Removable singularities for planar quasiregular mappings
Albert Clop
PhD Thesis
Memoria` presentada per aspirar al grau de doctor en Matematiques`
iii iv CERTIFIQUEM que aquesta memoria` ha estat realitzada per Albert Clop, sota la direccio´ dels Drs. Joan Mateu Bennassar i Joan Orobitg Huguet.
Bellaterra, octubre del 2006.
Joan Mateu Bennassar
Joan Orobitg Huguet
v
Introduction
A compact planar set E is said to be removable for bounded analytic functions if for any open set Ω ⊃ E, any bounded function f : Ω → C holomorphic on Ω \ E admits a holomorphic extension to Ω. It is enough to care about the case Ω = C. The Painleve´ problem consists on characterizing these sets in metric and geometric terms.
The analytic capacity of a compact set E, introduced by Ahlfors [3], is defined by
0 γ(E) = sup | f (∞)|; f ∈ H(C \ E), k f k∞ ≤ 1, f (∞) = 0 . This set function vanishes only on removable sets. Some excellent introductions to γ are given by H. Pajot [43] and J. Verdera [55]. As a matter of fact, H1(E) = 0 implies γ(E) = 0, while if dim(E) > 1 then γ(E) > 0. Thus, 1 is the critical dimension for the Painleve´ problem.
Some easier variants of this problem are obtained when replacing L∞ by other classes < of functions, such as BMO or Lipα for 0 α ≤ 1. To specify:
• Kaufman [31] showed that E is removable for BMO analytic functions if and only if M1(E) = 0.
• Dolzenkoˇ [16] proved that E is removable for Lipα analytic functions if and only if M1+α(E) = 0.
• It follows from Verdera’s work [54] that E is removable for VMO analytic func- 1 tions if and only if M∗(E) = 0.
1 As it can be seen, the main difference between these variants and the classical Painleve´ problem is that now removable sets are characterized by a Hausdorff content or, in some cases, other set functions, that in any case are not so complicated as analytic ca- pacity.
The main purpose of this thesis is the study of removable sets for quasiregular map- pings of the plane. Given a domain Ω ⊂ C, a function f : Ω → C is said to be µ- Ω 1,2 Ω quasiregular on if it belongs to the Sobolev space Wloc ( ) and its derivatives satisfy at almost every point the so called Beltrami equation,
∂ f = µ ∂ f
K−1 < for some measurable function µ, the Beltrami coefficient, such that kµk∞ = K+1 1. If we forget µ and just care about K, then the above PDE is equivalent to the distortion inequality, |D f (z)|2 ≤ K J(z, f ) a.e. z ∈ Ω, where |D f (z)| is the operator norm of the differential matrix of f at the point z, and J(z, f ) is its Jacobian determinant. Then we talk about K-quasiregular mappings. Bijec- tive quasiregular mappings are called quasiconformal. When K = 1 (i.e. if µ = 0) we recover, respectively, the classes of analytic functions and conformal mappings on Ω. For several reasons, this is considered one of the most convenient ways of extending the notion of analycity. As a matter of fact, notice that the distortion inequality has an analog in Rn. However, our interest comes not only from this point of view, but also from the fact that these functions are (at least in the plane) solution to a more or less gen- eral model of an elliptic PDE of first order. As introductory references to this subject, we cite Ahlfors [2], Lehto and Virtanen [37], Reshetnyak [48] and Rickman [49].
In this dissertation we will consider Painleve’s´ problem, as well as some of its vari- ants, in the quasiregular context. More precisely, a compact set E is said to be removable for bounded K-quasiregular mappings if for every open set Ω ⊃ E, any bounded function f : Ω → C, K-quasiregular on Ω \ E, admits a K-quasiregular extension to Ω. It also
2 makes sense to replace the boundedness by other properties, such as being in BMO or in Lipα. Moreover, we can even talk about removable sets for µ-quasiregular mappings, when our interest focusses on the Beltrami coefficient, rather than on K. In any case, as in the holomorphic setting, we just need to understand in all situations the case Ω = C.
It follows from Uy’s work [53] that if E has positive Lebesgue measure, then E is not removable, neither for K = 1. Conversely, given a compact set E of zero area, and a function f quasiregular on C \ E, in order to say that f can be quasiregularly extended to the whole plane, one just must make sure that f extends as a W 1,2 function, in a neighbourhood of E. To explain it, such an f already satisfies a Beltrami equation (or a distortion inequality) almost everywhere. Hence, only the integrability of its distri- butional derivatives has to be checked. In other words, we are facing a simplificated version of a Sobolev removability problem.
Chapter 2 is devoted to bounded and BMO K-quasiregular removability. In general, the Stoilow factorization Theorem (see for instance [37]) ensures that if f is K-quasiregular on C \ E, then f = h ◦ φ, where φ : C → C is K-quasiconformal and h is holomorphic on C \ φ(E). Moreover, both f and φ have the same Beltrami coefficient. Notice also that
E φ(E) φ
f h
Figure 1 f is bounded if and only if h is, and Reimann [46] showed that a similar thing happens with BMO. Therefore, extending f K-quasiregularly is equivalent to extendig h holo- morphically. Thus, in the bounded case we need to check under which conditions is it true that γ(φ(E)) = 0
3 whenever φ is K-quasiconformal. At the same time, for the BMO problem the condition H1(φ(E)) = 0 must be checked. In any case, is precisely at this point when the distor- tion problem appears. For instance, one is first led to the dimension distortion formula of K. Astala [4], which says that if dim(E) = t, then
2K dim(E) dim(φ(E)) ≤ t0 = 2 + (K − 1) dim(E) for any K-quasiconformal mapping φ. Using this inequality, it is shown in [4] that a 1 < 2 sufficient condition for H (φ(E)) = 0 (and therefore for γ(φ(E)) = 0) is t K+1 , or 0 < > 2 equivalently, t 1. In [4] it is also shown that for any real number t K+1 there exists a set E and a K-quasiconformal mapping φ such that dim(E) = t and dim(φ(E)) > 2 1. Hence, the number K+1 appears as the critical dimension for these K-quasiregular problems of removability. Afterwards, the natural following step to this dimensional setting consists on looking at the K-quasiconformal distortion of Hausdorff measures. For instance, it is well known (see for instance Ahlfors [2, p.33]) that
|E| = 0 ⇒ |φ(E)| = 0.
In this direction we prove the following Theorem.
Theorem 1 ([5]). Let E be a compact set, and φ a planar K-quasiconformal mapping. Then,
2 1 H K+1 (E) = 0 ⇒ H (φ(E)) = 0.
0 Therefore, there is absolute continuity of φ∗Ht with respect to Ht both for t0 = 1 and 0 2 t = 2, that is, t = K+1 and t = 2. For general t ∈ (0, 2) we cannot prove the absolute ∗ t0 t > 2 continuity of φ H with respect to H . However, for t K+1 we obtain a weaker result concerning Bessel capacities.
2 It follows from the above theorem and the BMO invariance that if H K+1 (E) = 0 then E is removable for BMO K-quasiregular mappings. Unexpectedly, for the L∞ problem
4 we can go further and obtain the following result.
Theorem 2 ([5]). 2 Let K > 1, and E be a compact set with σ-finite H K+1 (E). Then, E is removable for bounded K-quasiregular mappings.
In the proof of this Theorem we use several standard quasiconformal tools, together with the optimal integrability result for K-quasiregular mappings obtained by Astala and Nesi [10]. It also uses the fact that the lower 1-dimensional Hausdorff content behaves precisely as the analytic VMO capacity [54]. Moreover, there is a difficult the- orem appearing in [5] which asserts that if ψ is any K-quasiconformal mapping, and E is a subset of any rectifiable set, then
2 dim(ψ(E)) > . K + 1
In other words, these sets may not attain the maximal distortion. Furthermore, we can complete the proof by the results on analytic capacity from David [15] and Tolsa [51].
It can be seen that the construction of nonremovable sets for bounded (or BMO) K- quasiregular mappings is equivalent to some critical examples of extremal K-quasiconformal distortion. In that sense, we found sets E and K-quasiconformal mappings φ for which 2 > dim(E) = K+1 and γ(φ(E)) 0. As a consequence, one obtains the following result.
Theorem 3 ([5]). > 2 Let K 1. There exists compact sets E of dimension K+1 , non removable for bounded K- quasiregular mappings.
Our construction gives examples of compact sets E of dimension t and K-quasiconformal 0 mappings φ such that 0 < Ht (φ(E)) < ∞. However, we do not know if it is possible to 0 find E and φ for which 0 < Ht(E) < ∞ and 0 < Ht (φ(E)) < ∞.
In Chapter 3, we consider the removability problem for K-quasiregular mappings, but ∞ with the Lipα condition instead of the L norm. Recall that a function f belongs to Lipα
5 if | f (z) − f (w)| ≤ C |z − w|α whenever |z − w| < 1. Although we again go back through Stoilow factorization, there are in this new setting some differences with respect to se situation in Chapter 2. As we have explained, the latter is strongly related to the way that general K-quasiconformal 2 mappings distort sets of dimension K+1 . A good understanding of this distortion is enough for the complete solution of the removability problem, since both BMO and L∞ are K-quasiconformally invariant spaces of functions. However, it comes from Mori’s
Theorem [41] that every K-quasiconformal mapping is of class Lip 1 , and the radial K 1 −1 stretching φ(z) = z |z| K shows that this Holder¨ exponent cannot be improved. It then follows that, unfortunately, the space Lipα is not K-quasiconformally invariant. Because of this non invariance, techniques of Chapter 2 will need to be modified.
Our first result in this Lipα setting is the following.
Theorem 4. Let K > 1 and α ∈ (0, 1). Let E be compact, and assume that
2 (1+αK) H K+1 (E) = 0.
Then E is removable for α-Holder¨ continuous K-quasiregular mappings.
The proof of this result also uses strongly planar techniques. Roughly speaking, if C C C C f : → is a Lipα function on , K-quasiregular on \ E, then we show that ∂( f ◦ φ−1) = 0 in the sense of distributions (here φ is K-quasiconformal with the same Beltrami coefficient of f ). As our main tools, we use the integrability properties of K- quasiconformal mappings, as well as a classical result on partitions of unity.
2 To show that the index d = K+1 (1 + αK) is sharp, extremal K-quasiconformal distor- tion is not enough, as it was in the L∞ problem. Again the principal reason is the non-invariance of the class Lipα under K-quasiconformal changes of variables. More- over, a compact set of dimension d may be K-quasiconformally sent to another set of
6 dimension bigger than 1 + α, which is the the critical dimension in the analytic case. The way to overcome these obstacles comes from an explicit example of extremal K- quasiconformal distortion.
Lemma 5. > 0 2Kt Let t ∈ (0, 2) and K 1. Denote t = 2+(K−1)t . There exists a compact set E and a K- quasiconformal mapping φ satisfying:
(a) Ht(E) is positive and σ-finite.
(b) dim(φ(E)) = t0.
t 1 K−1 (c) φ is Holder¨ continuous with exponent t0 = K + 2K t.
We must remark that in this Lemma the extremal distortion of dimension is nothing new. The interesting fact here is that the Holder¨ exponent of this mapping φ is strictly 1 above the usual K coming from Mori’s Theorem. This higher regularity ensures that we do not loose so much regularity when composing with φ than with any other K- quasiconformal mapping. Therefore, a similar procedure to that in [4] for the L∞ prob- lem gives the following result.
Theorem 6. > 2 Let K ≥ 1 and α ∈ (0, 1). For any t K+1 (1 + αK), there exists a compact set E of dimension t, non removable for α-Holder¨ continuous K-quasiregular mappings.
2 We do not know what is happening for sets of dimension K+1 (1 + αK) and positive Hausdorff measure. Therefore, this remains as an open problem.
Among the consequences of the Main Lemma, we get not only the sharpness of the di- 2 mension K+1 (1 + αK) in the Lipα problem for K-quasiregular mappings, but also other applications. Recently, L. Kovalev [35] has found some of them in the setting of K- quasiregular gradient mappings. Furthermore, the techniques we used in the Lipα setting have been shown to apply in the more general field of finite distortion mappings (see Sec- tion 3.4). In that case, 2α is the corresponding critical dimension.
7 When dealing with any removability problem, it is not strange to look for some sort of selfimprovement results. Phylosophically, these results are used to ensure that weak solutions to some equation are actually strong solutions. For instance, in Chapters 2 and 3, we will use repeatedly the fact that K-quasiconformal mappings exhibit better 1,2 Sobolev regularity than the usual Wloc appearing in its definition. In this setting, Weyl’s Lemma is an interesting tool, but has an unavoidable limitation, which is that it only works for constant coefficient operators. Therefore, the Beltrami equation seems to be a bad model for that kind of distributional results. The reason is that if f is a function then the identity hµ ∂ f , ϕi = −h f , ∂(µ ϕ)i (0.1) makes no sense, unless both µ and ϕ are functions in the class D (compactly supported C∞ functions). If one wants to work with more general Beltrami coefficients µ, then it is sufficient to ask that ∂ f acts on a bigger class of testing functions. As a matter of fact, 2 notice that equality (0.1) turns out to make sense if f ∈ Lloc, µ is a Beltrami coefficient of class W1,2, and our testing functions ϕ have also W1,2 regularity. This question is treated in Chapter 4. We are able to prove the following result, which may be interpreted as a Weyl’s Lemma in the quasiconformal setting.
Theorem 7. p C > Let f ∈ Lloc( ) for some p 2. Let µ be a compactly supported Beltrami coefficient, such that µ ∈ W1,2(C). If h∂ f , ϕi = hµ ∂ f , ϕi
2,q C < for each ϕ ∈ D, then f ∈ Wloc ( ) for any q 2. In particular, f is µ-quasiregular.
One of the tools used in the proof of this result is that such good Beltrami coefficients 1,p C < 2K give µ-quasiconformal mappings not only in Wloc ( ) for any p K−1 (this holds for K−1 2,q C any µ with kµk∞ ≤ K+1 ) but also in the second order Sobolev spaces Wloc ( ) for every q < 2. We know that this regularity result is sharp.
8 Since the improvement result is known, we can go back to the removability prob- lem. Then we say that E is removable for bounded µ-quasiregular mappings if ev- ery bounded function f : C → C, µ-quasiregular to C \ E, is actually constant. As a difference, notice that analytic functions are not in general µ-quasiregular for every µ, while they are K-quasiregular for every K. So it means that this problem is refered precisely to the regularity of the PDE itself, more than to the concrete value of the con- stant K. Among the consequences of the Weyl’s Lemma, we find that both BMO and VMO problems of characterizing the removable sets for µ-quasiregular mappings have precisely the same solution that for the ∂ equation. When reading it in terms of qua- 1 1 1 siconformal distortion, we get that M (E) = 0 iff M (φ(E)) = 0, and M∗(E) = 0 iff 1 1,2 C M∗(φ(E)) = 0. In other words, for W ( ) compactly supported Beltrami coefficients, the corresponding quasiconformal mappings preserve the classes of removable sets for BMO (and VMO) analytic functions, just as any bilipschitz mapping does.
Of course, the following natural step is the removability L∞ problem. That is, we want to characterize removable sets for bounded µ-quasiregular mappings. We could also formulate the problem asking which are the compact sets E whose µ-quasiconformal images have always vanishing analytic capacity. This reformulation of the problem directly leds us to the bilipschitz invariance of analytic capacity of Tolsa [52]. More pre- cisely, given a compact set E and a planar homeomorphism f , the analytic capacities γ(E) and γ( f (E)) are comparable if and only if f is bilipschitz. As a consequence, the bilipschitz image of a removable set is also removable. One of the main goals in Chapter 4 is to show that these µ-quasiconformal mappings we are dealing with, although they are not bilipschitz in general, they preserve sets of zero analytic capacity and σ-finite length.
Theorem 8. Let µ ∈ W1,2(C) be a compactly supported Beltrami coefficient, and let φ be µ-quasiconformal. If E is a compact set such that H1(E) is σ-finite, then
γ(E) = 0 ⇔ γ(φ(E)) = 0
9 For the proof, it is very important that µ-quasiconformal mappings are diffeomor- phisms at every point in C, except on a set of zero length. The same happens to their inverses. This differentiability allows to show that rectifiable sets are mapped to recti- fiable sets. The characterization of G. David [15] does the rest.
This dissertation is divided into 4 chapters. In Chapter 1 we give an introduction and some general background. Each of the other 3 chapters contains different problems on removability for quasiregular mappings. Hence, each of them is independent from the others.
In Chapter 2 we consider questions on K-quasiconformal distortion and on removable singularities for bounded K-quasiregular mappings. It is based on the paper [5],
• K. ASTALA, A. CLOP, J. MATEU, J. OROBITG, I. URIARTE-TUERO, Distortion of Hausdorff measures and improved Painleve´ removability for quasiregular mappings, Sub- mitted.
In Chapter 3 we obtain some results related to the problem of characterizing the re- movable sets for α-Holder¨ continuous K-quasiregular mappings. There we include the results from [12] and [13]:
• A. CLOP, Removable singularities for Holder¨ continuous quasiregular mappings in the plane, To appear in Ann. Acad. Sci. Fenn. Ser. A I Math.
• A. CLOP, Nonremovable sets for Holder¨ continuous quasiregular mappings in the plane, To appear in Michigan Math. J.
Finally, Chapter 4 includes questions on removability for quasiregular mappings with the Beltrami coefficient µ given in the Sobolev space W 1,2. In other words, we study the removable sets for the solutions of a given PDE. Here we have included the results of the preprint [14]
• A. CLOP, J. MATEU, J. OROBITG, Beltrami equations with coefficient in the Sobolev space W1,2, Preprint (2006).
10 I have mostly used the notation I have found in the main references on quasiconformal mappings. In order to give consistence to the full text, I have made some changes with respect to the original versions of the papers.
11
Acknowledgements
I feel indebted to all those who have offered their support and backed me. For this reason I would like to express my sincere gratitude to them all.
To the Universitat Autonoma` de Barcelona, for its technical support. To Rosa Rodr´ıguez, who made all the drawings in these notes. To Miquel Brustenga, who made my days in Finland easier due to his technological support. To all the staff at the Departament de Matematiques` of the UAB, where I have always found a friendly and inspiring atmo- sphere.
To K. Astala, D. Faraco, P. Koskela, I. Prause, I. Uriarte-Tuero, J. Verdera and X. Zhong, for their time and suggestions.
To my friends. To Gerard, Dani, Juanje, Monica` and Isabel, because with you I have laughed very much, and hope to continue to do so. To my burbujitas, M´ıriam, Marta, Elisenda, Joan, Estrella, Carme and Laura. To Anna, Neus, Agust´ı, Pau, Mart´ı and Nausica, for being there. To my parents, who have always been at my side. To Xavi.
Finally to my advisors: Joan Mateu and Joan Orobitg, to which this dissertation is ded- icated. For all the many hours of discussion and for all the patience they have shown. They proposed the problems that are dealt with here. Their ideas, their advice, their constant support and their good sense of humor have been fundamental for the com- pletion of this thesis.
13
Contents
1 Preliminaries 17 1.1 Function spaces ...... 17 1.2 Hausdorff measures and contents ...... 21 1.3 The problem of Painleve´ ...... 23 1.4 Quasiconformal mappings ...... 25
2 Removable singularities for bounded quasiregular mappings 33 2.1 Introduction ...... 33 2.2 Absolute continuity ...... 36 2.3 Improved Painleve´ theorems ...... 48 2.4 Extremal distortion and nonremovable sets ...... 52
3 Removable singularities for Holder¨ continuous quasiregular mappings 61 3.1 Introduction ...... 61
3.2 Sufficient conditions for Lipα removability ...... 64 3.3 Extremal distortion and nonremovable sets ...... 69 3.4 Removability results for finite distortion mappings ...... 81
4 Beltrami equations with coefficient in the Sobolev space W1,2 91 4.1 Introduction ...... 91 4.2 Regularity of µ-quasiconformal mappings ...... 94 4.3 Distributional Beltrami equation ...... 98 4.4 µ-quasiconformal distortion of Hausdorff measures ...... 101 4.5 µ-quasiconformal distortion of analytic capacity ...... 106
15
1 Preliminaries
1.1 Function spaces
If Ω is any open set, we denote by D(Ω) the algebra of functions ϕ : Ω → C of class C ∞ with compact support contained in Ω. For these functions, and given a pair α = (α1, α2) of nonnegative integers, we denote
∂|α| ϕ ∂α ϕ = ∂xα1 ∂yα2
1 1 where |α| = α1 + α2. In particular, ∂ = 2 ∂x − i∂y and ∂ = 2 ∂x + i∂y . By Dϕ(z) we denote the differential matrix of ϕ at the point z, and J(z, ϕ) is the Jacobian determinant. The space of all distributions will be denoted by D0(Ω). We say that a distribution f vanishes on an open set Ω if h f , ϕi = 0 for every ϕ ∈ D(Ω). The following result is known as Weyl’s Lemma.
Theorem 1.1. Let f ∈ D0(Ω) be such that ∂ f = 0 on a planar domain Ω ⊂ C. Then, there exists a holomor- phic function h : Ω → C such that f = h.
For p ∈ [1, ∞] and k = 1, 2, . . . , the Sobolev space W k,p(Ω) consists of all functions f for which the distribution itself and its distributional derivatives up to order k are represented by Lp(Ω) functions. The formula
1 p α k f kk,p = ∑ k∂ f kp |α|≤k !
17 1 Preliminaries
k,p Ω ∞ Ω k,p Ω defines a complete norm in W ( ), for which C ( ) is dense. By W0 ( ) we denote Ω k,p Ω k,p C k,p C k,p Ω the closure of D( ) in W ( ). In particular, W ( ) = W0 ( ). By Wloc ( ) we k,p Ω p Ω α p Ω mean the local Sobolev space, that is, f ∈ Wloc ( ) iff f ∈ Lloc( ) and ∂ f ∈ Lloc( ) k,p Ω whenever |α| ≤ k. We say that a sequence { f j}j converges to f in Wloc ( ) if
α α k∂ fj − ∂ f kLp(F) → 0 as j → ∞, for any compact set F ⊂ Ω and every |α| ≤ k.
1 Ω Ω We say that a function f ∈ Lloc( ) has bounded mean oscillation on , and write f ∈ BMO(Ω), if for any disk D there is a constant fD such that
1 k f k∗ = sup | f (z) − f | dA(z) < ∞ |D| D D⊂Ω ZD We say that a function f ∈ BMO(C) has vanishing mean oscillation on C, and write f ∈ VMO(C) if 1 lim | f − f | = 0 |D| D ZD 1 ∞ as |D| + |D| → . Equivalently, VMO may be introduced as the BMO-closure of com- pactly supported continuous functions.
Ω Given α ∈ (0, 1), the Holder¨ space Lipα( ) is the class of locally Holder¨ continuous func- Ω Ω tions of exponent α on the set , that is, f ∈ Lipα( ) if the quantity
| f (z) − f (w)| k f kα = sup α |z−w|<1,z6=w |z − w|
Ω is finite. It turns out that for α ∈ (0, 1) a function f ∈ Lipα( ) iff
1 | ( ) − | ( ) < ∞ sup 1+ α f z fD dA z Ω | | 2 D⊂ D ZD
18 1.1 Function spaces
Moreover, for any p ∈ (1, ∞), the quantity
1 1 p sup | f (z) − f |p dA(z) 1+ αp D D⊂Ω |D| 2 ZD ! is comparable to k f k∗ if α = 0, and to k f kα if α ∈ (0, 1).
p k,p The Sobolev Embedding Theorem establish relations between L , Lipα, BMO and W .
Theorem 1.2. Let p ∈ [1, ∞), and let Ω ⊂ C be either a disk or the whole plane.
< 1,p Ω q Ω 1 1 1 (a) If 1 ≤ p 2, then W ( ) ⊂ L ( ), where q = p − 2 .
(b) If p = 2, then W1,p(Ω) ⊂ BMO(Ω).
1,p (c) If 2 < p < ∞, then W (Ω) ⊂ Lip 2 (Ω). 1− p
An Orlicz function is a continuous function P : [0, ∞) → [0, ∞), nondecreasing, such that P(0) = 0 and lim P(t) = ∞. The Orlicz space LP(Ω) consists in all the measurable t→∞ functions f : Ω → C satisfying
P(λ| f (z)|) dA(z) < ∞ ZΩ for some λ > 0. The quantity
1 k f k = inf : P(λ| f (z)|) dA(z) < P(1) P λ ZΩ is called Luxemburg funtional. If P is convex then k · kP defines a complete norm on LP(Ω). As examples,
α • P(t) = et − 1. In this case, we denote LP(Ω) = Exp(Ω).
• P(t) = tp logα(e + t), α ≥ 1 − p. Now we write LP(Ω) = Lp logα L(Ω).
We remark that if α > 0 then the norm in L p logα L(Ω) can be computed without going
19 1 Preliminaries through the Luxemburg functional,
1 p p α | f (z)| k f k p α ' | f (z)| log e + dA(z) L log L k f k ZΩ p Two Orlicz functions P, Q are mutually conjugated iff there exists a constant C > 0 such that
| f (z)| |g(z)| dA(z) ≤ Ck f kLP (Ω) kgkLQ (Ω) ZΩ for all f ∈ LP(Ω) and g ∈ LQ(Ω). As examples,
p α q −α p < < ∞ > • P(t) = t log (e + t) and Q(t) = t log (e + t), q = p−1 , 1 p , α 0.
1 • P(t) = t logα(e + t) and Q(t) = et α − 1, for α > 0;
A more detailed discussion on Orlicz spaces can be found, for instance, in the books of Krasnosel’ski˘ı and Ruticki˘ı [36], as well as in Rao and Ren [45].
The Beurling Transform is the operator defined on smooth functions as
−1 f (w) B f (z) = lim dA(w) (1.1) ε→0 π (w − z)2 Z|z−w|≥ε Equivalently, B may be defined as
[ (B f )(ξ) = m(ξ) f (ξ)
ξ2 b with m(ξ) = . It is a Calderon-Zygmund´ operator, so that B is bounded in L p(C) |ξ|2 for all p ∈ (1, ∞), and for p = 1 it satisfies a weak estimate. If f ∈ W 1,p(C), p > 1, then
B(∂ f ) = ∂ f because by Plancherel we have
ω2 hB(∂ f ), gi = c ω f (ω) g(ω) dA(ω) = c ω f (ω) g(ω) dA(ω) = h∂ f , gi |ω|2 Z Z b b b b
20 1.2 Hausdorff measures and contents
The Beurling Transform can be thought as the derivative of the Cauchy Transform. It is defined for smooth functions as
1 f (w) C f (z) = − dA(w) (1.2) π w − z Z If f ∈ Lp(C) for some p > 1 then C f ∈ W1,p(C) and
∂(C f ) = B f ∂(C f ) = f
One may normalize this operator in such a way that the image function fix the origin. Thus, we define as Ahlfors [2, p.85],
1 1 1 C˜ f (z) = − f (w) − dA(w) (1.3) π w − z w Z Actually, for smooth enough functions C˜ f (z) = C f (z) − C f (0). Thus, for any p ∈ (1, ∞), if f ∈ Lp(C) then C˜ f ∈ C + W1,p(C). In any case, if p > 2 then
|C f (z) − C f (w)| |C˜ f (z) − C˜ f (w)| = ≤ Cp k f kp 1− 2 1− 2 |z − w| p |z − w| p
1.2 Hausdorff measures and contents
We call measure function to any nondecreasing function h : [0, ∞) → [0, ∞) such that lim h(t) = 0. Given a measure function h, for every δ ∈ (0, ∞] we denote t→∞
∞ h Hδ (F) = inf ∑ h(diam(Dj)) (j=1 )
∞ where the infimum runs over all countable coverings of F, F ⊂ j=1 Dj, by disks Dj of ∞ h h diameter diam(Dj) ≤ δ. If δ = , then we write M (F) = HS∞(F). It is easy to see h h h that Hδ defines a monotonic and subadditive set function, such that Hδ (F) ≤ He (F) if
21 1 Preliminaries
0 < e ≤ δ. The limit h h H (F) = lim Hδ (F) δ→0 defines a Borel regular measure (see for instance Mattila [39]), satisfying
h h H (F) = 0 ⇐⇒ Hδ (F) = 0 for any δ ∈ (0, ∞]. In the particular case h(t) = td, d > 0, then we write Mh(F) = Md(F) and Hh(F) = Hd(F), and we call them d-dimensional Hausdorff content and d- dimensional Hausdorff measure of F, respectively. We can give, to every set F, a real num- ber defined by
dim(F) = inf{d > 0 : Md(F) = 0} = sup{d > 0 : Md(F) > 0} which will be called Hausdorff dimension (or simply dimension) of F.
The following result, known as Frostman’s Lema (see e.g. [39, p.112]), characterizes the compact sets with positive d-dimensional Hausdorff content.
Theorem 1.3. Let E be a compact set, and d ∈ (0, 2]. Then, Md(E) > 0 if and only if there exists a positive Radon measure ν, supported on E, such that ν(D(z, r)) ≤ C rs for any z ∈ E and r > 0. Further, ν can be chosen so that ν(E) & Md(E).
The lower d-dimensional Hausdorff content of F is
d h M∗(F) = sup M (F) h∈Fd
d where F = Fd is the set of all measure functions h(t) = t ε(t), 0 ≤ ε(t) ≤ 1, such that d d d < d lim ε(t) = 0. In general, M∗(F) ≤ M (F), but it may happen that M∗(F) = 0 M (F). t→0 1 1 For instance, if F is the line segment [0, 1] on the plane, then M∗(F) = 0 but M (F) = 1. d d d d For a disk D, M∗(D) = M (D), and for open sets U, M∗(U) ' M (U). The following is an old result due to Sion and Sjerve [50].
22 1.3 The problem of Painleve´
Theorem 1.4. C t t Let E ⊂ be a compact set, and let t ∈ (0, 2]. Then, M∗(E) = 0 if and only if H (E) is σ-finite.
We say that Γ ⊂ C is a rectifiable curve if Γ = f (I) for a Lipschitz map f : I → C, where R C R I ⊂ is a bounded interval. A set E ⊂ is called rectifiable if there exists subsets Ij ⊂ C and Lipschitz maps fj : Ij → such that the set
∞
E \ fj(Ij) j[=1 has zero length. A set E ⊂ C is called purely unrectifiable if
H1(E ∩ Γ) = 0 for any rectifiable curve Γ ⊂ C. It is well known that rectifiable sets have always σ- finite length. They are stable under countable unions, and any subset of a rectifiable set is also rectifiable. The following is a well known result due to Besicovitch (see e.g.[39, p.205]).
Theorem 1.5. If A ⊂ C has finite length, then A admits a decomposition
A = R ∪ P ∪ Z where R is rectifiable, P is purely unrectifiable, and H1(Z) = 0.
1.3 The problem of Painleve´
A set E is called removable for bounded analytic functions if for every open set Ω ⊃ E, every function f ∈ L∞(Ω) holomorphic on Ω \ E extends holomorphically to Ω. In this definition, one may consider Ω = C. The classical problem of Painleve´ consists on giving metric and geometric characterizations of removable sets for bounded analytic
23 1 Preliminaries functions. Related to this problem one has an important set function, called analytic capacity, introduced by Ahlfors in [3] and defined as
0 γ(E) = sup | f (∞)|; f ∈ H(C \ E), k f k∞ ≤ 1, f (∞) = 0 . (1.4) Here H(Ω) is the space of functions which are holomorphic on the open set Ω, f (∞) = lim f (z) and f 0(∞) = lim z ( f (z) − f (∞)). Notice that by means of Stokes Theorem |z|→∞ |z|→∞ one may show that if f ∈ H(C \ E) then
| f 0(∞)| = |h∂ f , 1i|
It turns out that γ(E) = 0 if and only if E is removable for bounded analytic functions. Moreover, the classical Painleve´ Theorem asserts that γ(E) . H1(E), while dim(E) > 1 implies γ(E) > 0. Therefore, only sets E with dim(E) = 1 are interesting.
The following result establishes the solution to Painleve’s´ Problem for those sets with finite length. It is due to G. David [15].
Theorem 1.6. Let E be compact, and suppose that H1(E) < ∞. Then,
γ(E) = 0 ⇔ E is purely unrectifiable
Later on, it comes from Tolsa [51] that this result turns out to be true even for sets E with σ-finite length. At the same time, for sets with Hausdorff dimension 1 but non σ-finite length we have the following nonremovability sufficient condition from P. Mattila [40].
Theorem 1.7. Let h be a measure function, and let E be a compact set such that Hh(E) > 0. If
1 h(t)2 dt < ∞ t3 Z0 then γ(E) > 0.
24 1.4 Quasiconformal mappings
A complete solution to Painleve’s´ problem is given by Tolsa in [51], in terms of Menger curvature. Another main result in [51] is the so called semiadditivity.
Theorem 1.8. There exists an absolute constant C such that
∞ ∞ γ Ej ≤ C ∑ γ(Ej) j=1 j[=1 for any family of sets Ej.
Recall that a function f : C → C is called bilipschitz if there exists a constant C such that
1 |z − w| ≤ | f (z) − f (w)| ≤ C|z − w| C
The following theorem is due to X. Tolsa [52].
Theorem 1.9. Let f : C → C be a bilipschitz mapping. Then, for any compact set E ⊂ C,
γ(φ(E)) ' γ(E)
1.4 Quasiconformal mappings
Let K ≥ 1. A homeomorphism φ : Ω → Ω0 between planar domains Ω, Ω0 ⊂ C is called 1,2 Ω K-quasiconformal if it belongs to the Sobolev space Wloc ( ) and satisfies
max |∂αφ(z)| ≤ K min |∂αφ(z)| (1.5) α α at almost every z ∈ Ω. This inequality is equivalent to the distortion inequality,
|D f (z)|2 ≤ K J(z, f ) (1.6)
25 1 Preliminaries
From this inequality, it follows that at small scales
max |φ(z) − φ(z0)| ≤ C min |φ(z) − φ(z0)| (1.7) |z−z0|=r |z−z0|=r for a constant C that depends only on K. Moreover, homeomorphisms satisfying (1.7) are K-quasiconformal for some K. In some cases, the following equivalent definition, given in terms of a partial differential equation, will be more convenient. We call Bel- ∞ trami coefficient to any measurable function µ ∈ L (Ω) such that kµk∞ < 1. Then, a Ω Ω 1,2 Ω µ-quasiconformal mapping is any homeomorphism φ : → φ( ) lying in Wloc ( ) that solves the Beltrami equation, ∂φ = µ ∂φ (1.8)
From this point of view, φ is K-quasiconformal if and only if it is µ-quasiconformal K−1 for some µ with kµk∞ ≤ K+1 . By quasiconformal we mean K-quasiconformal for some K ≥ 1. Here we give some examples:
• All the conformal mappings, which are precisely the only 1-quasiconformal ones.
• Given 1 ≤ K ≤ K0, every K-quasiconformal mapping is also K0-quasiconformal.
1 K −1 K−1 z • The radial stretching, φ(z) = z |z| , whose Beltrami coefficient is µ(z) = − K+1 z .
z 1 • φ(z) = z (1 − log |z|), with µ(z) = z 1−2 log |z| .
Among the above examples, we remark that the radial stretching is in some sense ex- tremal to obtain regularity of K-quasiconformal mappings. We show in the following example an explicite method for solving the Beltrami equation, under the additional assumption that µ is compactly supported (see e.g.in Ahlfors [2, p.90]).
Example 1.10. We show that if µ is a compactly supported Beltrami coefficient, then there exists exactly one µ-quasiconformal mapping φ such that
φ(z) − z = O(1/|z|) (1.9)
< as |z| → ∞. To construct it, first notice that kµBkL2(C)→L2(C) ≤ kµk∞ 1. Hence, the
26 1.4 Quasiconformal mappings formula ∞ Tg = ∑ (µB)n(g) = g + µB(g) + µB(µB(g)) + . . . n=o defines a bounded linear operator T : L2(C) → L2(C), such that
(I − µB)(Tg) = T(I − µB)(g) = g that is, T = (I − µB)−1. Actually, this holds also in Lp(C) for some values p > 2, since kBkLp(C)→Lp(C) is a continuous function of p [37, p.214]. Define h = Tµ (then h = µBh + µ) and let φ(z) = z + Ch(z)
1,p C 1,p C Clearly, φ ∈ Wloc ( ), ∂φ = 1 + Bh and ∂φ = h. Thus, φ is a Wloc ( ) solution to the Beltrami equation ∂φ = µ ∂φ. Moreover, an approximation argument [2] shows that in fact the function φ is globally injective. On the other hand, classical topology results ensure that φ is onto. Finally, since µ is compactly supported, h is too, thus it is clear that |Ch(z)| decreases at 1 infinity as |z| . In this construction, different normalizations are also allowed. For instance, there exists exactly one solution φ˜ satisfying
φ˜(0) = 0 and φ˜(z) − (z + a) ∈ W1,p(C) (1.10) for some constant a ∈ C and some p > 2. This may clearly be done just removing a constant,
φ˜(z) = z + C˜h(z) = φ(z) − φ(0)
Notice that normalizations (1.9) and (1.10) differ only in a linear mapping. More precisely, φ and φ˜ satisfy the relation φ˜ ◦ φ−1(z) = z − a. For compactly supported Beltrami coefficient µ, we call both φ and φ˜ principal solution to the Beltrami equation ∂φ = µ ∂φ, and we will specify the normalization (1.9) or (1.10) just when needed.
For general µ, the following result holds not only in the plane, but also on every simply connected domain (see for instance Lehto and Virtanen [37]).
27 1 Preliminaries
Theorem 1.11. Let µ be a Beltrami coefficient. There exists one µ-quasiconformal mapping φ : C → C. More- C C −1 over, if φ1, φ2 : → are µ-quasiconformal, then φ1 ◦ φ2 is conformal.
We are interested not only in homeomorphic, but also in more general solutions to Beltrami equations. Thus we say that a function f : Ω → C is a K-quasiregular mapping Ω C 1,2 Ω in an open set ⊂ if f ∈ Wloc ( ) and satisfies inequality (1.5). Of course, we may also talk about µ-quasiregular mappings, whenever a reference to the Beltrami coefficient is needed. As before, a quasiregular map is just a K-quasiregular one, for some K ≥ 1. It is clear that when µ = 0 (or when K = 1) one gets the class of analytic functions. 1,2 For general µ, it turns out that homeomorphic solutions characterize the rest of Wloc solutions. This fact is known as Stoilow’s Theorem.
Theorem 1.12. Let µ be a Beltrami coefficient. Let φ be any µ-quasiconformal mapping.
(a) If h is holomorphic, then f = h ◦ φ is µ-quasiregular.
(b) If f is µ-quasiregular, then h = f ◦ φ−1 is holomorphic.
Therefore, any K-quasiregular function f may be written as a K-quasiconformal change of variables φ of a holomorphic function h. Moreover, both f and φ have the same Bel- trami coefficient. Thus, all regularity properties of K-quasiregular functions are charac- terized by its K-quasiconformal component.
One of the typical features of quasiconformal mappings is that they exhibit some strong properties of normality (for an exhaustive discussion see the book of Lehto and Virtanen [37]). To keep an idea in mind, assume that F is an arbitrary family of K-quasiconformal mappings φ : C → C, normalized in some way. Normalization for quasiconformal mappings from C to itself can be given by either equation (1.9) or (1.10). Other possi- bilities of normalizing come from the condition diam(φ(D)) = 1, or even letting fixed 3 points in the Riemann sphere. Assume, for instance, that every mapping in F fixes 0, 1 and ∞. Then, one of the following properties implies the other two:
28 1.4 Quasiconformal mappings
• F is an equicontinuous family, when restricted to any compact set.
• F is a compact family, when restricted to any compact set.
• F is uniformly Holder¨ continuous, when restricted to any compact set.
It turns out that we always have these properties, as was shown by Mori [41] in the following theorem. The radial stretching shows that it is sharp.
Theorem 1.13. Let φ : C → C be a normalized K-quasiconformal mapping. Then,
1 |φ(z) − φ(w)| ≤ C |z − w| K where C depends only on K.
A K-quasidisk (resp. a K-quasicircle) is the image of the unit disk D (resp. the unit circle ∂D) under a K-quasiconformal mapping C → C. It follows from (1.7) that for any disk D and any K-quasiconformal mapping φ,
1 diam(φ(D)) ' |φ(D)| 2 with constants that may depend on the normalization. Therefore, quasidisks have area comparable to the square of its diameter. Now, for any disk D, Mori’s Theorem says that 1 |φ(D)| ≤ C |D| K and it is natural to ask wether this holds for a more general class of sets. After several years, this was shown to be true in a celebrated paper of K. Astala [4]. The principal result in this paper is kown as the Area Distortion Theorem .
Theorem 1.14. Let φ be a normalized K-quasiconformal mapping on a domain Ω ⊂ C. For any measurable set E ⊂ Ω, 1 |φ(E)| ≤ C |E| K
29 1 Preliminaries
Along these notes we will see how trascendental is this result. To start, observe that if φ is any homeomorphism with locally integrable jacobian determinant, then higher integrability of J(·, φ) is equivalent to some area distortion theorem. More precisely, 1 ∞ 1−α , J(·, φ) ∈ Lloc for some α ∈ (0, 1) if and only if
|φ(E)| ≤ C |E|α for every measurable set E. However, if we assume now that φ is quasiconformal, then 2 ∞ 1−α , this is equivalent to saying that Dφ ∈ Lloc . As a consequence, it follows the sharp integrability result for K-quasiconformal mappings.
Theorem 1.15. 2K ,∞ C C K−1 Let φ : → be K-quasiconformal mapping. Then, Dφ ∈ Lloc .
The sharpness comes again from the radial stretching. In particular, every K-quasiregular mapping has also this regularity.
It turns out that something better may be said under additional assumptions of con- formality inside some set. More precisely, this sharp integrability exponent may be at- tained in some cases. The following result is due to K. Astala and V. Nesi [10].
Theorem 1.16. Let φ : C → C be a K-quasiconformal mapping, conformal on C \ D, normalized with the 1 ∞ D condition φ(z) − z = O( |z| ) as |z| → . Assume that E ⊂ is such that ∂φ = 0 at almost every point of E. Then, J(z, φ)p dA(z) ≤ 1 ZE K with p = K−1 .
Notice that the normalization (1.9) is not the only one allowed, provided that the jaco- bian determinant does not change too much. For instance, under normalization (1.10) the above result still holds true.
30 1.4 Quasiconformal mappings
Ω C 1,p Ω We say that f : → is weakly K-quasiregular if and only if f ∈ Wloc ( ) for some p > 1, and f satisfies (1.5). The following is a dual version of Theorem 1.15.
Theorem 1.17. 1,p Ω > 2K Let f ∈ Wloc ( ) be weakly K-quasiregular. If p K+1 , then f is K-quasiregular.
It turns out that Theorems 1.15 and 1.17 are strongly related to the problem of finding the range of real numbers p for which the Beltrami operators
I − µB : Lp(C) → Lp(C)
K−1 are invertible whenever kµk∞ ≤ K+1 . This problem was solved by K. Astala, T. Iwaniec 2K 2K and E. Saksman in [8], where this range was shown to be precisely ( K+1 , K−1 ). More- over, S. Petermichl and A. Volberg [44] showed that one can get injectivity even at the lower borderline. From this, it follows that even weakly K-quasiregular mappings in 2K 1, K+1 Wloc are K-quasiregular.
Another consequence of the Area Distortion Theorem is the so called Dimension Dis- tortion Theorem , also proved by Astala in [4].
Theorem 1.18. Let φ be a K-quasiconformal mapping on a domain Ω. Then,
2K dim(E) dim(φ(E)) ≤ (1.11) 2 + (K − 1) dim(E) for any compact set E ⊂ Ω.
Observe that if φ is K-quasiconformal, then also φ−1 is. Thus, the above bounds may also be read as
1 1 1 1 1 1 1 − ≤ − ≤ K − (1.12) K dim(E) 2 dim(φ(E)) 2 dim(E) 2
The sharpness of these bounds at the level of dimensions is due to Astala [4].
31 1 Preliminaries
Theorem 1.19. Given t ∈ (0, 2) and K ≥ 1, there is a K-quasiconformal mapping φ : C → C and a compact set E of dimension t, such that
2Kt dim(φ(E)) = 2 + (K − 1)t
32 2 Removable singularities for bounded quasiregular mappings
2.1 Introduction
We say that a compact set E is removable for bounded K-quasiregular mappings, or sim- ply K-removable, if for every open set Ω ⊃ E, every bounded K-quasiregular mapping f : Ω \ E → C admits a K-quasiregular extension to Ω. In this definition, as in the analytic setting, we may replace L∞(Ω) by BMO(Ω). In this chapter we want to give metric and geometric sufficient conditions on E to be K-removable. At the same time, we want non trivial examples of non removable sets E.
The Stoilow factorization Theorem converts bounded (or BMO) quasiregular mappings in bounded (or BMO) analytic mappings. Therefore, a compact set is removable for the K-quasiregular problem if and only if all its K-quasiconformal images are removable for the corresponding analytic problem. One is then led to the celebrated Astala’s pa- per on quasiconformal distortion [4]. In particular, K-quasiconformal mappings φ send compact sets E of Hausdorff dimension t to compact sets φ(E) of dimension
2Kt dim(φ(E)) ≤ . 2 + (K − 1)t
The first question we study in this chapter is if this estimate may be improved to the level of Hausdorff measures Ht. In other words, if φ is a planar K-quasiconformal < < 0 2Kt mapping, 0 t 2 and t = 2+(K−1)t , is it true that
0 Ht(E) = 0 ⇒ Ht (φ(E)) = 0, (2.1)
33 2 Removable singularities for bounded quasiregular mappings
0 that is, φ∗Ht Ht. Classical results of Ahlfors [2] and Mori [41] assert that this is true 2 when t = 0 or t = 2. As a first main result of this chapter we prove (2.1) for t = K+1 , i.e. for the case of image dimension t0 = 1.
Theorem. Let φ be a planar K-quasiconformal mapping, and let E be a compact set. Then,
K+1 1 2 2K M (φ(E)) ≤ C M K+1 (E) . (2.2) As a consequence, 2 1 H K+1 (E) = 0 ⇒ H (φ(E)) = 0.
A key point in proving the above result is that we are able to show that for planar quasiconformal mapping h conformal on C \ E and normalized at ∞ the inequality (2.2) can be improved. Such mappings h essentially preserve the 1-dimensional Hausdorff content, 1 1 M (h(E)) ≤ CK M (E) (2.3) where CK depends only on K. For the area the same estimate was shown by Astala in [4]. In fact, as we will see later, a counterpart of (2.3) for the t-dimensional Hausdorff 0 content Mt is the only missing detail for proving the absolute continuity φ∗Ht Ht for general t. Towards solving (2.1) we conjecture that actually
Mt(h(E)) ≤ C Mt(E), 0 < t ≤ 2, whenever E ⊂ C compact and h normalized and conformal in C \ E, with a K-quasiconformal extension to C. For more detailed discussion and other formulations see Section 2.
The reason our methods work in the special case of dimension t = 1 only is that content M1 is equivalent to analytic BMO capacity [54]. For exponents 1 < t < 2, we do have interpolating estimates but unfortunately we have to settle for the Riesz capacities. We have
Cα,t(h(E)) ≤ C Cα,t(E)
34 2.1 Introduction
2 for any t ∈ (1, 2), and where α = t − 1. This fact has consequences towards the absolute continuity for general quasiconformal mappings, but these are not strong enough for (2.1) when 1 < t0 < 2.
Going back to the removability problem, Iwaniec and Martin previously conjectured n [29] that in Rn, n ≥ 2, sets with Hausdorff measure H K+1 (E) = 0 are removable for bounded K-quasiregular mappings. A positive answer for n = 2 was found in [6]. In the present work we show that, surprisingly, for K > 1 one can do even better.
Theorem. Let K > 1 and suppose that E is any compact set with
2 H K+1 (E) σ − finite.
Then E is removable for all bounded K-quasiregular mappings.
The theorem fails for K = 1, since for instance the line segment E = [0, 1] is not remov- > 2 able. Note that by [4], for any t K+1 there are non-K-removable sets with dim(E) = t. Furthermore, we improve also the converse direction.
Theorem. > 2 Given K 1, there exists compact sets E of dimension K+1 , nonremovable for bounded K- quasiregular mappings.
Via the classical Stoilow factorization, all our main results are closely connected. In- deed, the first step will be to show that for a general K-quasiconformal mapping φ one has 2 1 H K+1 (E) σ-finite ⇒ H (φ(E)) σ-finite.
However, this conclusion will not be enough since there are rectifiable sets of finite length, such as E = [0, 1], that are non-removable for bounded analytic functions. Therefore, we need to know how rectifiable sets are distorted by K-quasiconformal
35 2 Removable singularities for bounded quasiregular mappings mappings. The right answer to this question is given in Theorem 2.16. Essentially,
2 F 1-rectifiable ⇒ dim(φ(F)) > K + 1 whenever φ is K-quasiconformal.
This chapter is structured as follows. In Section 2 we deal with the quasiconformal dis- tortion of Hausdorff measures and some Riesz capacities. In Section 3 we apply these results in the removability problems for bounded (or BMO) quasiregular mappings. Finally, Section 4 is devoted to the construction of nonremovable sets.
2.2 Absolute continuity
Here we call principal K-quasiconformal mappings to any K-quasiconformal mappings that are conformal outside a compact set and normalized by equation (1.9), that is, 1 ∞ φ(z) − z = O |z| as |z| → .
In the work [4] of K. Astala, it is shown that for all K-quasiconformal mappings φ : C → C, |φ(E)| ≤ C |E|1/K (2.4) where C is a constant that depends on K and the normalizations. By scaling we may always arrange diam(φ(E)) = diam(E) ≤ 1 (2.5) and then C = C(K) depends only on K. In order to achieve the result (2.4), one first reduces to the case where the set E is a finite union of disks. Secondly, the mapping φ is written as φ = h ◦ φ1, where both h, φ1 : C → C are K-quasiconformal mappings, such that φ1 is conformal on E and h is conformal in the complement of the set F = φ1(E).
Here one obtains the right conclusion for φ1,
1 |φ1(E)| ≤ C |E| K
36 2.2 Absolute continuity
by including φ1 in a holomorphic family of quasiconformal mappings. Further, one shows in [4, p. 50] that under the special assumption that h is conformal outside of F, we have |h(F)| ≤ C |F|, (2.6) where the constant C still depends only on K.
In searching for absolute continuity properties of other Hausdorff measures under qua- siconformal mappings, such a decomposition appears unavoidable, and leads one to look for counterparts of (2.6) for Hausdorff measures Ht or Hausdorff contents Mt. Here we establish the result for the dimension t = 1.
Lemma 2.1. Suppose E ⊂ C is a compact set, and let φ : C → C be a principal K-quasiconformal mapping, such that φ is conformal on C \ E. Then,
M1(φ(E)) ' M1(E), with constants depending only on K.
For the proof of this Lemma, we need some previous results. We say that a set E is removable for BMO analytic functions if for every open set Ω ⊃ E, every function f ∈ BMO(Ω), analytic on Ω \ E, admits a holomorphic extension in the whole of Ω. These sets were characterized by Kaufman [31] with the condition M1(E) = 0. The analytic BMO capacity is defined by 0 γ0(E) = sup | f (∞)| where the supremum runs over all functions f ∈ BMO(C) analytic on C \ E, such that k f k∗ ≤ 1. Verdera [54] showed the following estimate.
Lemma 2.2. For every compact set E ⊂ C, 1 M (E) ' γ0(E).
We recall now a result from Reimann [46], asserting that the space of functions of
37 2 Removable singularities for bounded quasiregular mappings bounded mean oscillation, BMO, is invariant under quasiconformal changes of vari- ables. More precisely, if φ is K-quasiconformal and f ∈ BMO(C), then f ◦ φ ∈ BMO(C) and
k f ◦ φk∗ ≤ C(K) k f k∗.
Proof of Lemma 2.1. Due to Lemma 2.2, it is sufficient to show that
γ0(φ(E)) ≤ C(K) γ0(E).
So let f be an admissible function for γ0(E). Then, f ∈ BMO(C) is a holomorphic −1 mapping of C \ E such that k f k∗ ≤ 1 and f (∞) = 0. The function g = f ◦ φ is then also in BMO(C), with
kgk∗ ≤ C(K).
C C g Moreover, since φ is conformal on \ E, g is holomorphic on \ φ(E). Thus, C(K) is an admissible function for γ0(φ(E)). The normalization on φ ensures that g(∞) = 0 and
|g0(∞)| = lim |z g(z)| = lim |φ(w) f (w)| = | f 0(∞)| |z|→∞ |w|→∞
Hence, γ0(E) ≤ C(K) γ0(φ(E)). The converse inequality follows by symmetry.
This lemma is a first step towards the results on absolute continuity.
Theorem 2.3. Let E be a compact set and φ : C → C K-quasiconformal, normalized by (2.5). Then
K+1 1 2 2K M (φ(E)) ≤ C M K+1 (E) 2 where the constant C = C(K) depends only on K. In particular, if H K+1 (E) = 0 then H1(φ(E)) = 0.
Proof. There is no restriction if we assume E ⊂ D. We can also assume that φ is a principal K-quasiconformal mapping, conformal outside of D. Now, given ε > 0, there
38 2.2 Absolute continuity
is a finite covering of our set E by open disks Dj = D(zj, rj), j = 1, ..., n, such that
n 2 2 K+1 K+1 ∑ rj ≤ M (E) + ε. j=1
Ω n Denote by = ∪j=1Dj. As in [4], we use a decomposition φ = h ◦ φ1, where both φ1, h are principal K-quasiconformal mappings. Moreover, we may require that φ1 is confor- mal in Ω ∪ (C \ D) and that h is conformal outside of φ1(Ω).
By Lemma 2.1, we see that
1 1 1 1 M (φ(E)) ≤ M (φ(Ω)) = M (h ◦ φ1(Ω)) ≤ C M (φ1(Ω)).
1 Hence, the problem has been reduced to estimating M (φ1(Ω)). For this, K-quasidisks have area comparable to the square of the diameter,
1 2 1/2 diam(φ1(Dj)) ' |φ1(Dj)| = J(z, φ1) dA(z) ZDj ! with constants which depend only on K. Thus, using Holder¨ estimates twice, we obtain for p > 1 small enough
1 1 2p 1− 2p n n n p−1 p 2p−1 ∑ diam(φ1(Dj)) ≤ C(K) ∑ J(z, φ1) dA(z) ∑ |Dj| . j=1 j=1 ZDj ! j=1 !
Now we are in the special situation of [10, Lemma 5.2]. Namely, as φ1 is conformal in Ω K the subset , we may take p = K−1 to obtain
n p p ∑ J(z, φ1) dA(z) ' J(z, φ1) dA(z) ≤ 1. Ω j=1 ZDj Z
39 2 Removable singularities for bounded quasiregular mappings
p−1 1 With the above choice of p one has 2p−1 = K+1 . Hence we get
K+1 2 n n 2 K K+1 2 2K K+1 K+1 ∑ diam(φ1(Dj)) ≤ C(K) ∑ rj ≤ C(K) M (E) + ε . (2.7) = = ! j 1 j 1
But ∪jφ1(Dj) is a covering of φ1(Ω), so that actually we have
K+1 1 1 2 2K M (φ(E)) ≤ CM (φ1(Ω)) ≤ C(K) M K+1 (E) + ε Since this holds for every ε > 0, the result follows.
At this point we want to emphasize that for a general quasiconformal mapping φ we p < K K have J(·, φ) ∈ Lloc only for p K−1 . The improved border line integrability (p = K−1 ) under the extra assumption that φ Ω is conformal, was shown in [10, Lemma 5.2]. This phenomenon was crucial for our argument, since we are studying Hausdorff measures rather than dimension. Actually, the same procedure shows that inequality (2.7) works in a much more general setting. That is, still under the special assumption that φ1 is n conformal in ∪j=1Dj, we have for any t ∈ [0, 2]
1 1 1 n d n t K d t ∑ diam(φ1(Dj)) ≤ C(K) ∑ diam(Dj) (2.8) j=1 ! j=1 !
2Kt where d = 2+(K−1)t . On the other hand, another key point in our proof is the estimate
M1(h(E)) ≤ C M1(E) whenever h is a planar K-quasiconformal mapping which is conformal outside of E. We believe that finding the counterpart to this estimate is crucial for understanding distortion of Hausdorff measures under quasiconformal mappings.
Problem 2.4. Suppose we are given a real number d ∈ (0, 2]. Then for any compact set E ⊂ C and for any
40 2.2 Absolute continuity
K-quasiconformal mapping h which is conformal on C \ E, we have
Md(h(E)) ' Md(E) (2.9) with constants that depend on K and d only.
We already know that (2.9) is true for d = 1 and d = 2. An affirmative answer for general d, combined with the optimal integrability bound proving (2.8), would provide ∗ d t 2Kt the absolute continuity of φ H with respect to H , where d = 2+(K−1)t , 0 ≤ t ≤ 2 and K ≥ 1. Therefore, (2.9) would have important consequences in the theory of quasicon- formal mappings.
The positive answer for the dimension d = 1 was based on Lemma 2.2, as well as on the quasiconformal invariance of BMO. However, this is not the only quasiconfor- mal invariant class. Actually, as we will see, the invariance of VMO has interesting consequences. Here we start with the first one.
Theorem 2.5. 2 Let E ⊂ C be a compact set, and φ : C → C a K-quasiconformal mapping. If H K+1 (E) is finite (or even σ-finite), then H1(φ(E)) is σ-finite.
1 Due to Theorem 1.4, this result may be equivalently expressed in terms of M∗. Theorem 2.6. 2 K+1 Let E ⊂ C be a compact set, and φ : C → C a K-quasiconformal mapping. If M∗ (E) = 0, 1 then M∗(φ(E)) = 0.
We say that a set E is removable for analytic VMO functions if for every open set Ω ⊃ E, every function f ∈ VMO(Ω), analytic on Ω \ E, extends holomorphically to Ω. These 1 sets were characterized by Verdera [54] with the condition M∗(E) = 0. The analytic VMO capacity is defined by 0 γ∗(E) = sup | f (∞)| where the supremum runs over all VMO(C) functions f , analytic on C \ E, such that k f k∗ ≤ 1. Verdera [54] showed the following Lemma.
41 2 Removable singularities for bounded quasiregular mappings
Lemma 2.7. For every compact set E ⊂ C, 1 M (E)∗ ' γ∗(E).
Notice that VMO is quasiconformally invariant, as BMO. This is due to the fact that VMO can be seen as the BMO closure of compactly supported continuous functions. Consequently, we can prove the following auxiliar result.
Lemma 2.8. Let E be a compact set. For any principal K-quasiconformal mapping φ : C → C, conformal on C \ E, we have 1 1 M∗(φ(E)) ' M∗(E).
Proof. Again, as in Lemma 2.1, we will use the equivalence given at Lemma 2.7 between 1 γ∗ and M∗. Thus, let f be an admissible function for γ∗(φ(E)). Then f ∈ VMO is an- alytic on C \ φ(E), k f k∗ ≤ 1 and f (∞) = 0. Set g = f ◦ φ. Therefore g ∈ VMO, 0 0 g is analytic on C \ E, kgk∗ ≤ C(K) and |g (∞)| = | f (∞)| since φ is a principal 1 K-quasiconformal mapping. Consequently, C(K) g is admissible for γ∗(E) and hence γ∗(φ(E)) ≤ C(K) γ∗(E).
Proof of Theorem 2.6. Naturally, the argument is similar to that in Theorem 2.3. Without loss of generality, we may assume that E ⊂ D and that φ is a principal K-quasiconformal 2 mapping. Since H K+1 (E) is finite, for any δ there is a family of disks Di such that 2 2 K+1 K+1 < E ⊂ ∪iDi, ∑i diam(Di) ≤ H (E) + 1 and diam(Di) δ. Set Ω = ∪iDi. Again, we have a decomposition φ = φ2 ◦ φ1, where both φ1 and φ2 are principal K-quasiconformal mappings, and where we may require that φ1 is conformal in (C \ D) ∪ Ω, and φ2 is con- formal outside φ1(Ω). Thus,
1 1 M∗(φ(E)) ≤ M∗(φ(Ω)).
By Lemma 2.7, the lower content can be replaced by the VMO capacity,
1 M∗(φ(Ω)) ≤ C γ∗(φ(Ω)).
42 2.2 Absolute continuity
Now, since φ2 is conformal outside of φ1(Ω), by Lemma 2.8 we get
γ∗(φ2 ◦ φ1(Ω)) ' γ∗(φ1(Ω)).
Again, by Lemma 2.7, 1 γ∗(φ1(Ω)) ≤ C M∗(φ1(Ω)).
1 Now it remains to bound M∗(φ1(Ω)). For this, take h ∈ F, h(t) = t ε(t), and argue as in Theorem 2.3. Since K-quasiconformal mappings are Holder¨ continuous with exponent 1/K,
h 1/K M (φ1(Ω)) ≤ ∑ diam(φ1(Dj))ε(diam(φ1(Dj))) ≤ ε(CK δ ) ∑ diam(φ1(Dj)) j j K−1 2K 1/K K 1/2K ≤ ε(CK δ ) ∑ J(z, φ1) K−1 dm(z) |Dj| j ZDj ! K+1 2K K+1 1/K 2 1/K 2 2K ≤ ε(CK δ ) CK ∑ diam(Dj) K+1 ≤ ε(CK δ ) CK H K+1 (E) + 1 . ! j
Finally, taking δ → 0 we get Mh(φ(E)) = 0. This holds for any h ∈ F, and the theorem follows.
2 One might think of extending the preceeding results from the critical index K+1 to arbi- trary ones by using other capacities that behave like a Hausdorff content. For instance, the analytic Lipα capacity γα introduced by O’Farrell in [42] satisfies
1+α M (E) ' γα(E),
whenever α ∈ (0, 1). Unfortunately, the space Lipα is not quasiconformally invariant. Thus, other procedures are needed. It turns out that the homogeneous Sobolev spaces provide suitable tools, basically since W˙ 1,2(C) is invariant under quasiconformal map- pings. Here recall that for 0 < α < 2 and p > 1, the homogeneous Sobolev space
43 2 Removable singularities for bounded quasiregular mappings
W˙ α,p(C) is defined as the space of Riesz potentials
f = Iα ∗ g
p 1 1,p ∈ (C) ( ) = k k α,p = k k = ˙ (C) where g L and Iα z |z|2−α . The norm f W˙ (C) g p. When α 1, W agrees with the space of functions f whose first order distributional derivatives are given by Lp(C) functions. Let f ∈ W˙ 1,2(C) and let φ be a K-quasiconformal mapping on C. Defining g = f ◦ φ we have
|Dg(z)|2dA(z) = |D f (φ(z))|2 |Dφ(z)|2 dA(z) C C Z Z ≤ K |D f (φ(z))|2 J(z, φ) dA(z) C Z = K |D f (w)|2 dA(w) C Z so that g ∈ W˙ 1,2(C). In other words, every K-quasiconformal mapping φ induces a bounded linear operator
T : W˙ 1,2(C) → W˙ 1,2(C), T( f ) = f ◦ φ (2.10) with norm depending only on K. As we have mentioned before, this operator T is also bounded on BMO(C) ([46]). Moreover, Reimann and Rychener [47, p.103] proved that 2 ,q W˙ q (C), q > 2, may be represented as a complex interpolation space between BMO(C) 2 ,q and W˙ 1,2(C). It follows that T is bounded on the Sobolev spaces W˙ q (C), q > 2. More precisely, there exists a constant C = C(K, q) such that
k f ◦ φk 2 ,q ≤ Ck f k 2 ,q , (2.11) W˙ q (C) W˙ q (C) for any K-quasiconformal mapping φ on C. These invariant function spaces provide us with related invariant capacities. Recall (e.g. [1, pp.34 and 46]) that for any pair α > 0, p > 1 with 0 < αp < 2, one defines the Riesz capacity of a compact set E by
p C˙ α,p(E) = sup{µ(E) }
44 2.2 Absolute continuity
where the supremum runs over all positive measures µ supported on E, such that kIα ∗ 1 1 µkq ≤ 1, p + q = 1. We get an equivalent capacity if we replace positive measures µ by p distributions T supported on E, kIα ∗ Tkq ≤ 1, and take the supremum of |hT, 1i| .
To see the connection with equation (2.11) consider the set functions
0 ∞ C ∞ γ1−α,q(E) = sup{| f ( )|; f ∈ H( \ E), k f kW˙ 1−α,q ≤ 1, f ( ) = 0}.
Observe again that | f 0(∞)| = |h∂ f , 1i| where this action must be understood in the sense of distributions. With this terminology we have the analogous result to Lemmas 2.2 and 2.7.
Lemma 2.9. Suppose that E is a compact subset of the plane. Then, for any p ∈ (1, 2),
C˙ α,p(E) ' γ1−α,q(E),
2 p where α = p − 1 and q = p−1 .
q Proof. On one hand, take µ an admissible measure for C˙ α,p. Then, Iα ∗ µ is in L with 1 ∞ norm at most 1. Define f = z ∗ µ. Clearly, f is analytic outside E, f ( ) = 0 and | f 0(∞)| = µ(E). Moreover, proceeding roughly,
1 ξ 1 f (ξ) ' µ(ξ) = µ(ξ) = R I µ ξ |ξ| |ξ| 1 b b b and consequently we can write b b b
1 f = ∗ µ = R(I ∗ µ) = I − ∗ R(I ∗ µ) z 1 1 α α where R is a Calderon-Zygmund´ operator and k f kW˙ 1−α,q = kR(Iα ∗ µ)kq . kIα ∗ µkq.
For the converse, let f = I1−α ∗ g be an admissible function for γ1−α,q. We have that
T = ∂ f is an admissible distribution for C˙ α,p because
t Iα ∗ T = R (g)
45 2 Removable singularities for bounded quasiregular mappings
t 1/p 0 where R is the transpose of R. Thus, C˙ α,p(E) ≥ |hT, 1i| = | f (∞)| and the proof is complete.
We end up with new invariants built on the Riesz capacities.
Theorem 2.10. Let φ : C → C be a principal K-quasiconformal mapping of the plane, which is conformal on C < < 2 \ E. Let 1 p 2 and α = p − 1. Then
C˙ α,p(φ(E)) ' C˙ α,p(E), with constants that depend only on K and p.
Proof. By the preceding Lemma it suffices to show that γ1−α,q (φ(E)) ≤ γ1−α,q(E).
Let f be an admissible function for γ1−α,q(φ(E)). This means that f is holomorphic on C ∞ \ φ(E), f ( ) = 0 and that k f kW˙ 1−α,q ≤ 1. Then, we consider the function g = f ◦ φ. ∞ 2 Clearly, ∂( f ◦ φ) is conformal outside E, and g( ) = 0. Moreover, for α = p − 1 we 2 have 1 − α = q . Hence, because of equation (2.11),
kgkW˙ 1−α,q ≤ CKk f kW˙ 1−α,q ≤ CK
1 so that g is an admissible function for γ − (E). Hence, as φ is a principal K- CK 1 α,q quasiconformal mapping, 1 0 γ1−α,q(E) ≥ | f (∞)| CK and we may take supremum over f .
The above theorem has direct consequences towards the absolute continuity of Haus- dorff measures, but unfortunately these are slighly weaker than one would wish for. In h fact, there are compact sets F such that Cα,p(F) = 0 and H (F) > 0 for some measure function h(t) = tp ε(t) [1]. Thus, Theorem 2.10 does not help for Problem 2.4. We have to content with the following setup.
46 2.2 Absolute continuity
Given 1 < d < 2 consider the measure functions h(t) = td ε(t) where
1 1 dt ε(t) d−1 < ∞ (2.12) t Z0 Typical examples are h(t) = td| log t|−s or h(t) = td| log t|1−p log−s(| log t|) where s > d − 1.
Corollary 2.11. Let E be a compact set on the plane, and φ : C → C a principal K-quasiconformal mapping, conformal outside of E. Let 1 < d < 2. Then,
Mh(φ(E)) ≤ C Md(E) for any measure function h(t) = td ε(t) satisfying (2.12). Moreover, if Hd(E) < ∞ then Hh(φ(E)) = 0 for every such h.
Proof. By [1, Theorem 5.1.13], given a measure function h satisfying (2.12) there is a constant C = C(h) with
2 Mh(φ(E)) ≤ C C (φ(E)), α = − 1 α,d d
By Theorem 2.10, Cα,d(φ(E)) ≤ C Cα,d(E) and using again [1, Theorem 5.1.9] we have d finally Cα,d(E) ≤ M (E).
Arguing now as in Theorems 2.3 and 2.6, we arrive at the following conclusion.
Corollary 2.12. Let E be a compact set of the plane and φ : C → C a K-quasiconformal mapping. Let t ∈ 2 2Kt ( K+1 , 2) and d = 2+(K−1)t . Then,
1 Mh(φ(E)) ≤ C Mt(E) Kt for any measure function h satisfying (2.12).
2 < < < < Here note that for K+1 t 2 we always have 1 d 2.
47 2 Removable singularities for bounded quasiregular mappings
2.3 Improved Painleve´ theorems
A set E is said to be removable for bounded (resp. BMO) K-quasiregular mappings, if ev- ery K-quasiregular mapping in C \ E which is in L∞(C) (resp.BMO(C) ) admits a K- quasiregular extension to C. Obviously, when K = 1, in both situations L∞ and BMO, we recover the original Painleve´ problem (see Section 1.3). Moreover, by means of the Stoilow factorization, one can represent any bounded K-quasiregular function as a composition of a bounded analytic function and a K-quasiconformal mapping. The corresponding result holds true also for BMO since this space, like L∞, is quasiconformally invariant. Therefore, when we ask ourselves if a set E is K-removable, we just need to analyze how it may be distorted under quasiconformal mappings, and then apply the known results for the analytic situation. With this basic scheme, it is shown in [4, Corollary 1.5] 2 that every set with dimension below K+1 is K-removable. Indeed, the precise formulas for the distortion of dimension ensure that such sets will always be mapped to a set with dimension strictly smaller than 1, therefore with vanishing analytic capacity.
2 Iwaniec and Martin [29] had earlier conjectured that sets of zero K+1 -dimensional mea- sure are K-removable. A preliminary answer to this question was found in [6], and actually it was that argument which suggested Theorem 2.3. Using our results from 2 Section 2.2 we can now prove that sets of zero K+1 -dimensional measure are removable for BMO K-quasiregular mappings.
Corollary 2.13. 2 Let E be a compact subset of the plane. Assume that H K+1 (E) = 0. Then E is removable for all BMO K-quasiregular mappings.
Proof. Assume that f ∈ BMO(C) is K-quasiregular on C \ E. Denote by µ the Beltrami coefficient of f , and let φ be a principal solution to ∂φ = µ ∂φ. Then, F = f ◦ φ−1 is holomorphic on C \ φ(E) and F ∈ BMO(C). On the other hand, as we showed in Theorem 2.3, H1(φ(E)) = 0. Thus, φ(E) is a removable set for BMO analytic functions. In particular, F admits an entire extension. Hence, f = F ◦ φ extends quasiregularly to
48 2.3 Improved Painleve´ theorems the whole plane.
We believe that Corollary 2.13 is sharp, in the sense that we expect a positive answer to the following question.
Problem 2.14. 2 Does there exist for every K ≥ 1 a compact set E with 0 < H K+1 (E) < ∞ such that E is not removable for all K-quasiregular functions in BMO(C).
> 2 Here we observe that by [4, Corollary 1.5], for every t K+1 there exists a compact set E with dimension t, nonremovable for bounded (and hence for BMO) K-quasiregular mappings.
Next we return back to the L∞ problem by noticing that Theorem 2.3 proves the con- 2 jecture of Iwaniec and Martin that sets with H K+1 (E) = 0 are removable for bounded K-quasiregular mappings. This is so because those sets E satisfy H1(φ(E)) = 0 when- ever φ is K-quasiconformal and, hence, γ(φ(E)) = 0. Nevertheless, analytic capacity is somewhat smaller than length, and hence with Theorem 2.5 we may go even further. To 2 be precise, if a set has finite (or even σ-finite) K+1 -dimensional Hausdorff measure, then all its K-quasiconformal images have at most σ-finite length. Thus, such images might still be removable for analytic bounded functions, provided that they do not contain any rectifiable subset with positive length. For this the following result, which appears in [5], gives the correct tools.
Theorem 2.15. Suppose that φ : C → C is a K-quasiconformal mapping. Let E ⊂ ∂D be such that dim(E) = 1. Then, we have the strict inequality
2 dim(φ(E)) > K + 1
As a direct consequence, one obtains the following version, which applies to any recti- fiable set.
49 2 Removable singularities for bounded quasiregular mappings
Corollary 2.16. Suppose that E is a rectifiable set, and let φ : C → C be a K-quasiconformal mapping. Then, there exists a subset E0 ⊂ E of zero length such that
2 dim φ(E \ E ) > . 0 K + 1
It points out that the proofs of the above results actually give more than what here is stated. Actually, there exists constants c1, c2 > 0 such that if E ⊂ ∂D, then
2 2 dim(E) ≥ 1 − c1 ε ⇒ dim(φ(E)) ≥ 1 − c2 ε (2.13) whenever φ is (1 + ε)-quasiconformal and ε > 0 is small enough. We end up with the following improved version of Painleves´ Theorem for quasiregular mappings.
Theorem 2.17. 2 Let E be a compact set in the plane, and let K > 1. Assume that H K+1 (E) is σ-finite. Then E is removable for all bounded K-quasiregular mappings.
Proof. Let f : C → C be bounded, and assume that f is K-quasiregular on C \ E. Let ∂ f µ = ∂ f be the Beltrami coefficient of f and let φ be a principal homeomorphic solution to ∂φ = µ ∂φ. Then F = f ◦ φ−1 is a bounded holomorphic mapping on C \ φ(E). If we can extend F holomorphically to the whole plane, then we are done. Thus, we have to show that φ(E) has zero analytic capacity. By Theorem 2.5, φ(E) has σ-finite length, 1 that is, φ(E) = ∪nFn where each H (Rn) < ∞. From Besicovitch’s Theorem (Theorem
1.5), each set Fn can be decomposed as
Fn = Rn ∪ Un ∪ Bn
where Rn is a rectifiable set, Un is a purely unrectifiable set, and Bn is a set of zero length. Because of Tolsa’s Theorem on semiadditivity of analytic capacity [51],
γ(Fn) ≤ C (γ(Rn) + γ(Un) + γ(Bn))
50 2.3 Improved Painleve´ theorems
1 Now, γ(Bn) ≤ H (Bn) = 0 and γ(Un) = 0 since purely unrectifiable sets of finite length have zero analytic capacity [15]. On the other hand, Rn is a rectifiable image, under 2 a K-quasiconformal mapping, of a set of dimension K+1 . Thus according to Theorem −1 1 2.15 and Corollary 2.16, applied to φ , we must have H (Rn) = 0. Therefore we get
γ(Fn) = 0 for each n. Again by semiadditivity of analytic capacity we conclude that γ(φ(E)) = 0.
As pointed out earlier, the above theorem does not hold for K = 1, any rectifiable set such as E = [0, 1], of finite and positive length gives a counterexample. In the above proof the improved distortion of rectifiable sets was the decisive phenomenon allowing the result. A similar procedure to that in the proof of Theorem 2.15 may be used to show the following expansion result (compare with the compression result of equation (2.13)).
Corollary 2.18.
There exist constants c1, c2 > 0 such that if φ : C → C is K-quasiconformal, then
2 2 dim(E) ≤ 1 − c1 ε ⇒ dim(φ(E)) < 1 − c2 ε whenever K = 1 + ε, E ⊂ ∂D and ε > 0 is small enough.
2 This Corollary drives to removable sets with Hausdorff dimension above K+1 . Corollary 2.19. There exist a constant c ≥ 1 such that if E ⊂ ∂D is compact and
K − 1 2 dim(E) ≤ 1 − c K + 1 then E is removable for bounded K-quasiregular mappings, K = 1 + ε, whenever e > 0 is small enough.
Proof. This is a consequence of Corollary 2.18. If ε > 0 is small enough and K = 1 + ε, then the K-quasiconformal images of E will always have dimension below 1, so that γ(φ(E)) = 0 for each K-quasiconformal mapping φ.
51 2 Removable singularities for bounded quasiregular mappings
2.4 Extremal distortion and nonremovable sets
The discussion of Section 2.3 on removable sets for bounded K-quasiregular mappings 2 2 still leaves open the case where dim(E) = K+1 but Edoes not have σ-finite K+1 -dimensional Hausdorff measure. In view of Theorem 2.17 one might think that maybe everything at the critical dimension is removable for bounded K-quasiregular mappings. In this section we show that our results are sharp in a quite strong sense. The main goal of this section consists in finding some kind of analog result to Mattila’s Theorem (Theorem 1.7) for the K-quasiregular problem.
Theorem 2.20. Let K ≥ 1. Let h(t) = t ε(t) be such that:
• ε is continuous and nondecreasing, ε(0) = 0 and ε(t) = 1 whenever t ≥ 1. tδ • lim = 0 for all δ > 0. t→0 ε(t)
2 There exists a compact set E ⊂ D, a measure function g(s) = s K+1 η(s) and a K-quasiconformal mapping φ such that:
• Hh(φ(E)) ' 1
• Hg(E) ' 1 sδ • The function η is such that lim η(s) = 0, and lim = 0 for all δ > 0. As a s→0 s→0 η(s) 2 consequence, dim(E) = K+1 .
1 2 1 1+ 1 ε(t) 1 η(s) K • dt ≥ ds whenever this makes sense. t K s Z0 Z0
Proof. We will construct a K-quasiconformal mapping φ as the limit of a sequence φN of K-quasiconformal mappings, and E will be a Cantor-type set. The main idea consists on changing, at every step in the construction of E, both the quantity m j and the size of the generating disks. So we will construct a sequence of integers m j wich will go very fastly to ∞, and two sequences of real numbers σj, Rj ∈ (0, 1) that will converge to 0 very quiclky.
52 2.4 Extremal distortion and nonremovable sets
1 Step 1. Consider m1 disjoint disks of radius R1, centered at zi , i = 1, ..., m1, uniformly distributed inside D, such that 1 c = m R2 > 1 1 1 2
The function f (t) = m1 h(tR1) is continuous in the variable t and f (0) = 0. Moreover, for any t > 0
ε(tc1/2 m−1/2) ( ) = ( ) = 1/2 1/2 ( 1/2 −1/2) = 1 1 2 f t m1tR1 ε tR1 t c1 m1 ε tc1 m1 1/2 −1/2 t c1 tc1 m1 so that lim f (t) = ∞ m1→∞
Hence, for any t < 1, we may choose m1 big enough so that the equation m1 h(σ1 R1) = 1 has a unique solution σ1 ∈ (0, t). Under this situation, an easy computation shows that
2 2K K−1 K K+1 K+1 K+1 m1 (σ1 R1) ε(σ1 R1) (c1) = 1
1 1 K Denote r1 = R1, and for each i = 1, . . . , m1, let ϕi (z) = zi + σ1 R1 z, and
1 = 1(D) = ( ) Di K ϕi D zi, r1 σ1 0 1 D K Di = ϕi ( ) = D(zi, σ1 r1)
0 Notice that Di and Di are concentric disks. We define
1−K 0 σ1 (z − zi) + zi z ∈ Di 1 K −1 z−zi 0 g1(z) = (z − zi) + zi z ∈ Di \ D r1 i z otherwise
m1 0 This mapping is K-quasiconformal, conformal outside of i=1(Di \ Di), maps Di onto 0 00 1 itself and Di onto Di = D(zi , σ1r1), while the rest of the Splane remains fixed. Define then φ1 = g1.
53 2 Removable singularities for bounded quasiregular mappings
Step 2. We have already fixed m1, R1, σ1 and c1. Consider m2 disjoint disks of radius R2, 2 D centered at zj , j = 1, . . . , m2, uniformly distributed inside , such that
1 c = m R2 > 2 2 2 2
Then we repeat the above procedure. So the function f (t) = m1m2h(tR1 R2) is continu- ous in t, f (0) = 0 and lim f (t) = ∞ m2→∞ for any t > 0. Thus, we may choose m2 big enough, so that the equation
m1m2h(σ1σ2R1R2) = 1 has a unique solution σ2 ∈ (0, 1) as small as we wish. Then,
2 2K K−1 K K K+1 K+1 K+1 m1m2 (σ1 σ2 R1R2) ε(σ1σ2R1R2) (c1c2) = 1
2 2 K Denote r2 = R2 σ1r1 and ϕj (z) = zj + σ2 R2 z, and define the auxiliary disks
1 1 2 D = φ1 ϕ ◦ ϕ (D) = D(z , r2) ij σK j i ij 2 0 1 2 D 0 K Dij = φ1 ϕj ◦ ϕi ( ) = D (zij, σ2 r2) 0 for i = 1, . . . , m1 and j = 1, . . . , m2. Notice again that Dij and Dij are concentric. Now let 1−K 0 σ2 (z − zij) + zij z ∈ Dij 1 −1 z−zij K 0 g2(z) = (z − zij) + zij z ∈ Dij \ D r2 ij z otherwise
0 Clearly, g2 is K-quasiconformal, conformal outside of i,j(Dij \ Dij), maps Dij onto itself 0 00 and Dij onto Dij = D(zij, σ2r2), while the rest of the Splane remains fixed. Now define φ2 = g2 ◦ φ1.
54 2.4 Extremal distortion and nonremovable sets
φ
D D Figure 2.1
Nth Step. We take mN uniformly distributed disjoint disks of radius RN, giving at least half of the area of D, 1 c = m R2 > N N N 2
Since the continuous function f (t) = m1 . . . mN h(tR1 . . . RN) satisfies f (0) = 0 and
lim f (t) = ∞, for any t > 0, then we may choose mN big enough so that the equation mN→∞
m1 . . . mN h(σ1 . . . σN R1 . . . RN) = 1
has a unique solution σN as small as we wish. Let us remark that we can take lim σN = N→∞ 0. Then, for these values,
2 2K K−1 K K K+1 K+1 K+1 m1 . . . mN (σ1 . . . σN R1 . . . RN) ε(σ1 . . . σN R1 . . . RN) (c1 . . . cN ) = 1
N N K Denote then ϕj (z) = zj + σN RN z and rN = RN σN−1rN−1. Define, for any multiindex
J = (j1, ..., jN), 1 ≤ jk ≤ mk, k = 1, ..., N,
1 1 N D DJ = φN−1 ϕ ◦ · · · ◦ ϕ ( ) = D(zJ, rN) σK j1 jN N 0 = φ ϕ1 ◦ · · · ◦ ϕN (D) = 0( , σK ) DJ N−1 j1 jN D zJ NrN
55 2 Removable singularities for bounded quasiregular mappings
0 As before, DJ and DJ are concentric. Let
1−K 0 σN (z − zJ ) + zJ z ∈ DJ 1 −1 z−zJ K 0 gN (z) = (z − zJ) + zJ z ∈ DJ \ D rN J z otherwise
Clearly, g is K-quasiconformal, conformal outside of (D \ D0 ), maps D N J=(j1,...,jN) J J J 0 00 onto itself and DJ onto DJ = D(zJ, σNrN), while the restS of the plane remains fixed. Now define φN = gN ◦ φN−1.
D Since each φN is K-quasiconformal, and all agree outside the unit disk , there exists a limit K-quasiconformal mapping
φ = lim φN N→∞
1,p C < 2K with convergence in Wloc ( ) for any p K−1 . On the other hand, φ maps the compact set ∞ E = ϕ1 ◦ · · · ◦ ϕN (D) j1 jN N\=1 j1,...,[jN to the set ∞ φ(E) = ψ1 ◦ · · · ◦ ψN (D) j1 jN N\=1 j1[,...,jN i i where here we defined ψj(z) = zj + σi Ri, j = 1, ..., mi, i = 1, ..., N. To end the proof, denote K K sN = (σ1 R1) . . . (σN RN)
tN = (σ1R1) . . . (σN RN)
With this notation, observe that our parameters RN, mN, σN have been chosen so that
2 2K K−1 K+1 K+1 K+1 m1 . . . mN h(tN ) = 1 and m1 . . . mN sN ε(tN ) (c1 . . . cN) = 1
56 2.4 Extremal distortion and nonremovable sets
2 Set g(s) = s K+1 η(s), where η is defined by
2K K−1 η(s) = ε(tN ) K+1 (c1 . . . cN) K+1 whenever s ∈ [sN, sN−1).
Claim 1. Hh(φ(E)) ' 1. We start by fixing δ > 0 very small. The covering (ψ1 ◦ · · · ◦ ψN (D)) is admissible j1 jN j1,...,jN when computing Hh(φ(E)) if diam(ψ1 ◦ · · · ◦ ψN (D)) ≤ δ. Thus, δ j1 jN
Hh(φ( )) ≤ (diam(ψ1 ◦ · · · ◦ ϕN (D))) = . . . ( ) = 1 δ E ∑ h j1 jN m1 mN h tN j1,...,jN because of our choice of σN. For the converse inequality, take a finite covering (Uj)j of
φ(E) by open disks of diameter diam(Uj) ≤ δ. Now let
δ0 = inf(diam(Uj)) > 0 j
Denote by N the minimal integer such that t ≤ δ . By construction, the family (ψ1 ◦ 0 N0 0 j1 N0 D h · · · ◦ ψ ( ))j1 ,...,jN is a covering of φ(E) with the M -packing condition. Thus, jN0 0
N0 1 D ∑ h(diam(Uj)) ≥ C ∑ h(diam(ϕ ◦ · · · ◦ ϕj ( ))) = C jN0 1 j j1,...,jN0
h h Hence, Hδ (φ(E)) ≥ C and now letting δ → 0, we get that C ≤ H (φ(E)) ≤ 1 and our first claim follows.
Claim 2. Hg(E) ' 1. The proof of this fact is analogous to that of Claim 1.
δ sN Claim 3. One can choose σj so that lim = 0 for every δ > 0. N→∞ η(sN )
57 2 Removable singularities for bounded quasiregular mappings
Fix any δ > 0. We first notice that
2K K+1 K+1 K−1 δ δ δ 2K δ sN sN tN (σ1 . . . σN ) = 2K K−1 = 1 η( ) +1 +1 ε( ) +1 sN ε(tN ) K (c1 . . . cN) K tN (c1 . . . cN) K ! tδ Now, since lim = 0 for every δ > 0, it will be sufficient to find N big enough such t→0 ε(t) that δ 1 (σ1 . . . σN) < (c1 . . . cN) K+1
However, if δ > 0 is fixed, then 1 1 K+1 σδ < j 2 for every j > N = N(δ, K). Thus, the result follows.
1 2 1 1+ 1 ε(t) 1 η(s) K Claim 4. dt ≥ ds. t K s Z0 Z0
From the definition of η and since c1 . . . cN < 1 we have
tN−1 2 ε(t) 2 tN−1 1+ 1 sN−1 1 dt ≥ ε(t ) log > η(s ) K log − (K − 1) log t N t N s σ ZtN N N N Now notice that 1 1 1 1 s = ≤ N−1 log log K log σN K σN K sN Therefore t 2 s 1+ 1 N−1 ε(t) 1 N−1 η(s) K dt ≥ ds t K s ZtN ZsN and then Claim 4 follows.
We would like to remark at this point that the construction in the above theorem also works if we start with h(t) = t. In that case, we get a compact set E, a K-quasiconformal
58 2.4 Extremal distortion and nonremovable sets
2 mapping φ and a measure function g(s) = s K+1 η(s) satisfying
sδ lim η(s) = 0 and lim = 0 ∀δ > 0 s→0 s→0 η(s)
g 1 2 and such that H (E) ' 1 and H (φ(E)) ' 1. Again, E has dimension K+1 , also with non σ-finite measure, but now φ(E) has positive finite length.
Corollary 2.21. There exists a compact set E, such that:
• E is non removable for bounded K-quasiregular mappings.
2 • Hg(E) ' 1, where g(s) = s K+1 η(s) is such that
sδ lim η(s) = 0 and lim = 0 ∀δ > 0. s→0 s→0 η(s)
2 In particular, E has Hausdorff dimension K+1 .
Proof. Let h(t) = t ε(t) be any measure function in the hypothesis of the above example and such that 1 h(t)2 dt < ∞ t3 Z0 2 Now construct a compact set E of dimension K+1 and a K-quasiconformal mapping φ such that Hh(φ(E)) ' 1. By [40], φ(E) has positive analytic capacity, so there ex- ists a bounded function f , holomorphic off φ(E), which is not a constant. Therefore the composition g = f ◦ φ is K-quasiregular off E, bounded on C and cannot extend quasiregularly to C. Thus E is non removable.
It should be pointed out here that this nonremovable set E supports a positive measure ν (actually one may choose ν = Hg) with growth ν(D(z, r)) ≤ g(r) and such that
1 1 1 1+ 1 g(s) p−1 ds η(s) K = ds < ∞ s2−αp s s Z0 Z0
59 2 Removable singularities for bounded quasiregular mappings
2K 2K+1 where α = 2K+1 and p = K+1 . Therefore, by [1, Th. 5.1.13] we get that for these values of α and p we actually have an estimate for a concrete Riesz capacity,
Cα,p(E) > 0
> > 3 However, K 1 implies p 2 so that in fact what we have is
2 0 < Cα,p(E) ≤ C 2 2 3 (E) ' γ 2 (E) . M K+1 (E) 3 (2− K+1 ), 2 K+1
In the expression above, the equivalence C 2 2 3 (E) ' γ 2 (E) comes from a result 3 (2− K+1 ), 2 K+1 by J. Mateu, L. Prat and J. Verdera [38]. Recall that for any β ∈ (0, 2], γβ is the signed Riesz capacity, associated to the signed Riesz kernels z . It is shown in [38] that if |z|1+β β ∈ (0, 1) then
γβ(E) ' C 2 3 (E) 3 (2−β), 2 β < ∞ 2 β < ∞ In particular, H (F) implies γβ(F) = 0. Notice that for β = K+1 , then if H (F) then F is removable for bounded K-quasiregular mappings. Furthermore, if β = 1 then we obtain the capacity γ1, and it follows from Tolsa’s work [51] that it behaves precisely as analytic capacity,
γ1(E) ' γ(E)
This encourages us to write the following problem.
Problem 2.22.
Let E be compact. Assume that γ 2 (E) = 0. Then, E is removable for bounded K-quasiregular K+1 mappings.
If this were true, then this would give interesting relations between the quasiconformal world and the potential theory of signed Riesz kernels.
60 3 Removable singularities for Holder¨ continuous quasiregular mappings
3.1 Introduction
Let α ∈ (0, 1). Recall that a function f : C → C is said to be locally Holder¨ continuous C with exponent α, that is, f ∈ Lipα( ), if
| f (z) − f (w)| ≤ C |z − w|α (3.1) whenever z, w ∈ C are such that |z − w| < 1.
We say that a compact set E is removable for α-Holder¨ continuous K-quasiregular mappings C C C if every Lipα function f : → , K-quasiregular on \ E, is actually K-quasiregular on the whole plane. In this chapter we study the geometric conditions that make a com- C pact set E ⊂ removable (or non removable) for Lipα K-quasiregular mappings.
Some results are already known related to this subject. For instance, Koskela and Martio 1 α [33] showed that if d = K 1 + K , then every compact set with vanishing d-dimensional
Hausdorff measure is removable for Lipα K-quasiregular mappings. Moreover, they also gave some sufficient conditions for removability in Rn. In particular, they proved that if Hd(E) = 0 for d = min{1, nα}, then E is removable for α-Holder¨ continuous K- quasiregular mappings. From another point of view, while dealing with second order quasilinear elliptic equations, Kilpelainen¨ and Zhong [32] showed that H α(n−1)(E) = 0 is enough. In the present chapter we improve the above results and get in Theorem 3.2 the following.
61 3 Removable singularities for Holder¨ continuous quasiregular mappings
Theorem. Let K ≥ 1 and α ∈ (0, 1), and let E be compact. If
Hd(E) = 0
2 with d = K+1 (1 + αK), then E is removable for α-Holder¨ continuous K-quasiregular mappings.
This index d was suggested to us by K. Astala.
Our techniques do not work in Rn, since there is no factorization theorem when n > 2. However, the above result may be rewritten in the planar class of finite distortion map- Ω C C 1,1 Ω pings. Recall that f : ⊂ → is said to be a mapping of finite distortion if f ∈ Wloc ( ), 1 Ω J(·, f ) ∈ Lloc( ) and |D f (z)|2 ≤ K(z) J(z, f ) at almost every z ∈ Ω. Here K : Ω → [1, ∞] is a measurable function, finite at almost every z ∈ Ω. Equivalently, |µ(z)| < 1 for almost every z ∈ Ω. If K ∈ L∞, we get the class of bounded distortion (that is, quasiregular) functions. During the last years, there has been deep progresses in this field. In particular, removability of singulari- ties for bounded mappings of finite distortion have been studied (see, for instance, [7], [18], [27] or [30]). In general, relatively weak assumptions on K are sufficient to ex- tend some basic quasiregular properties, such as continuity, discreteness and openness. Furthermore, in the plane we have existence of normalized solutions (as Theorem 1.11) and factorization theorems (as Theorem 1.12). Althoug the natural regularity for these 1,P C t2 mappings is the Orlicz-Sobolev space Wloc ( ), P(t) = log(e+t) (which is larger than the 1,2 C usual Wloc ( )), we can study the removability problem for Lipα mappings of finite dis- tortion, which appears as a limiting situation of the quasiregular case.
To look for results in the converse direction, we recall from [16] that any compact set E with H1+α(E) > 0 is nonremovable for holomorphic functions and, thus, neither for
K-quasiregular mappings in Lipα. Hence, our interest is centered on sets of dimension 1+αK between d = 2 1+K and 1 + α. We show in Corollary 3.6 that the index d is sharp in the following sense.
62 3.1 Introduction
Theorem. Given α ∈ (0, 1) and K ≥ 1, for any t > d there exists a compact set E of dimension t, and a C C function f ∈ Lipα( ) which is K-quasiregular in \ E, and with no K-quasiregular extension to C.
> 1+αK In other words, given any dimension t d = 2 1+K , we will construct a compact set E C C C of dimension t and a Lipα( ) function which is K-quasiregular on \ E but not on .
Our procedure starts with a compact set E of dimension t and a K-quasiconformal map- 0 2Kt 0 > ping φ such that dim(φ(E)) = t = 2+(K−1)t (notice that t 1). Then, we will show C > that there are Lipβ( ) functions, for some β 0, analytic outside of φ(E), which in turn induce (by composition) K-quasiregular functions on C \ E with some global Holder¨ continuity exponent. This construction will encounter two obstacles. First, the extremal 1+αK dimension distortion of sets of dimension 2 1+K under K-quasiconformal mappings is not exactly 1 + α, the critical number in the analytic setting (notice that this was so for α = 0, i.e. in the removability problem for BMO K-quasiregular mappings). Second, the composition of β-Holder¨ continuous functions with K-quasiconformal mappings is C only in Lipβ/K( ), so there is some loss of regularity that might be critical. To avoid these troubles, we will construct in an explicit way the mapping φ. This concrete con- struction allows us to show in Corollary 3.5 that our mapping φ exhibits an exponent of Holder¨ continuity given by t 1 K − 1 = + t t0 K 2K 1 which is larger than the usual K obtained by Mori (Theorem 1.13). This idea was sug- gested to us by X. Zhong. Observe that if dim(E) = t and dim(φ(E)) = t0 it is natural to expect φ to be Lipt/t0 . This regularity will be sufficient for our purposes.
This chapter is structured as follows. In Section 3.2 we prove Main Theorem 3.1. In Section 3.3, we construct nonremovable sets by means of extremal distortion mappings.
Finally, in Section 3.4 we study the Lipα removability problem for the class of finite distortion mappings.
63 3 Removable singularities for Holder¨ continuous quasiregular mappings
3.2 Sufficient conditions for Lipα removability
In this section, if µ is compactly supported, we mean by principal solution the unique µ-quasiconformal mapping φ normalized as in (1.10), that is, such that φ(0) = 0 and φ(z) − z ∈ C + W1,p(C). We start with an auxiliary compactness result.
Lemma 3.1.
Suppose that µn, µ are Beltrami coefficients, all of them compactly supported on D, and assume K−1 C C that kµnk∞, kµk∞ ≤ K+1 . Let φn, φ : → be, respectively, the principal solutions to the corresponding Beltrami equations. If µn → µ at almost every point, then:
p K (a) lim J(·, φn) = J(·, φ), with convergence in L , for every p ∈ (1, ). n→∞ loc K−1
(b) lim φn = φ uniformly on compact sets. n→∞
(c) lim φ−1 = φ−1 uniformly on compact sets. n→∞ n
Proof. Since µn, µ are compactly supported on D, we can proceed as in Example 1.10 2p to represent φ and φn as φn(z) = z + C˜hn(z), φ(z) = z + C˜h(z). Here hn, h ∈ L (C) −1 −1 are defined, respectively, by hn = (I − µnB) (µn) and h = (I − µB) (µ), C˜g(z) = Cg(z) − Cg(0) is the variant of the classical Cauchy transform defined in equation (1.3), K and p is any real number in (1, K−1 ). A computation shows that
(I − µnB)(h − hn) = (µ − µn) (1 + Bh)
Astala, Iwaniec and Martin [8, Theorem 1] showed that I − µnB is a bilipschitz isomor- 2p C K−1 phism from L ( ) to itself, whose bilipschitz constant depends only on kµnk ≤ K+1 and p. Therefore,
−1 kh − hnk2p ≤ k(I − µnB) k2p k(µ − µn) (1 + Bh)k2p ≤ CK k(µ − µn) (1 + Bh)k2p
2p Hence limn khn − hk2p = 0, because h ∈ L (C) and µn, µ have support uniformly 2p C bounded. It then follows that Dφn converges in Lloc( ) to Dφ, and hence J(·, φn) con-
64 3.2 Sufficient conditions for Lipα removability
p C verges in Lloc( ) to J(·, φ). Furthermore [2, p. 86], there exists a constant C p for which
1 1− p |C˜ f (z) − C˜ f (w)| ≤ Cp k f k2p |z − w| for any pair of points z, w ∈ C. We also have C˜ f (0) = 0 whenever f ∈ L2p(C). Thus,
1 1− p |φ(z) − φn(z)| = |C˜(h − hn)(z)| ≤ Cp kh − hnk2p |z|
−1 φn and φn → φ uniformly on compact sets. Finally, notice that the mappings ψn = −1 φn (1) form a normal familly, since they all are K-quasiconformal and fix 0, 1 and ∞. Therefore, 1 they are locally uniformly K -Holder¨ continuous. Thus, if φ(w) = z,
−1 −1 −1 −1 −1 −1 |φ (z) − φn (z)| = |φ φ(w) − φn φ(w)| = |φn φn(w) − φn φ(w)| 1 −1 −1 K = |φn (1)| |ψn (φn(w)) − ψn(φ(w))| ≤ CK |φn (1)| |φn(w) − φ(w)|
In particular, if here we take z = 1 then
−1 φ (1) 1 K −1 − 1 ≤ CK |φn(w) − φ(w)| → 0 φn (1)
−1 −1 −1 Since φn(0) = 0 for all n, necessarily limn φn (1) = φ (1). Therefore φn converges locally uniformly to φ−1.
Theorem 3.2. C 2 Let E ⊂ be a compact set, α ∈ (0, 1) and K ≥ 1. Set d = K+1 (1 + αK). If
Hd(E) = 0
then, E is removable for K-quasiregular mappings in Lipα.
C C C C Proof. Assume that f : → is a Lipα( ) mapping, K-quasiregular on \ E. We will see that f is K-quasiregular in the whole plane C. ∂ f Let µ = ∂ f be the Beltrami coefficient of f . Obviously, there is no restriction if we assume that E ⊂ D and that µ is supported on D. Let φ be the principal solution to ∂φ = µ ∂φ. Then the function F = f ◦ φ−1 is holomorphic on C \ φ(E) and Holder¨
65 3 Removable singularities for Holder¨ continuous quasiregular mappings continuous on C with exponent α/K. In particular, ∂F defines a distribution compactly 2 C supported on φ(E). Indeed, since F is continuous, it is an Lloc( ) function, so that its distributional derivatives define continuous linear functionals on W 1,2(C) functions. From Weyl’s lemma, it is enough to show that ∂F = 0.
For each ε > 0 we can consider a finite family of disks Dj = D(zj, rj) such that Ωε =
j Dj covers E and n d S ∑ diam(Dj) ≤ ε j=1
We define µε = µ χC\Ωε . If φε is the principal solution to ∂φε = µε ∂φε, then by Lemma −1 C 3.1 the functions Fε = f ◦ φε converge uniformly to F as ε → 0, because f ∈ Lipα( ). Thus, for any compactly supported C ∞ function ϕ,
h∂F, ϕi = h∂(F − Fε), ϕi + h∂Fε, ϕi = −h(F − Fε), ∂ϕi − h∂Fε, ϕi
Here the first term converges to 0 as ε → 0. Thus, we just care about the second term. ∞ Consider a partition of unity ψj subordinated to the covering Dj, that is, each ψj is a C C n function, compactly supported on 2Dj, |Dψj| ≤ , and ∑ ψj = 1 on Ωε. We diam(Dj) j=1 −1 1,2 C now define ϕj = ψj ◦ φε . Then, each ϕj is in W ( ), has compact support in φε(2Dj), n and ∑j=1 ϕj = 1 on φε(Ωε). For any constants cj we have
φε
E φε(E)
Dj φε(Dj)
ψj ϕj
n n n h∂Fε, ϕi = h∂Fε, ϕ ∑ ϕji = ∑h∂Fε, ϕ ϕji = ∑h∂(Fε − cj), ϕ ϕji j=1 j=1 j=1 n n (3.2) = − ∑h(Fε − cj), ∂ϕ ϕji − ∑h(Fε − cj), ϕ ∂ϕji = I + II j=1 j=1
66 3.2 Sufficient conditions for Lipα removability and now we have just to bound I and II independently. Notice that both terms may now be written as integrals. After a change of coordinates, for I we get
n n ∑h(Fε − cj), ∂ϕ ϕji ≤ ∑ |Fε(z) − cj| |∂ϕ(z)| |ϕj (z)| dA(z) φε(2D ) j=1 j=1 Z j
n ≤ kDϕk∞ ∑ | f (w) − cj| J(w, φε)dA(w). j=1 Z2Dj
K p C Now observe [4] that for any p ∈ (1, K−1 ), J(·, φε) ∈ Lloc( ). Using this, together with the fact that f ∈ Lipα, we obtain for I the bound
1/p n p α+2/q |I| ≤ C kDϕk∞ ∑ J(w, φε) dA(w) diam(2Dj) j=1 Z2Dj ! 1/p 1/q n n p qα+2 ≤ C kDϕk∞ ∑ J(w, φε) dA(w) ∑ diam(2Dj) j=1 Z2Dj ! j=1 !
p where q = p−1 . But both sums converge to 0 as ε tends to 0, so that we get I → 0. For the term II in (3.2),
n n ∑h(Fε − cj), ϕ ∂ϕji ≤ ∑ |Fε(z) − cj| |ϕ(z)| |∂ϕj(z)| dA(z) φε(2Dj) j=1 j=1 Z n
≤ kϕk∞ ∑ | f (w) − cj| |∂ϕj(φε(w))| J(w, φε )dA(w). j=1 Z2Dj
−1 Now, φε is also K-quasiconformal. Thus, by the chain rule,
−1 |Dψj| |Dϕj(φε)| ≤ |Dψj| |Dφε (φε)| ≤ 1 . J(·, φε) 2
Therefore, the Lipα condition for f and the decay of Dψj give
n 1 α−1 |II| . kϕk∞ ∑ J(w, φε) 2 dA(w) diam(2Dj) . j=1 Z2Dj !
67 3 Removable singularities for Holder¨ continuous quasiregular mappings
1 Now, we recall that J(·, φε) 2 defines a doubling measure, with constants depending only on K, so that
n 1 α−1 |II| . kϕk∞ ∑ J(w, φε) 2 dA(w) diam(Dj) . j=1 ZDj !
On the other hand, inside the union Ωε = ∪j Dj the mapping φε is conformal, so that K J(·, φε) attains the optimal integrability K−1 , as in Theorem 1.16, and
K J(w, φε) K−1 dA(w) ≤ 1 ZΩε Hence, as before, we use Holder¨ ’s inequality two times (one for the integral and one for the sum) to get
K−1 n 2K K α+ 1 |II| . kϕk∞ ∑ J(w, φε) K−1 dA(w) diam(Dj) K j=1 ZDj ! K−1 K+1 n 2K n 2K K 2(1+αK) ≤ kϕk∞ ∑ J(w, φε) K−1 dA(w) ∑ diam(Dj) K+1 . j=1 ZDj ! j=1 !
From the optimal integrability results, the integral above remains uniformly bounded, while the sum is bounded by ε from our choice of the covering.
We wish to emphasize that the improved integrability properties for K-quasiconformal mappings of Astala and Nesi [10] is precisely what allows us to jump from sets of dimension strictly smaller than d to sets of null d-dimensional Hausdorff measure. Note also that Theorem 3.2 may be stated in a more general sense, in terms of the modulus of continuity of f . Assume that f is K-quasiregular in C \ E and, instead of inequality (3.1), suppose that | f (z) − f (w)| ≤ ω(|z − w|)
C < for any z, w ∈ with |z − w| 1, where ω(r) = ω f (r) is any bounded continuous non-decreasing function ω : [0, 1) → [0, ∞) such that ω(0) = 0. If Hh(E) = 0 for
68 3.3 Extremal distortion and nonremovable sets
2 2K h(t) = t K+1 ω(2t) K+1 , then f is K-quasiregular on the whole plane. As particular cases, α C ω f (r) ω(r) = C r gives the previous theorem, while if f ∈ lipα( ), that is, if rα → 0 as d 1 −α > r → 0, then M∗(E) = 0 is enough. If ω(t) = log t , α 0, then the removability g result has a consequence in terms of quasiconformal distortion: H (φ(E)) = 0, where 2 1 −α K+1 g(t) = t log t . Our argument also shows that if M∗ (E) = 0 then E is removable for continuous K-quasiregular mappings.
The results obtained in Chapter 2 encourage us to make the following question.
Problem 3.3. 1+αK > d Let d = 2 1+K , K 1. Is it true that if E is a compact set, such that H (E) is σ-finite, then E is removable for α-Holder¨ continuous K-quasiregular mappings?
3.3 Extremal distortion and nonremovable sets
One of the main results in [4] is the sharpness of the dimension distortion equation (1.12). To obtain the equality there, the author moved holomorphically a fixed Cantor- type set F. This movement defined actually a holomorphic motion on F (see for instance [9] to have a quick introduction to holomorphic motions). An interesting extension result, known as the λ-lemma, allowed to extend this motion quasiconformally from F to the whole plane. This procedure avoided most of the technicalities and gave the desired result in a surprisingly direct way. However, since we look both for extremal dimension distortion and higher Holder¨ continuity, we need to construct φ explicitely. 0 2Kt Thus, let t ∈ (0, 2) and K ≥ 1 be fixed numbers, and denote t = 2+(K−1)t . As Astala did in [4], we first give a K-quasiconformal mapping φ that maps a regular Cantor set E of dimension t, to another regular Cantor set φ(E) for which dim(φ(E)) is as close as we want to t0. Later, again as in [4], we will glue a sequence of such mappings in the convenient way.
Proposition 3.4. Given t ∈ (0, 2), K ≥ 1 and ε > 0, there exists a compact E and a K-quasiconformal mapping φ : C → C, with the following properties:
69 3 Removable singularities for Holder¨ continuous quasiregular mappings
(a) φ is the identity mapping on C \ D.
(b) E is a self-similar Cantor set, constructed with m = m(ε) similarities.
(c) dim(φ(E)) ≥ t0 − ε.
p C K (d) J(·, φ) ∈ Lloc( ) if and only if p ≤ K−1 .
1 − 1 t (e) |φ(z) − φ(w)| ≤ C m t t0 |z − w| t0 whenever |z − w| < 1.
Proof. Following the scheme of Astala, Iwaniec, Koskela and martin in [7], we will ob- tain φ as a limit of a sequence of K-quasiconformal mappings
φ = lim φN N→∞ while E will be a regular Cantor set. The mapping φN will act only on the first N steps in the construction of E. Usually, radial stretchings like
1 −1 f (z) = z|z| K are extremal for some typical quasiconformal features. Thus, it is natural to construct the mappings φN using dilations and translations of f . However, we look for mappings C φ living in better Holder¨ spaces than Lip1/K( ) (this fails for f ), so that we will replace f by a linear mapping in a small neighbourhood of its singularity. This change will not affect the integrability index, and at the same time will give some improvement on the Holder¨ exponent.
D Take m ≥ 100, and consider m disjoint disks inside of , D(zi, r), uniformly distributed, all with the same radius r = r(m). By taking m big enough, we may always assume that 2 1 cm = mr ≥ 2 . Given any σ ∈ (0, 1) to be determined later, we can consider m simili- tudes D ϕi(z) = zi + σr z, z ∈
70 3.3 Extremal distortion and nonremovable sets and denote, for every i = 1, . . . , m,
1 D = ϕ (D) = D(z , r ), i σ i i 1 0 D Di = ϕi( ) = D(zi, σr1), where r1 = r. We define
1 K −1 0 σ (z − zi) + zi z ∈ Di 1 K −1 z−zi 0 g1(z) = (z − zi) + zi z ∈ Di \ D r1 i z otherwise. It may be easily seen that g1 defines a K-quasiconformal mapping, which is conformal 0 everywhere except on each ring Di \ Di. Moreover, if we put D D
0 Di 00 Di 1 Di g Di
Figure 3.1
1 K D ψi(z) = zi + σ r z, z ∈
0 00 D then g1 maps every Di to itself, while each Di is mapped to Di = ψi( ), as is shown in Figure 3.1. Now we denote φ1 = g1.
00 At the second step, we repeat the above procedure inside of every Di , and the rest
71 3 Removable singularities for Holder¨ continuous quasiregular mappings
remains fixed. That is, we define g2 on the target set of φ1, and then construct φ2 as
φ2 = g2 ◦ φ1
More explicitly, 1 D = φ ϕ (D) = D(z , r ) ij σ 1 ij ij 2 0 D Dij = φ1 ϕij( ) = D(zij, σr2)