lim Ft(z) = r, and t--+oo

L lim Re f'(z) Z->T

=f. 0, where f

= - ddF lt=o+.

t

Then there are .X E C, 0 < I.XI < 1, and h E

Hol(~, C)

such that for all t

~

0

(5.9.13)

h(Ft) = .Xh.

In other words, there is a solution of Schroder's equation which does not depend on t ~ 0.

In addition, the solution of (5.9.12) can be found by solving the differential equations (5.9.11) with J.L = -log .X.

Example 1. Let
=

z

yfz2-

e2(z 2 -

Koenigs embedding property with Ft(z) =

t

1)

. It is easy to verify that

yfz2- e2t(z2 -1)

, Ft(O) = 0, for all

~

0. Therefore, instead of (5.9.10) one can try to solve (5.9.12) with .X= cp'(O) = 1/e or (5.9.11) with J.L = 1. . dFt(z) 3 . Smce f(z) = -~ lt=o+= z- z equat10n (5.9.11) becomes h(z) = h'(z)(z- z 3 ).

Solving the latter equation we obtain that h(z) =

az

.J'f'=Z2' 1 - z2

a E C,

is a solution of equations (5.9.10) and (5.9.12). The converse scheme does work if we know in advance the solution of Schroeder's equation. In particular, it can always be solved for a fractional linear transformation of the unit disk with an interior fixed point. Indeed, let

STARLIKE AND SPIRALLIKE FUNCTIONS

203

fractional linear mapping of!:::.. with a fixed point T E !:::... Without loss of generality we can consider the case T = 0. Then cp can be written in the form az cp(z) =-b-. 1- z

(5.9.14)

We will investigate the sufficient and necessary conditions on coefficients a and b such that cp will satisfy the Koenigs embedding property. First we note that the condition lcp(z)l < 1 for alllzl < 1 imply lal

+ lbl :S

(5.9.15)

1.

In addition, we have cp'(O) =a. Then it is easy to see that the function z h(z) = 1- kz'

(5.9.16)

where k = _b-, is the Koenigs function for cp, i.e., h( cp( z)) = >..h( z) with ).. = a 1-a and h'(O) = 1. Thus we tend to find a condition which will force h to be univalent 1-1-spirallike function with 1-1 -log a. In other words, we have to check the inequality 1-1h(z) (5.9.17) Re h'(z) > 0. By (5.9.15) inequality (5.9.17) becomes Re1-1z(1- kz) > 0,

(5.9.18)

since by (5.9.15) lkl ::; 1 the latter condition can be rewritten in the form (5.9.19)

cos arg(11) ::=: k. In particular, this condition always holds when 1-1 is real. Thus we have proved

Proposition 5.9.5 Let cp(z) = ~b with lal + lbl :S 1, and 0 < lal < 1. Then 1- z cp satisfies the Koenigs embedding property if and only if cos ( arg (log

~))

::=: 11

~ a I·

(5.9.20)

In particular, if a is a real number then cp can be always embedded in a continuous semigroup of holomorphic self-mappings of!:::...

If we consider the mapping ¢( z) = ~b , as in Example 1 of Section 4.3, with 1- z a= ~ exp(i3.1) and b = ~'then it is easy to see that lal + lbl = 1 whilst condition (5.9.20) does not hold. Thus ¢ : !:::.. ~-----> !:::.. cannot be embedded in a continuous semigroup of self-mappings of!:::.. (see Figure 4.1).

204

Chapter V Exercise 1. Prove the equivalence of conditions (5.9.17) and (5.9.18).

Exercise 2. Under condition (5.9.19) find the semigroup Ft : A---+ A, t? 0, such that F 1 = cp, where cp has the form (5.9.14).

More detailed discussion of this approach for higher dimensions can be found in [76]. Because of our intention in a modest text to emphasize the dynamical flavor of the subject a big part of the theory which is primarily geometrical or functional analytic in nature has not been included. We refer the reader to books of A.W. Goodman [57], P. Duren [33], and J.H. Shapiro [131], which could be good guides to complete the knowledge in these topics. Also, to advance to futher study on the boundary behavior of holomorphic mappings we mention an excellent book of C. Pommerenke [107]. Finally, we point out that many applicable subjects as branching processes, optimization theory, functional calculus etc., mentioned in the Preface may motivate an investigation in this direction.

Bibliography [1] M. Abate, Horospheres and iterates of holomorphic maps, Math. Z. 198 (1988), 225-238. [2] M. Abate, Converging semigroups of holomorphic maps, Atti Accad. Naz. Lincei Rend. Cl. Sci. Fis. Mat. Natur. 82 (1988), 223-227. [3] M. Abate, Iteration Theory of Holomorphic Maps on Taut Manifolds, Mediterranean Press, 1989. [4] M. Abate, Iteration theory, compactly divergent sequences and commuting holomorphic maps, Ann. Scuola Norm. Sup. Pisa 18 (1991), 167-191. [5] M. Abate, The infinitesimal generators of semigroups of holomorphic maps, Ann. Mat. Pura Appl. 161 (1992), 167-180. [6] D. Aharonov, M. Elin, S. Reich and D. Shoikhet, Parametric representations of semi-complete vector fields on the unit balls in en and Hilbert space, Rend. Mat. Ace. Lincei, 10 4 (1999), 229-253. [7] D. Aharonov, S. Reich and D. Shoikhet, Flow invariance conditions for holomorphic mappings in Banach spaces, Math. Proceedings of the Royal Irish Academy, 99A (1999), 93-104. [8] L. Aizenberg, S. Reich and D. Shoikhet, One-sided estimates for the existence of null points of holomorphic mappings in Banach spaces, J. Math. Anal. Appl. 203 (1996), 38-54. [9] L. Aizenberg and A. Yuzhakov, Integral representations and residues in multidimensional complex analysis AMS, Providence, 1983.

[10] J.W. Alexander, Functions which map the interior of the unit circle upon simple regions, Annals of Math. 17 (1915), 12-22. [11] LA. Aleksandrov, Parametric Extensions in the Theory of Univalent Functions, Nauka, Moscow, 1976.

205

206

BIBLIOGRAPHY

[12] J. Ara:z.y, An application of infinite dimensional holomorphy to the geometry of Banach spaces, Lecture Notes in Math. 1267 (1987), 122-150. [13] V. Barbu, Nonlinear Semigroups and Differential Equations in Banach Spaces, Noordhoff, Leyden, 1976. [14] R.W. Barnard, C.H. FitzGerald and S. Gong, The growth and 1/4-theorems for starlike mappings in en, Pacific J. Math. 150 (1991), 13-22. [15] D.F. Behan, Commuting Analytic Functions Without Fixed Points, Proc. of the Americ. Math. Society 37 1 (1973), 114-120. [16] E. Berkson, R. Kaufman and H. Porta, Mobius transformations on the unit disk and one-parametric groups of isometries of HP, Transections A mer. Mathern. Soc. 199 (1974), 223-239. [17] E. Berkson and H. Porta, Semigroups of analytic functions and composition operators, Michigan Math. J. 25 (1978), 101-115. [18] S.D. Bernardi, Bibliography of Schlicht Functions, Mariner Publ. Co., Tampa, FL, 1983. [19] H. Brezis, Operateurs Maximaux Monotenes, North Holland, Amsterdam, 1973. [20] L. Brickman, IP-like analytic functions, I, Bull. Amer. Math. Soc. 79 (1973), 555-558. [21] C. Caratheodory, Untersuchungen tiber die konformen Abbildungen von festen und veranderlichen Gebiten, Math. Ann., 72 (1912), 107-144. [22] C. Caratheodory, Uber die Winkelderivierten von beschrankten Analytischen Funktionen, Sitzungsber.Preuss. Acad. Viss. Berlin, Phys.-Math. Kl.(1929), 39-54. [23] C. Caratheodory, Theory of functions of a complex variable, Chelsea, 1954. [24] G.-N. Chen, Iteration for Holomorphic Maps of the Open Unit Ball and the Generalized Upper Half-Plane of en, Journ. of Mathemat. Analysis and Appl. 98 (1984), 305-313. [25] P. Chernoff and J .E. Marsden, On continuity and smoothness of group actions, Bull. Amer. Math. Soc. 16 (1970), 1044-1049. [26] M. Chuaqui, Applications of subordination chains to starlike mappings in en, Pacific J. Math., 168 (1995), 33-48.

BIBLIOGRAPHY

207

[27] C.C. Cowen, Iteration and the solution of functional equations for functions analytic in the unit disk, Trans. Amer. Math. Soc. 265 (1981), 69-95. [28] C.C. Cowen and B.D. MacCluer, Composition Operators on Spaces of Analytic Functions, CRC Press, Boca Raton, FL, 1995. [29] M.G. Crandall and T.M. Ligett, Generation of semigroups of nonlinear transformations on a general Banach space, Amer. J. Math. 93 (1971), 265-298. [30] L. de Branges, A proof of the Bieber bach conjecture, Acta Math. 154 (1985), 137-152. [31] A. Denjoy, Sur !'iteration des fonctions analytiques, C. R. Acad. Scie. 182 (1926), 255-257. [32] S. Dineen, The Schwartz Lemma, Clarendon Press, Oxford, 1989. [33] P. Duren, Univalent Functions, Springer, 1983. [34] C.J. Earle and R. S. Hamilton, A fixed point theorem for holomorphic mappings, Proc. Symp. Pure Math., Vol. 16, Amer. Math. Soc., Providence, R.I. (1970), 61-65. [35] E. Egervary, Abbildungseigenschaften der arithmetischen Mittel der geometrischen Reihe, Math. Z. 42 (1937), 221-230. [36] M. Elin, S. Reich and D. Shoikhet, Complex Dynamical Systems and the Geometry of Domains in Banach Spaces, Technion preprint series. [37] M. Elin, S. Reich and D. Shoikhet, Dynamics of Inequalities in Geometric Function Theory, Journ. of Inequalities and Applications (excepted for publication). [38] M. Elin, S. Reich and D. Shoikhet, Holomorphically Accretive Mappings and Spiral-shaped Functions of Proper Contractions, Nonlinear Analysis Forum 5 (2000) 149-161. [39] M. Elin, S. Reich and D. Shoikhet, Asymptotic behavior of semigroups of p-nonexpansive and holomorphic mappings on the Hilbert ball, Technion preprint series, (2000). [40] M. Elin, S. Reich and D. Shoikhet, Asymptotic behavior of semigroups of holomorphic mappings, Progress in Nonlinear Differential Equations and Their Applications, Birkhauser Verlag, Basel, 42 (2000), 249-258.

208

BIBLIOGRAPHY

[41] M. Elin, S. Reich and D. Shoikhet, A semigroup approach to the geometry of domains in complex Banach spaces, Nonlinear Analysis, accepted for publication. [42] M. Elin and D. Shoikhet, Dynamic extension of the Julia-Wolff-Caratheodory Theorem, Dynamic Systems and Applications (accepted for publication). [43] Ky Fan, Julia's lemma for operators, Math. Ann. 239 (1979), 241-245.

[44] Ky Fan, Iterations of analytic functions of operators, Math. Z. 179 (1982), 293-298. [45] Ky Fan, Iterations of analytic functions of operators II, Linear and Multtilinear Algebra 12 (1983), 295-304. [46] F. Forelli, The isometries of HP, Canad. J. Math. 16 (1964), 721-728. MR 29 #6336. [47] T. Franzoni and E. Vesentini, Holomorphic Maps and Invariant Distances, North-Holland, Amsterdam, 1980. [48] I.I. Gikhman and A.V. Skorokhod, Theory of Random Processes, Nauka, Moskow, 1973. [49] I. Glicksberg, Julia's lemma for function algebras, Duke Mathematical Journal 43 2 (1976). [50] K. Goebel, Fixed points and invariant domains of holomorphic mappings of the Hilbert ball, Nonlinear Analysis 6 (1982), 1327-1334. [51] K. Goebel and S. Reich, Iterating holomorphic self-mappings of the Hilbert ball, Proc. Japan Acad. 58 (1982), 349-352. [52] K. Goebel and S. Reich, Uniform Convexity, Hyper·bolic Geometry and Nonexpansive Mappings, Marcel Dekker, New York and Basel, 1984. [53] K. Goebel, T. Sekowski and A. Stachura, Uniform convexity of the hyperbolic metric and fixed points of holomorphic mappings in the Hilbert ball, Nonlinear Analysis 4 (1980), 1011-1021. [54] J.L. Goldberg, Functions with positive real part in a half-plane, Duke Math. J. 29 (1962), 335-339. [55] G.M. Golusin, Geometric Theory of Functions of a Complex Variable, Amer. Math. Soc., Providence, RI, 1969.

BIBLIOGRAPHY

209

[56] S. Gong, Convex and starlike mappings in severol complex variables, Science Press, Beijing-New York & Kluwer Acad. Publ., Dordrecht-Boston-London. [57] A.W. Goodman, Univalent Functions, Vols. I, II, Mariner Publ. Co., Tampa, FL, 1983. [58] K.R. Gurganus, 4>-like holomorphic functions in en and Banach space, Trons. Amer. Math. Soc. 205 (1975), 389-406. [59] J. Hadamard, Two works on iteration and related questions, Bull. Amer. Math. Soc. 50 (1944), 67-75. [60] L.A. Harris, The numerical range of functions and best approximation, Proc. Camb. Phil. Soc. 76 (1974), 133-141. [61] L.A. Harris, On the size of balls covered by analytic transformations, Monatshefte fur Mathematik 83 (1977), 9-23. [62] L.A. Harris, S. Reich and D. Shoikhet, Dissipative holomorphic functions, Bloch radii, and the Schwarz Lemma, J. Analyse Math. 82 (2000), 221-232. [63] T.E. Harris, The Theory of Bronching Processes, Springer, Berlin, 1963. [64] U. Helmke and J.B. Moore, Optimization and Dynamical Systems, Springer, London, 1994. [65] J.A. Hummel, Multivalent starlike functions, J. Analyse Math. 18 (1967), 133-160. [66] J.A. Hummel, Extremal properties of weakly starlike p-valent functions, Trons. Amer. Math. Soc. 130 3 (1968), 544-551. [67] J.A. Hummel, The coefficients of starlike functions, Proc. Amer. Math. Soc. 22 (1969), 311-315. [68] A. Hurwitz and R. Courant, Funktionentheorie, Springer Verlag, Berlin, 1929. [69] M.E. Jacobson, Computation of extinction probabilities for the BellmanHarris branching process, Math. Biosciences 77 (1985), 173-177. [70] F. Jafari and K. Yale, Cocycles, coboundaries and spectra of compositions operators, Proc. of the Second Big Sky Conf. in Analysis, Eastern Montana College, Billings, MT (1994). [71] G. Julia, Extension nouvelle d'un lemme de Schwarz, Acta Math. 42 (1920), 349-355.

210

BIBLIOGRAPHY

[72] G. Julia, Principes geometriques d'analyse, Gauthier-Villars, Paris, Part I (1930), Part II (1932). [73] W. Kaplan, Close-to-convex schlicht functions. Michigan Math. J. 1 (1952), 169-185. [74] S. Karlin and J. McGregor, Embeddability of discrete time simple branching processes into continuous time processes, Trans. Amer. Math. Soc. 132 (1968), 115-136. [75] V. Khatskevich, S. Reich and D. Shoikhet, Complex dynamical systems on bounded symmetric domains, Electronic J. Differential Equations 19 (1997), 1-9. [76] V. Khatskevich, S. Reich and D. Shoikhet, Schroder's functional equation and the Koenigs embedding property, Nonlinear Analysis, accepted for publication. [77] V. Khatskevich and D. Shoikhet, Differentiable Operators and Nonlinear Equations, Birkhiiuser, Basel, 1994. [78] G. Koenigs, Recherches sur les integrales de certaines equations fonctionnelles, Ann. Sci. Ecole Norm. Sup. 1 (1884), 2-41. [79] M.A. Krasnoselskii and P.P. Zabreiko, Geometrical Methods of Nonlinear Analysis, Springer, Berlin, 1984. [80] Y. Kubota, Iteration of Holomorphic Maps of the unit Ball into itself, American Mathern. Society, 88 it 3 (1983). [81] T. Kuczumow, S.Reich and D.Shoikhet, The existence and non-existence of common fixed points for commuting families of holomorphic mappings, Nonlinear Anal., 43 (2001), 233-251. [82] T. Kuczumow and A. Stachura, Convexity and fixed points of holomorphic mappings in Hilbert ball and polydisc, Bull. Acad. Polon. Sci. 34 (1986), 189-193. [83] T. Kuczumow and A. Stachura, Common fixed points of commuting holomorphic mappings, Kodai J. Math. 12 (1989), 423-428. [84] T. Kuczumow and A. Stachura, Iterates of holomorphic and kD-nonexpansive mappings in convex domains in en, Advances in Math. 81 (1990), 90-98. [85] Y.D. Latuskin and A.M. Stepin, Weighted shift operators and linear extensions of dynamical systems, Uspekhi Mat. Nauk 46 (1991), 85-143.

BIBLIOGRAPHY

211

[86] E. Landau and G. Valiron, A deduction from Schwarz's Lemma, ].London Math. Soc. 4 (1929), 162-163. [87] N.A. Lebedev, The square principle in the Theory of Univalent Functions, Nauka, Moskow, 1975. [88] E. Lindelof, Memoire sur sertaines inegalites dans la theorie des fonctions monogenes et sur quelques properietes nouvelles de ces fonctions dans le voisinage d'un point singulier essentiel, Acta Soc. Sci. Fennicae 35 7 (1909). [89] K. Lowner, Untersuchungen iiber schlichte konforme Abbildungen des Einheitskreises, I, Math. Ann. 89 (1923), 103-121. [90] A. Lyzzaik, On a conjecture of M.S. Robertson, Proc. Amer. Math. Soc. 91 (1984), 108-110. [91] B.D. MacCluer, Iterates of Holomorphic Self-Maps of the unit Ball in Michigan Math. J. 30 (1983).

en,

[92] P. Mazet and J.P. Vigue, Points fixes d'une application holomorphe d'un domaine borne dans lui-meme, Acta Math. 166 (1991), 1-26. [93] P. Mazet and J.P. Vigue, Convexite de la distance de Caratheodory et points fixes d'applications holomorphes, Bull. Sc. Math. 16 (1992), 285-305. [94] P. Mellon, Another look at results of Wolff and Julia type for J* -algebras, J.Math. Anal. Appl. 198 (1996), 444-457. [95] P.R. Mercer, Another look at Julia's Lemma, Preprint, 1991. [96] P.R. Mercer, Proper Maps, Complex Geodesics and Iterates of Holomorphic Maps on Convex Domains inC, Contemporary Mathematics, 137 (1992). [97] P.R. Mercer, Sharpened Versions of the Schwarz Lemma, J. of Math. Anal. and Appl. 205 (1997), 508-511 Article No. AY975217. [98] P.R. Mercer, Note on a strengthened Schwarz-Pick inequality, Journ. Math. Anal. Appl. 234 (1999), 735-739. [99] P. Montel, Surles familles normales de fonctions analytiques, Ann. Sci. Ecole Norm. Sup. 33 3 (1916), 223-302. [100] P. Montel, Ler;ons suT les families noTmales de fonctions analytiques et leurs applications, Gauthier-Villars, Paris, 1927. [101] P. Montel, Ler;ons sur les Fonctions Univalentes au Multivalentes, GauthierVillars, Paris, 1933.

212

BIBLIOGRAPHY

[102] R. Nevanlinna, Uber die konforme Abbildung Sterngebieten, Oeversikt av Finska- Vetenskaps Societeten Ferhandlinger 63{A), No. 6, 1921. [103] R. Nevanlinna, Analytic Functions, Springer, New York, 1970. [104] J.A. Pfaltzgraff, Subordination chains and univalence of holomorphic mappings in en, Math. Ann. 210 (1974), 55-68. [105] J .A. Pfaltzgraff, Subordination chains and quasiconformal extension of holomorphic maps in en, Ann. Acad. Sci. Fen. Ser. A Math. 1 (1975), 13-25. [106] J.A. Pfaltzgraff and T.J. Suffridge, Close-to-starlike holomorphic functions of several variables, Pacific J. Math. 57 (1975), 271-279. [107] Ch. Pommerenke, Gottingen, 1975.

Univalent Functions,

Vandenhoech and Ruprecht,

[108] Ch. Pommerenke, On the iteration of analytic functions in a half plane I, J. London Mat. Soc. 19 (1979), 439-447. [109] T. Poreda, On generalized differential equations in Banach space, Dissert. Math. Vol. 310, Warsaw, 1991.

[110] V.P. Potapov, The multiplicative structure of J-contractive matrix functions, Amer. Math. Soc. Transl. 2 15 (1960), 231-243. [111] S. Reich, Averaged mappings in the Hilbert ball, J. Math. Anal. Appl., 109 (1985), 199-206. [112] S. Reich, Nonliear semigroups, holomorphic mappings, and integral equations, in Proc. Symp. Pure Math., vol. 45, Part II Amer. Math. Soc., Providence, RI (1986), 307-324. [113] S. Reich, The asymptotic behavior of a class of nonlinear semigroups in the Hilbert ball, J. Math. Anal. Appl. 157 (1991), 237-242. [114] S. Reich and I. Shafrir, Nonexpansive iterations in hyperbolic spaces, Nonlinear Analysis 15 (1990), 137-158. [115] S. Reich and D. Shoikhet, Generation theory for semigroups of holomorphic mappings in Banach spaces, Abstr. Appl. Anal. 1 (1996), 1-44. [116] S. Reich and D. Shoikhet, Semigroups and generators on convex domains with the hyperbolic metric, Atti. Acad. Naz. Lincei (9) 8 (1997), 231-250.

[117] S. Reich and D. Shoikhet, The Denjoy-Wolff theorem, Ann. Univ. Mariae Curie-Sklodowska 51 (1997), 219-240.

BIBLIOGRAPHY

213

[118] S. Reich and D. Shoikhet, Metric domains, holomorphic mappings and nonlinear semigroups, Abstr. Appl. Anal. 3 (1998), 203-228. [119] M.S. Robertson, On the theory of univalent functions, Ann. of Math., 37 (1936), 374-408. [120] M.S. Robertson, Applications of the subordination principle to univalent functions, Pacific J. Math. 11 (1961), 315-324. [121] M.S. Robertson, Univalent functions starlike with respect to a boundary point, J. Math. Anal. Appl. 81 (1981), 327-345. [122] W. Rudin, Real and Complex Analysis, McGraw-Hill, New York, 1974. [123] D. Sarason, Angular derivatives via Hilbert space, Complex Variables, 10 (1988), 1-10. [124] E. Schroeder, Uber itierte Funktionen, Math. Ann. 3 (1871), 296-322. [125] L. Schwartz, Analyse Mathematique, Hermann, 1967. [126] J. Serrin, A note on harmonic functions defined in a half-plane, Duke Math. J. 23 (1956), 523-526. [127] B.A. Sevastyanov, Branching Processes, Nauka, Moscow, 1971. [128] B.V. Shabat, Introduction to complex analysis, Nauka, Moscow, 1976. [129] I. Shafrir, Common fixed points of commuting holomorphic mappings in the product of n Hilbert balls, Michigan Math. J. 39 (1991), 281-287. [130] I. Shafrir, Coaccretive operators and firmly nonexpansive mappings in the Hilbert ball, Nonlinear Analysis 18 (1992), 637-648. [131] J.H. Shapiro, Composition Operators and Classical Function Theory, Springer, Berlin, 1993. [132] A.L. Shields, On fixed points of commuting analytic functions, May 16, 1963. [133] D. Shoikhet, Some properties of Fredholm mappings in Banach analytic manifolds, Integral Equations Operator Theory 16 (1993), 430-451. [134] H. Silverman and E.M. Silvia, Subclasses of univalent functions starlike with respect to a boundary point, Houston J. Math. 16 (1990), 289-299. [135] R. Sine, Behavior of iterates in the Poincare metric, Houston Journal of Mathematics 15 2 (1989).

214

BIBLIOGRAPHY

[136] A. Siskakis, Composition semigroups and the Cesaro operator on HP, Journ. London Math. Soc. 36 (1987), 153-164. [137] L. Spacek, Prispevek k teorii funkci prostych, Casopis Pest. Mat., 62 (1933), 12-19. [138] G. Srekeres, Regular iteration of real and complex functions, Acta Math. 100 (1958), 203-258. [139] J.F. Steffensen, Om sandsynligheden for at afkommet uddor, Matematisk Tidsskrift B1 (1930), 19-23. [140] T.J. Suffridge, Starlike and convex maps in Banach space, Pacific J. Math. 46 (1973), 575-589. [141] T .J. Suffridge, Star likeness, convexity and other geometric properties of holomorphic maps in higher dimensions, Complex Analysis (Proc. Conf. Univ. Kentucky, Lexington, KY, 1976}, Lecture Notes in Math. 599 (1977), 146159. [142] H. Upmeier, Jordan Algebras in Analysis, Operator Theory and Quantum Mechanics, CBMS, Reg. Conf. Ser. in Math., Vol. 67, Amer. Math. Soc., Providence, RI, 1986. [143] G. Valiron, Sur !'iteration des fonctions holomorphes dans un demi-plan, Bull. des Sci. Math. 55 2 (1931), 105-128. [144] E. Vesentini, Su un teorema di Wolff e Denjoy, Rend. Sem. Mat. Fis. Milano 53 (1983), 17-25. [145] E. Vesentini, Iterates of holomorphic mappings, Uspekhi Mat. Nauk 40 (1985), 13-16. [146] E. Vesentini, Semigroups of holomorphic isometries, Advances in Math. 65 (1987), 272-306. [147] E. Vesentini, Krein spaces and holomorphic isometries of Cartan domains, Geometry and Complex Variables, :Marcel Dekker, New York, 1991, 409-413. [148] E. Vesentini, Sernigroups of linear contractions for an indefinite metric,Met. Mat. ACC. Lincei 9 2 (1994),53-83. [149] G. Vitali, Sopra le serie di funzione analitiche, Rend. del R. Instituto Lombardo di Scienze e Lettere 36 2 (1903), 772-774.

BIBLIOGRAPHY

215

[150] J.K. Wald, On starlike functions, (Ph.D. Thesis), Univ. of Delaware, Newark, Delaware, 1978. [151] H.W. Watson and F. Galton, On the probability of the exinction of families, Jour. Anihrop. Inst. 4 (1874), 138-144. [152] K. Wlodarczyk, Iterations of Holomorphic Maps of Infinite Dimensional Homogeneous Domains, Mh. Math. 99 (1985), 153-160. [153] K. Wlodarczyk, Studies of iterations of Holomorphic maps in j-Algebras and complex Hilbert Spaces Oxford University Press, 1986.

[154] K. Wlodarczyk, Julia's Lemma and Wolff's Theorem for j-Algebras, American Math. Society 99 3 March 1987. [155] J. Wolff, Sur !'iteration des fonctions holomorphes dans une region, et dont les valeurs appartiennent a cette region, C. R. Acad. Sci. 182 (1926), 42-43. [156] J. Wolff, Sur !'iteration des fonctions bornees, C. R. Acad. Sci. 182 (1926), 200-201. [157] J. Wolff, Sur une generalisation d'un theoreme de Schwarz, C. R. Acad. Sci. 182 (1926), 918-920. [158] J. Wolff, L'E~quation differetielle dzjdt = w(z) =fonction holomorphe a partie reelle positive dans un demi-plan, Compos. Math. 6 (1938), 296-304. [159] K. Yosida, Functional Analysis, Springer, Berlin, 1968. [160] P. Yang, Holomorpic curves and boundary regularity of biholomorpic maps of pseudoconvex domains, ?reprint (1978).

Author and Subject Index Abate, M., vii, 101, 205 Alexander, J. W., 158, 205 Behan, D.F., 37, 206 Berkson, E., vii, 63, 93, 95, 101, 206 Bernardi, S.D., viii, 206 Brickman, L., ix, 206

Latuskin, Y., vii, 210 Lyzzaik, A., viii, 171, 177, 211 MacCluer, B., vii, viii, 207, 211 Mazet, P., vii, 211 Mellon, P., vii, 211 Mercer, P., vii, 211 Mantel, P., 6, 211 Nevanlinna, R., viii, 94, 158, 194, 212 Pjaltzgraff, J.A., ix, 212 Pick, G., 9 Poincare, H., 7, 43, 44 Porta, H., vii, 63, 93, 95, 101, 206 Potapov, V.P., vii, 212 Reich, S., vii, 205, 207-210, 212 Riemann, B., 7 Riesz, F., 5 Robertson, M.S., viii, ix, 159, 171, 176,213 Schroeder, E., 198, 213 Schwarz, H.A., 9 Sekowski, T., vii, 208 Sevastyanov, B.A., vi, 213 ShafTi.r, I., vii, 213 Shields, A.L., 36, 213 Silverman, H., viii, 177, 179, 213 Silvia, E.M., viii, 177, 179, 213 Siskakis, A.G., vii, 214 Srekeres, G., 200, 214 Stachura, A., vii, 208, 210 Steffensen, J.F., vi, 214 Stepin, M., vii, 210

Caratheodory, C., 9, 17, 206 Chen, G.N., vii, 206 Cowen, C. C., viii, 200, 207 de Branges, L., ix, 194, 207 Denjoy, A., vi, 9, 32, 35, 207 Earle, C., vii, 207 Egervary, E., 176, 207 Fan, K., vii, 208 Forelli, F., vii, 208 Galton, F., v, 215 Glicksberg, 1., vii, 208 Goebel, K., vii, 208 Gurganus, K.R., ix, 209 Hadamard, J., 200, 209 Hamilton, R., vii, 207 Harris, L.A., 13, 209 Harris, T.E., vi, 200, 209 Herglotz, G., 5 Hummel, J.A., 171, 172, 174, 209 Jafari, F., vii, 209 Julia, G., vii, 9, 17, 209 Koenigs, G., vi, 200, 210 Kubota, Y., vii, 210 Kuczumov, T., vii, 210 Lowner, K., ix, 211 216

AUTHOR AND SUBJECT INDEX Suffridge, T.J., ix, 212, 214 Vesentini, E., vii, 101, 208, 214 Vigue, J.P., vii, 211 Wald, J.K., viii, 159, 215 Watson, H. W., v, 215 Wlodarczyk, K., vii, 215 Wolff, J., vi, vii, 9, 32, 35, 101, 215 Yale, K., vii, 209

admissible curve, 42, 45-47 angular derivative, 23, 32, 35, 135, 140, 145, 150, 164 angular limit, 20, 23, 186 approximating curve, vii, 53, 55, 56 asymptotic behavior, vii, 83, 98, 103, 113, 135, 142-145, 150, 177 automorphism, vi, 9, 15, 19, 26, 30, 35, 56, 83, 146 elliptic, 27, 30, 36, 103, 105, 112, 113 hyperbolic, 27-29, 32, 37, 105, 109, 111 parabolic, 27, 28, 30, 32, 105, 111,112 Banach Fixed Point Principle, 8, 31, 52, 99 Berkson-Portarepresentation, 95, 98, 99, 139, 162, 170, 171, 174, 181 Bieberbach conjecture, ix, 194 Bohl-Poincare theorem, 81 boundary behavior, 17, 163 branching process, v, vi Brouwer's Fixed Point Principle, 8, 31 Caratheodory Kernel Theorem, viii, 6, 166, 177, 183 Cauchy Integral Formula, 4

217 Cauchy problem, 66, 68, 69, 72, 76, 82, 83, 87, 103, 161 Cauchy-Schwarz formula, 5 class of Caratheodory, 89, 178 commuting family, 36, 101 complete metric space, 7, 43, 44 complete vector field, 104, 156 contraction, 7, 17 strict, 8, 31, 52, 70, 99, 122 Denjoy-Wolff Theorem, vii, 30, 32, 36, 101, 102 embedding property, 109, Chapter V Euclidean distance, 7, 13, 114 exponential formula, 68, 70, 97, 122 fixed point, 8, 27, 31, 32, 53, 54, 57, 99, 111 attractive, 32, 35, 156, 199, 200 boundar~ 26, 29, 30, 32, 35 common, 36, 101 free mapping, 33, 37, 54 interior, vi, vii, 25, 30-32, 56 of automorphisms, 26 of holomorphic self-mapping, 25, 31 of nonexpansive mapping, 52 flow invariance condition, vii, 82, 98 fractional linear transformation, 10, 29, 40, 107, 109, 200, 202 function p-monotone, 79-82, 98, 141, 195 strongly, 98, 99, 169 close-to-convex, ix, 153, 177 fanlike, 157, 176, 192, 198 harmonic, 5, 85, 86, 97, 175 holomorphic, 4, 9 spirallike, see spirallike function starlike, see starlike function

218

AUTHOR AND SUBJECT INDEX univalent, ix, 4, 6, 153, 154, 158, 159,161, 162,170, 177,178, 180

generator, 66, 68, 81, 82, 144, 147, 150, 160, 162 of a one-parameter group, 82, 83, 104 of a one-parameter semigroup, 70, 76, 83 geodesic segment, 15, 45, 46 geodesics, 44, 47 group of automorphisms, 82, 104 one-parameter, 60 growth estimate, 194, 196 Harnack inequality, 89, 146, 149 strong, 90, 119 horocycle, 18, 19, 32, 37, 54, 97, 136, 167 Hummel's multiplier, 171, 172, 177 Hurwitz convergence theorem, 6 hyperbolic ball, 136, 174 distance, 45 length, 39, 42, 45 metric, vii, 7, 8, 39, 72, 79, 98, 99, 122 Implicit Function Theorem, 72, 82 infinitesimal generator, 66, 70, 76, 83 involution property, 11 isometry, p-isometry, 52, 60 Julia number, 19, 33, 35 Julia's Lemma, vi, 18, 19, 24, 32, 33, 35, 151 Julia-CaratheodoryTheorem, 22, 23, 32, 35, 150, 201

Julia-Wolff-Caratheodory Theorem, 33, 164, 171, 202 Koebe distortion theorem, 182, 194, 195 Koebe function, 168, 194 Koebe One-Quarter Theorem, 153 Koenigs embedding property, 198, 201203 Koenigs function, 200, 201 Lebesgue's Bounded Convergence Theorem, 185 Lindelof's inequality, 12 Lindelof's Principle, 23, 182 lower bound, 80, 135, 144 appropriate, 137, 140, 142, 143, 148 Mobius transformation, vii, 10, 11, 15,40,46,83, 168,172,173, 195, 196 mapping conformal at the boundary point, 164 identity mapping, 11 isogonal at the boundary point, 164 nonexpansive, 7, 8, 39, 43 p-nonexpansive, 52, 56, 57, 6870, 75-77, 79, 81, 82, 136, 144 fixed point free, 54 fixed point of, 52 maximum principle for harmonic functions, 5, 85, 86, 96 maximum modulus principle, 5, 10, 35, 47 Monte! Theorem, 6

AUTHOR AND SUBJECT INDEX Nevanlinna's condition, ix, 163, 167 nontangential approach region, 20, 23, 150, 182, 198 nontangentiallimit, 20, 21, 34 normal family, 6 null point, 98, 104, 135, 144, 156-158, 161, 171 Open Mapping Theorem, 154 Poincare metric, 7, 8, 39, 43-46, 70, 72, 79, 98 infinitesimal, 44, 45, 122 power convergence, 32, 35, 36, 56, 57 pseudo-hyperbolic ball, 17, 25, 118, 167 disk, 32 distance, 14, 17, 39, 114, 196 metric, 7 range condition, 69, 70, 72, 74, 77, 79, 82, 88 rate of convergence, 29, 35, 98, 113, 135, 139, 142, 144, 147, 149 resolvent, 73-78, 81, 82, 96, 99, 193 nonlinear, 67, 69, 72, 79, 82 resolvent identity, 74-78 retraction, 56, 78 Riemann Mapping Theorem, viii, 154, 166 Riesz-Herglotz representation, 5, 93, 178, 184 Schroeder's equation, 198-201 Schwarz Lemma, vi, 9, 12, 27, 87, 89, 91, 149, 167, 188 Schwarz-Pick inequality, 11, 13, 17, 19, 43, 82 Schwarz-Pick Lemma, 11, 14, 17, 19, 25, 26, 31, 32, 57 boundary version, 35

219 semi-complete vector field, 91, 95, 156, 174, 192, 195 semigroup, ix, 9, 60, 68, 97, 98, 135, 150, 155, 187, 190 one-parameter, v, 70 continuous, 60, 63, 66, 109 discrete, 26, 60, 109 sink point, 33, 35-37, 55-57, 78, 97, 105, 110-112, 135, 139, 142, 144, 149, 150, 171, 182, 185, 192 spirallike function, viii, ix, 159, 160, 167, 190, 191, 195 strongly, 169, 196 with respect to a boundary point, 159, 162 with respect to an interior point, 159, 161, 162 spiralshaped set, 153, 159, 160, 166, 167 starlike function, viii, ix, 157, 158, 160, 162,167,169,170,190, 191, 196 of order .\, 159, 179, 197 with respect to a boundary point, viii, 157, 158, 171, 176, 177, 180, 185, 192 with respect to an interior point, 157, 159, 161, 170, 171, 176, 177 starshaped set, 153, 157, 159-161, 166, 167, 172, 178 stationary point, 101, 104, 136, 138, 142 Stolz angle, 21 strict contraction, 8, 31, 52, 70, 99, 122 support functional, 136 Taylor representation, 5, 10, 154, 194

220 uniform convergence, 6, 54, 142 uniform Lipschitz condition, 13 vector field, 2 complete, 83, 85, 93, 96, 104, 110, 112, 156, 171 semi-complete, 83, 91, 95, 97, 98, 113, 156, 171, 174, 177, 180, 182, 190, 192, 195 Visser-Ostrowski condition, 165, 186, 192 Visser-Ostrowski quotient, 163, 165, 186, 192 Vitali theorem, 6, 31 Wald's condition, ix, 163 Weierstrass convergence theorem, 6 Wolff's Lemma, 32, 33, 35, 54

AUTHOR AND SUBJECT INDEX

List of Figures function w = f(z). . . . . . . . . . . translation f(z) = z +a, a= 4 + 2i. rotation f(z) = ei1r 8 z, (} = -27r/3 .. contraction f(z) = kz, k = 1/3. vector field w = f(z) . . . . . . . . .

0.1 0.2 0.3 0.4 0.5

The The The The The

1.1 1.2 1.3 1.4 1.5 1.6 1. 7 1.8 1.9 1.10

An orthogonal circle to a/:1 and its image under an automorphism. A Mobius transformation of t1r. . . . . . . . . . . . . A horocycle at the point ( E a/:1. . . . . . . . . . . . . A nontangential approach region at a boundary point. A Stolz angle at a boundary point. . . . . . Boundary behavior of a self-mapping of /:1. Elliptic automorphism. . . . Hyperbolic automorphism. . Parabolic automorphism. Uniqueness of a point (. . .

14 16 18 20 21 22 27 28 29 34

2.1

The points of anharmonic relation.

40

3.1 3.2 3.3

Boundary condition for f E aut(/:1). Boundary flow invariance condition. Values of functions of Caratheodory's class.

84 86 90

4.1 4.2 4.3 4.4

Fractional iterations of the self-mapping F(z) . . The circle f(z) and the sectors Sand S. The nontangential convergence toT. The asymptotic behavior of the flow.

110 126 130 140

5.1 5.2 5.3

A starshaped domain (0 En) . . . . . A starshaped domain (0 E an) . . . . The spiralshaped domain (0 E an) ..

157 158 160

221

2 2 3 3 3

LIST OF FIGURES

222 5.4 5.5 5.6

The set D .. . . . . . . . . . . . . . . . . . . Condition (A). . . . . . . . . . . . . . . . . An approximation of a starshaped domain.

168 175

183