Compact composition operators on Bergman-Orlicz spaces

We construct an analytic self-map $\phi$ of the unit disk and an Orlicz function $\Psi$ for which the composition operator of symbol $\phi$ is compact on the Hardy-Orlicz space $H^\Psi$, but not compact on the Bergman-Orlicz space ${\mathfrak B}^\Psi$. For that, we first prove a Carleson embedding theorem, and then characterize the compactness of composition operators on Bergman-Orlicz spaces, in terms of Carleson function (of order 2). We show that this Carleson function is equivalent to the Nevanlinna counting function of order 2.


Introduction
Due to the Littlewood subordination principle, the boundedness of composition operators C ϕ , defined by C ϕ (f ) = f • ϕ, on Hardy spaces H p , as well as on Bergman spaces B p , 1 ≤ p ≤ ∞, is automatic. Their compactness is something much more subtle, but is well understood now, and there are two well-separated cases. First, the case p = ∞, for which C ϕ : H ∞ → H ∞ is compact if and only if ϕ ∞ < 1 (note that B ∞ = H ∞ ). Secondly, the case p < ∞, for which the compactness does not depend on p. For Hardy spaces, this fact, proved by J. Shapiro and P. Taylor ( [16]), is not completely trivial, and is due to the good factorization properties of functions in H p . For the scale of Bergman spaces B p , the factorization properties are not so good, but the independence with respect to p follows from the following characterization ( [12], Corollary 4.4): for 1 ≤ p < ∞, C ϕ : B p → B p is compact if and only if the pull-back measure of the area-measure A by ϕ is a vanishing 2-Carleson measure. The case p = 2 (proved in [1]) gives then, for 1 ≤ p < ∞: In both cases (Hardy and Bergman), a brutal change of situation occurs when we pass from finite values of p to the value p = ∞, and the need was felt for an intermediate scale between H p and H ∞ , or between B p and B ∞ . This is what we did ( [7]), in full detail, with Hardy-Orlicz spaces H Ψ associated with an Orlicz function Ψ (and began to do for Bergman-Orlicz spaces B Ψ ). We introduced a generalization of the notion of Carleson measure, and proved a contractivity property of those Carleson measures m ϕ = ϕ * (m), attached to an analytic self-map ϕ : D → D, which turned out to be central to obtain a necessary and sufficient condition for the compactness of C ϕ : H Ψ → H Ψ . In that paper, we also began a study of the compactness of composition operators on B Ψ . We proved, in particular, but implicitly (see the comments at the beginning of Section 4), that, if the Orlicz function Ψ grows very fast (satisfying the so-called ∆ 2 condition), then the compactness of C ϕ : H Ψ → H Ψ implies its compactness as an operator C ϕ : B Ψ → B Ψ . On the other hand, it is well-known that the compactness on H p implies the compactness on B p because it is easy to see that the right-hand side of (1.1) is implied by the compactness on H p . One might think that it is generally easier to achieve compactness on B Ψ than on H Ψ . The main result of the present work is the existence of an analytic self-map ϕ of D and an Orlicz function Ψ such that the composition operator C ϕ is compact on H Ψ but not on B Ψ . For that, we first have to characterize the compactness of composition operators on Bergman-Orlicz spaces. More precisely, the paper is organized as follows.
In Section 2, given two Orlicz functions Ψ 1 and Ψ 2 , and a finite positive measure µ on the unit disk D, we investigate under which conditions the canonical inclusion I µ : B Ψ1 → L Ψ2 (µ), defined by I µ (f ) = f , is either bounded, or compact. In Theorem 2.1, we give a necessary condition and a sufficient condition, in terms of the Carleson function ρ µ of µ, for the boundedness of I µ . Analoguously, we have a similar statement (Theorem 2.5) for the compactness of I µ . In general, these necessary and those sufficient conditions do not fit.
In Section 3, we prove one of the main results of this paper (Theorem 3.1) under the form of a contractivity principle for the pull-back measure A ϕ of the planar Lebesgue measure A on D by ϕ. The proof is rather long and uses a Calderón-Zygmund decomposition, as well as an elementary, but very useful, inequality due to Paley and Zygmund. This contractivity principle eliminates the absence of fitness mentioned above and allows us to have a necessary and sufficient condition for the compactness of C ϕ : B Ψ → B Ψ in terms of the same Carleson function ρ Aϕ = ρ ϕ,2 (Theorem 3.2).
In Subsection 3.2, we consider the Nevanlinna counting function N ϕ,2 (initiated in [15]), adapted to the Bergman case, and we compare it with the 2-Carleson function ρ ϕ,2 of ϕ. These two functions turn out to be equivalent, in the sense precised in Theorem 3. 10. This extends to the Bergman case (and fol-lows from) such an equivalence for the Hardy case, that we recently established in [9], Theorem 1.1.
Finally, in Section 4, we exploit the necessary and sufficient conditions that we established, either on H Ψ and on B Ψ , to give (Theorem 4.2) an example of an analytic self-map ϕ : D → D and of a fairly irregularly varying Orlicz function Ψ such that, contrary to the general intuition, C ϕ : H Ψ → H Ψ is compact, whereas C ϕ : B Ψ → B Ψ is not compact. This is due to the fact that we can evaluate, in an accurate way, the two Carleson functions ρ ϕ and ρ ϕ,2 of ϕ.
Acknowledgement. The fourth-named author is partially supported by a Spanish research project MTM2006-05622.

Notation
We shall denote by D the open unit disk {z ∈ C ; |z| < 1} of the complex plane, and its boundary, the unit circle, by T. The normalized area measure dA = dx dy/π on D will be denoted by A.
For any finite positive measure µ on D, we define, for 0 < h ≤ 1, the Carleson function of µ by: and we set: , one says that µ is a 2-Carleson measure; we also say that µ is a Bergman-Carleson measure, to insist that the order 2 is adapted to the Bergman spaces. When µ = A ϕ is the pull-back measure of A by an analytic self-map ϕ : D → D, we shall simply write ρ ϕ,2 and K ϕ,2 instead of ρ Aϕ and K Aϕ,2 respectively. We shall say that ρ ϕ,2 the 2-dimensional Carleson function of ϕ.
The Hastings-Luecking sets of size 2 −n are defined by: An Orlicz function Ψ is a positive increasing convex function Ψ : [0, ∞) → [0, ∞) such that Ψ(0) = 0 (and Ψ(∞) = ∞). If µ is a positive measure on D, the Orlicz space L Ψ (µ) is the space of (classes of) measurable functions f : D → C such that D Ψ(|f |/C) dA < ∞, for some constant C > 0, and the norm f Ψ is defined as the infimum of all constants C > 0 for which D Ψ(|f |/C) dA ≤ 1. The Bergman-Orlicz space is the subspace of L Ψ (A) whose members are analytic in D.
The Hardy-Orlicz space H Ψ is the subspace of H 1 whose boundary values are in the Orlicz space L Ψ (T, m).
We refer to [3] (see also [5], and [17]) for the theory of Bergman spaces and to [14] for more information about Orlicz spaces.

Carleson embeddings
We consider in this Section the "embedding" map I µ : B Ψ1 → L Ψ2 (µ), defined by I µ (f ) = f , where µ is an arbitrary finite positive Borel measure on D and Ψ 1 and Ψ 2 are two Orlicz functions.

Boundedness
Theorem 2.1 Given µ a finite positive Borel measure on D and Ψ 1 and Ψ 2 two Orlicz functions, let I µ : B Ψ1 → L Ψ2 (µ) be the canonical map defined by I µ (f ) = f . One has: 1) If I µ is bounded, then there is a constant A > 0 such that: , for all 0 < h < 1.
2) In order that I µ is bounded, it suffices that there is a constant A > 0 such that: , for all 0 < h < 1.
Note that condition (2.1) reads as is bounded (by 1/A) and is bounded.
When Ψ 1 = Ψ 2 = Ψ and the Orlicz function Ψ satisfies the usual condition ∆ 2 : Ψ(2x) ≤ C Ψ(x) for some constant C > 1 and x large enough, it is clear that conditions (2.1) and (2.2) are equivalent. However, they are not equivalent in general; and even condition (2.1) is not sufficient and condition (2.2) is not necessary: the examples 1.b and 2. of [7], Chapter 4, §3, given in the Hardy case, work also for the Bergman case. For the sake of completeness, we are going to sketch them. Example 1. For every Orlicz function which does not satisfy the ∆ 2 condition, there exists a finite positive measure µ on D such that I µ : B Ψ → L Ψ (µ) is continuous, though µ is not a 2-Carleson measure, and a fortiori does not verify (2.2). Proof. Since Ψ does not satisfy ∆ 2 , there exists an increasing sequence (a n ) n≥1 such that Ψ(2a n )/n is increasing and Ψ(2a n )/Ψ(a n ) ≥ n2 n . Define the discrete measure µ: On the other hand, for every f in the unit ball of B Ψ , one has there exists a finite positive measure µ on D such that (2.1) holds but I µ : Proof. Let ν be a probability measure on T, supported by a compact set L of Lebesgue measure zero, such that ν(I) ≤ |I| 1/2 , for each interval I. We can associate to ν the measure on D defined byν(E) = ν(E ∩ T). By Rudin-Carleson's Theorem, for every integer n, there exists a function g n in the unit ball of the disk algebra such that |g n | = 1 on L and g n H Ψ ≤ 4 −n . As L is compact, there exists some r n ∈ (1/2, 1) such that |g n (r n z)| ≥ 1/2 for every z ∈ L. Now, define the measure µ by: where: ν n (E) = ν {z ∈ T ; r n z ∈ E} .
If W is a Carleson window of size h then, for each n ≥ 1, we have: , and the condition (2.1) is fulfilled. Nevertheless, the identity from B Ψ to L 1 (µ) is not continuous since this would imply that the identity from H Ψ to L 1 (µ) were continuous as well, which is not the case: g n L 1 (µ) ≥ 1/2 n+1 .
In order to prove Theorem 2.1, we shall need some results. They are analogous to Proposition 4.9, Theorem 4.13 and Lemma 4.14 of [7], but their proofs require different arguments 1 .
1 By the way, we seize the opportunity to correct here the proof of Theorem 4.13 given in [7], where some argument had been put awkwardly. In that proof, we first had to set M = {z ∈ D ; |z| > 1 − h and |f (z)| > t}. Then, M f being the non-tangential maximal function of f ∈ H 1 , the open set {M f > t} is the disjoint union of a countable family of open arcs I j ⊆ T, and we had to say that every z such that |f (z)| > t belongs to some window W (I j ) (see [2], page 39).
We first introduce the following maximal function: One has: But, for every Carleson window W , we can write: and, since ∆ k ∩W =∅ ∆ k is contained in the windowW with the same center as W , but with size two times that of W , one has µ(W ) ≤ A(W ) = 4 A(W ). Hence µ is a Bergman-Carleson measure. By [4], it follows that, for some constant C 0 > 0, which does not depend on f , one has, using the subharmonicity of Ψ(|f |/C): We shall, as we may, assume that C 0 ≥ 1. Now, by convexity of Ψ, we get: 3 For every f ∈ B 1 and every finite positive Borel measure µ on D, one has, for 0 < h < 1/2 and t > 0: Proof. Remark that when z ∈ ∆ k and |z| > 1 − h, we must have If W k is the smallest Carleson window containing ∆ k , we have: and Lemma 2.3 is proved.
Lemma 2.4 Let µ be a finite Borel measure on D and Ψ 1 and Ψ 2 two Orlicz functions.
We suppose that, for some positive constant A, there is 0 Then, for every f ∈ B Ψ1 such that f B Ψ 1 ≤ 1 and every Borel subset E of D, one has, with Proof. For every s > 0, the inequality |f (z)| > s implies that the norm of the evaluation δ z at z is greater than s. But this norm is ( [7], Lemma 5.2) Ψ 1 1/(1 − |z|) 2 , up to constants; more precisely: Hence, we have: (by convexity) We get therefore: But, as Ψ 1 and Ψ 2 are Orlicz functions, one has tΨ ′ 2 (t) ≤ Ψ 2 (2t) and and the proof of Lemma 2.4 is finished.
2) By Lemma 2.2, there is a constant C > 0, that we may, and shall do, which means that g L Ψ 2 (µ) ≤ 64CC/A.

Compactness
Theorem 2.5 Let µ be a finite positive Borel measure on D, Ψ 1 and Ψ 2 two Orlicz functions, and let I µ : B Ψ1 → L Ψ2 (µ) be the canonical map defined by I µ (f ) = f . One has: 1) If I µ is compact, then: 2) In order that I µ is compact, it suffices that As for the boundedness case, conditions (2.6) and (2.7) are equivalent if Ψ 1 = Ψ 2 = Ψ is sufficiently regular, but not in general. We shall give examples after the proof of the theorem, at the end of the section.
To prove the first part of this theorem, we shall need the following lemma.
is compact if and only if for every bounded sequence (f n ) n in B Ψ1 converging to 0 uniformly on compact subsets of D, the sequence I µ (f n ) n converges to 0 in the norm of L Ψ2 (µ).
Proof of Lemma 2.6. Assume that I µ is compact, and let (f n ) n be a bounded sequence in B Ψ1 which converges to 0 uniformly on compact subsets of D. Since I µ is compact and (f n ) n is bounded, we have a subsequence (g n ) n such that g n = I µ (g n ) converges to some g ∈ L Ψ2 (µ). Then some subsequence of (g n ) n converges µ-a.e. to g. Since (g n ) n converges to 0 uniformly on compact subsets of D, we must have g = 0 µ-a.e., that is g = 0 as an element of L Ψ2 (µ). Now, the compactness of I µ implies that f n L Ψ 2 (µ) tends to 0. Conversely, assume that the condition of the lemma is satisfied, and let (f n ) n be an arbitrary bounded sequence in B Ψ1 . Since the evaluation map is continuous on B Ψ1 ([7], Lemma 5.2), (f n ) n is a normal family, and Montel's Theorem gives a subsequence (g n ) n which converges uniformly on compact subsets, to some holomorphic function g. By Fatou's lemma, g belongs to B Ψ1 . Now, (g n − g) n is a bounded sequence of B Ψ1 which converges to 0 uniformly on compact subsets of D. By hypothesis, g n − g L Ψ 2 (µ) tends to 0, and it follows that I µ is compact.
Proof of Theorem 2.5. 1) Assume that the map I µ : B Ψ1 → L Ψ2 (µ) is compact. Consider, for every a ∈ D, the Berezin kernel (2.4). It follows from (2.4) and (2.5) that H a / H a B Ψ 1 converges uniformly to 0 on compact subsets of D as |a| goes to 1; hence, by Lemma 2.6, the compactness of I µ implies that (H a / H a B Ψ 1 ) L Ψ 2 (µ) tends to 0. That means that, for every ε > 0, one has: for |a| close enough to 1, depending on ε. Now, let ξ ∈ T and 0 < h < 1 with a = (1 − h)ξ. As already said in the proof of Theorem 2.1, z ∈ W (ξ, h) implies that |H a (z)| ≥ 1/625h 2 . Therefore: We get, for h > 0 small enough: Since ξ ∈ T is arbitrary, it follows that, for h > 0 small enough: , which reads: , we have obtained (2.6). 2) Assume now that (2.7) is satisfied. By Lemma 2.6, we have to show that for every sequence (f n ) n in the unit ball of B Ψ1 which converges uniformly to 0 on compacts subsets of D, I µ (f n ) n converges to 0 for the norm of L Ψ2 (µ). So, let (f n ) n be such a sequence, and let ε > 0. By For 0 < r < 1, we may therefore apply, for every n ≥ 1, Lemma 2.4 to f n /C (which is in the unit ball of B Ψ1 ), with E = D \ rD; we get: But this last quantity is ≤ 1/2 for r small enough. Fix such an r < 1. Since (f n ) n converges uniformly to 0 on compacts subsets of D, one has rD Ψ 2 (|f n |/ε) dµ ≤ 1/2 for n large enough.
That ends the proof of Theorem 2.5.
As we said, in general, condition (2.6) is not sufficient to ensure compactness, and condition (2.7) is not necessary. They are equivalent (and hence necessary and sufficient for compactness) if Ψ 1 = Ψ 2 = Ψ and Ψ is a regular Orlicz function. Here, regular means that Ψ satisfies the condition we called ∇ 0 : for some x 0 > 0 and some [7], Theorem 4.11, whose proof works as well in the Bergman case). However, we gave in [7], examples showing, in the Hardy case, that this is not always the case (examples 3 and 4 in [7], Chapter 4, §3). These examples work in the Bergman case and we are going to recall them sketchily.
Example 1. For every Orlicz function Ψ not satisfying ∇ 0 , there exists a measure µ such that I µ : B Ψ → L Ψ (µ) is compact but for which (2.7) is not satisfied.
This actually shows that condition Ψ ∈ ∇ 0 is necessary and sufficient in order to ensure that the identity from B Ψ to L Ψ (µ) is compact if and only if µ satisfies (2.7).

Example 2.
There exist an Orlicz function Ψ and a measure µ on D such that (2.6) holds, but for which I µ : B Ψ → L Ψ (µ) is not compact.

Carleson function
We know ( [7], Proposition 5.4), that every analytic self-map ϕ : D → D induces a bounded composition operator C ϕ : B Ψ → B Ψ . The main result of this section is that, for the pull-back measure A ϕ of A by ϕ, the necessary and the sufficient conditions of Theorem 2.5 are equivalent. The same kind of result occurs for Hardy-Orlicz spaces ( [7], Theorem 4.18 and Theorem 4.19), but the proofs must be different (because we use the analytic functions themselves, and not their boundary values).
We have the following contractivity (or homogeneity) result, which can be viewed as a "multi-scaled" fact that A ϕ is a 2-Carleson measure.

Some lemmas
Lemma 3.4 There is some constant C 1 > 0 such that for every analytic function f : D → Π + and for every λ > 0.
In particular, there is a constant K 1 > 0 such that f L 1 (D) ≤ K 1 |f (0)| for every such a function.
The second assertion follows from the first one: To prove the first assertion, remark first that the left-hand side of inequality (3.3) is ≤ 1, so (3.3) is obvious for λ ≤ 2 (with C 1 ≥ 4). Assume that λ > 2, and set ϕ( . Then |f (z)| > λ implies |ϕ(z) − 1| = 2 |Re f (0)|/|f (z) + f (0)| ≤ 2/(λ − 1) ≤ 4/λ. But ϕ maps D into itself and hence induces a bounded composition operator C ϕ : B 2 → B 2 . It follows from [4] that A ϕ is a Bergman-Carleson measure, and hence (see the proof of Theorem 2.1, 1), with Ψ 1 (x) = Ψ 2 (x) = x 2 ): , Theorem 11.6, page 308), and the result follows. Let: By applying Lemma 3.4 to f 2 , we get: Lemma 3.5 There exists a constant C 2 > 0 such that for every analytic function f : D → G, one has, for every λ > 0 In particular there is a constant K 2 > 0 such that f L 2 (D) ≤ K 2 |f (0)| for every such a function. Now, we shall have to replace D by some conformal copies of D. This will be possible by the following version of Lemma 3.5.

Lemma 3.6
Let Ω be a bounded Jordan domain bounded by a C 1 Jordan curve J and let h : D → Ω be a Riemann map which extends to a bi-Lipschitz homeomorphism h of D to Ω such that: Then, there exists a constant C > 0, depending only on b/a, such that, for any λ > 0 and any analytic function f : Ω → G, one has: where c = h(0). In particular, there is a constant K = K(Ω, c) > 0 such that f 2 ≤ K A(Ω) |f (c)| for all such functions.
Moreover C > 0 can be taken so that, for every positive harmonic function u : Ω → R + , one has: Proof. By the change-of-variable formula, we have, setting F = f • h and w = h(z): We implicitly used the fact that F maps D to G and that |F (0)| = |f (c)|, so we were allowed to use the previous lemma. Moreover, we have so that finally: This ends the proof of the first part of Lemma 3.6 for Ω, with C = C 2 b 2 /a 2 .
Finally, one has: which gives the right-hand side of (3.7), since A(Ω) ≥ a 2 . The left-hand side is proved in the same way, since A(Ω) ≤ b 2 .
The last lemma is of a different kind.

Lemma 3.7
For every analytic function f : D(z 0 , r) → G, one has: where K 2 is the constant given by Lemma 3.5.
Proof. Recall Paley-Zygmund's inequality (see [11], Proposition III.3), usually stated in probabilistic language: for any positive random variable X on some probability space, one has, for 0 < a < 1: where E stands for expectation. For our problem, we will take as probability space the disk D(z 0 , r) equipped with the probability measure dA/r 2 , and as a random variable X = Re f := u. We get from (3.8) that, since u(z 0 ) = E(X) by the mean value property of harmonic functions: We get: which gives the desired result by taking a = 1/2.

Proof of Theorem 3.3.
For technical reasons, we are going to work with functions with range in the set G. Proving Theorem 3.3 amounts to prove: Proposition 3.8 There exist constants K ′ > 0, α 1 > 0 and λ 1 > 1 such that every analytic function f : D → G with |f (0)| ≤ α 1 satisfies, for λ ≥ λ 1 : It will be useful to note that √ 2 Re w ≥ |w| when w ∈ G.
Proof. The idea of the proof is to split F λ = {|f | > λ} (actually, F λ will be conditionned by the more regular set {M d f > 1}) in parts on which we shall be able to apply Lemma 3.6. In order to construct these parts, we shall use a Calderón-Zygmund type decomposition adapted to the geometry of the unit disk. We are going to recall the principle of this decomposition for the convenience of the reader. Before that, we have to say that we shall begin to work not with the function f as is the statement of Proposition 3.8, but with this function multiplied by a constant (as we shall specify at the end of the proof). Nevertheless, we shall denote this new function by f yet. Now, remark that for f : D → G with |f (0)| ≤ α < 1, one has: for β = (1−α 2 )/(1+α 2 ). In fact, setting a = f (0), the function g = f 2 −a 2 f 2 +a 2 maps D into itself and vanishes at 0. By Schwarz's lemma, |z| ≤ β implies |g(z)| ≤ β, and since f 2 = a 2 g+a 2 1−g , we get: The goal of the remark (3.9) is that, in order to make the Calderón-Zygmund decomposition, we have to avoid the center 0 of D, which is not covered by the sets S ∈ S defined below.
In order to perform the Calderón-Zygmund decomposition, we have to make a splitting of the annulus Γ. For that, consider the rectangle R 0 = {x + iy ; log(1/2) ≤ x < 0 and 0 ≤ y < 2π} and the family R = n∈N R n of all dyadic half-open rectangles, where R n is the family of sets R = {x + iy ∈ C ; x j ≤ x < x j+1 and y k ≤ y < y k+1 } , where x j = (1 − j2 −n ) log(1/2) and y k = 2kπ/2 n , 0 ≤ j, k ≤ 2 n − 1, n ≥ 0.
We now maps the annulus Γ onto the unit disk D, using the exponential map x + iy → z = e x+iy . We get a family S = n∈N S n .
Note that the jacobian e 2x of the transformation is between β 2 = 1/4 and 1; hence the area of a set in S n−1 is less than 16 times that of a set in S n .
Recall now the Calderón-Zygmund decomposition. Consider the conditional expectation E n |f | of |f |, given the σ-algebra generated by S n , namely: The dyadic maximal function M d f is defined by: Note that: where the supremum is taken over all S z ∈ S containing z.
By Lebesgue's differentiation Theorem (or by the martingale convergence Theorem), we know that (E n |f |)(z) converges to |f (z)|, A-almost everywhere, so M d f (z) ≤ 1 implies |f (z)| ≤ 1 for almost all z ∈ D, and we can write: Since E n |f | is constant on the sets S ∈ S n , each E n can be in its turn decomposed into a disjoint union E n = k S n,k , where S n,k ∈ S n . By definition, for z ∈ E n , one has (E n |f |)(z) ≥ 1 and hence But, on the other hand, (E n−1 |f |)(z) ≤ 1, and we have, if z ∈ S n,k : where S n−1,j is the set of rank (n − 1) containing S n,k . Finally, reindexing the sets S k,n , we can write {M d f > 1} as a disjoint union for which: Equations (3.10), (3.11) and (3.12) define the Calderón-Zygmund decomposition of the function f .
In order to apply Lemma 3.6, we have to control (from above and from below ) the values of |f | at the "center" of the sets S l . One might think of doing this by using (3.12), but it is not always possible (for example, the function with positive real part 1/z 2 is not integrable on the square of vertices 0, 1 + i, 1 − i, 2). Nevertheless, since our function is holomorphic on a domain bigger than S l , we may enlarge S l in order to use (3.12).
Let R l ∈ R be the rectangle which is mapped onto S l by the exponential, and let us round R l inR l , by adding half-disks, as indicated in Figure 1. Let S l = exp(R l ). It is essential, when using Lemma 3.6 for Ω =Ŝ l , that the constant C given by this lemma does not depend on l. This will be checked in the following way. The setsR l can be performed by making similarities fromR 0 , where R 0 = {x + iy ; log(1/2) ≤ x < 0 and 0 ≤ y < 2π} .
Therefore, one has, for every l ≥ 1, if γ l is the center (defined in an obvious way) of S l (and hence ofŜ l ): Now, let D l be the greatest open disk with center γ l and contained in S l (see Figure 3). We have, by the last part of Lemma 3.6: Using the fact that A(Ŝ l ) ≤ 4 A(S l ) and that F λ ⊆ {M d f > 1} ∪ N when λ ≥ 1, we get:

It remains to control
For that, we shall use Lemma 3.7. By harmonicity and Lemma 3.6, one has, with u = Re f : We now apply Lemma 3.7 and we get: We obtain hence: The proof of Theorem 3.3 is now finished, because (3.15) and (3.16) give, for λ ≥ 1: and if f is as in the statement of Proposition 3.8, we can apply (3.17) to f 1 = 64C f and we get, for λ ≥ λ 1 = 64C: when |f (0)| ≤ 1/64C.

Nevanlinna counting function
The Nevanlinna counting function is defined, for every analytic function ϕ : D → D, and for every w ∈ ϕ(D) \ {ϕ(0)}, by: where each term log 1 |z| is repeated according to the multiplicity of z, and by N ϕ (w) = 0 for the other w ∈ D.
Recall (see [9]) that, if m is the normalized Lebesgue measure on T, then the Carleson function of ϕ is the Carleson function of the pull-back measure m ϕ of m by ϕ. We proved in [9] (Theorem 3.1 and Theorem 3.7) that the behaviour of N ϕ is equivalent to that of the Carleson function ρ ϕ in the following way. Theorem 3.9 ([9]) There exist two universal constants C, c > 1, such that, for every analytic self-map ϕ : D → D, one has: for 0 < h < 1 small enough.
We are going to deduce from Theorem 3.9 the same result in the 2-dimensional case. The Nevanlinna counting function of order 2 is defined (see [15], § 6.2), for w ∈ D \ {ϕ(0)}, by: where each preimage z of w appears as often as its multiplicity. The partial Nevanlinna counting function is defined, for 0 < r ≤ 1 by: and we have ( [15], Proposition 6.6, where a misprint occurs): One has: Theorem 3.10 There exists a universal constant C > 1, such that, for every analytic self-map ϕ : D → D, one has: for 0 < h < 1 small enough.
Remark. Actually, the proof shows that for some constant C > 1, one has: for every ξ ∈ T and 0 < h < 1 small enough. Since the ℓ 2 -norm is less than the ℓ 1 -norm, one has N ϕ,2 (w) ≤ [N ϕ (w)] 2 , it follows hence from Theorem 3.9 and Theorem 3.10 (actually (3.25)) that, for some constant C > 1, one has: for every ξ ∈ T and every 0 < h < 1 small enough, a fact which does not seem easy to proved in a straightforward way. In particular, for some constant C > 0, one has, for h > 0 small enough: Corollary 3.11 For every analytic self-map ϕ : D → D and for every Orlicz function Ψ, the composition operator C ϕ : B Ψ → B Ψ is compact if and only if: Proof. If C ϕ is compact, Theorem 3.2 gives, for every A > 0, an h A > 0 such that, for 0 < h ≤ h A : Then Theorem 3.10 gives: i.e. A Ψ −1 (1/C 2 h 2 ) ≤ Ψ −1 [C/ν ϕ,2 (h)]; and then, by concavity, since C > 1: which implies (3.27). The converse follows the same lines.
Corollary 3.12 The compactness of the composition operator C ϕ : B Ψ → B Ψ implies that: This corollary was proved in [7], Theorem 5.7, by a more direct method; and we also showed that the condition is sufficient when Ψ grows fast enough (namely, satisfies the condition ∆ 2 ). However, we do not know whether it is not sufficient for some symbol ϕ and some Orlicz function Ψ (see the remark at the end of the paper). Nevertheless, it follows easily from Corollary 3.11 that this condition is sufficient when ϕ is finitely-valent (see [9], proof of Theorem 5.3).

Comparison of the compactness of composition operators on Hardy-Orlicz spaces and on Bergman-Orlicz spaces
In the classical case (Ψ(x) = x p , 1 ≤ p < ∞), it is known ( [12], Theorem 3.5, with Proposition 2.7) that the compactness of C ϕ : H p → H p implies the compactness of C ϕ : B p → B p . On the other hand, we implicitly proved in [7], Theorem 5.7, that when Ψ grows very fast (namely, satisfies the so-called ∆ 2 condition), then the compactness of C ϕ : H Ψ → H Ψ implies the compactness of C ϕ : B Ψ → B Ψ . Let us write why: it is easy to show (see [9], proof of Theorem 4.3) that the compactness of C ϕ : H Ψ → H Ψ implies that ] tends to 0 as |z| goes to 1, and we actually proved in [7], Theorem 5.7, that, when Ψ ∈ ∆ 2 , this last condition implies the compactness of C ϕ : B Ψ → B Ψ . The next proposition gives a condition on Ψ which, though not very satisfactory, includes the cases Ψ(x) = x p and Ψ ∈ ∆ 2 and for which compactness on H Ψ implies compactness on B Ψ . Proposition 4.1 Assume that the Orlicz function Ψ satisfies the following condition: for every A > 0, there exist x A > 0 and B ≥ A such that: for every x ≥ x A . Then every analytic map ϕ : D → D defining a compact composition operator C ϕ : H Ψ → H Ψ also defines a compact composition operator It is clear that Ψ satisfies (4.1) if Ψ(x) = x p . Recall that condition ∆ 2 means that there exists some α > 1 such that [Ψ(t)] 2 ≤ Ψ(αt) for t large enough. For such a function, one has Ψ −1 ( 2 for x large enough, since it tends to infinity. Therefore, (4.1) holds if Ψ has ∆ 2 . Classical examples of Orlicz functions with ∆ 2 are Ψ(x) = e x q − 1, where q ≥ 1.
By (4.1), we get: Corollary 3.11 ensures that C ϕ is compact on B Ψ .
However, we are going to see that the conclusion of Proposition 4.1 does not hold for an arbitrary Orlicz function, by proving the following theorem.
Theorem 4.2 There exist an analytic self-map ϕ : D → D and an Orlicz function Ψ such that C ϕ : In order to prove it, we shall show and use the following result. Theorem 4.3 There exists an analytic self-map ϕ : D → D such that, for some constants c 2 ≥ π and π/4 ≥ c 1 > 0, one has, for some constant C > 0 and for h > 0 small enough: and (by Theorem 3.2) that the compactness of C ϕ : B Ψ → B Ψ is equivalent to Hence, it suffices to construct an Orlicz function Ψ such that: We shall actually construct an increasing concave function f : [0, ∞) → R + such that f (0) = 0 and f (∞) = ∞ which satisfies (4.6) and (4.7) (with f instead of Ψ −1 ) and we shall then take Ψ = f −1 .
Putting the two cases together, we get that lim x→+∞ f (x) f (e c1x ) = 0, so (4.6) is satisfied, and Theorem 4.2 is fully proved.
Proof of Theorem 4.3. The analytic map ϕ will be a conformal mapping from D to the domain G, edged by three circular arcs of radii 1/2, and which is represented in Figure 4. More precisely, let G 0 = D∩{Re z > 0} and let f : D → G 0 be the conformal map such that f (−1) = 0 ; f (1) = 1 ; f (i) = i ; f (−i) = −i .
2) In the opposite direction of Theorem 4.2, it would be interesting to have, for every Orlicz function Ψ, a composition operator which is compact on B Ψ but not compact on H Ψ . This is the case for Ψ(x) = e x 2 − 1 ([7], Theorem 5.8). Theorem 3.1 of [10] could give such an example. It could also give an example where the condition of 3.12 is not sufficient to have compactness.