Spectral Analysis of Julia Sets

Thesis by

St anislav K. Smirnov

In Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

California Institute of Technology Pasadena, California 1996 (Submitted May 10, 1996)
Typeset by A M T E X

Acknowledgements

I am deeply grateful to my advisor, Prof. N. G. Makarov for his support and supervision, without which this research would not have been completed. It is a great pleasure to thank my advisor at St.-Petersburg State University Prof. V. P. Havin, who introduced me to analysis, and taught me many things; and Prof. D. Sullivan, from whom I have learned a lot, sometimes indirectly. I am indebted to the California Institute of Technology and everyone in the Mathematics Department, especially M. D'Elia, for the wonderful time spent here. I wish to thank my friends I. Binder, A. Epstein, J . Graczyk, and A. Poltoratski for many hours of interesting discussions.

I am also grateful to V. Baladi, G. Jones, M. Lyubich, F. Przytycki,

D. Ruelle, M. Urbariski, A. Volberg, and many others for valuable conversations and suggest ions. This work was completed with a very much appreciated support by an Alfred P. Sloan Foundat ion Doctoral Fellowship.

Abstract
We investigate different measures defined geometrically or dynamically on polynomial Julia sets and their scaling properties. Our main concern is the relationship between harmonic and Hausdorff measures. We prove that the fine structure of harmonic measure at the more exposed points of an arbitrary polynomial Julia set is regular, and di-

mension spectra or pressure for the corresponding (negative) values of

parameter are real-analytic. However, there is a precisely described class of polynomials , where a set of preperiodic critical points can generate a unique very exposed tip, which manifests in the phase transition for some kinds of spectra. For parabolic and subhyperbolic polynomials, and also semihyperbolic quadratics we analyze the spectra for the positive values of parameter, establishing the extent of their regularity. Results are proved through spectral analysis of the transfer (PerronF'robenius-Ruelle) operat or.

Contents
. Introduction Chapter 1

...................
...............

1. Overview and general discussion

l a . Measures and transfer operator
l b . Fine structure of measures

3 4

..............

. . . . . . . . . . . . . . . . 20

2. Results

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

Chapter 2 Negative spectra 1. Quasicompactness

2. Analyticity

................

45

. . . . . . . . . . . . . . . . . . . . . . . . 47 . . . . . . . . . . . . . 78
84

. . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

3. Transfer operator on Sobolev spaces

Chapter 3 Parabolic case

1. Transfer operator

..................

. . . . . . . . . . . . . . . . . . . . . . . . 85

2. Analyticity

. . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

Chapter 4 Semihyperbolic case 1. Yoccoz puzzle and Hofbauer tower

2. Analysis of the transfer operator 3 . Analyticity of spectrum

..............

102

. . . . . . . . . . . . . . 103

. . . . . . . . . . . . . . . 108

References

............................

List of Figures
. . . . . . . . . . 29
. . . . . . . . . 35
2 > 2-w

Figure 1. Spectra of hyperbolic polynomials Figure 2. Phase transition in negative spectra Figure 3. Spectra of parabolics with HDim JF

. . 37 . . 37

2 Figure 3. Spectra of parabolics with HDim JF 5 2 - j ; T I

Chapter 1 Introduction

The main topic of this dissertation lies in the study of the dynamics of polynomial iterations and the geometry of corresponding Julia sets. We investigate different measures defined geometrically or dynamically on Julia sets and their scaling properties. This is done via describing a collection of characteristics, which we will call spectra.

Holomorphic dynamics. Considering a holomorphic dynamical system, such as Julia set, one notices that dynamics yields self-similarity. Hence geometry is "homogeneous" in different places, and the logical way to describe it is to count "how often" certain behavior occurs in different scales. It is also logical to expect that all geometrical objects have also dynamical/ergodic meaning, and the same set of parameters will describe both geometry and dynamical properties. As usual in dynamics, one finds that hyperbolic Julia sets with expanding dynamics are easier to study and have very nice properties.

CHAPTER 1. INTRODUCTION

Non-hyperbolic dynamics is more difficult to understand, but still some remnants of expanding can be observed, hence one can expect to find difficult, but interesting behavior.

Complex analysis. Approaching the problem of describing the geometry of domains and their boundaries in the complex plane, one notices that many questions can be reduced to understanding the structure of harmonic measure. However, harmonic measure will have good scaling properties only for self-similar sets, so one can obtain more interesting results in such a case. On the other hand, the extremal behavior of harmonic measure can be approximated on self-similar fractals, which also motivates studying of their properties.

1. Overview and general discussion

Julia sets represent a class of dynamical systems which are defined seemingly easily but can exhibit very difficult properties. They were studied, starting with the works of L. Bottcher, P. Fatou, and G. Julia throughout this century, and very intensively over the last two decades. Partially this interest was evoked by beautiful computer pictures, which showed difficult "fractal" structure and resemblance to many physical phenomena arising in nature. For a rational function F one can define the Julia set JF

as the

complement to the set of points in whose neighborhoods iterates of F form a normal family:

JF

:=

{ z : 3 U 3 z , { F n1 v ) is normal family)

another possible definition is the closure of all repelling periodic points:

JF := Clos { z : F n ( z ) = z , I(Fn)'(z)I > 1 for some n ) .

We mainly will be interested in the case when F is a polynomial, then

JF coincides with the boundary of domain of attraction to infinity A(oo)

:=

{ z : F n ( z ) + oo as n +

00)

which is fully invariant under the action of F. The nicest class of Julia sets are the so called hyperbolic ones, for which dynamics F on the

1 Julia set is expanding, i.e. ((Fn)'(z)

z E JF and n E

> CQn, C > 0, Q > 1 for any

Z+. For these definitions and basic properties of the

Julia sets see monograph [DH] of A. Douady and J. Hubbard; books

[Be] by A. Beardon, [Mil] by J. Milnor and [CG] by L. Carleson

and T. Gamelin, the latter follows a more analytical approach. The expository paper [ELy] of A. Eremenko and M. Lyubich gives a good present ation of their ergodic properties.

la. Measures and transfer operator

Trying to understand the properties of a Julia set, one can start looking at the different measures (defined geometrically or dynamically), and at their behavior under dynamics.

1.1. Entropy and Balanced Measures

Balanced measures. A first logical class of measures to look at are

the invariant ones. However, this class is too broad; so we can start with considering the so called balanced measures, where the mass is uniformly distributed among the preimages. Namely deg F . p (A) = p (F(A)) , if F is injective on A

i.e. Jacobian of p is equal to deg F. H. Brolin in [Bro] has proved existence of a balanced measure for polynomials and has shown that it coincides with the equilibrium (harmonic) measure from potential theory, which is well-defined since capacity of the Julia set is equal to
1. He has also shown that it has strong mixing property, namely

Uniqueness of the balanced measure was established later by A. F'reire, A. Lopes, and R. Maii6 in [FLM], [Maiil]; also in [BGH] by

M. F. Barnsley, J. S. Geronimo, and A. N. Harrington. See the latter

paper, [BH],and [Lo21 for the properties of potential, generated by the balanced (equilibrium) measure and its connections to Pade approximations and theory of moments. Another way to view the balanced measure is to notice that preimages F-"z as n tends to infinity have uniform distribution with respect to it, hence one can construct the balanced measure as a weak limit of the sums w - lim
n-oo

d-"

6,

Here 6, denotes the 6-measure supported at point y. This process can be viewed as considering operator L* acting on measures:

and analyzing its iterates. In such a way M. Lyubich has constructed balanced measures for rational functions (see [Ly2-31). One can play with these sums (or operator) in a different way, putting a unit mass at some point, l / d masses at its preimages, l/d2 at their preimages, etc., and taking a weak limit of measures obtained at the n-th step, properly normalized. A similar approach to another measure will appear to be useful later. H a r m o n i c measure and symbolic dynamics. One would like to have a good model of the dynamics on the Julia set. If it is connected and locally connected, then (via the Riemann uniformization map) dynamics z
H

zd (d := deg F) o n the unit circle gives us a topological

model (modulo some lamination). Hence it is logical to expect that it will also be a proper metric model.

A more sophisticated way, which also works in a non-connected case,

is symbolic dynamics, which models the dynamics F on the Julia set by It was first introduced equal-weighted one-sided Bernoulli shift on Zr. for Julia sets by M. Jakobson and J. Guckenheimer (see [Jl-41 and

[Gu]) and used extensively later. Bernoulli shift indeed appears to

be a proper metric model: R. Maii6 has shown in [Mafi2] that for some iterate N, (FN ,p ) is equivalent to the equal-weighted one-sided Bernoulli shift on dN symbols. Length plays an important role in the dynamics on the unit circle,

particularly it is balanced and maximizes the entropy (with a value of log d). The same is true for its analogue - d-adic measure on

ZT.

In the connected case mapping the length from the unit circle we obtain harmonic measure on the Julia set. As was mentioned, H. Brolin proved that it coincides with the balanced measure, moreover A. Lopes has shown that this property characterizes the polynomials among the rational functions (see [Loll and also paper of M. Lyubich and A. Volberg [LVl-21). Harmonic measure is a very important tool in understanding the structure of some set. Hence the unique balanced measure, except for dynamical, plays quite an important geometric role. Harmonic measure also admits a probabilistic approach, since it measures the probability of a set being hit by a Brownian motion, which is invariant under the conformal maps. Hence it naturally fits in the framework of holomorphic dynamics. This connection of probability to dynamics hasn't remained unnoticed: some of the mentioned theorems were proved in this way by S . P. Lalley in [La]. Entropy. Another logical thing to expect is that this measure will maximize the entropy. Indeed, this was proved by M. Lyubich (see [Lyl-31): particularly, there exists a unique invariant measure maximizing the entropy, it coincides with balanced measure and has entropy log deg F, which is the topological entropy of this dynamical system. The latter equality was conjectured earlier by R. Bowen in [Boll, and

partial results in establishing it were obtained by M. Jakobson, J. Guckenheimer, M. Misiurewicz, and F. Przytycki (see [Jl-31,[Gu], [MP]). Moreover, this balanced measure has very nice ergodic properties: in the case of totally disconnected Julia set it is Gibbs (see e.g. [C]and

[MV]for more general setting of conformal Cantor sets), for hyperbolic

polynomials the mixing is exponentially fast.

1.2. Geometric Measures

One knows that self-similar sets (e.g. Cantor sets, snowflakes, etc.
-

see [Fa])usually carry some geometric measure

-

like Hausdorff or

Minkowski

since they have nice scaling properties. It is logical to

expect the same from the Julia set. Since the Julia set has more complicated origins of self-similarity than, say, an affine cantor set, it is hard to evaluate (even estimate) its Hausdorff dimension and understand geometrically what will be the proper Hausdorff measure. One has to approach this question dynamically, as was first done by D. Sullivan. If we assume that the Julia set has Hausdorff dimension t and, moreover 0

< fit ( J F ) < oo, then the

Jacobian of the Hausdorff measure fit (or normalized v := fit / X t ( J F ) ) will be equal to

IF'^^:
du = u ( F ( A ) ),

1 F' lt

if F is injective on A .

Therefore one can look for a probability measure on JF with such property, and hope that it will be a t-dimensional Hausdorff or some other geometric measure. We will call such measures t-conformal.

Patterson-Sullivan construction. A constructive method to find

conformal measure, comes from the theory of Kleinian groups, where it was introduced by S. J. Patterson [Pal] and then used by D. Sullivan

[Su1,3] (see also exposition of related results in [Pa2]). It is pretty

much the same as the described method for constructing a balanced

i x t, pick some point z outside the Julia set, and put a measure: we f
unit mass on it. To "make" this measure conformal, we have to put appropriate masses on preimages of z , then on their preimages, etc. and consider a weak limit. Roughly speaking, we take

and set v to be the weak limit of the normalized measures vn/Var (vn). In practice the process is more,difficult: one has to chose t depending on the convergence of the sum above, and sometimes "correct" the terms in it to make the weak limit conformal. D. Sullivan has shown that for any rational map there exists t for which we can construct a conformal measure in such a way.

Particularly, there exists a 6-conformal measure for

for some (any)

outside J F . The latter series is an analogue of the

Poincar6 series for a Kleinian group. G e o m e t r i c properties of conformal measures. Unfortunately it is difficult to deduce much about the conformal measure from the construction or conformal property itself in the general case. There might not be unique S with existing conformal measure, hypothetically conformal measure for given S might not be unique, it might be atomic and does not have much of a geometric meaning.

] , hyperbolic case, First results were obtain by D. Sullivan, [ S U ~in

following the work [Bo2] of R. Bowen on the dimension of quasicircles (see about Bowen's formula later). Since every small ball is mapped eventually by some iterate of F to large scale with bounded distortion, and after some work one obtains that

for any ball B, of radius r centered on the Julia set. Thereafter v is (up to a constant) equal to a &dimensional Hausdorff measure, 6 being the Hausdorff dimension of J F . In this case there is a unique conformal measure for a unique exponent.

In the general case, results are harder and many things are still unknown. If we set 6 to be a minimum exponent with an existing conformal measure, it is unknown whether 6 is always equal to Hausdorff, Minkowski or hyperbolic dimension of JF.The latter dimension is defined as a supremum of the Hausdorff dimensions of hyperbolic subsets of JF,where dynamics is expanding. However, some partial results are known. M. Denker and M. Urbariski have shown in [DU5]that the following numbers coincide: 1) minimal zero of the pressure function, 2) supremum of Hausdorff dimensions of ergodic invariant measures with positive entropy (dynamical dimension), and 3) the minimal exponent 6' for which a measure conformal except on some finite set exists, coincide. Moreover, for many rational

' and the minimal exponent 6 for conformal measures are the maps 6
same (some kind of "expanding" on critical orbits is sufficient). If there are no recurrent critical points in the Julia set (but we allow parabolic points), the situation is even nicer
-

M. Urbariski has shown

in [U] (partial results were obtained earlier by him and M. Denker in [DU2-4,7]) that the exponent S will coincide with Hausdorff and packing dimensions of the Julia set and the 6-conformal measure will be equal to normalized &dimensional Hausdorff or /and packing measure. The latter is determined by the existence of parabolic points, which can produce interesting dichotomies for the possible properties of conformal

and invariant measures, see [ADU]. It remains to mention that work [Sh] of M. Shishikura implies that for topologically generic quadratics with parameter in the boundary of Mandelbrot set all mentioned dimensions will be equal to 2. However, we don't know what will be the 2-conformal measure in this case if the Julia set has zero area.

1.3. Transfer Operator

It is plausible to have another approach, maybe not working for a general Julia set, but giving more properties of a conformal measure. First we introduce a notion of (A, t)-conformal measure, i.e. such measure v that

F " i d v = v ( F ( A ) ), if F is injective on A .

Note that it generalizes two previous definitions: t-conformal measure is (1,t)-conformal, whereas balanced is (d, 0)-conformal. This notion is very closely related to one of a transfer (PerronFrobenius-Ruelle) operator. Transfer operator with weight

4 is defined

in a proper functional space. Choice of it is very important and will be discussed later.

The main role is played by the family Lt of transfer operators with weights
#t :=

IF'J-t.

If the Julia set does not contain critical points,

Lt acts on the space of functions, continuous on JF. Hence its formal

adjoint acts on the space of Bore1 measures supported on J F :

It is easy to see that a measure v is (A, t)-conformal if and only if it is an eigenmeasure of LZ; with eigenvalue A. Moreover, if Lt has an eigenfunction f with the same eigenvalue, the measure f v is an invariant measure equivalent to v.
Pressure. For hyperbolic polynomial let rF(t) denote the spectral

radius of the transfer operator Lt in a proper space (e.g. C(JF)). We define the pressure by PF(t) = log rF(t). As we shall see, by carrying out the spectral analysis of the transfer operator, one can establish nice properties of pressure and connect it to other characteristics of the Julia set. For the Julia sets, pressure was first used by D. Ruelle in [Ru2] to establish a conjecture of D. Sullivan that HDim JF depends real analytically on a hyperbolic rational function F . Particularly, he applied the machinery of thermodynamic formalism (see [Ru1,4])and showed that when the Julia set JF of a rational function F is hyperbolic, the pressure function PF(t) is real analytic as a function of t and F. The

remaining ingredient of his proof was

Bowen's formula. In the hyperbolic case the only zero of PF(t)

is the Hausdorff dimension HDim JF of the Julia set

the intuitive

reason is that conformal measure for t = HDim JF should coincide with Hausdorff in the same dimension and hence At = 1 and PF(t) = 0. This formula was noticed by R. Bowen for quasicircles, [Bo2]; D. Ruelle established it for hyperbolic Julia sets in [Ru2] and [Ru3] (see also
[DS] for another proof), it is equivalent to the mentioned result [Su2]

of D. Sullivan on conformal measures.

Spectral analysis of the transfer operator. In the hyperbolic case

the map F is expanding, so the transfer operator makes "smooth" functions even "smoother" which implies its quasicompactness in the spaces of smooth functions (e.g. Holder continuous - see later), and rF( t )is a simple isolated eigenvalue with eigenfunction
ft

and eigenmeasure

vt :

Operator Ltdepends real analytically on t , thus by perturbation theory

PF(t) is real analytic as a function of t. Actually, in "infinitely smooth"

spaces (e.g. functions real analytic in the neighborhood of the Julia set) the transfer operator behaves even nicer: D. Ruelle proved that it is nuclear in the sense of A. Grot hendieck ( [Grl]) , see [Ru2]. Eigenfunction
ft

and eigenmeasure u t also depend real-analytically

(in a proper sense) on t and play an important role:

vt

is (r,t)-conformal

(e.g. for t = 0 it is balanced and for t = HDim JF is equal to the normalized t-dimensional Hausdorff measure), invariant measure.
ft vt

gives us an equivalent

Quasicompactness. Another met hod of making spectral analysis

and establishing quasicompactness is due to C. T. Ionescu-Tulcea and

G. Marinescu (see [IM] and also the paper [N] of R. Nussbaum): the
"smoothing" property of the transfer operator leads us to the inequality of the type

IILnf

IIsrnooth space

&In IIf

IIsrnooth space

An

IIf

IIusual space

which "pushes" the essential spectral radius in the "smooth space" down to (A - E ) and makes operator L quasicompact. By quasicompactness we mean that the essential spectral radius

re,, (L) is strictly less than the spectral radius r(L) of the transfer operator L. Therefore the spectrum outside of the disk of radius re,,(L) consists of finite number of isolated eigenvalues. We are interested in the main eigenvalue r (L), determining the pressure; under nice circumstances it appears to be an isolated eigenvalue of multiplicity 1, which provides good properties of pressure.

Zeta-function. To prove the real analyticity of PF (t) in F, D. Ruelle

considered another approach, which gives the same pressure: he defined

PF(t) as the inverse to the pole of the dynamical 5-function

with the smallest modulus. Real analyticity follows then from nice spectral properties of the transfer operator and general theory of Fredholm determinants (see [Gr2]), to which C-function is closely connected. Variational principle. Variational approach works in the general setting too. For $ E C(JF) , the pressure is defined as

where the supremum is taken over all probability measures p on JF invariant under F. In the hyperbolic case, this variational problem has a unique solution for smooth $ and the pressure function is PF(t) =

P (-t log IF' I). Moreover, the maximizing measure for $ = -t log IF' 1
is equal to ftvt.

On c o m p u t e r experiments. Nice spectral properties of the transfer

operator imply that, iterating it, we converge to the main eigenfunction. Particularly, in the hyperbolic case, the partition function 2, satisfies

giving us an opportunity to estimate the main eigenvalue numerically (with an error of const '/" - 1 x 1I n , if we compute Ln by taking the preimages of z under F-n).

Hence whenever the transfer operator behaves "nicely," particularly in the hyperbolic case, one can make rigorous computer estimates of all the parameters involved: spectra, dimensions, etc.
- see

[STV] and

[V] by G. Servizi, G. Turchetti, S. Vaienti. Moreover, reasonable algorithms should converge exponentially fast, depending on the "expansion" constants for the dynamics. See, e.g., paper [GI of L. Garnett for computations of Hausdorff dimension for hyperbolic quadratics z2

+c

with small c. Roughly speaking, she evaluated zero of the pressure via considering the finite rank approximations to the transfer operator and computing their spectral radii. Unfortunately, for most interesting cases with low hyperbolicity, the exponent is close to 1 and can have a nasty constant in front of it, spoiling the situation. So, with the present abilities of computers, it still seems favorable to use "unrigorous" methods (e.g. Monte-Carlo), speeding up the computations: consult the paper [BZ] of 0. Bodart and M. Zinsmeister.

1.4. Non-hyperbolic situation

The key to analysis of spectra and properties of JF lies in the spectral analysis of the transfer operator. If the dynamics is "expanding" in some sense, the transfer operator has a nice spectrum (in a proper space), its eigenvalues and eigenfunctions (eigenmeasures of its adjoint)

behave "nicely" and all mentioned objects are "nicely" connected and have "good" properties. The main problem is hence to introduce a proper notion of "expanding," and analyze the spectrum of transfer operator in a proper space - the rest will follow. In a non-hyperbolic situation there are many difficulties, since the dynamics is not expanding in the usual sense. Moreover, some of the previously mentioned definitions must be modified in this case (e.g. if there are critical points on the Julia set, the transfer operator does not act on C ( J F ) for t

> 0) and a priori they can lead to different objects.

Since the first work of Ruelle, some progress has been made for the case of expansive maps, which corresponds to the Julia sets with parabolic points, see e.g. [HR], [ R u ~ ] [ADU] , or [DU2,3,6,7]. In the case of an arbitrary rational function M. Denker, M. Urbariski, and F. Przytycki (see [Dull,[Prl],[DPU]) showed that transfer operator L$ with a Holder continuous, positive 1C, is almost periodic on the space of Holder continuous functions (under an additional assumption

(0)-see the Subsection 2.3). This is a weaker property than quasicompactness, and if there are critical points on the Julia set, Lt = L I P,,- t for t

0 does not satisfy their assumptions. Recently N. Haydn has

proved a stronger theorem, establishing the quasicompactness of the transfer operator in this case - see [Ha]. An interesting partial case consists of polynomials z2 - c with real

< -2. Then the Julia set is a Cantor subset of the real line, and

the potential 1 F' ( z )l M t , restricted to it, is equal to a holomorphic function ~ ' ( z ) - ~ This . was used by A. Eremenko, G. Levin, M. Sodin, and P. Yuditski'i in [ELS] and [LSYl-21 to thoroughly investigate the spectrum of the transfer operator with t = 2. Substantial progress has been made on similar questions in onedimensional dynamics, mainly using bounded variation spaces, see, e.g., [HKl-21, [Pol], [ R u ~ ] [Ry] , (see book [RuS] of D. Ruelle for exposition of this subject). There is no good analogue of the bounded variation space for the complex plane; B V on the interval is invariant under monotone transformations, while in the complex plane we can only hope to find a space invariant under conformal maps, and still we will have some problems with critical points. Nevertheless, considering the space BV2 of functions whose second partial derivatives are measures, D. Ruelle [Ru6] proved the quasicompactness of L$ with

I I , E BV2 assuming certain conditions. His theorem implies some partial results for t he negative spectrum. Another way is in using Markov extensions, introduced by F. Hofbauer in [Ho]and developed by G. Keller in [K].This technique is very much related to the "jump transformation," when instead of given dynamics for every point one considers a proper iterate, carrying enough expansion (such method usually works in expansive case
-

see, e.g.

[ADU]). The idea is to do the same thing, but "separating" different iterates - we make a "tower" of countably many copies of the original dynamical system: the point goes up, till it accumulates enough expansion, then returns to the first floor. Hofbauer towers were used very successfully in analyzing maps of the interval, see papers [BK], [KN], [HK3-41, and thesis [Bru]of H. Bruin, which contains very instructive exposition of the tower construction. We will use Markov extensions build on the Yoccoz puzzle t o analyze the non-recurrent quadratics. Relating the pressure to the C-function is more difficult (see, e.g., [Ru5-6]), consult [Ba] and [Ru8] for the account of results in onedimensional dynamics. For non-hyperbolic Julia sets, we are aware only of the work [Hi21 of A. Hinkkanen, who deduced an explicit formula for the C-function with constant weights (i.e. t = 0), which appeared to be always rational.

l b . Fine structure of measures

There are a few other approaches of analytical and geometrical nature, rather than dynamical, which lead to the same objects.

1.5. Complex analysis

H a r m o n i c measure. One of the most important notions in the study

of the subsets of complex plane (more precisely, boundaries of domains) is harmonic measure. For a given domain (and a point inside) one can define harmonic measure as an equilibrium distribution for logarithmic potential, probability of being hit by Brownian motion, or as an image of the length on the unit circle under the Riemann uniformization map for simply connected domains. There are other definitions, generating more applications of this notion; we refer the reader to the monograph

[GM] of J. Garnett and D. Marshall. The choice of the point is not

important for the local behavior, so we will consider the harmonic measure on the boundary of the basin of attraction to infinity with respect to infinity, which we will denote by w . In the past decade there was a lot of progress in understanding geometric behavior and fine structure of harmonic measure, see exposition and references in the lectures [Mak2] of N. Makarov. Of particular interest to us are the papers [CJ], [JM], and [CM] of L. Carleson,

P. Jones, and N. Makarov, where they obtained bounds on the extremal behavior of harmonic measure. These papers show that it can be approximated on self-similar sets, which have good scaling properties and hence nice fine structure of harmonic measure. One should note, that many of its properties were first (and easier) observed and proved for self-similar sets: e.g. the dimension of harmonic measure on the boundary of any simply connected domain by Makarov's theorem is

equal to 1 ([Makl]); the proof for domains of attraction to infinity (for connected hyperbolic polynomial Julia sets) is easier and has more intuitive reasons, than in general case - see paper [Man] of A. Manning. This provides an additional motivation to investigate harmonic measure on self-similar fractals, since still there are many open questions about its behavior in general case. Distortion of the R i e m a n n map. Having a simply connected domain, one can learn about the geometry of the boundary by measuring the distortion of the Riemann uniformization map. One possible way is to consider the rate of growth of the integral means of its derivative, this makes even more sense if domain has self-similar boundary. Particularly, for a connected polynomial Julia set we can define the

conformal spectrum in the following way:

PF(t) := limsup log

log,

hzIZr l9'lt ,tER,

1

where cp is the Riemann map from the outside of the unit disk to the basin of attraction to oo. Considering the conjugation of F outside

JF with z

zd outside the unit disk, given by properly normalized

cp, it is easy to see that in hyperbolic case PF(t) = (PF(t) - t

+ 1)

log (deg F). There are also ways to extend this definition to the multiply connected case, e.g. considering mean values of distance to the boundary

(or modulus of gradient of the Green's function), taken to some power, along the Green's lines. H a r m o n i c Measure on J u l i a Sets. Many geometric properties of Julia sets are closely related to the behavior of harmonic measure and different spectra under discussion: The dichotomy of A. Zdunik (see [Zd], for hyperbolic Julia sets it was proved by F. Przytycki, M. Urbariski, and A. Zdunik in [PUZ]) for the boundary of attractive basin to be either an analytic curve or have Hausdorff dimension strictly more than 1, is proved using LIL (Law of Iterated Logarithm) principle for harmonic measure. Her theorem is in fact stronger, and has the following meaning in terms of conformal @-spectrum: either the boundary is an analytic curve (then the second derivative P" (0) is strictly positive. The Pommerenke-Yoccoz-Levin-Petersen inequality, which estimates the multipliers of periodic points in connected Julia sets (see [Porn], [ELel-21, [Le], [Pel),is related to the A. Beurling's estimate for the possible concentration of harmonic measure at a point when we know how "twisted" is our domain. By [GS] and [CJY] we know that Collet-Eckmann condition (i.e. exponential expansion on critical orbits) forces Fatou components to be Holder, particularly semihyperbolic polynomials have Fatou components even John domains. Holder property is closely related to spectra

P r 0), or

and their regularity: simply connected domain is Holder if and only if for the conformal spectrum limt-t+oo,B(t)/t and only if PF(t)

< 1 (equivalently if

< 0 for large t), the value of the limit depending on

the Holder exponent. In this case, by [Mak2], the only root of the equation ,B(t) = t - 1 is the Minkowski dimension of the boundary (an analogue of the R. Bowen's formula in non-dynamical context). We want also to note that many things known about the geometric regularity of the Julia sets imply some nice properties of harmonic measure. R. Mafib and L. F. da Rocha have shown in [MR] (see also [Hill) that Julia sets of rational functions are uniformly perfect, this property ensures some kind of regularity usually needed for the study of harmonic measure on disconnected sets. O n rational functions and conformal Cantor sets. We will mainly discuss the harmonic measure on the basin of attraction to infinity for polynomial Julia sets, which coincides with the balanced measure of maximal entropy. In the general case one has to consider, instead of a pressure as function of t, P = P ( t log IF'I), a pressure of two parameters: P = P ( t log IF'I

+ s log J,), taking into account the

non-constant Jacobian J, of the harmonic measure. However, many of our constructions do apply to conformal Cantor sets, or (super)attractive/parabolic basins of attraction for rational functions. It appears that the situation is still good enough to estab-

lish the regularity of the fine structure, though some other things, like rigidity properties, may "go wrong" (see e.g. [C], [MV], [LVl-21 for questions concerning conformal Cantor sets).

1.6. Multifractal Analysis - Chaotic Sets

Multifractal analysis is an intensively developing subject on the border between mathematics and physics. It was introduced by T. Halsey,

M. Jensen, L. Kadanoff, I. Procaccia, and B. Shraiman in a physical

paper [HJKPS], where they tried to understand and describe scaling laws of physical measures on different fractals of physical nature (strange attractors, stochastic fractals like DLA, etc.) . They considered the dimension spectrum of those measures - a continuum of parameters characterizing the size of the set where certain power law applies to the mass concentration. Since then there appeared a number of papers with rigorous multifractal analysis of different dynamically defined measures, see e.g.

[PW] for references. For a physical approach to multifractal analysis

, [Mand], and its mathematical counterpart, consult [BS], [Fa], [Fe]

and [T2]. It is natural to perform (and expect to perform well) multifractal analysis for self-similar sets and for measures behaving nicely under rescaling. Generally it is done for expanding dynamical systems

(- hyperbolic Julia sets) and for Gibbs measures.

It appears, that the fine structure of harmonic measure is best described via its multifractal analysis, see [Mak2]. A. Lopes

LO^])

made mult ifract a1 analysis of equilibrium (harmonic) measure for hyperbolic Julia sets, when all reasonable definitions of spectra work and lead to the same objects. He also pointed out in [Lo5-61some irregularities (to be discussed later), which may occur in subhyperbolic situation. Also see papers of Y. Pesin and H. Weiss [PW], [PI,who, in particular, analyze equilibrium measures with Holder potentials for conformal expanding case.
Dimension spectra. Dimension spectra will estimate the size of the

set where our measure has certain power law behavior. It makes sense to consider two kinds of dimension spectra, of "Minkowski" and "Hausdorff" types. Roughly speaking, we define (for rigorous definitions see [Mak2]) Hausdorff dimension spectrum as
J(a) := HDim { z : wB,,~ % S", 6 - 4 )

To define box dimension spectrum, we not just change the Hausdorff dimension to box-counting, but also "flip" limits in the definitions of spectrum and dimension:
f (a) := lim BDima { z : w B ~ x, Sa) ~ .
6-0

Here BDima is the box-counting dimension, estimated with disks of

radii 6: BDims E := inf { p : 3 disjoint {B,,a), z E E with E S P =: 1) .

B

Of course, in the general situation there will be many points, where measure behaves differently at different scales, so we will have to add lim sup's and lim inf's to our definitions. By analyzing the multifractal structure of some measure we mean investigating relationship of dimension spectra with other objects and proving some kind of its regularity
-

like the possibility of taking real

limits (instead of upper or lower) in the definitions.

Packing and covering spectra. Sometimes it is more convenient to

play with disks, having certain concentration of harmonic measure, in a different fashion, analogous to considering grand ensemble in statistical mechanics (cf. [Rul]) . Resulting spectra appear (under nice circumstances) to be connected via Legendre transform to dimension spectra and behave more like pressure in dynamical situations. In fact, pressure evaluates the same quantities with Lyapunov exponent s instead of rescaling coefficients for measures; in dynamical context those are closely related, as was noticed by J.-P. Eckmann and I. Procaccia in

[EP] (see also [FPT],[ST],and [TI]).

We define the packing spectrum ~ ( t as)
q : VS> 0 3 6-packing

{B) with ) : S ( B ) ~ ~ ( B ) ~ >1

where S(B)is the diameter of disk B. Covering spectrum c ( t ) is defined similarly, as

q:

V6 > 0 3 S - cover {B) with

~ ( B ) ~ U5 (B 1) ~

In nice situations (e.g. hyperbolic Julia sets, there, in fact, also f and .rr

c ) , spectra and dimension spectra are related via Legendre-

type transforms (see Figure 1):

n ( t ) = sup
a>O

( a )- t a

, f( a ) =
7

inf ( t
t

+ ax ( t ) ) , + ac(t)) .

c(t)

= sup
a>O

?(a) - t a

f ( a ) = inf (t
t

For the properties in general case, see [Mak2].

h, = logd f ( a ) / a

f (4
HD JF -

-t log d / P(t)

Figure 1. Spectra of hyperbolic polynomials

2. Results
Summing it up, one observes that spectra for negative values of

t (or small a ) describe the geometry of the set at "more exposed"

points
-

tips

with higher concentration of harmonic measure. The

more negative is t , the higher the concentration. On the other hand, spectra for positive t (big a) describe the geometry of "fjords," where the harmonic measure is low, but the set is "more dense" (here the Hausdorff dimension "lives" ) .

A physical interpretation of this is discussed by T. Bohr, P. CvitanoviC, and M. H. Jensen in [BCJ],where they suggest that spectra for small a should be robust under small parameter perturbation, when large a ' s are noisy and poorly convergent - the fjords are screened, and this can manifest in a "phase transition" at the Hausdorff dimension. We prove nice behavior of the spectra for negative values of t for
any polynomial Julia set. All of them coincide, nicely converge and are

real analytic except for the rare class of polynomials with very specific combinatorics of the critical orbits, which causes their Julia sets to have points with unique geometry, unrepeated elsewhere. In some cases this phenomenon can cause the "Hausdorff" and "Minkowski" spectra to be

different, with latter having a "phase transition." The reason is that "Hausdorff" definitions neglect the input from individual, not repeated patterns, while "Minkowski" take them into account. For positive spectra the situation is indeed more difficult
-

lack of

expansion has more impact: when the critical point belongs to the Julia set the weight J F ' ( z ) ~ is-not ~ even bounded, and we have troubles defining the pressure. However, using different methods, we were able to work out three particular cases. For parabolic Julia sets (dynamics is not expanding, but is expansive, and there are no critical points on the Julia set) we prove nice behavior of spectra up to the value t = HDim JF,where the phase transition happens. In this case it is caused not by a unique pattern, but by infinitely duplicated cusp at the parabolic point. Then work [ADU] of J. Aaronson, M. Denker, and M. Urbaliski implies an intriguing dichotomy for discontinuity of the derivative at the phase transition point and the value of HDim JF. For subhyperbolic Julia sets (M. Misiurewicz - W. Thurston sets where there can be preperiodic critical points and the dynamics is expanding with respect to a metric with finite number of singularities) we were able to transfer the problem via the Riemann uniformization map to the unit circle, where preperiodicity of the critical point allows to multiply the weight function by a proper homology, thus getting rid

of its singularities. Resulting pressure behaves nicely and has proper connections to other spectra. In a more difficult semihyperbolic case (M. Misiurewicz's Julia sets where dynamics is expanding with respect to some singular metric
-

see paper [CJY] of L. Carleson, P. Jones, and J.-C. Yoccoz for other equivalent definitions and proof of the John property for their Fatou components) we constructed, for non-recurrent quadratics, Markov extension (analogous to F. Hofbauer tower) on the J.-C. Yoccoz puzzle (see J . Hubbard's [Hu] and J . Milnor's [Mi21 expositions of the J.-

C. Yoccoz's results). Dynamics on this tower is expanding, hence spectra for tower behave nicely and have good relation to some spectra for the original dynamical system.

2.1. Negative spectra

Establishing the quasicompactness of the transfer operat or on Sobolev spaces (see Chapter 2 for precise formulation), we prove the following results, which almost completes the analysis of the negative spectra.

Theorem A. A l . For any polynomial F and negative t either

(i) PF(t) is real analytic on (-oo,O), or

(ii) there exists a "hase transition'' point to < 0 such that

0) , and PF ( t ) is real analytic on [to, = - P-, t on (-oo,to] .
In fact PF (t) = max(- P-, alytic on
(-00,0).

. t ,PF( t ) ) for some FF ( t )real an-

A2. In the case (i) all mentioned definitions of pressure give the same
function P F ( t ) . In the case (ii) function P F ( t ) corresponds to "Box" definitions, when "hidden" spectrum & ( t )
-

to "HausdorEn All defi-

nitions converge nicely, i.e. one can take lim instead of limsup, etc.

A3. For the occurrence of a "hase transition" it is necessary that

either

(iia) F is conjugate to a Chebyshevpolynomial (then JF is an interval), or

(iib) there are a fixed point a, F a = a, and a positive number E such

that

where p(b) : = (F")' ( b )

'

denotes the multiplier of a periodic

=

point b. This implies that a E J F , P-, ~-'a\{a)

logp(a) and

c Crit F ( =

zeroesof F ' ) .

A4. For quadratic polynomials (iib) cannot happen (for com binatorial
reasons), and for cubics it only happens for some polynomials with disconnected Julia set. However, there is a degree 4 su bhyperbolic polynomial with connected Julia set for which it occurs.

A5. The pressure function depends continuously on F as a function in

the space Cm(-00, 0)) or Coo((-00, to - E) U (to transition occurs.

+ E, 0)) if the phase

Parts A1 and A2 state that the distribution of the harmonic measure at the "more exposed" points is very regular, and multifractal analysis does apply. Part A3 and A4 show that the phase transition is a rather rare phenomenon. The Julia set JF itself does not depend continuously on F , nevertheless A5 shows' that "more exposed" parts of the Julia set are "rigid" and depend continuously on the dynamics. The geometrical meaning of the phase transition is that the tip at point a is unique and "significantly more exposed" than any other point of the Julia set. Thus for t

> to the spectrum is determined by the

combined influence of many similar parts, but for t

2 the input from the

"very exposed" tip at point "a" "screens" the rest of the real-analytic spectrum (which, nevertheless, continues to exist and can be calculated as Hausdorff spectrum). The "thermodynamical" meaning of the phase transition is that for each t > to we have a unique equilibrium state, supported on the whole

Julia set. For t = to we obtain another equilibrium state, given by a 6-measure at the point a , which dominates for t

<

to. However,

the original equilibrium state "continues" to exist being hidden. For Chebyshev quadratic (F(z) = z2 - 2 and Julia set is an interval) this phenomenon was observed by E. Ott, W. Withers,and J. A. Yorke in
[ O W ] and A. Lopes in [Lo5,6]. Also in the latter paper maps having

the so called gap (see [Lo51 and compare to the maps falling into the case (ii) of Theorem A) are discussed, including Chebyshev polynomial
= and the Lattes example (F(z)
((2

-2)/~)~ Julia , set is the whole

sphere).

Figure 2. Phase transition in negative spectra

In other words, the transfer operator is always quasicompact, and the maximal eigenvalue is isolated, but for to two eigenvalues (one generated by a nice measure, and another
-

by a 6-measure at the tip)

"cross" and the maximal eigenvalue has multiplicity 2. Note that this differs from the phase transition in positive spectrum (e.g. for parabolics), or one described by M. Feigenbaum, I. Procaccia, and T. T61 in

[FPT],when the essential spectral radius reaches the maximal eigenvalue. In our case the transfer operator stays quasicompact and hence spectra behave nicely.

2.2. Positive spectrum

Parabolic case. Assume that all critical points are attracted parabolic or (super) attractive cycles. Establishing the quasicompactness of transfer operator for the parameter values t the following

< HDim JF,we prove

Theorem B. B1. For a parabolic polynomial F, the function PF(t)

is real analytic on [0,HDim JF)and PF(t) = 0, t E [HDim JF, +oo).

B2. The derivative of PF(t) is discontinuous a t the point HDim JF if

and only if HDim JF

> 2

- -

p+l

'

where p is the maximal number of petals a t the parabolic points.

The reason for the phase transition is that, while for t

< 6 transfer >6

operator is quasicompact and conformal measure is "nice", for t

the situation changes, conformal measure becomes atomic - supported on the preimages of the parabolic cycle. Another interpretation is that main input in the spectra for t

> HDim JF comes from the infinitely

many times duplicated cusp at the parabolic point. The dichotomy comes from the work [ADU]of J. Aaronson, M. Denker, and M. Urbaliski on the existence of the invariant measures equivalent to 6-conformal.

Figure 3. Spectra of parabolics with HDJf > 2 -

&

2 Figure 4. Spectra of parabolics with HD J f 5 2 - 3

The inequality

(4) is true

for z2

+ 114 (the "cauliflower" Julia set

indeed, its dimension is greater than I), and hence there is a discontinuity of the derivative at the phase transition point. We do not know which case of the dichotomy occurs even for other parabolic quadratics; and we were not able to prove the discontinuity of the derivative for
z2

+ 1/4 directly. Note that computer experiments in [BZ] suggest that

(4)fails.
Note that as an application of the quasicompactness result we obtain

for z2 - 314 the inequality

an improved estimate on the radius of essential spectrum of the transfer operator (in hyperbolic or parabolic case), which depends only on t, P ( t ) , and P(f oo) (see the Remark 1.7 in the corresponding Chapter).
Subhyperbolic case. Assume that all critical points in the Julia set

are strictly preperiodic. Transferring the problem to the unit circle, we prove for the critically finite polynomials the following
Theorem C. For a su bhyperbolic polynomial F with connected Julia

set, the function PF (t) is real analytic on [O, +m).

Semihyperbolic case. For semihyperbolic quadratic polynomials (i.e.

those with non-recurrent critical point) we build a Markov extension of the original system (analogous to the Hofbauer tower), taking the Yoccoz puzzle as a base. Then we establish the quasicompactness of the corresponding transfer operator, and prove the following

CHAPTER 1. INTRODUCTION
Theorem D. For a non-recurrent quadratic polynomial F ( z ) = z 2
either

39

+c

(i) PF(t) is real analytic on [0, + m ) , or (ii) A phase transition occurs: there exists to > HDim JF such that PF(t) is real analytic on [O,to), and PF(t) = - l 2 t log (liminf,,,

(F~)'(C)''")

on [to,+ m ) .

We do not know whether case (ii) can occur. By the work [CJY] of

L. Carleson, P. Jones and J.-C. Yoccoz the polynomials under consideration have domain of attraction to infinity A(m) John, thus one can speculate that spectrum should be "nice" and (ii) impossible. On the other hand, by [GS],Collet-Eckmann polynomials have Holder A(m), so P(t)

< 0 for large t and one can expect to have (ii) for Collet-

Eckmann recurrent polynomials.

2.3. Transfer operator on Sobolev spaces

We will apply the methods used for the analysis of negative spectrum to investigating transfer operators with general Sobolev weights. Considering functional spaces on the complex sphere, or some open set

fl with

taken with spherical metric, we define transfer operator on the space of continuous functions by

Lf ( z )

:=

( Y )g ( y )

where preimages y of z are counted with multiplicities. Note that

L n f ( z ) :=
where g,(y) radius:
=

C
yEF-*y

f ( ~ ) g n ( 7~ )

g ( y ) g ( F y ) .. . g ( ~ n - l ( y ) Denote . b y X the spectral

We prove the following theorem, establishing the quasicompactness

of L:
Theorem E. Suppose that

for some (any) large n. Then for p

> 2 sufficiently close to 2 operator

L acts on W 1 ,and

ress ( L ) Wl,,) < r ( L ,W1,,)

= X

where r and re,, are spectral and essential spectral radii of L as an operator on Wl,,. Moreover, X is an isolated eigenvalue of multiplicity one.

This theorem is analogous to the results [Dull, [Prl],[DPU] of

M. Denker, M. Urbariski, and F. Przytycki for the transfer operator

on the space of Holder continuous functions They have established the (weaker) property of almost periodicity under the assumption

(0).Re-

cently N. Haydn has extended their results, proving the quasicompactness (see [Ha]).

2.4. Open problems

Negative s p e c t r u m . Some work still remains to be done for negative values of t. If we establish a connection between the spectrum and

) , might be the C-function (the similar problem is stated in [ R u ~ ] we

able to learn more about the dependence of PF(t) on F and get some necessary and sufficient conditions for the phase transition to occur. Particularly it is interesting whether or not (iib) is sufficient for the phase transition (it is so in degrees 2 and 3). Positive s p e c t r u m . A very optimistic statement is Conjecture. For any polynomial F either

(I) PF(t) is real analytic on [0, +m), or

(11) PF(t) is real analytic on [O, tii) and PF(t) = P+t, t E [to,+w), or

(111) PF ( t )is real analytic on [O, tiii) and PF(t) = 0, t E [to,+oo).

A possible step in this direction is to establish

Conjecture.
For any polynomial F, PF(t)is real analytic in some

neighborhood of 0.

Work [Zd] of A. Zdunik implies that for any polynomial (except ones with Julia set being an interval or a circle) the pressure at zero has strictly positive second derivative, which advances us in that direction. Next step will be to find some necessary and (or) sufficient conditions for (I), (11), (111) to hold: in terms of the orbits of the critical points of F or geometry of the Julia set J F . There are some euristic reasons to expect that case of John domain of attraction to infinity (i.e. semihyperbolic polynomials
-

see [CJY]) corresponds to (I), while Holder

(but not John, e.g. Collet-Eckmann polynomials fall into this category, see [GS]) case - to (11). If the phase transition occurs, we need to analyze its nature. One example is the question about the meaning of the first zero of the pressure function (the "phase transition point" in case (111)).It is interesting to compare it with such parameters as the dynamical, Hausdorff and hyperbolic dimensions of J F . There is some hope that in cases (I), (11) all of these dimensions will coincide. On the other hand, the geometry of the Julia sets with the phase transition (case (111)) is likely to be "bad" and it might happen that the dimensions above will differ.

Dependence on F. The logical statement to prove is that PF(t) depends continuously on F whenever it is positive. As in the case of the negative spectrum, it might follow from the "sufficiently good" spectral analysis of the transfer operator. If one wants to prove some stronger results, like analyticity, he will probably need to study (-function.

2.5. About methods and organization

The general scheme of the proofs is to find a proper notion of expanding, establish the quasicompactness of the transfer operator, and deduce the results about the spectra. The thesis is organized as three (independent) chapters, devoted to negative spectrum, parabolic and semihyperbolic polynomials. Some arguments are similar in different cases, but st ill they have (sometimes substantial) differences, so we repeat them for the sake of completeness. The Chapter 2 is devoted to the analysis of negative spectra for arbitrary polynomials (Theorem A). In the arbitrary case F is not expanding and Lt does not act on the space of Holder continuous functions. This problem for t

< 0 is solved by considering the Sobolev

space, where bad behavior of F near critical points is compensated by the zeroes of the weight function IF / I -t . Then we apply the same technique to the study of transfer operators with arbitrary Sobolev weights, establishing the Theorem E.

In the Chapter 3 we consider the parabolic case (Theorem B), when dynamics is not expanding but expansive, and careful computation shows that non-expanding branches of F-" do not contribute much to the transfer operator for t corresponding parameters. For other Julia sets and t

< HDim J F , hence spectra are good for the > 0 the situation is more complicated,

since the function I F ' I - ~ is unbounded and thus Lt does not preserve the space of bounded functions. In the subhyperbolic case we "cheat" by multiplying the weight by a precisely defined homology and cancelling its singularities. Then the problem is transferred via the Riemann uniformization map to the unit circle, where the corresponding dynamics is expanding. The pressure for the new operator gives the same spectra. The corresponding result (Theorem C) is proved in the paper [MS] by N. Makarov and the author. We will also rely heavily on this paper during our analysis of the phase transition phenomenon in the Chapter 2 (negative spectrum).

In the Chapter 4 we work (establishing Theorem D) with non-recurrent quadratics, building a Markov extension of original system
-

an

analogue of a Hofbauer tower in one-dimensional dynamics, which gives the desired expansion.

CHAPTER 2. NEGATIVE SPECTRA

Chapter 2 Negative spectra

In this chapter we will consider the spectra for negative values of parameter t, in which case transfer operator preserves the space of continuous functions. In the first Section we prove Ionescu-Tulcea and Marinescu inequality and establish the quasicompactness of transfer operator in the Sobolev space, while the second is devoted to the analyticity of spectra and their properties, which result in the following

Theorem A. A l . For any polynomial F and negative t either

(i) PF(t) is real analytic on
(-00,

0) , or

(ii) there exists a phase transition" point to < 0 such that PF(t) is real analytic on [tO,O) , and - - P-, t on (-00, to] .

In fact PF(t) = max(- P-, alytic on ( -00, 0) .

t , pF( t ) ) for some

&(t) real an-

A2. In the case (i) all mentioned definitions of pressure give the same
function PF ( t ) . In the case (ii) function PF( t ) corresponds to "Box" definitions, when "hidden" spectrum

PF ( t )- to

('HausdorK " All defi-

nitions converge nicely, i.e. one can take lim instead of limsup, etc.

A3. For the occurrence of a (@phasetransition" it is necessary that

either

(iia) F is conjugate to a Chebyshev polynomial (then JF is an interval), or

(iib) there are a fixed point a, F a = a, and a positive number E such

that

where p(b) : = ( F ~ )( b ')

denotes the multiplier of a periodic

=

point b. This implies that a E JF,P-,

logp(a) and

la \ {a) c Crit F ( =

zeroes of F' )

(6)

A4. For quadratic polynomials (iib) cannot happen (for combinatorial

reasons), and for cubics it only happens for some polynomials with disconnected Julia set. However, there is a degree 4 subhyperbolic polynomial with connected Julia set for which i t occurs.

A5. The pressure function depends continuously on F as a function in

to - E ) U (to the space Cw(-00,0), or Cw ((-00,

E,

0)) if the phase

transition occurs.

Finally, in the last Section we apply the developed technique to the study of transfer operators with arbitrary Sobolev weights. This results in the Theorem E (see the Subsection 3.1 for more details).

1. Quasicompactness
1.1. Notation
Let S1 be a big disk compactly containing JF such that

and orbits of the critical points do not intersect 80. For t

< 0 we define operator Lt on C (a) by

where preimages y of z are counted with multiplicities. Note that

(here and later we write Fn = Fn). Denote

s ( t ) := logd r

(L~ C, (n)) (spectral radius) .

It will be shown later that for connected JF

We will also consider this operator in various Sobolev spaces W1,,(0). Observe that for p

> 2, W1,, (0) c C (a) (see [Zi] for this and other

properties of Sobolev spaces).

1.2. Quasicompactness theorem

Theorem. For any negative t for p Lt acts on W1,,(O) and
ress

> 2 suficiently close to 2 operator

(Lt , Wl,, (0)) < r (Lt WI,, (a)) = ds(t) ,

where r and re,, are spectral and essential spectral radii of Lt as an operator on Wl,, ( 0 ) .

1.3. Sobolev spaces

We consider Sobolev space W1,,(O) with the norm

If p > 2, then
Wl,P(fl)

C(Q)

and

llfllm 5 Ilfll,

Moreover, Wl,p-functions are Holder continuous: Wl lp More precisely, if 1 z - y 1 = 6, then

c Hall- 2 . P

For the proofs of these results see W.Ziemer "Weakly Differentiable Functions", particularly [Zi, 2.4.41.

Lemma. I f t

< -2(1

z),

then LtWllp c W I , ~ .

Proof. It is sufficient to estimate

by const 11 f 11 l,p. Clearly

Here

< const
since t

/ a f iP

< -2(1- ) : implies -tp+2-p

> 0 and supn IF'I --tp+a-p <

00.

The second term can be estimated by

so we just need to prove the convergence of the last integral which is non-trivial only at the critical points of F. Consider some critical point (say zero), let F' have singularity (zero) of order k :

Then

Since k

2 1, t < -(1

:)(I+

i )thus
>
-2

-p(l

+ kt) + k ( 2 - p)

which implies the convergence and hence the desired estimate.

1.4. Integral means ( t 5 0)

Lemma. For any (all) z E dS1

In JF is connected, then

CHAPTER 2. NEGATIVE SPECTRA

Remark. Lemma implies that

P(t) is well-defined and

Proof. Clearly IILyllco = IILylllco. The function z

harmonic. Therefore

I-+

L r l (z) is sub-

Fix zo E DR. For any z E dR we can choose a domain 9 3 z, zo without forward iterates of critical points. Then any branch of FWn is conformal on Q, so

where y, yo are images of z , zo under the same branch of F+. We can estimate the distortion by a constant independent of z, hence sup L; 1 x L,nl(zo) x L; 1 (2) an which implies the first statement of the Lemma. Note that similar estimates (and F-n 0 imply that

\ KF

= filled Julia set)

I F ; (yo)l 2 1 for yo E FVnzo and large n

thus for t

<u <0

and

Moreover area estimate shows that

implying that for fixed d satisfy IF A ( y )1

> q > 1 for large n most of the y

E F-n~o

> qn. Therefore for n >> 1

- -- n logq L Y l ( z o ) .
A

So for

+n(t)
we have +,(t)

:=

logd ( L y l ( 2 0 ) )

< -$ log q < 0 which implies that s ( t ) is strictly de-

creasing because

+,

-+ s. n-oo

To prove the second part note that without loss of generality cp conjugates F with dynamics T : z
H

z d on ID)-:

Differentiating the identity Fn o cp = cp o T n , we obtain

Applying this equality to the preimages

JE

< E T-n of some fixed point

ID- we notice that the right side is =: dn, thus (taking to power t )

where y = p(C) is a corresponding preimage of z = p(c) under F n . The points [ are equidistributed on the circle

dn with (rn - 1) =: -

Therefore summing over all

E TVn[ ( y E

F - n ~ we ) , have

which together with the first statement completes the proof.

1.5. Ionescu-Tulcea and Marinescu

inequality
Lemma. Suppose t < -2(1
-

g), so Lt acts on W1,,(SZ). Then

CHAPTER 2. NEGATIVESPECTRA
with

Proof. Clearly

We can estimate the right-hand side by

(in (*) we used H6lder inequality: and

(Caibi)' < C a: (C b f )

'

),

IIP

A( C

f (Y) ~ Y I F ; ( Y ) I - ~ I F ; ( Y ) I - ~

d m 2 (2)

yFWn(z)

Here Cn is finite since we assumed t

< -2(1 - $)

use the same

estimate as in 1.3. It remains to notice that (see 1.4)

1.6. Finite rank approximation

Lemma. For any
E

> 0 there exists a finite rank operator M in

IIMII1,, 5 abs. const ,

W1,&2) such that

Proof. Cover 0 with a grid of triangles

a continuous function satisfying

4 of

size S << 1. Define M f as

Mf =

{f

at all vertices , is linear in each triangle

4.

For a fixed

4, we have

and by the Holder continuity (see 1.3) this is

Taking to the power p and integrating over R (i.e. summing up over all

4's) we have

which proves the first inequality. The second one follows from the Holder continuity.

1.7. Analysis of w(p,t)

We show here that for any negative t < -2 function w (p, t ) satisfies

(1 - $1 with p > 2

Clearly s(t) satisfies following two properties: s(lct) 5 lc s(t) for lc s(t) is decreasing .

> 1 and

CHAPTER 2. NEGATIVE SPECTRA

.o(n>

Proof. In fact, d N S @ ) =:

11 L ~ ~ I I so obvious
ml

implies the first property. For the second see Proof of the Lemma 1.4. There we proved even that s ( t ) is strictly decreasing but we use the simpler property for the sake of 2.7. Now we can write

If the second inequality is strict, we are done. Otherwise for any k E

[ I ,P'I k s ( t ) = - s (p't) 5 s ( k t ) 5 k s ( t ) , P'
hence s ( k t ) = k s ( t ) and s ( t ) is strictly decreasing on ip't, t ] . This implies that the first inequality is strict and again we are done.
k

1.8. Quasicompactness
Claim. Suppose t < - 2 ( 1 -

g) and p > 2. Then

= dS@) ,

r ( L t ,Wl,,(R))

CHAPTER 2. NEGATIVESPECTRA
and

Remark. Clearly p E (2, &)

for t E (-2,O) and any p

> 2 for t 5 -2

satisfies the condition of this Claim and therefore it completes the Proof of our Theorem.

Proof. First recall that

Also llL;llw and ds(t) llL;lllm

5 llL:llll,p 5 llL~lll,p
7

< r(Lt, Wl,p(R)).

To prove the converse, by the Ionescu-Tulcea and Marinescu inequality choose E > 0 and N such that

IILNf

5 dN(s(t)-') llf Ill,,

llf ,l

Define inductively an increasing sequence { M k ) by

Then

CHAPTER 2. NEGATIVE SPECTRA

and we can choose a constant C satisfying

C dN(s(t)-")
By induction,

+1< -

C d N S @ ) and K 5 C Ml

11 ~
therefore

" 5 f dN"s(t)-') llf Ill,, +

N

CMk

llf ,l

< dN"(t) 11f Ill,P

which implies the desired inequality and thus the first statement. To prove the second statement choose q small such that

for any approximation operator ME. Since w ( p ,t ) < s ( t ) , we have

for some N and K. Take

< & d N S @ ) , and define M

:= ME. Then

CHAPTER 2. NEGATIVESPECTRA
and

2. Analyticity
2.1. Remark on the proofs
We will prove Theorem A1 in the Subsections 2.3 - 2.8 below, last two of which are also devoted to the analysis of the phase transition phenomenon (Theorem A3). Multifractal analysis (Theorem A2) is completed in the Subsections 2.2 and 2.9. In the remaining part of this Subsection we prove Theorem A4-5.

Proof of Theorem Ad. Discussion in [MS, 6.21 shows that for combinatorial reasons (4) cannot happen for quadratic polynomials. Comparison of multipliers in [MS, 6.51 for cubic polynomials satisfying (4) implies that iib) cannot be true for F with connected J F . See [MS,6.61 for an example of a degree 4 subhyperbolic polynomial with a phase transition.

Corollary. I f F is quadratic or cubic polynomial, it has a phase transition if and only if (iia) or (iib) is true.

Proof of Theorem A5. Discussion below shows that the main eigenvalue
of the transfer operator is isolated and of multiplicity one for all t E (-oo,O), or for t E ( - w , t 0 - E ) U (to

+ E,O),

if the phase transition

occurs. Using perturbation theory for transfer operators with weights

I F ' I - ( ~ + we '~) conclude , that for specified values of t

extended to a holomorphic function of t
6.

pressure can be
-6

+ is in a thin strip

<s<
H

By the perturbation theory again, this extended pressure depends

continuously on F, implying Theorem A5 via Cauchy formula.

2.2. Variational principle

For t

< 0 define pressure by

where supremum is taken over all probability measures p on JF invariant under F. We also define P ( t ) as the same supremum, taken over all probability non-atomic measures (or equivalently, with positive entropy)
*

Lemma. s(t)

= P ( t ) / log d

and

5(t) = p ( t ) / log d

Proof. Equivalent property r (Lt) = exp (P(t)) is mentioned in [Ru6,

6.31, where he refers to [Pr]. The same method works for the second

formula - see section 2.7 for the definition of S ( t ) .

2.3. Eigenvalues and multiplicity

Fix t and denote

Lemma. ker (Lt

A) # (0).

Proof. See [Ru6, Theorem 2.21.

Idea of the proof: since re,, ( L t ,W1,,(fl)) < r ( L t ,W1,,(fl)) = A, we have some eigenvalues Aj of absolute value A, and a decomposition

where Xo is the subspace corresponding to the rest of the spectrum. Then

Here LTXo = o ( A n ) and L r X j = A? X j , implying that one of the

Xj7sis positive and hence is equal to X = r ( L ~ C, ( a ) ) .

Remark. This reasoning also implies that Lyl

An and therefore for

z E d o we have Lr 1 x An.

2.4. Conformal measures

Let L,* denote the adjoint of Lt : C ( J F )6.

Lemma. There exists probability measure v on JF such that

Proof. Fix a point z E afl and denote

Clearly L;pn = X pn+> Since

we have a following formula for p,:

and Var (p,) Set

A-" L?l(z) x 1.

Then Var (v,) =: n and Var (Lfvn - X v n ) = Var (X/L,+~ - Xpo)

1.

Following Patterson and Sullivan choose a subsequence of normalized measures vn / Var (vn) weakly converging to some measure v. Clearly v is probability measure supported on J F . Since

Var (LTv, - XU,)

/ Var

(u,)

n+oo

--+

measure v satisfies Lfv = Xu.

Lemma. Suppose v is a measure from the previous Lemma. Then

suppv = JF unless F satisfies iia) or ( 4 ) .
Proof. Clearly v satisfies

hence

z E suppv F z E suppv

F z E suppv and
E suppv

, z 51 Crit F + z

Then the same reasoning as in [MS, 6.11 proves this Lemma. Main idea: preimages of any point z E JF are dense in J F ,
SO

if z E suppv does not belong to the forward orbits of critical points, suppv = JF and we are done. The only cases when v can be supported only on the forward orbits of critical points are iia) and (4).

2.5. Jordan cells

Example. Operator

f t ) and acting on IK2 depends real analytically on t. But a ( Q t ) = {

therefore r ( Q t ) = I t 1 is not a real analytic function.

Lemma. Suppose there exists an eigenmeasure v, L* v = X v with

suppv = J F . Then dim ker ( L t - x value.

)~ =

1, i.e. X is a simple eigen-

Proof. First we will prove the following Sublemma. For f

E

Wl,p(52)
=

Ltf

Xf

implies f = 0 .

Proof of Sublemma. For any z E 52 distortion estimates in 1.4 imply

~ ~ ; ( y ) l - l d-n
x d-"

--

lp'(C)l
d-n

x d-"

dist (Y J F ) dist aID-)

(c,

dist ( Y , J F )

= dist

( y ,J F ) ,

where notation is taken from 1.4. Using Wl,p(fl)

c Hol,

with small

a < 1 - $ we obtain

because s ( t ) is strictly decreasing by 1.4. Consequently f = 0 and the Sublemma is proved. If we consider only functions in C (JF), the same reasoning as in

[MS, 3.61 gives us that dim ker (Lt - A) = 1, and then Sublemma
implies this for f E Wl,p(0). Also [MS, 53.61 shows that we have an eigenfunction f non-negative

If A is not a simple eigenvalue, then there exists function h such that

Therefore

Thus v-a.e. we have f = 0, hence everywhere on JF and (by the Sublemma) in R.

2.6. Analyticity
Lemmas 1.8 and 2.5 show that if for some t' there exists eigenmeasure v, L;, v =
Xtt

v with suppv

JF, then for p > 2 close to 2

number ds@) is a simple isolated eigenvalue of Lt : W

( )6. Clearly

Lt depends real analytically on t therefore s(t) and hence P(t) is real

analytic in the neighborhood of t' (see [MS, 4.11). So 2.5 implies that s(t) is real analytic on happens. In the case iia) JF is an interval, without loss of generality
(-00,

0), unless iia) or (4)

and simple calculation shows that

It remains to deal with the case

(4)when
=

for some t' there is no

eigenmeasure supported on the whole J F . Then analysis in [MS, 61 implies that in fact
vtt =

6, and At/

p(a)-t (or equivalently s (t') =

-s-t'

with s- = l o g d p ( a ) ). Set

By the proof of 1.4 s ( t )

> 1, thus to < 0.

For t E (to, 0 ) by the discussion above s ( t ) is real analytic on (to, 0). Since

L,*S, = ,u(a)-t St = d-s-t 6,

we have s ( t )

> -s-t.

On the other hand by 1.7 for t 5 to

s ( t ) 5 - s ( t o ) = s- t
which implies that s ( t ) = -s-t for t E
(-00,

t to

to].

Remembering the relation P ( t ) = s ( t ) t - 1 we notice that to prove the main Theorem it remains to show that s ( t ) can be extended from

[to, 0) to a real analytic function on ( - 0 0 , 0 ) , and occurrence of the

phase transition together with

(4)implies condition

iib) .

2.7. Hidden spectrum

We want to analyze the case when F satisfies sition occurs.

(4) and a phase tran-

Let

:= min {multiplicity c E

F-'(a)). We will consider functions

I F ' I - ~ Ha ,t H,,t F ' Ha,&) := Iz - alAat ,

:=

G,,t

and associated transfer operator

For all

a!

<

=*

amax function

G,,t belongs to W1,JR) for p > 2

close to 2 and has zeroes of multiplicity

2 -tkmin

with

at the critical points of F . Also denote

kmax := max {multiplicity c E Crit F)

Set

r,(t)

:=

(~a,t,C

(a)),
a!

s, ( t ) := logd r , ( t ) .
We can repeat all previous arguments for the operator L,,t with
E

0 )

(or we can use results of the Section 3).

A few keypoints:

0bviously
n-1

(G,,t),

:=

G,,t

Fj

(2)

jqq

j=o 1

I F ' I - ~ Ha ,t Ha,t 0 Fn

and

(L:,$H f) ( z ) ,

this relation helps to reformulate properties of L,,t in terms of Lt. It means that considering L,,t we in fact restrict Lt to co-dimension one subspace W1,, (0)fl { f :

f (a) = 0) of W1,, (0) , "forgetting" the

eigenvalue of Lt which gives us the phase transition. Then note that the same reasoning as in 1.4 shows that for 0 5 a 5
A and ~ + l

d0

Since Ga,t has smaller orders of zeroes, to make La,t act on W1,,(fl) we need a stronger condition, particularly

will be sufficient. But this inequality is equivalent to

which is clearly true for p An analog of w (p, t)

> 2 close to 2.

< s(t) which we need for the proof of Ionescu-

Tulcea and Marinescu inequality is

<

s,(t)

with

This can be proved following 1.7, using the inequality sat ( t ) 5 s, ( t ) , which is true since
eo(n>

T,l

(t)n X
=

L Z I ,t 1( z )
LZ,,Hal-, ( z )

for fixed z E do. Like in 2.6 we obtained that s ( t ) is real analytic whenever s ( t )
-s-t,

>

here we conclude in the same way that whenever

s,(t)

>

-say-t

with

:= - logd G,,t(a)

1 -t

= (1 - a ) s-

all our theory is applicable to

hence the function s,(t)

is real

analytic and adjoint operator has eigenmeasure

with support dense in J F .

CHAPTER 2. NEGATIVE SPECTRA

Then for any a' E [a, a,,,] p we have H :=
E

72

Wl,,

(a)for

> 2 close to 2 and

implying for fixed z E 30 that

Thus (fix z E 30)

x (L;,,H) ( z ) =: r,(t)"

This leads us to

Claim. Let a E [0,a,,,)

and

t,

:= inf

{ t : s,(t) > -s,,-t)

Then s,(t) is real analytic on the interval (t,, 0 ) and is equal (on this interval) to s,, ( t )for any a' E [a, amax].
Set S ( t ) := s,,,

(t) . Noticing that so ( t )= s ( t ) , we conclude that

1) S ( t ) = s ( t ) for t E [to, 0).

2) S ( t ) is real analytic whenever 5 ( t ) > -s,,,,

So it remains to prove the following

,-t.

Lemma. For any t the condition S(t)

>

-s,,,,,

-t holds, i.e.

Proof. Throughout this proof for simplicity we denote L :=

and G := G,,,,

,t.

Function G (unlike G,,t for smaller a ) does not have zeroes at all critical points of F , moreover for any z E 0 at least one of its preimages is not a zero of G. Therefore we can consider the following operator on probability measures on JF: v
I-+

Var (L*V)-'

L*v

and by the Schauder theorem it fixes some measure which is an eigenmeasure of L*:

Clearly X 5 r,,,,

(t).

Consider some neighborhood of the point a which is mapped by F onto itself. Take small disk U inside it such that it's preimages Un under the branches of F-" preserving a are disjoint. Reasoning from [MS, 6.11 shows that suppv = JF thus v(U)

> 0.

Since v is an eigenmeasure of L* and simple distortion estimates imply

we have

=
Therefore

A-"

Gn (a) v(U)

>

( )

which implies G(a) < X and hence the desired inequality.

2.8. Phase transition and multipliers

We want to prove that the occurrence of a phase transition together with

(4) implies

Suppose it is not true and a phase transition occurs, though p(a) 5 p. For any cycle b = {bl, ..bn) for N = k n we have

{a) of order n (i.e. F n b l = bl) and

Thus S(t)

-tlogdp

-tlogdp(a) = -ts-

Together with real analyticity of S(t) this implies

But if a phase transition occurs at a point to then S(to) = s(to) = -to% and this condition is false. We got a contradiction therefore iib) follows from the phase transition together with

(4).

2.9. Multifractal analysis

For simplicity we will work out the case of connected Julia set. We already established the equality s(t) = this implies SUP

P(t) - t + 1, and by

[Mak2]

f (a)a

= n(t) =

a>O

1 P(t) = P(t) - t s(t) = log d

+1.
p
it will

Moreover, since we proved nice convergence of the limits for imply the same for n (t) and f ( a ) .

It remains to analyze the Hausdorff spectra. Following the Subsection 2.7 we notice that for some (any) z E A(oo) we have

and following 1.4 this implies dqt) where

K,

- dn(1-t) -

:= -a,,,t

> 0.

zl=l+d-n

lv'lt Iv - alK ,

Hence the spectrum S(t) represents the @-spectrum with the neutralized input from the point a. Particularly, that implies that if

w' := WLUis the restriction of w to some open set U not containing a, then

But (unlike the packing spectrum) the covering spectrum has "Hausdorff" properties, particularly if w = Therefore we deduce that

w j , then c, (t) = supj cUj(t) .

To establish the inverse inequality it is sufficient to consider the Hausdorff dimension spectrum f(a) and prove (cf. [Mak2, 2.21) that for
at := - l/Sr (t)

we have the inequality

7

f (at) >

inf (r

+ atS(r)) = t + atS(t)

= t -

S(t) . 5' (t)

=

By the Subsection 2.2 for any t there exists a probability measure p ,ut := ft ut with positive entropy such that

From the variational principleand convexity of g(t) we also deduce that for any r

<0

hence, by the smoothness of g(r), log d Z(t) = and also logd s(t) = h,

log ~ F ' d p ,

+ t logd

gt(t) .

Then, by Pesin's theory (see, e.g., result [Mafi3] of R. Maiie) for palmost every z the local behavior of w can be evaluated as lim log w Bg log w B6 log pBs = lim lim log6 6-0 10gll.B~6 3 0 log6

6-0

which together with the estimate of dimension (see [Mafi3] and [Y])

= (logd. S(t) - t logd

gt(t)) (-logd

gt(t))-l

gives the desired inequality !(at) proof.

2 t - S ( t )/St ( t ) and completes the

4

3. Transfer operator on Sobolev spaces

3.1. Main theorem
We will consider functional spaces on the complex sphere, or some open set 0 with
F-In

cR,

taken with spherical metric. Transfer operator on the space of continuous functions is defined by

where preimages y of z are counted with multiplicities. Note that

where g n ( y ) = g ( y ) g ( F y ) . . . g ( ~ n - l ( y ) . Denote by X the spectral radius:

X := lim llLn llml / n .

72-00

We assume that the weight g belongs to Sobolev space W17, for some p

> 2 close to 2, is non-negative and vanishes exactly at the

critical points of the dynamics F. Equivalently, we can assume that

g = h

IF'I~

for small r

> 0 and g, h, IF'IT

E W17, for some (any)

= h,

number p

> 2 sufficiently close to 2. Note that g,

1 FA IT.

We will assume that an additional condition is satisfied, namely

for some (any large) n, equivalently const qn An for any n and a constant q

>

SUPS,

< 1.

Theorem E. Under the assumption 2 operator L acts on W 1 , and

(0)for p > 2 sufficiently close to

ress (L, WI,,) < r (L, W17p) = .A

where r and re,, are spectral and essential spectral radii of L as an operator on Wl,, . Moreover, X is an isolated eigenvalue of multiplicity one.

Proof. After some technical lemmas contained in the Subsection 3.2 we

will establish Ionescu-Tulcea and Marinescu inequality in the Subsection 3.3. As in the Subsection 1.8, together with finite rank approximation it will imply quasicompactness of the transfer operator.

CHAPTER 2. NEGATIVE SPECTRA

80

Following the Subsection 2.3, we obtain that X is an isolated eigenvalue of L, and also an eigenvalue of the adjoint operator L* with some eigenmeasure v:

It remains to prove that suppv = J F , and then the argument of the Subsection 2.5 will yield that multiplicity of X is one. But considerations from [MS, 61 show that if this is not the case, then measure v must be supported on a single fixed point or an orbit of order 2 (the latter might happen only for Chebyshev polynomials). Let a be this point (or one of these two points), then for even n we have Fn(a) = a and hence

which contradicts the condition

(0).

3.2. Technical lemmas

Lemma. There exists q l

< 1 such that for small E > 0 we have

sup (g, 1 FA1

Proof. If we take
E

-') < const (ql

)n

to be small enough so that q

< (qX/ sup h) / t a u,

then there exists ql

< 1 such that

Therefore, noting that sup hn

< (sup h)n we can write

<

(sup h)n+ (const qn A ~ ) ?

proving the desired estimate.

Lemma. If 2 < p

2 < i=-; a n d g , h,

E WlYp, then L W 1 , c W I , ~ .

Proof. As before, it is sufficient to estimate

by const 11 f 11

Clearly

Here

and since we consider 2

< p < &,then r p + 2 - p > 0 and

The second term can be estimated by

n

llflI'(SUPl~l

TP+~--P

j(ah)pdm2

SUP(^)^

1(a

(IF/I~ IF^^^^+^-^ ))~

2 . I r ;

so it remains to show the convergence of the last integral, which, as before, is implied by the condition p <
W

3.3. Ionescu-Tulcea and Marinescu

inequality
Lemma. For p

> 2 close to 2, L acts on W 1 , and

IILZ"fIll,,

const qn An

llf lll,p + Cnllf llm

for some q < 1 and constant Cn.

Proof. Clearly

Choosing a small

E,

we can estimate the right hand side by

<

const (ql)pn

x~~

/ P('++-l) sup (1 F, 1

I const (ql)pn X

P ~

Ilaf l l i
- p)

if we assume that ( p ~ 2

> 0, i.e.

p is sufficiently close to 2.

Following the previous Subsection we obtain

which completes the proof.

Chapter 3 Parabolic case

In this chapter we will be concerned with parabolic polynomials, i.e. those where all critical points are attracted either to (super)attractive cycles or parabolic cycles. In the first section we will establish the quasicompactness of the transfer operator, and in the second apply it to the analyticity of spectrum, proving the following Theorem B. B1. For a parabolic polynomial F, the function PF(t)
is real analytic on [0,HDim JF)and PF(t) = 0, t E [HDim JF, +oo).

B2. The derivative of PF(t) is discontinuous a t the point HDim JF if and only if

HDim JF

> 2

- -

p+l '

where p is the maximal number of petals a t the parabolic points.

Dynamics in the parabolic case is not expanding, but expansive;

x 3 fi' x

I(fi)S- (4SI

R-xl

dns

=: ( S ) x d g

~ J O U SU!MOI~J ayl

y w

(x)3

JO

aaredsqns

sre (x)dg a x d s ayl auyap arn 3 3

x JOJ

-uo!?!ugaa

-[nav] u!
arn

paysyqre?sa sa!lxado~d :,!x)aur auros asn o s p

I ~ M

' ( [ ~aas) a ]slas reqny papauuo:, J!

papauuo:, L11~301 anrey 03 u ~ o u y

OM)

axre s~s~uroulC~od y:,ns 'l~red't! JBJ 08 L~~renluana sluurod

Lure 'L~aurreu

111. There exists constant C such that any x, y E U can be joined

inside U by the arc of length less than CIx - y 1.

Remark. The third property implies that for any 6 > 0, f E C(U)

Idea of the Proof. As in [ D H ,X.2.41, one can take some neighborhood

of the Julia set and then cut out a few angles, inside which critical points approach parabolic cycles.
H

1.2. Transfer operator

For z E U and t E R we define

Clearly for bounded subsets X

cC

for any n gn E Lip(X). This

allows us to consider the transfer operator on the Lip(U):

where preimages y of z are counted with multiplicities. Note that

Definition. For t E R and positive integer n we define the partition function

Denote also

X*
Theorem. Fix t

:= limsup
n--+w

v w .
:= vt on

R.
JF,and a positive

I. There exist a probability measure v

number X := At such that
L*v = Xu,

suppv = JF ,

and for any positive integer n and z E U

thus X is the spectral radius of the operator Lg in C(JF).

then X is the spectral radius and an isolated eigenvalue ofmultiplicity one of the operator Lt : Lip(U) + Lip(U). Furthermore for any positive integer n and z E JF

The first part of the theorem is proved in the sections 1.3-1.4 below, and the second in the sections 1.5-1.9. Throughout these sections we often omit t , denoting vt by v, At by A, etc.

1.3. Existence of conformal measure

Clearly, for p E Prob(JF), L*p is a positive non-zero measure. Thus we can consider the following operator on Prob:

P: p

I-+

L*P VarL*p

'

It fixes some measure v which is the desired one. Condition suppv = JF holds since z E suppv implies F-n(z) any point z E JF are dense in JF.

c suppv

and preimages of

1.4. Partition function

Lemma. For any positive integer n and x, y E U

Proof. Since U is compact it is sufficient to prove this for x, y E V,

where V is some small connected subset of U . Denote by nents of F - ~ v . By the Proposition 1.1. we know that supdiam
j %k

%k

compo-

o(l), k

--+oo

thus
Wk

:=

sup
j

SUPy 9 infvg

- ,o(n)

because g is separated from zero on U . Therefore

This implies our Lemma.

Remark. We also proved that for x,y in one component of connectivity

Now we can prove ( 2 ) . In fact for any x E JF we have

So ( 2 ) holds and the spectral radius of L, in C ( J F ) is equal to

CHAPTER 3. PARABOLIC CASE

We turn now to the proof of the second part of Theorem 1.1.

1.5. Prevalence of the expanding branches

Lemma. Fix Q > 1. For z E U denote

Then

Also
ZEU

inf l(Fn)' (z)l

2 e- o ( n ) .

Proof. Take V to be a neighborhood of z in U. Then

hence by the first part of the Theorem 1.2

which implies

The second statement can be proved in the same way.

1.7. Ionescu-Tulcea and Marinescu inequality

Suppose the condition (3) is satisfied. Normalize g ( z ) by setting

Denote

then

Lemma. In the above notation there exist 0 < q

< 1 and Mn = eo@)

such that

Remark. Note that in our construction, we obtain

" eO(l) ((Q2+t 1 ) + Q - ~ lln )

Choosing Q properly and taking large n, we can obtain some value of

q, depending only on X and t . This method, applied in parabolic or

hyperbolic case, will give an estimate on the radius of essential spectrum of the transfer operator Lt , in terms of t , P ( t ), and P ( f oo) only.

CHAPTER 3. PARABOLIC CASE

92

However, this does not provide us with an estimate for the absolute value of the second eigenvalue.
Proof. First we want to estimate

:=

# N n ( z ) inf

zEU

I (F")' (z)l

SUP

hn(y)

yen/,( z )

IILnlll, Q-n .

If t 2 0 then for h ( z )

and
SUP hn(y) Y ENn ( z )

5 llhnll, < - eo(n)pn

Choose Q

> 1 such that pQ2 < 1. Then

If t < 0 then

Clearly Lemma 1.5 implies that X that inequality

> 1, so we can choose Q > 1 such

Q2+t / X < 1 holds. Then

In any case we can pick large n (denote it by N) such that

for some q

(C is the constant from the Proposition 1.1).

For f E Lip consider such continuation

f E C(U) that

for example it is true for

Now if points z , z' E U are sufficiently close we have

where we denote by y' the preimage of z' obtained by the branch of

F-" for which F-"(2)

= y.

Using ( 6 ) we can estimate A by

and B (again, we suppose z and z' are sufficiently close) by

sup
Ix-y1<6

lx

-9 1

Therefore

with

Now by induction we have

with Ck Too,

This implies (5).

CHAPTER 3. PARABOLIC CASE

1.8. Quasicompactness
A Banach space linear operator L is called quasicompact if

for some compact operator K and some integer N. An equivalent stateof] L~ in the Calkin algebra (= bounded ment is that the image [ L ~ operators modulo compact) has norm strictly less than one. This implies that the spectral radius of [L] in the Calkin algebra is strictly less than one, and therefore there is r

< 1 such that the part of the

spectrum of L lying outside of the disk {lzl 5 r ) consists of a finite number of eigenvalues that all have finite geometric multiplicity.

Lemma. Any operator L on Lip satisfying ( 5 ) is quasicompact.

Proof. Cover J by some equilateral triangle B. By the Whitney extension theorem there exists bounded operator W : Lip (JF)+ Lip (B) with ( W f ) IJ,
= JF.

Set K := IIWIILi, 2 1 and fix N such that

IILNf

lLip

llf llLip + M If lloo, f

E Lip

Also fix small E (what we need is M K E 5

A) and take partition of B

E.

into equilateral triangles with sides less then on Lip(B) by the requirement

Define an operator P

Pf (x)is a linear function of

Pf

f on the vertices of our small triangles,

Rex ,Im x inside any of them.

Lip(B)

Then d i m P ( ~ i ~ (< ~ 00, ) ) and P

< 2.
< oo

Set now P := PW, P : Lip + Lip. It is clear that dim P (Lip) thus to prove that L is quasicompact it remains to check that

since llPllLip < -

11'

Lip

wLip 5 2~

and

we have

Returning to the proof of the second part of Theorem 1.2, we see that the last two lemmas imply that 1 is the spectral radius and an

isolated eigenvalue of L =: Lip 6, therefore X is the spectral radius and an isolated eigenvalue of L, : Lip 6 . It remains to show that dim

n
n>l

ker(Lg - A)"

= 1.

1.9. Multiplicity
In this section we denote by L the unnormalized operator L,. First we prove dim ker(L - A) = 1. Observe that if f satisfies Lf = Xf with X Indeed,

(8)

> 0 then Llfl

= Xlfl.

(Xlf 1 7 ) = (ILf 1 7 ) I (Llf 1 7 4

=

(If l,L*v) = (If I,

Xv)

and Ll f 1 = XI f 1 v-a.e. and hence everywhere because suppv = J F . Next observe that if f is a real-valued eigenfunction, L f = X f , then either f

2 0 or f I 0.

Otherwise, we can find a nontrivial eigenfunction The equation

f 2 0 (e.g. f = If 1 - f ) which is zero at some point.

then implies that

f must vanish on a dense set, hence everywhere.

Note that we also proved that eigenfunction f is strictly positive. Thus equality

implies (4): in fact, for any n and z E JF

Therefore we also have

Suppose now that we have two real functions

fl, f2

satisfying

Normalizing them by the condition

we have

and

fl = f2.

This proves (8).

It remains to show that ker(L = ker(L - A).

CHAPTER 3. PARABOLIC CASE

Suppose this is not true. Then there are f

# 0 and h such that

Lf

=X

f , L h = Xh+cf, ( c # 0).

By induction, we have

Lnh

An (h+ncX-'f)

which contradicts the relation (9).

2. Analyticity
2.1. Analyticity
Define P ( t ) := log At. Area estimate in the Subsection 1.5 shows that P ( t ) 5 0 for t

2 2, hence A+ 5

1. On the other hand, existence

of the parabolic cycle with multiplier 1 gives us A+

2 1. Thus A+

= 1,

and we can reformulate Theorem 1.2 in the following form:

As long as P ( t ) is positive, At is the spectral radius and an isolated eigenvalue of multiplicity one of the operator Lt : Lip(U) + Lip(U).

Let 6 be the first root of P ( t ) . Then At for t

< 6 is an isolated

eigenvalue of multiplicity one of the operator Lt which depends realanalytically on t, hence, by perturbation theory, the pressure P ( t ) is

real analytical on this interval. On the other hand, P ( t ) is decreasing and non-negative, thus P ( t ) = 0 for t

> 6.

To complete the proof of the Theorem B it remains to analyze the phase transition at the point 6.

2.2. Phase transition

By the paper [DU2] the Bowen's formula holds for parabolic maps, hence 6 = HDim J F . By [ADU] (see especially Theorem 9.9) there is a 6-conformal measure v = vb (which is either Hausdorff or Packing, depending on the dimension), and it admits an equivalent (a-finite) invariant measure p, which is finite if and only if the inequality holds. Now, if

(4)

(4) holds,

such a measure p is finite, thus (being equiva-

lent to the conformal measure v) it maximizes the expression in the variational principle:

and has a positive entropy, hence (by the variational principle)

The last integral is positive (it is equal to h,/S), therefore Pt(S-) <

0 = Pt(S+) and the derivative of the pressure is discontinuous at S.

On the other hand, suppose that Pt(S-)

< 0. Then for each t < S

there is an invariant measure pt (equivalent to the ( A t , t)-conformal measure vt: particularly, pt = ftvt, where ft is the eigenfunction of the transfer operator), for which Considering the limit of ft as t
-+

S log lFtldpt >

-Pt(6-)

> 0.

6 (and normalizing them properly)

it is easy to construct a finite invariant measure p equivalent to v = us and therefore deduce the inequality

(4).

Chapter 4 Semihyperbolic case

In this chapter we restrict ourselves to the study of non-recurrent quadratic polynomial F (z) = z 2

+ C:

We prove the following

Theorem D. For a non-recurrent quadratic polynomial F(2) = z2 c

either
(i) PF(t) is real analytic on [0,+oo), or (ii) A phase transition occurs: there exists to > HDim JF such that

PF(t) is real analytic on [0,t o ) , and

1 PF(t) = - 2 t log (lim inf,,,

1 (F")'(c) 1 'In)

on [to, +m) .

The chapter is organized as follows: in the Section 1 we introduce Hofbauer tower, changing the dynamics and forcing expansion. In the

Section 2 we prove quasicompactness of the transfer operator, and in the Section 3 investigate its eigenvalues and their connection to spectrum,

1. Yoccoz puzzle and Hofbauer tower

1.1. Preliminaries
Define
1

s(t)
for some (any
-

:= lim sup - logz

n+oo

C I( ~ ~ 9 ) ' l
-t
7

yF-*(z)

there is no difference, see Proof of the Lemma) z E

Lemma. For any z E A(m)

Therefore

Proof. Without loss of generality cp conjugates F with dynamics T :

z2 on

ID-:
Fop = poT.

Differentiating the identity Fn o cp = cp o Tn, we obtain

Applying this equality to the preimages

c E T-* of some fixed point

]ID- we notice that the right side is =: 2n, thus (taking to power t)

where y = p(C) is a corresponding preimage of z = p(<)under F n . The points

5 are equidistributed on the circle

2" with (r, - 1) =: -

Therefore summing over all

T-n< ( y E

F-%), we have

which completes the proof.

1.2. Tower construction

For non-recurrent polynomials dynamics on the orbit of the critical point is expanding, i.e

Q , := liminf IF"'(c)~' > 1 .

n+oo

CHAPTER 4. SEMIHYPERBOLIC CASE

105

Consider Yoccoz puzzle (see [Mi21 and [Hu] for the construction and its properties) and denote by Pn (0) the puzzle-piece of depth n

+ no

containing 0 (in this case 0 cannot lie on the boundary of a puzzlepiece). Since for such Julia sets diameter of puzzle-pieces tends to 0 when depth increases, we can choose no in such a way that Po(0) is disjoint from wO. Denote by 0 formal disjoint union of all puzzle pieces of depth no

n l , n l to be chosen later. Our tower will be a union of floors equivalent to subsets of

a:

where Fkr c 0. Define floors by

1.3. Metrics on the tower

We define metrics on the tower by setting

Q ~ - ( ~ ' - ' )Izl - z21, k1 = k2 > 1 kj = 1, Zj are in Izl p ((a, k l ) , ( ~ 2 k2)) , = the same puzzle-piece ( oo otherwise ,
where Qp is some constant satisfying Q,z
1

> Qp, to be chosen later. To

simplify notation we will write Ix - yl instead of p (x, y) for x, y E I.

1.4. Dynamics on the tower

Set

uk' := Fk+l' ,
and

z ) ~ ':= F k ' \ U k r ,

u
Define

:= (

k )

Dk := (Z)kl,k)

by

Note that

F is undefined for some points on the first level of the tower,

however transfer operator and its' formal adjoint are still well-defined (we adopt an agreement that sum of empty set of terms is zero). To simplify the formulas we will write F, in place of n-th iterate

pn.

1.5. Properties of the tower

Proposition.
i] Dynamics

maps 1-to-l

U k to Fk+1 Vk, k

,
< oo, then
they have

22

to a union of some components of Fl .

ii] If x, y are in one component, i.e. I x - yl preimages in the same components.

iii]

iv] By the choice of nl we can make and Q

expanding, i.e. for some C

>1
I F A I > CQn .

v] For some Q

> 1 for any n and C we can choose T in such a

way that y E .Fk,k down

2 T which after less than n

iterations goes

i],ii]. Follow directly from the construction. The property ii] means
that we can estimate IV (Lf (z))l through estimating IVf (y)l for y E

F-l.
iii]. First note that

6, := diam (F(Pk(0))) x IF;(C)~-' .

Take y E Pk \ Pk+1 = Dk , k

2 n1, from the distortion theorems

~ ~ ' ( = 9 ) I F ~ ( YF )( I~-~)'F(Y)

= =
n

IyI

IF(~-')'(C)~

diam ( F ( P(o)))$ ~ F(~~)'(c)

1

/F(kl)'(C)T

Therefore ~ ' ( y , k )= l Q,-("~)IF~'(~)~

Q~

F(k-l)'(C)

1 2

Q ~ - ~ Q ~ - T

iv], v ] . Follow from Qp < Q , 1 .

2. Analysis of the transfer operator

2.1. Transfer operator on the tower
We will consider two function spaces on the tower: C (I), Lip (7) with norms

:=

SuPllfllW,Fk
k

For t E R we define operator Lt on C ( I ) by

note that

Denote by At the spectral radius r ( L t ,C ( I ) ) .

Lemma. Lt is bounded operator on C ( I )and , At

2S(t) .

Proof. The boundness of L f follows from the property v] of the tower.

To prove the inequality take some point z E A(m) such that

For such z and n and

2 nl

all points y E F-"z satisfy pn ( y , 1 ) = ( z , ~ )

F~~( y , 1 ) = ~

~ 'Thus y . by Lemma 1.1 for n

2 nl

2.2. Ionescu-Tulcea and Marinescu inequality

Lemma. Suppose At > Q,-'~. Then for n >> 1 and T >> n we have

Proof.
For z E
Fk,

< n we will use the estimate (since all functions are

Lipschitz the gradients are defined almost everywhere, and later we will be able to consider an essential supremum)

ye@-" ( t )

5 sup Lip (f,.Fk) Q-n A; eo(n) + C ILn f ( z )I ,

k

where C is an absolute constant.

Note that z E &, k

2 n has only one preimage y E Fk-,

and in its

neighborhood Fn increases all distances exactly by a multiplier of Q p n , thus

Using the latter inequality for z E Fk, k

2n

we obtain that for

> > 1 for sufficiently large T >> n

sup Lip (Ln f ) = sup ess-supFkV L nf

k k

<

'
5

A; Lip ( f )

+ C IILnf,1
5(C+1)

Now estimate in the same way 1 f (z) 1. Since At

> Q,-~, for large n we have Q p-tn <

An. Then for

z E F k , k 2 n we can estimate

where y E Fn-k is the only preimage of z. For z E Fk, k < n number of preimages of z on each level of the tower is bounded by some constant K (depending only on n). We can

CHAPTER 4. SEMIHYPERBOLIC CASE

choose T in such a way that for some q

112

< 1 and any y

Fk, k 2T

which after less then n iterations goes down

Then

Together these estimates give us

CHAPTER 4. SEMIHYPERBOLIC CASE

Therefore

< - A," supLip(f)

-

<
-

1 5 1 - A," sup Lip (f) 5

' '

( C + l ) IILnfll,

2.3. Finite rank approximation

Lemma. For any integer T and 6 > 0 there exists a finite rank oper-

ator M in Lip (7)such that

IIMIILip

5 1

, 5 6 IIfllLip,Fk
Mf(z) = 0

I ~ < T ,

k>T.

Proof. Cover .Fk, k

< T with a grid of triangles q5 of size 6 << 1. Define

M f as a continuous function vanishing on Mf =

on F', k

F ' , k 2 T and satisfying

f at all vertices is linear in each triangle q5

< T.The properties are obvious.

CHAPTER 4. SEMIHYPERBOLIC CASE

2.4. Quasicompactness
Claim. If At > Q ~ then - ~
r(Lt, Lip (7)) =
and
At

Proof.

First recall that r (Lt , C (7)) = Also

At

and At 5 r (Lt ,Lip (7)) To prove the converse, by the Ionescu-Tulcea and Marinescu inequality choose E

> 0 and N such that

IILNf

l Lip

'F-'

llf llLip + K llf,l

Define inductively an increasing sequence { M k ) by

Then

and we can choose a constant C satisfying

cA,"-&
By induction,

+ 1 5 C A ; and K 5 C M l .

1ILNkffIlLiP 5
N

',"k-ek

llf

llLip
7

'Mk

l l f ll,

<

'tNk llf llLip

therefore

r ( L t ,Lip ( l ) I) At ,
which implies the desired inequality and thus the first statement. To prove the second statement choose n and T such that

and a finite rank approximation M for S

<

&X r.

Then

and

3. Analyticity of spectrum
In the subsections 3.1 Claim 2.4 holds.
-

3.3 we will assume that At > Q ~ - , , i.e.

3.1. Eigenvalues and multiplicity

Fix t and denote X = At := r (Lt, C ( I ) ) .

Lemma. ker (Lt - A)

eigenfunction f = ft.

# (0).

Moreover there is a strictly positive

Proof. See [Ru6, Theorem 2.21.

Idea of the proof: since re,, ( L t ,Lip ( I ) ) < r ( L t ,Lip ( I ) ) = A, we have some eigenvalues X j of absolute value A, and a decomposition

where Xo is the subspace corresponding to the rest of the spectrum. Then

Here L r X o = o ( A n ) and L r X j = X y X j , implying that one of the

X j ' s is positive and hence is equal to

X = r ( L t ,C ( I ) ) .

Also it follows that there is non-negative (and hence strictly positive: if f ( z ) = 0 then f ( y ) = 0 for Fny = z, from continuity f s 0 on JF and by Sublemma everywhere) eigenfunction f with eigenvalue A.

Sublemma. For f E Lip (7)

Ltf

Xf

implies f = 0

.
distortion estimates

Proof of Sublemma. For fixed z E 7 and y

imply

l~:(~)l-'

x dist

( y , j F ).

Therefore

< X-"
YEP-"

IF;(Y)I-~

(y,jF)

< X-"

< A-"

yF-" ( z ) 1 ~ ~ ; ( y ) l5 - ~eO(")/Q-" CQn y F - n ( z )

(z)

I~A(Y)I-~-~

72-00

0.

Consequently f = 0 and Sublemma is proved.

3.2. Conformal measures

Consider the formal adjoint L* to L, acting on the space of measures

Lemma. There is a probability measure v = vt supported on 7 with

L*v = X v .
Proof. Clearly

CHAPTER 4. SEMIHYPERBOLIC CASE

119

Take some positive measure p supported on the first floor of the tower and define

Then

thus to find v it is sufficient to prove that some subsequence {p,) has a weak limit. The only trouble is that 7 is not compact. But one can easily check by induction that for small Q > 1 and large positive C

which is sufficient.

Remark. Denote
ant under

SF

:= Uk { ( z ,k ) E Fk :

z E J F ) . Then jF is invari-

and suppv

> SF.

3.3. Pressure and spectrum

Lemma. We have X = 2'@).
Proof. We know that X

2 2S(t) 2

Q,-5.

Assume the claim of the

Lemma is not true. Then X

> 2S(t). Chose such X1 and Qp that

By the definition of s ( t ) the series

is convergent, hence we can define a finite measure uo on JF by

Note that duo(Fy) = X1 ~ F ' ~ l - ~ d u ~ ( y ) . Now transfer this measure to the tower by setting

we obtain a finite measure again (since Xy QMnt check that the mentioned property of uo implies

< 1). It is easy to

and hence L* ul = X1 u1. Therefore for the positive eigenfunction f we can write

Since f is strictly positive this implies X = XI, and we obtain a contradiction.

rn

Claim. W e have dim ker (Lt - x

)~ =

1, i.e. X is a simple eigenvalue.

Proof. If we consider only functions in C JF

( - 1, the same reasoning as

in [MS, 83.61 gives us that dim ker (Lt - A) = 1, and then Sublemma implies this for f E Lip ( 7 ) . Lemma 3.1 shows that we have a non-negative eigenfunction f : ( L , - A) f = 0 .

If X is not a simple eigenvalue, then there exists function h such that

(Lt - A) h = f . Therefore

Thus v-a.e. we have f = 0, hence everywhere on jF and (by the Sublemma) in 7.

3.4. Analyticity
Claims 2.4 and 3.3 show that if for some t' we can choose Q p in such a way that At

> QPwt > Q,-5 then 2S(t) is a simple isolated

eigenvalue of Lt : Lip (7) 0.Clearly it implies that Lt depends real

CHAPTER 4. SEMIHYPERBOLIC CASE

122

analytically on t therefore s(t) and hence P(t) is real analytic in the neighborhood of t' (see [MS, 4.11). By Lemma 2.1 At

2'@), thus inequality 2S(t) >

Q,-g

is sufficient

to find Qp satisfying our conditions. On the other hand we always have 2s(t)

2 Q C - i thus to prove the main Theorem we only have to check

First note that 6,

:= diam

(F(Pn(0))) x

I FA(C)

1-l

We can choose a point z E A(oo) in each part of R, so that for any

n one of them has a preimage y E Pn\ Pn+l under the F-". From the
distortion theorems

x diam (F(pn (0)))

F("-')'(c)

Then for the corresponding z:

and taking limsup as n

-+ oo

we obtain the desired inequality.

References

[ADU]

Aaronson, Jon; Denker, Manfred; Urbafiski, Mariusz, Ergodic theory for Markov fibred systems and parabolic rational maps, Trans. Amer. Math. Soc. 337 (1993)) no. 2, 495-548.

[Bal

Baladi, Viviane, Dynarnical zeta functions, Real and complex dynamical systems (Hiller~d, 1993), NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., 464, Kluwer Acad. Publ., Dordrecht, 1995, pp. 1-26.

[BK]

Baladi, V.; Keller, G., Zeta functions and transfer operators for piecewise monotone transformations, Comm. Math. Phys. 127 (1990), no. 3, 459477.

[BGH]

Barnsley, M. F.; Geronimo, J. S.; Harrington, A. N., Orthogonal polynomials associated with invariant measures on Julia sets, Bull. Amer. Math. Soc. (N.S.) 7 (1982), no. 2, 381-384.

[BH]

Barnsley, M. F.; Harrington, A. N., Moments of balanced measures on Julia sets, Trans. Amer. Math. Soc. 284 (1984), no. 1, 271-280.

[Be]

Beardon, Alan F., Iteration of rational functions. Complex analytic dynamical systems, Graduate Texts in Mathematics, 132, Springer-Verlag, New York, 1991.

Beck, Christian; Schlogl, Friedrich, Thermodynamics of chaotic systems. An introduction, Cambridge Nonlinear Science Series, 4, Cambridge University Press, Cambridge, 1993. Bodart, 0 . ; Zinsmeister, M., Quelques re'sultats sur la dimension de Hausdorfl des ensembles de Julia des polyndmes quadratiques, Preprint Universitb d'orlbans, Dbpartement de Mathhmatiques URA 1803, 95-08 (1995). (French) Bohr, T.; CvitanoviC, P.; Jensen, M. H., Fractal aggregates in the complex plane, Europhys. Lett. 6 (1988), no. 5, 445-450. Bowen, Rufus, Equilibrium states and the ergodic theory of Anosov diffeomorphisms, Lecture Notes in Mathematics, Vol. 470, Springer-Verlag, Berlin-New York, 1975.

Hausdorfl dimension of quasicircles, Inst. Hautes ~ t u d e s Sci.

Publ. Math. No. 50 (1979), 11-25. Brolin, Hans, Invariant sets under iteration of rational functions, Ark. Mat. 6 (1965), 103-144. Bruin, Henk, Invariant measures of interval maps, Thesis, Technische Universiteit te Delft, 1995. Carleson, Lennart, On the support of harmonic measure for sets of Cantor tgpe, Ann. Acad. Sci. Fenn. Ser. A I Math. 10 (1985), 113-123. Carleson, Lennart; Gamelin, Theodore W., Complex dynamics, Universitext: Tracts in Mathematics, Springer-Verlag, New York, 1993.

Carleson, Lennart ; Jones, Peter W . , O n coefficient problems for univalent functions and conformal dimension, Duke Math. J . 66 (1992),no. 2 , 169206. Carleson, Lennart; Jones, Peter W . ; Yoccoz, Jean-Christophe, Julia and John, Bol. Soc. Brasil. Mat. (N.S.) 25 (1994), no. 1, 1-30. Carleson, Lennart; Makarov, Nikolai G., Some results connected with Brennan's conjecture, Ark. Mat. 32 (1994), no. 1 , 33-62. Denker, Manfred; Seck, Christoph, A short proof of a theorem of Ruelle, Monatsh. Math. 108 (1989), no. 4 , 295-299. Denker, Manfred; Urbariski, Mariusz, Ergodic theory of equilibrium states for rational maps, Nonlinearity 4 (1991), no. 1, 103-134.

, The capacity
73-86.

of parabolic Julia sets, Math. Z . 211 (1992), no. 1 ,

, Hausdorff and conformal measures o n Julia sets with a rationally

indifferent periodic point, J . London Math. Soc. ( 2 ) 43 (1991),no. 1 , 107118.

, Hausdorff measures o n Julia sets of subexpanding rational maps,

Israel J . Math. 76 (1991), no. 1-2, 193-214.

, O n Sullivan's conformal measures for rational maps of the Riem a n n sphere, Nonlinearity 4 (1991), no. 2, 365-384.

, Absolutely continuous invariant measures for expansive rational

maps with rationally indiflerent periodic points, Forum Math. 3 (1991))

no. 6, 561-579. [DU7]

, Geometric measures for parabolic rational maps, Ergodic Theory

Dynamical Systems 12 (1992), no. 1, 53-66.

[DPU]

Denker, Manfred; Urbariski, Mariusz; Przytycki, Feliks, O n the transfer

operator for rational functions o n the Riemann sphere, Preprint SFB 170

Gottingen, 4 (1990); Ergodic Theory Dynamical Systems (to appear). [DH]

dynamique des polyn6mes complezes. Douady, A.; Hubbard, J . H . , ~ t u d e Partie I, Publications Mathematiques d'Orsay, 84-2, Universite de Paris-

Sud, Departement de Mat hematique, Orsay, 1984 (French) ; Partie 1 1 , with the collaboration of P. Lavaurs, Tan Lei and P. Sentenac, Publications Mathematiques d'Orsay, 85-4, Universite de Paris-Sud, Departement de Mathematique, Orsay, 1985. (French) [Epl Eckmann, Jean-Pierre; Procaccia, Itamar, Fluctuations of dynamical scaling indices in nonlinear systems, Phys. Rev. A (3) 34 (1986), no. 1, 659-

661. [ELI] Eremenko, A. E.; Levin, G. M., Periodic points of polynomials, Ukrain. Mat. Zh. 41 (1989), no. 11, 1467-1471, 1581; (Russian); English translation in Ukrainian Math. J. 41 (1989), no. 11, 1258-1262 (1990). [EL21

, Estimation of the characteristic

exponents of a polynomial, Teor.

Funktsii Funktsional. Anal. i Prilozhen. No. 58 (1992), 30-40 (1993). (Russian) [ELS] Eremenko, A.; Levin, G.; Sodin, M., O n the distribution of zeros of a

Ruelle zeta-function, Comm. Math. Phys. 159 (1994), no. 3, 433-441. Eremenko, A. E.; Lyubich, M. Yu., The dynamics of analytic transformations, Algebra i Analiz 1 (1989), no. 3, 1-70; (Russian); English translation in Leningrad Math. J. 1 (1990)) no. 3, 563-634. Falconer, Kenneth, Fractal geometry. Mathematical foundations and applications, John Wiley & Sons, Ltd., Chichester, 1990. Feder, Jens, Fractals, Physics of Solids and Liquids, Plenum Press, New York-London, 1988. Feigenbaum, Mitchell J. ; Procaccia, Itamar; TQ1, Tam&, Scaling properties of multifractals as an eigenvalue problem, Phys. Rev. A (3) 39 (1989), no. 10, 5359-5372. Freire, Alexandre; Lopes, Artur; Ma%, Ricardo, An invariant measure for rational maps, Bol. Soc. Brasil. Mat. 14 (1983), no. 1, 45-62. Garnett, John; Marshall, Don, Harmonic Measure, Monograph (in preparation). Garnett, Lucy, A computer algorithm for determining the Hausdorfl dimension of certain fractals, Math. Comp. 51 (1988)) no. 183, 291-300. Graczyk, Jacek; Smirnov, Stas, Collet, Eckmann, and Holder, Preprint (1996). Grot hendieck, Alexandre, Produits tensoriels topologiques et espaces nucle'aires, 1955. (French)

, La thkorie de Fredholm, Bull. Soc. Math. France 84 (1956), 319-

384. (French) [Gul Guckenheimer, John, Endomorphisms of the Riemann sphere, 1970 Global Analysis (Proc. Sympos. Pure Math. Vol, XIV, Berkeley, Calif., 1968), Amer. Math. Soc., Providence, R.1, pp. 95-123. [HJKPS] Halsey, Thomas C.; Jensen, Mogens H.; Kadanoff, Leo P.; Procaccia, Itamar; Shraiman, Boris I., Fractal measures and their singularities: the characterization of strange sets, Phys. Rev. A (3) 33 (1986), no. 2, 11411151; Erratum, Phys. Rev. A (3) 34 (1986), no. 2, 1601. [Hal Haydn, Nicolai, Convergence of the transfer operator for rational maps, Preprint (1996). [HR] Haydn, N. T . A,; Ruelle, D., Equivalence of Gibbs and equilibrium states for homeomorphisms satisfying expansiveness and specification, Comm. Math. Phys. 148 (1992), no. 1, 155-167. [Hill Hinkkanen, A,, Julia sets of rational functions are uniformly perfect, Math. Proc. Cambridge Philos. Soc. 113 (1993), no. 3, 543-559. [Hi21 Hinkkanen, A., Zeta functions of rational functions are rational, Ann. Acad. Sci. Fenn. Ser. A I Math. 19 (1994), no. 1, 3-10. [H0 1 Hofbauer, Franz, Piecewise invertible dynamical systems, Probab. Theory Relat. Fields 72 (1986), no. 3, 359-386. [HKl] Hofbauer, Franz; Keller, Gerhard, Ergodic properties of invariant measures for piecewise monotonic transformations, Math. Z. 180 (1982), no. 1, 119-140.

[HK2]

, Equilibrium states for piecewise

monotonic transformations, Er-

godic Theory Dynamical Systems 2 (1982), no. 1, 23-43. [HK3]

, Zeta-functions

and transfer operators for piecewise linear trans-

formations, J . Reine Angew. Math. 352 (1984), 100-1 13. [HK4]

Equilibrium states and Hausdorfl measures for interval maps,

Math. Nachr. 164 (1993)) 239-257. [Hul Hubbard, J. H., Local connnectivity of Julia sets and bifurcation loci: three theorems of J.-C. Yoccoz, Topological methods in modern mathematics (Stony Brook, NY, 1991), Publish or Perish, Houston, T X , 1993, pp. 467-511. [IMl Ionescu-Tulcea, C. T.; Marinescu, G., The'orie ergodique pour des classes d'opirations non compl4tement continues, Ann. of Math. (2) 52 (1950), 140-147. (French)
[Jl]

Jakobson, M. V., The structure of polynomial mappings on a special set, Mat. Sb. (N.S.) 77 (119) (1968), 105-124. (Russian)

[J21

On the question of classijication of polynomial endomorphisms

of the plane, Mat. Sb. (N.S.) 80 (122) (1969), 365-387. (Russian) 1531

, On the

question of the topological classification of rational map-

pings of the Riemann sphere, Uspehi Mat. Nauk 28 (1973), no. 2(170), 247-248. (Russian)

[541

, Markov

partitions for rational endomorphisms of the Riemann

sphere, Multicomponent random systems, Adv. Probab. Related Topics,

6, Dekker, New York, 1980, pp. 381-396. Jones, P. W.; Makarov, Nikolai G., Density properties of harmonic measure, Ann. of Math. (2) 142 (1995), no. 3, 427-455. Keller, G., Markov extensions, zeta functions, and Fredholm theory for piecewise invertible dynamical systems, Trans. Amer. Math. Soc. 314 (1989), no. 2, 433-497. Keller, Gerhard; Nowicki, Tomasz, Spectral theory, zeta functions and the distribution of periodic points for Collet-Eckmann maps, Comm. Math. Phys. 149 (1992), no. 1, 31-69. Lalley, Steven P., Brownian motion and the equilibrium measure on the Julia set of a rational mapping, Ann. Probab. 20 (1992), no. 4, 19321967. Levin, G. M., Bounds for multipliers of periodic points of holomorphic mappings, Sibirsk. Mat. Zh. 31 (1990), no. 2, 104-110, 224; (Russian); English translation in Siberian Math. J. 31 (1990), no. 2, 273-278. [LSY1 1 Levin, G.; Sodin, M.; Yuditskii, P., Ruelle operators with rational weights for Julia sets, J . Anal. Math. 63 (1994), 303-331. [LSY2 1

,A

Ruelle operator for a real Julia set, Comm. Math. Phys. 141

(1991), no. 1, 119-132. [Loll Lopes, Artur Oscar, Equilibrium measures for rational maps, Ergodic Theory Dynamical Systems 6 (1986), no. 3, 393-399. [Lo21

The complex potential generated by the maximal measure for a

family of rational maps, J. Statist. Phys. 52 (1988), no. 3-4, 571-575. [Lo31

, The dimension spectrum of the maximal measure, SIAM J . Math.

Anal. 20 (1989), no. 5, 1243-1254.

~041 546. [Lo51

, Entropy

and large deviation, Nonlinearity 3 (1990), no. 2, 527-

, Dimension

spectra and a mathematical model for phase transi-

tion, Adv. in Appl. Math. 11 (1990), no. 4, 475-502. [Lo61

, Dynamics

of real polynomials on the plane and triple point phase

transition, Math. Comput. Modelling 13 (1990), no. 9, 17-31. [Lyl] Lyubich, M. Yu., Entropy of analytic endomorphisms of the Riemann sphere, Funktsional. Anal. i Prilozhen. 15 (1981), no. 4, 83-84. (Russian) [LY 2 1

The measure of maximal entropy of a rational endomorphism

of a Riemann sphere, Funktsional. Anal. i Prilozhen. 16 (1982), no. 4, 78-79. (Russian) [LY~I

, Entropy properties

of rational endomorphisms of the Riemann

sphere, Ergodic Theory Dynamical Systems 3 (1983)) no. 3, 351-385. [LVl] Lyubich, M.; Volberg, A,, A comparison of harmonic and maximal measures for rational functions, Approximation by solutions of partial differential equations (Hanstholm, 1991), NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., 365, Kluwer Acad. Publ., Dordrecht, 1992, pp. 127-139. [LV21

,A

comparison of harmonic and balanced measures on Cantor

repellors, Proceedings of the Conference in Honor of Jean-Pierre Kahane,

J. Fourier Anal. Appl. (1995), Special Issue, 379-399.

[Makl] Makarov, N. G., On the distortion of boundary sets under conformal mappings, Proc. London Math. Soc. (3) 51 (1985), no. 2, 369-384. [Mak2]

, Fine structure of harmonic measure (Talks given a t the Strobl

Winter School in Analysis, 1994), Preprint (1995).

[MSI

Makarov, N.; Smirnov, S., Phase transition in subhyperbolic Julia sets, Ergodic Theory Dynamical Systems 16 (1996), no. 1, 125-157.

[MVI

Makarov, N.; Volberg, A., On the harmonic measure of discontinuous fractals, Preprint LOMI E-6-86 (1986).

[Mand]

Mandelbrot , B. B., An introduction to rnultifractal distribution functions, Random fluctuations and pattern growth (CargBse, 1988), NATO Adv. Sci. Inst. Ser. E Appl. Sci., 157, Kluwer Acad. Publ., Dordrecht, 1988, pp. 279-291. Ma%, Ricardo, On the uniqueness of the maximizing measure for rational maps, Bol. Soc. Brasil. Mat. 14 (1983), no. 1, 27-43.

, On the Bernoulli property for rational maps, Ergodic Theory

Dynamical Systems 5 (1985), no. 1, 71-88.

, The Hausdorfl dimension of invariant probabilities of rational

maps, Dynamical systems (Valparaiso 1986), Lecture Notes in Math., 1331, Springer, Berlin-New York, 1988, pp. 86-117. MafiQ,R.; da Rocha, L. F., Julia sets are uniformly perfect, Proc. Amer. Math. Soc. 116 (1992), no. 1, 251-257.

[Man1

Manning, Anthony, The dimension of the maximal measure for a polynomial map, Ann. of Math. (2) 119 (1984), no. 2, 425-430.

[Mil]

Milnor, J., Dynamics in One Complex Variable: Introductory lectures, SUNY Stony Brook Institute for Mathematical Sciences Preprints 5 (1990).

[Mi21

Local Connectivity of Julia Sets: Expository Lectures, SUNY

Stony Brook Institute for Mathematical Sciences Preprints 11 (1992). [MPI Misiurewicz, Michal; Przytycki, Feliks, Topological entropy and degree of smooth mappings, Bull. Acad. Polon. Sci. SQr.Sci. Math. Astronom. Phys.
25 (1977), no. 6, 573-574.

Nussbaum, Roger D., The radius of the essential spectrum, Duke Math. J. 37 (1970), 473-478. Ott, E.; Withers, W. D.; Yorke, J. A., Is the dimension of chaotic attractors invariant under coordinate changes? J . Statist. Phys. 36 (1984), no. 5-6, 687-697. Patterson, S. J., The limit set of a Fuchsian group, Acta Math. 136 (1976), no. 3-4, 241-273.

Lectures on measures on limit sets of Kleinian groups, Ana-

lytical and geometric aspects of hyperbolic space (Coventry/Durham, 1984), London Math. Soc. Lecture Note Ser., 111,Cambridge Univ. Press, Cambridge-New York, 1987, pp. 281-323. Pesin, Yakov; Weiss, Howard, A multzfractal analysis of equilibrium measures for conformal expanding maps and Moran-like geometric construc-

tions, Preprint (1996). Petersen, Carsten Lunde, On the Pommerenke-Levin- Yoccox inequality, Ergodic Theory Dynamical Systems 13 (1993), no. 4, 785-806. Pollicott , Mark, A complex Ruelle-Perron-Frobenius theorem and two counterexamples, Ergodic Theory Dynamical Systems 4 (1984), no. 1, 135-146. Pommerenke, Ch., On conformal mapping and iteration of rational functions, Complex Variables Theory Appl. 5 (1986), no. 2-4, 117-126. Przytycki, Feliks, On the Perron-Frobenius-Ruelle operator for rational maps on the Riemann sphere and for Holder continuous functions, Bol. Soc. Brasil. Mat. (N.S.) 20 (1990), no. 2, 95-125. Przytycki, Feliks; Urbariski, Mariusz; Zdunik, Anna, Harmonic, Gibbs and Hausdorfl measures on repellers for holomorphic maps. I, Ann. of Math. (2) 130 (1989), no. 1, 1-40; 11, Studia Math. 97 (1991), no. 3, 189-225. Ruelle, David, Thermodynamic formalism. The mathematical structures of classical equilibrium statistical mechanics, Encyclopedia of Mathematics and its Applications, 5, Addison-Wesley Publishing Co., Reading, Mass., 1978.

, Repellers for real analytic maps, Ergodic Theory Dynamical Systems 2 (1982), no. 1, 99-107.

Bowen's formula for the Hausdorfl dimension of self-similar

sets, Scaling and self-similarity in physics (Bures-sur-Yvette, 1981/1982), Progr. Phys., 7, Birkhuser Boston, Boston, Mass., 1983, pp. 351-358. [Ru~]

, The thermodynamic formalism for expanding maps, Comm. Math.

Phys. 125 (1989), no. 2, 239-262.

[Ru5]

, Spectral properties of a class of operators associated with maps

in one dimension, Ergodic Theory Dynamical Systems 11 (1991), no. 4, 757-767.

[RUG]

, Spectral properties of a class of operators associated with conformal maps in two dimensions, Comm. Math. Phys. 144 (1992), no. 3, 537-556.

~ u 7 1

, Thermodynamic formalism for maps satisfying positive expansiveness and specification, Nonlinearity 5 (1992), no. 6, 1223-1236.

[Rug]

, Dynamical zeta functions for piecewise monotone maps of the interval, CRM Monograph Series, 4, American Mat hematical Society, Providence, RI, 1994.

[RY

Rychlik, Marek, Bounded variation and invariant measures, Studia Math.

76 (1983), no. 1, 69-80.

[STV]

Servizi, G. ; Turchetti, G.; Vaienti, S., Generalized dynamical variables and measures for the Julia sets, Nuovo Cimento B (11) 101 (1988), no. 3, 285-307.

[Shl

Shishikura, M., The Hausdorfl Dimension of the boundary of the Mandelbrot set and Julia sets, Ann. of Math. (2) (to appear); See also The

Hausdorff Dimension of the boundary of the Mandelbrot set and Julia sets, SUNY Stony Brook Institute for Mathematical Sciences Preprints 7 (1991). [Sul] Sullivan, Dennis, The density at infinity of a discrete group of hyperbolic motions, Inst. Hautes ~ t u d e Sci. s Publ. Math. No. 50 (1979), 171-202. [SU~]

, Conformal dynamical systems, Geometric dynamics (Rio de Janeiro, 1981), Lecture Notes in Math., 1007, Springer, Berlin-New York, 1983, pp. 725-752.

[Su31

, Entropy,

Hausdorff measures old and new, and limit sets of ge-

ometrically finite Kleinian groups, Acta Math. 153 (1984), no. 3-4, 259277. [ST] SzQpfalusy, P.; TQ1,T., Dynamical fractal properties of one-dimensional maps, Phys. Rev. A (3) 35 (1987), no. 1, 477-480. [TI] TQ1, T . , Dynamical spectrum and thermodynamic functions of strange sets from an eigenvalue problem, Phys. Rev. A (3) 36 (1987)) no. 5, 2507-2510. [T21

Fractals, multifractals, and thermodynamics. An introductory

review, Z. Naturforsch. A 43 (1988), no. 12, 1154-1174.

[u] [vl

Urbaliski, Mariusz, Rational functions with no recurrent critical points, Ergodic Theory Dynam. Systems 14 (1994), no. 2, 391-414. Vaienti, S., Lyapunov exponent and bounds for the Hausdorfl dimension of Julia sets of polynomial maps, Nuovo Cimento B (11) 99 (1987), no. 1,

77-9 1.

[wl

Weiss, Howard, The lyapunov and dimension spectra of equilibrium measures for conformal repellers, Preprint (1996).

[y1

Young, Lai Sang, Dimension, entropy and Lyapunov exponents, Ergodic Theory Dynamical Systems 2 (1982), no. 1, 109-124.

[Zdl

Zdunik, Anna, Harmonic measure versus H a u s d o r -measures on repellers for holomorphic maps, Trans. Amer. Math. Soc. 326 (1991), no. 2, 633652.

[zi]

Ziemer, William P., Weakly diflerentiable functions. Sobolev spaces and functions of bounded variation, Graduate Texts in Mathematics, 120, Springer-Verlag, New York, 1989.