Sep 24, 2018 - (the substitution of n by n1/N is mandatory as shown by the results of [4]). We show in this ...... http://gamma.im.uj.edu.pl/ blocki/p...

0 downloads 13 Views 361KB Size

Pluricapacity and approximation numbers of composition operators Daniel Li, Hervé Queffélec, L. Rodríguez-Piazza September 25, 2018

Abstract. For suitable bounded hyperconvex sets Ω in CN , in particular the ball or the polydisk, we give estimates for the approximation numbers of composition operators Cϕ : H 2 (Ω) → H 2 (Ω) when ϕ(Ω) is relatively compact in Ω, involving the Monge-Ampère capacity of ϕ(Ω). Key-words: approximation numbers ; composition operator; Hardy space ; hyperconvex domain ; Monge-Ampère capacity ; pluricapacity ; pluripotential theory ; Zakharyuta conjecture MSC 2010 numbers – Primary: 47B33 – Secondary: 30H10 – 31B15 – 32A35 – 32U20 – 41A25 – 41A35 – 46B28 – 46E20 – 47B06

1

Introduction

Let D be the unit disk in C, H 2 (D) the corresponding Hardy space, ϕ a non-constant analytic self-map of D and Cϕ : H 2 (D) → H 2 (D) the associated composition operator. In [40], we proved a formula connecting the approximation numbers an (Cϕ ) of Cϕ , and the Green capacity of the image ϕ(D) in D, namely, when [ϕ(D)] ⊂ D, we have: (1.1) β(Cϕ ) := lim [an (Cϕ )]1/n = exp − 1/Cap [ϕ(D)] , n→∞

where Cap [ϕ(D)] is the Green capacity of ϕ(D). A non-trivial consequence of that formula was the following: (1.2)

kϕk∞ = 1

=⇒

an (Cϕ ) ≥ δ e−nεn where εn → 0+ .

In other terms, as soon as kϕk∞ = 1, we cannot hope better for the numbers an (Cϕ ) than a subexponential decay, however slowly εn tends to 0. In [41], we pursued that line of investigation in dimension N ≥ 2, namely on H 2 (DN ), and showed that in some cases the implication (1.2) still holds ([41, Theorem 3.1]): (1.3)

kϕk∞ = 1

=⇒

1/N

an (Cϕ ) ≥ δ e−n 1

εn

where εn → 0+ ,

(the substitution of n by n1/N is mandatory as shown by the results of [4]). We show in this paper that, in general, for non-degenerate symbols, we have similar formulas to (1.1) at our disposal for the parameters: − (1.4) βN (Cϕ ) = lim inf [anN (Cϕ )]1/n n→∞

+ and βN (Cϕ ) = lim sup[anN (Cϕ )]1/n . n→∞

These bounds are given in terms of the Monge-Ampère (or Bedford-Taylor) capacity of ϕ(DN ) in DN , a notion which is the natural multidimensional extension of the Green capacity when the dimension N is ≥ 2 ([41, Theorem 6.4]). We − + show that we have βN (Cϕ ) = βN (Cϕ ) for well-behaved symbols.

2 2.1

Notations and background Complex analysis

Let Ω be a domain in CN ; a function u : Ω → R ∪ {−∞} is said plurisubharmonic (psh) if it is u.s.c. and if for every complex line L = {a + zw ; z ∈ C} (a ∈ Ω, w ∈ CN ), the function z 7→ u(a + zw) is subharmonic in Ω ∩ L. We denote PSH(Ω) the set of plurisubharmonic functions in Ω. If f : Ω → C is holomorphic, then log |f | and |f |α , α > 0, are psh. Every real-valued convex function is psh (convex functions are those whose composition with all R-linear isomorphisms are subharmonic, though plurisubharmonic functions are those whose composition with all C-linear isomorphisms are subharmonic: see [30, Theorem 2.9.12]). ¯ and (ddc )N = ddc ∧· · ·∧ddc (N times). When u ∈ PSH(Ω)∩ Let ddc = 2i∂ ∂, 2 C (Ω), we have: 2 ∂ u (ddc u)N = 4N N ! det dλ2N (z) , ∂zj ∂ z¯k where dλ2N (z) = (i/2)N dz1 ∧d¯ z1 ∧· · ·∧dzN ∧d¯ zN is the usual volume in CN . In c N general, the current (dd u) can be deﬁned for all locally bounded u ∈ PSH(Ω) and is actually a positive measure on Ω ([5]). Given p1 , . . . , pJ ∈ Ω, the pluricomplex Green function with poles p1 , . . . , pJ and weights c1 , . . . , cN > 0 is deﬁned as: g(z) = g(z, p1 , . . . , pJ ) = sup{v(z) ; v ∈ PSH(Ω) , v ≤ 0 and v(z) ≤ cj log |z − pj | + O (1) , ∀j = 1, . . . , J} . In particular, for J = 1 and p1 = a, c1 = 1, g( · , a) is the pluricomplex Green function of Ω with pole a ∈ Ω. If 0 ∈ Ω and a = 0, we denote it by gΩ and call it the pluricomplex Green function of Ω; hence: ga (z) = g(z, a) = sup{u(z) ; u ∈ PSH(Ω) , u ≤ 0 and u(z) ≤ log |z−a|+O (1)} . 2

Let Ω be an open subset of CN . A continuous function ρ : Ω → R is an exhaustion function if there exists a ∈ (−∞, +∞] such that ρ(z) < a for all z ∈ Ω, and the set Ωc = {z ∈ Ω ; ρ(z) < c} is relatively compact in Ω for every c < a. A domain Ω in CN is said hyperconvex if there exists a continuous psh exhaustion function ρ : Ω → (−∞, 0) (see [30, p. 80]). We may of course replace the upper bound 0 by any other real number. Without this upper bound, Ω is said pseudoconvex. Let Ω be a hyperconvex domain, with negative continuous psh exhaustion function ρ and µρ,r the associated Demailly-Monge-Ampère measures, deﬁned as: (2.1)

µu,r = (ddc ur )N − 1Ω\BΩ,u (r) (ddc u)N ,

for r < 0, where ur = max(u, r) and: BΩ,u (r) = {z ∈ Ω ; u(z) < r} . The nonnegative measure µu,r is supported by SΩ,u (r) := {z ∈ Ω ; u(z) = r}. If: Z (ddc ρ)N < ∞ , Ω

these measures, considered as measures on Ω, weak-∗ converge, as r Rgoes to 0, to a positive measure µ = µΩ,ρ supported by ∂Ω and with total mass Ω (ddc ρ)N ([16, Théorème 3.1], or [30, Lemma 6.5.10]). For the pluricomplex Green function ga with pole a, we have (ddc ga )N = (2π)N δa ([16, Théorème 4.3]) and ga (a) = −∞, so a ∈ BΩ,ga (r) for every r < 0 and 1Ω\BΩ,ga (r) (ddc ga )N = 0. Hence the Demailly-Monge-Ampère mesure µga ,r N N = is equal to ddc (ga )r . By [51, Lemma 1], we have (1/|r|) ddc (ga )r ¯Ω,ga (r) = {z ∈ Ω ; ga (z) ≤ r} uB¯Ω,ga (r),Ω , the relative extremal function of B in Ω (see (3.2) for the deﬁnition), and this measure is supported, not only by ¯Ω,ga (r) (see Section 2.2.1 for SΩ,ga (r), but merely by the Shilov boundary of B the deﬁnition). Since (ddc ga )N = (2π)N δa has mass (2π)N < ∞, these measures weak-∗ converge, as r goes to 0, to a positive measure µ = µΩ,ga supported by ∂Ω 1 with mass (2π)N . Demailly ([16, Déﬁnition 5.2] call the measure (2π) N µΩ,ga the pluriharmonic measure of a. When Ω is balanced (az ∈ Ω for every z ∈ Ω and |a| = 1), the support of this pluriharmonic measure is the Shilov boundary of Ω ([51, very end of the paper]). A bounded symmetric domain of CN is a bounded open and convex subset Ω of CN which is circled (az ∈ Ω for z ∈ Ω and |a| ≤ 1) and such that for every point a ∈ Ω, there is an involutive bi-holomorphic map γ : Ω → Ω such that a is an isolated ﬁxed point of γ (equivalently, γ(a) = a and γ ′ (a) = −id: see [52, Proposition 3.1.1]). For this deﬁnition, see [13, Deﬁnition 16 and Theorem 17], or [14, Deﬁnition 5 and Theorem 4]. Note that the convexity is automatic 3

(Hermann Convexity Theorem; see [27, p. 503 and Corollary 4.10]). É. Cartan showed that every bounded symmetric domain of CN is homogeneous, i.e. the group Γ of automorphisms of Ω acts transitively on Ω: for every a, b ∈ Ω, there is an automorphism γ of Ω such that γ(a) = b (see [52, p. 250]). Conversely, every homogeneous bounded convex domain is symmetric, since σ(z) = −z is a symmetry about 0 (see [52, p. 250] or [26, Remark 2.1.2 (e)]). Moreover, each automorphism extends continuously to Ω (see [22]). The unit ball BN and the polydisk DN are examples of bounded symmetric domains. Another example is, for N = p q, bi-holomorphic to the open unit ball of M(p, q) = L(Cq , Cp ) for the operator norm (see [27, Theorem 4.9]). Every product of bounded symmetric domains is still a bounded symmetric domain. In particular, every product of balls Ω = Bl1 × · · · × Blm , l1 + · · · + lm = N , is a bounded symmetric domain. If Ω is a bounded symmetric domain, its gauge is a norm k . k on CN whose open unit ball is Ω. Hence every bounded symmetric domain is hyperconvex (take ρ(z) = kzk − 1).

2.2 2.2.1

Hardy spaces on hyperconvex domains Hardy spaces on bounded symmetric domains

We begin by deﬁning the Hardy space on a bounded symmetric domain, because this is easier. The Shilov boundary (also called the Bergman-Shilov boundary or the distinguished boundary) ∂S Ω of a bounded domain Ω is the smallest closed set F ⊆ ∂Ω such that supz∈Ω |f (z)| = supz∈F |f (z)| for every function f holomorphic in some neighborhood of Ω (see [13, § 4.1]). When Ω is a bounded symmetric domain, it is also, since Ω is convex, the Shilov boundary of the algebra A(Ω) of the continuous functions on Ω which are holomorphic in Ω (because every function f ∈ A(Ω) can be approximated by fε with fε (z) = f εz0 + (1 − ε)z , where z0 ∈ Ω is given: see [20, pp. 152–154]). The Shilov boundary of the ball BN is equal to its topological boundary, but the Shilov boundary of the bidisk is ∂S D2 = {(z1 , z2 ) ∈ C2 ; |z1 | = |z2 | = 1}, whereas, its usual boundary ∂D2 is (T × D) ∪ (D × T); for the unit ball BN , the Shilov boundary is equal to the usual boundary SN −1 ([13, § 4.1]). Another example of a bounded symmetric domain, in C3 , is the set Ω = {(z1 , z2 , z3 ) ∈ C3 ; |z1 |2 + |z2 |2 < 1 , |z3 | < 1} and its Shilov boundary is ∂S Ω = {(z1 , z2 , z3 ) ; |z1 |2 + |z2 |2 = 1 , |z3 | = 1}. For p ≥ q, the matrix A is in the topological boundary of M(p, q) if and only if kAk = 1, but A is in the Shilov boundary if and only if A∗ A = Iq ; therefore the two boundaries coincide if and only if q = 1, i.e. Ω = BN (see [14, Example 2, p. 30]). Equivalently (see [24, Corollary 9], or [13, Theorem 33], [14, Theorem 10]), ∂S Ω is the set of the extreme points of the convex set Ω. The Shilov boundary ∂S Ω is invariant by the group Γ of automorphisms of Ω and the subgroup Γ0 = {γ ∈ Γ ; γ(0) = 0} act transitively on ∂S Ω (see [22]). A theorem of H. Cartan states that the elements of Γ0 are linear trans4

formations of CN and commute with the rotations (see [24, Theorem 1] or [26, Proposition 2.1.8]). It follows that the Shilov boundary of a bounded symmetric domain Ω coincides with its topological boundary only for Ω = BN (see [35, p. 572] or [36, p. 367]); in particular the open unit ball of CN for the norm k . kp , 1 < p < ∞, is never a bounded symmetric domain, unless p = 2. The unique Γ0 -invariant probability measure σ on ∂S Ω is the normalized surface area (see [22]). Then the Hardy space H 2 (Ω) is the space of all complexvalued holomorphic functions f on Ω such that: kf kH 2 (Ω) :=

sup 0

Z

∂S Ω

1/2 |f (rξ)|2 dσ(ξ)

is ﬁnite (see [22] and [23]). It is known that the integrals in this formula are non-decreasing as r increases to 1, so we can replace the supremum by a limit. The same deﬁnition can be given when Ω is a bounded complete Reinhardt domain (see [1]). The space H 2 (Ω) is a Hilbert space (see [22, Theorem 5]) and for every z ∈ Ω, the evaluation map f ∈ H 2 (Ω) 7→ f (z) is uniformly bounded on compacts subsets of Ω, by a depending only on that compact set, and of Ω ([22, Lemma 3]). For every f ∈ H 2 (Ω), there exists a boundary values function f ∗ such that kfr − f ∗ kL2 (∂S Ω) −→ 0, where fr (z) = f (rz) ([9, Theorem 3]), and the map r→1

f ∈ H 2 (Ω) 7→ f ∗ ∈ L2 (∂S Ω) is an isometric embedding ([22, Theorem 6]).

2.2.2

Hardy spaces on hyperconvex domains

For hyperconvex domains, the deﬁnition of Hardy spaces is more involved. It was done by E. Poletsky and M. Stessin ([47, Theorem 6]). Those domains are associated to a continuous negative psh exhaustion function u on Ω and the deﬁnition of the Hardy spaces uses the Demailly-Monge-Ampère measures. The space Hu2 (Ω) is the space of all holomorphic functions f : Ω → C such that: Z sup |f |2 dµu,r < ∞ r<0

SΩ,u (Ω)

and its norm is deﬁned by: kf kHu2 (Ω) = sup r<0

1 (2π)N

Z

2

SΩ,u (Ω)

|f | dµu,r

1/2

.

We can replace the supremum by a limit since the integrals are non-decreasing as r increases to 0 ([16, Corollaire 1.9]. The space H ∞ (Ω) of bounded holomorphic functions in Ω is contained in 2 Hu (Ω) (see [47], remark before Lemma 3.4). These spaces Hu2 (Ω) are Hilbert spaces ([47, Theorem 4.1]), but depends on the exhaustion function u (even when N = 1: see for instance [49]). Nevertheless, they all coincide, with equivalent norms, for the functions u for which the measure (ddc u)N is compactly supported ([47, Lemma 3.4]); this is the case 5

when u(z) = g(z, a) is the pluricomplex Green function with pole a ∈ Ω (because then (ddc u)N = (2π)N δa : see [16, Théorème 4.3], or [30, Theorem 6.3.6]). When Ω is the ball BN and u(z) = log kzk2, then (ddc u)N = C δ0 and µu,r = (2π)N dσt , where dσt is the normalized surface area on the sphere of radius t := er (see [47, Section 4] or [17, Example 3.3]). When Ω is the polydisk DN and u(z) = log kzk∞, then (ddc u)N = (2π)N δ0 ([18, Corollary 5.4]) and 1 N (see [17, Example 3.10]). (2π)N µu,r is the Lebesgue measure of the torus rT i ¯ c Note that in [17] and [18], the operator d is deﬁned as 2π (∂ − ∂) instead of ¯ i(∂ − ∂), as usually used. In these two cases, the Hardy spaces are the same as the usual ones (see [2, Remark 5.2.1]). In the sequel, we only consider the exhausting function u = gΩ ; hence we will write BΩ (r), SΩ (r) and H 2 (Ω) instead of BΩ,u (r), SΩ,u (r) and Hu2 (Ω). The two notions of Hardy spaces for a bounded symmetric domain are the same: Proposition 2.1. Let Ω be a bounded symmetric domain in CN . Then the Hardy space H 2 (Ω) coincides with the subspace of the Poletski-Stessin Hardy space HgΩ (Ω), with equality of the norms. Proof. First let us note that if k . k is the norm whose open unit ball is Ω, then gΩ (z) = log kzk (see [7, Proposition 3.3.2]). Let µΩ be the measure which is the ∗-weak limit of the Demailly-MongeN Ampère measures µr = ddc (gΩ )r . We saw that it is supported by ∂S Ω. By the remark made in [16, pp. 536-537], since the automorphisms of Ω continuously extend on ∂Ω, the measure µΩ is Γ-invariant. By unicity, the harmonic measure µ eΩ = (2π)−N µΩ at 0 hence coincides with the normalized area measure on ∂S Ω. We have, for f : Ω → C holomorphic and 0 < s < 1: Z Z Z 1 |f (sz)|2 dµr (z) , |f (sz)|2 de µΩ (z) = lim |f (sz)|2 de µΩ (z) = r→0 (2π)N S (r) ∂Ω ∂S Ω Ω because z 7→ |f (sz)|2 is continuous on Ω. Now, since gΩ (z) = log kzk, we have SΩ (r) = er ∂Ω and (gΩ )r (z)+t = (gΩ )r+t (sz); hence µr (sA) = µr+t (A) for every Borel subset A of ∂Ω, where t = log s. It follows that: Z Z |f (ζ)|2 dµr+t (ζ) . |f (sz)|2 dµr (z) = SΩ (r+t)

SΩ (r)

By letting r and t going to 0, we get: Z 1 kf k2H 2 (Ω) = lim |f (ζ)|2 dµr+t (ζ) = kf k2Hg2 ; r,t→0 (2π)N S (r+t) Ω Ω hence f ∈ H 2 (Ω) if and only if f ∈ Hg20 (Ω), with the same norms. We have ([47, Theorem 3.6]): 6

Proposition 2.2 (Poletsky-Stessin). For every z ∈ Ω, the evaluation map f ∈ H 2 (Ω) 7→ f (z) is uniformly bounded on compacts subsets of Ω, by a constant depending only on that compact set, and of Ω. Hence H 2 (Ω) has a reproducing kernel, deﬁned by: (2.2)

for f ∈ H 2 (Ω) ,

f (a) = hf, Ka i ,

and for each r < 0: (2.3)

Lr :=

sup kKa k2 < ∞ .

a∈BΩ (r)

2.3

Composition operators

A Schur map, associated with the bounded hyperconvex domain Ω, is a non-constant analytic map of Ω into itself. It is said non degenerate if its Jacobian is not identically null. It is equivalent to say that the diﬀerential ϕ′ (a) : CN → CN is an invertible linear map for at least one point a ∈ Ω. In [4], we used the terminology truly N -dimensional. Then, by the implicit function theorem, this is equivalent to saying that ϕ(Ω) has non-void interior. We say that the Schur map ϕ is a symbol if it deﬁnes a bounded composition operator Cϕ : H 2 (Ω) → H 2 (Ω) by Cϕ (f ) = f ◦ ϕ. Let us recall that although any Schur function generates a bounded composition operator on H 2 (D), this is no longer the case on H 2 (DN ) as soon as N ≥ 2, as shown for example by the PnSchur map ϕ(z1 , z2 ) = (z1 , z1 ). Indeed (see [3]), if say N = 2, taking f (z) = j=0 z1j z2n−j , we see that: √ kf k2 = n + 1 while kCϕ f k2 = k(n + 1)z1n k2 = n + 1 . The same phenomenon occurs on H 2 (BN ) ([43]; see also [11], [12], and [15]; see also [47]).

2.4

s-numbers of operators on a Hilbert space

We begin by recalling a few operator-theoretic facts. Let H be a Hilbert space. The approximation numbers an (T ) = an of an operator T : H → H are deﬁned as: (2.4)

an =

inf

rank R

kT − Rk ,

n = 1, 2, . . .

The operator T is compact if and only if limn→∞ an (T ) = 0. According to a result of Allahverdiev [10, p. 155], √ an = sn , the n-th singular number of T , i.e. the n-th eigenvalue of |T | := T ∗ T when those eigenvalues are rearranged in non-increasing order. The n-th width dn (K) of a subset K of a Banach space Y measures the defect of ﬂatness of K and is by deﬁnition: sup dist (f, E) , (2.5) dn (K) = inf dim E

f ∈K

7

where E runs over all subspaces of Y with dimension < n and where dist (f, E) denotes the distance of f to E. If T : X → Y is an operator between Banach spaces, the n-th Kolmogorov number dn (T ) of T is the nth-width in Y of T (BX ) where BX is the closed unit ball of X, namely: sup dist (T f, E) . (2.6) dn (T ) = inf dim E

f ∈BX

In the case where X = Y = H, a Hilbert space, we have: (2.7)

an (T ) = dn (T ) for all n ≥ 1 ,

and ([40]) the following alternative deﬁnition of an (T ): sup dist (T f, T E) . (2.8) an (T ) = inf dim E

f ∈BH

In this work, we use, for an operator T : H → H, the following notation: − βN (T ) = lim inf [anN (T )]1/n

(2.9)

n→∞

and: (2.10)

+ βN (T ) = lim sup[anN (T )]1/n . n→∞

When these two quantities are equal, we write them βN (T ).

3 3.1

Pluripotential theory Monge-Ampère capacity

Let K be a compact subset of an open subset Ω of CN . The Monge-Ampère capacity of K has been deﬁned by Bedford and Taylor ([5]; see also [30, Part II, Chapter 1]) as: Z c N Cap (K) = sup (dd u) ; u ∈ PSH(Ω) and 0 ≤ u ≤ 1 on Ω . K

When Ω is bounded and hyperconvex, we have a more convenient formula ([5, Proposition 5.3], [30, Proposition 4.6.1]): Z Z c ∗ N (3.1) Cap (K) = (dd uK ) = (ddc u∗K )N , Ω

K

(the positive measure (ddc u∗K )N is supported by K; actually by ∂K: see [17, Properties 8.1 (c)]), where uK = uK,Ω is the relative extremal function of K, deﬁned, for any subset E ⊆ Ω, as: (3.2)

uE,Ω = sup{v ∈ PSH(Ω) ; v ≤ 0 and v ≤ −1 on E} , 8

and u∗E,Ω is its upper semi-continuous regularization: u∗E,Ω (z) = lim sup uE,Ω (ζ) , ζ→z

z ∈ Ω,

called the regularized relative extremal function of E. For an open subset ω of Ω, its capacity is deﬁned as: Cap (ω) = sup{Cap (K) ; K is a compact subset of ω} . When ω ⊂ Ω is a compact subset of Ω, we have ([5, equation (6.2)], [30, Corollary 4.6.2]): Z (3.3) Cap (ω) = (ddc uω )N . Ω

The outer capacity of a subset E ⊆ Ω is: Cap ∗ (E) = inf{Cap (ω) ; ω ⊇ E and ω open} . If Ω is hyperconvex and E relatively compact in Ω, then ([30, Proposition 4.7.2]): Z ∗ Cap (E) = (ddc u∗E,Ω )N . Ω

Remark. A. Zeriahi ([57]) pointed out to us the following result. Proposition 3.1. Let K be a compact subset of Ω. Then: Cap (K) = Cap (∂K) . Proof. Of course uK ≤ u∂K since ∂K ⊆ K. Conversely, let v ∈ PSH(Ω) non-positive such that v ≤ −1 on ∂K. By the maximum principle (see [30, Corollary 2.9.6]), we get that v ≤ −1 on K. Hence v ≤ uK . Taking the supremum over all those v, we obtain u∂K ≤ uK , and therefore u∂K = uK . By (3.1), it follows that: Z Z (3.4) Cap (K) = (ddc u∗K )N = (ddc u∗∂K )N = Cap (∂K) . Ω

Ω

3.2

Regular sets

Let E ⊆ CN be bounded. Recall that the polynomial convex hull of E is: b = {z ∈ C ; |P (z)| ≤ sup |P | for every polynomial P } . E E

b is called regular if u∗ (a) = −1 for an open set Ω ⊇ E b A point a ∈ E E,Ω (note that we always have uE,Ω = uE,Ω = −1 on the interior of E: see [17, b are Properties 8.1 (c)]). The set E is said to be regular if all points of E regular. 9

The pluricomplex Green function of E, also called the L-extremal function of E, is deﬁned, for z ∈ CN , as: VE (z) = sup{v(z) ; v ∈ L ,

v ≤ 0 on E} ,

where L is the Lelong class of all functions v ∈ PSH(CN ) such that, for some constant C > 0: v(z) ≤ C + log(1 + |z|) for all z ∈ CN .

b is called L-regular if V ∗ (a) = 0, where V ∗ is the upper semiA point a ∈ E E E b are continuous regularization of VE . The set E is L-regular if all points of E L-regular. By [28, Proposition 2.2] (see also [30, Proposition 5.3.3, and Corollary 5.3.4]), b for E bounded and non pluripolar, and Ω a bounded open neigbourhood of E, we have: (3.5)

m(uE,Ω + 1) ≤ VE ≤ M (uE,Ω + 1)

b is equivalent for some positive constants m, M . Hence the regularity of a ∈ E to its L-regularity. Recall that E is pluripolar if there exists an open set Ω containing E and v ∈ PSH(Ω) such that E ⊆ {v = −∞}. This is equivalent to say that there exists a hyperconvex domain Ω of CN containing E such that u∗E,Ω ≡ 0 (see [30, Corollary 4.7.3 and Theorem 4.7.5]). By Josefson’s theorem ([30, Theorem 4.7.4]), E is pluripolar if and only if there exists v ∈ PSH(CN ) such that E ⊆ {v = −∞}. Recall also that E is pluripolar if and only if its outer capacity Cap ∗ (E) is null ([30, Theorem 4.7.5]). When Ω is hyperconvex and E is compact, non pluripolar, the regularity of E implies that uE,Ω and VE are continuous, on Ω and CN respectively ([30, Proposition 4.5.3 and Corollary 5.1.4]). Conversely, if uE,Ω is continuous, for some hyperconvex neighbourhood Ω of E, then uE,Ω (z) = −1 for all z ∈ E; hence VE (z) = 0 for all z ∈ E, by (3.5); but VE = VEb when E is compact b by (3.5) again, we obtain ([30, Theorem 5.1.7]), so VE (z) = 0 for all z ∈ E; b that uE,Ω (z) = −1 for all z ∈ E; therefore E is regular. In the same way, the continuity of VE implies the regularity of E. These results are due to Siciak ([50, Proposition 6.1 and Proposition 6.2]). Every closed ball B = B(a, r) of an arbitrary norm k . k on CN is regular since its L-extremal function is: VB (z) = log+ kz − ak/r) ([50, p. 179, § 2.6]).

3.3

Zakharyuta’s formula

We will need a formula that Zakharyuta, in order to solve a problem raised by Kolmogorov, proved, conditionally to a conjecture, called Zakharyuta’s conjecture, on the uniform approximation of the relative extremal function uK,Ω 10

by pluricomplex Green functions. This conjecture has been proved by Nivoche ([45, Theorem A]), in a more general setting that we state below: Theorem 3.2 (Nivoche). Let K be a regular compact subset of a bounded hyperconvex domain Ω of CN . Then for every ε > 0 and δ small enough, there exists a pluricomplex Green function g on Ω with a finite number of logarithmic poles such that: 1) the poles of g lie in W = {z ∈ Ω ; uK (z) < −1 + δ}; 2) we have, for every z ∈ Ω \ W : (1 + ε) g(z) ≤ uK (z) ≤ (1 − ε) g(z) . In order to state Zakharyuta’s formula, we need some additional notations. Let K be a compact subset of Ω with non-empty interior, and AK the set of restrictions to K of those functions that are analytic and bounded by 1, i.e. those functions belonging to the unit ball BH ∞ (Ω) of the space H ∞ (Ω) of the bounded analytic functions in Ω, considered as a subset of the space C(K) of complex functions deﬁned on K, equipped with the sup-norm on K. Let dn (AK ) be the nth-width of AK in C(K), namely: (3.6) dn (AK ) = inf sup dist (f, L) , L

f ∈AK

where L runs over all k-dimensional subspaces of C(K), with k < n. Equivalently, dn (AK ) is the nth-Kolmogorov number of the natural injection J of H ∞ (Ω) into C(K) (recall that K has non-empty interior). It is convenient to set, as in [56]: (3.7) and: (3.8)

τN (K) =

1 Cap (K) (2π)N "

ΓN (K) = exp −

N! τN (K)

1/N #

,

i.e.: (3.9)

ΓN (K) = exp − 2π

N! Cap (K)

1/N

.

Observe that Cap (K) > 0 since we assumed that K has non-empty interior. Now, we have ([56, Theorem 5.6]; see also [55, Theorem 5] or [54, pages 30–32], for a detailed proof): Theorem 3.3 (Zakharyuta-Nivoche). Let Ω be a bounded hyperconvex domain and K a regular compact subset of Ω with non-empty interior, which is holoe Ω ). Then: morphically convex in Ω (i.e. K = K 1/N N! (3.10) − log dn (AK ) ∼ n1/N . τN (K) 11

e Ω is the holomorphic convex hull of K in Ω, that is: Here K

e Ω = {z ∈ Ω ; |f (z)| ≤ sup |f | for every f ∈ O(Ω)} , K K

where O(Ω) is the set of all functions holomorphic in Ω. Relying on that theorem, which may be seen as the extension of a result of Erokhin, proved in 1958 (see [19]; see also Widom [53] which proved a more general result, with a diﬀerent proof), to dimension N > 1, and as a result on the approximation of functions, we will give an application to the study of approximation numbers of a composition operator on H 2 (Ω) for a bounded symmetric domain of CN .

4

The spectral radius type formula In [41, Section 6.2], we proved the following result.

Theorem 4.1. Let ϕ : DN → DN be given by ϕ(z1 , . . . , zN ) = (r1 z1 , . . . , rN zN ) where 0 < rj < 1. Then: βN (Cϕ ) = ΓN ϕ(DN ) = ΓN ϕ(DN ) .

The proof was simple, based on result of Blocki [8] on the Monge-Ampère capacity of a cartesian product, and on the estimation, when A → ∞, of the number νA of N -tuples α = (α1 , . . . , αN ) of non-negative integers αj such that PN j=1 αj σj ≤ A, where the numbers σj > 0 are ﬁxed. The estimation was: (4.1)

νA ∼

AN · N ! σ1 · · · σN

As J. F. Burnol pointed out to us, this is a consequence of the following elementary fact. Let λN be the Lebesgue measure on RN , and let E be a compact subset of RN such that λN (∂E) = 0. Then: λN (E) = lim A−N |(A × E) ∩ ZN | . A→∞

PN Then, just take E = {(x1 , . . . , xN ) ; xj ≥ 0 and j=1 xj σj ≤ 1}. In any case, this lets us suspect that the formula of Theorem 4.1 holds in much more general cases. This is not quite true, as evidenced by our counterexample of [41, Theorem 5.12]. Nevertheless, in good cases, this formula holds, as we will see in the next sections. In remaining of this section, we consider functions ϕ : Ω → Ω such that ϕ(Ω) ⊆ Ω. If ρ is an exhaustion function for Ω, there is some R0 < 0 such that ϕ(Ω) ⊆ BΩ (R0 ), and that implies that Cϕ maps H 2 (Ω) into itself and is a compact operator (see [47, Theorem 8.3], since, with their notations, for r > R0 , we have T (r) = ∅ and hence δϕ (r) = 0) . 12

4.1

Minoration

Recall that every hyperconvex domain Ω is pseudoconvex. By H. CartanThullen and Oka-Bremermann-Norguet theorems, being pseudoconvex is equivalent to being a domain of holomorphy, and equivalent to being holomorphically convex (meaning that if K is a compact subset in Ω, then its holomorphic hull e is also contained in Ω): see [33, Corollaire 7.7]. Now (see [32, Chapter 5, K Exercise 11], a domain of holomorphy Ω is said a Runge domain if every holomorphic function in Ω can be approximated uniformly on its compact subsets by polynomials, and that is equivalent to saying that the polynomial hull and the holomorphic hull of every compact subset of Ω agree. By [32, Chapter 5, Exercise 13], every circled domain (in particular every bounded symmetric domain) is a Runge domain. Definition 4.2. A hyperconvex domain Ω is said strongly regular if there exists a continuous psh exhaustion function ρ such that all the sub-level sets: Ωc = {z ∈ Ω ; ρ(z) < c} (c < 0) have a regular closure. For example, every bounded symmetric domain Ω is strongly regular since if k . k is the associated norm, its sub-level sets Ωc (with ρ(z) = log kzk) are the open balls B(0, ec ), and the closed balls are regular, as said above. Theorem 4.3. Let Ω be a strongly regular bounded hyperconvex and Runge domain in CN , and let ϕ : Ω → Ω be an analytic function such that ϕ(Ω) ⊆ Ω, and which is non-degenerate. Then: − (Cϕ ) . (4.2) ΓN ϕ(Ω) ≤ βN

Recall that if Ω is a domain in CN , a holomorphic function ϕ : Ω → CM (M ≤ N ) is non-degenerate if there exists a ∈ Ω such that ranka ϕ = M . Then ϕ(Ω) has a non-empty interior.

Proof. Let (rj )j≥1 be an increasing sequence of negative numbers tending to 0. The set Hj = Ωrj is a regular compact subset of Ω, with non-void interior cj its polynomial convex hull; this compact set is (hence non pluripolar). Let H cj = H fj , and since H fj ⊆ contained in Ω, since Ω being a Runge domain, we have H cj is Ω, because Ω is holomorphically convex (being hyperconvex). Moreover H regular since VE = VEb for every compact subset of CN ([50, Corollary 4.14]). cj and let G be a subspace of H 2 (Ω) with dimension < nN . Let Kj = ϕ H

The set Kj is regular because of the following result (see [30, Theorem 5.3.9], [46, top of page 40], [29, Theorem 1.3], or [44, Theorem 4], with a detailed proof). Theorem 4.4 (Pleśniak). Let E be a compact, polynomially convex, regular and non pluripolar, subset of CN . Then if Ω is a hyperconvex domain such that E ⊆ Ω and if ϕ : Ω → CN is a non-degenerate holomorphic function, the set ϕ(E) is regular. 13

cj of Kj is contained in Ω and is also As before, the polynomial convex hull K cj also. regular. Since ϕ is non-degenerate, Kj has a non-void interior; hence K cj . We can hence use Zakharyuta’s formula (Theorem 3.3) for the compact set K cj ). By By restriction, the subspace G can be viewed as a subspace of C(K Zakharyuta’s formula, for 0 < ε < 1, there is nε ≥ 1 such that, for n ≥ nε : 1/N N! ) ≥ exp − (1 + ε) (2π) n dnN (AK · cj cj ) Cap (K Hence, there exists f ∈ BH ∞ ⊆ BH 2 such that, for all g ∈ G: 1/N N! · kg − f kC(K cj ) ≥ (1 − ε) exp − (1 + ε) (2π) n cj ) Cap (K cj = K fj and, by deﬁnition k . k f = k . kC(K ) , we have: Since K j C(Kj )

kg − f kC(K cj ) = kg − f kC(Kj ) = kCϕ (g) − Cϕ (f )kC(H fj ) .

Equivalently, since, by deﬁnition k . kC(H fj ) = k . kC(Hj ) , we have, for all g ∈ G:

kCϕ (g) − Cϕ (f )kC(Hj ) ≥ (1 − ε) exp − (1 + ε) (2π) n

N! cj ) Cap (K

This implies, thanks to (2.3), that, for all g ∈ G: (1 − ε) exp − (1 + ε) (2π) n kCϕ (g) − Cϕ (f )kH 2 (Ω) ≥ L−1 rj Using (2.8), we get, since the subspace G is arbitrary: (1 − ε) exp − (1 + ε) (2π) n anN (Cϕ ) ≥ Lr−1 j

1/N

N! cj ) Cap (K

N! cj ) Cap (K

1/N

·

1/N

·

·

Taking the nth-roots and passing to the limit, we obtain: 1/N N! − βN (Cϕ ) ≥ exp − (1 + ε) (2π) · cj ) Cap (K and then, letting ε go to 0: − βN (Cϕ )

≥ exp − (2π)

N! cj ) Cap (K

1/N

cj ) . = ΓN (K

S c cj )j≥1 is increasing and Now, the sequence (K j≥1 Kj ⊇ ϕ(Ω); hence, by [5, S cj ≥ Cap [ϕ(Ω)], so: cj ) −→ Cap K Theorem 8.2 (8.3)], we have Cap (K j→∞

− βN (Cϕ ) ≥ ΓN [ϕ(Ω)] ,

and the proof of Theorem 4.3 is ﬁnished. 14

j≥1

4.2

Majorization

For the majorization, we assume diﬀerent hypotheses on the domain Ω. Nevertheless these assumptions agree with that of Theorem 4.3 when Ω is a bounded symmetric domain. 4.2.1

Preliminaries

Recall that a domain Ω of CN is a Reinhardt domain (resp. complete Reinhardt domain) if z = (z1 , . . . , zN ) ∈ Ω implies that (ζ1 z1 , . . . , ζN zN ) ∈ Ω for all complex numbers ζ1 , . . . , ζN of modulus 1 (resp. of modulus ≤ 1). A complete bounded Reinhardt domain is hyperconvex if and only if log jΩ is psh and continuous in CN \ {0}, where jΩ is the Minkowski functional of Ω (see [7, Exercise following Proposition 3.3.3]). In general, the Minkowski functional jΩ of a bounded complete Reinhardt domain Ω is usc and log jΩ is psh if and only if Ω is pseudoconvex ([7, Theorem 1.4.8]). Other conditions for a bounded complete Reinhardt domain to being hyperconvex can found in [34, Theorem 3.10]. For a bounded hyperconvex and complete Reinhardt domain Ω, its pluricomplex Green function with pole 0 is gΩ (z) = log jΩ (z), where jΩ is the Minkowski functional of Ω ([7, Proposition 3.3.2]), and SΩ (r) = er ∂Ω. Since ∂Ω is in particular invariant by the pluri-rotations z = (z1 , . . . , zN ) 7→ (eiθ1 z1 , . . . , eiθN zN ), with θ1 , . . . , θN ∈ R, the harmonic measure µ eΩ at 0 (see the proof of Proposition 2.1) is also invariant by the pluri-rotations (note that it is supported by the Shilov boundary of Ω: see [51, very end of the paper]). We have, as in the proof of Proposition 2.1, for f ∈ H 2 (Ω): Z sup |f (sz)|2 de µΩ (z) = kf k2H 2 (Ω) < ∞ . 0

∂Ω

Since µ eΩ is in particular invariant by the rotations z 7→ eiθ z, θ ∈ R, there exists, by [9, Theorem 3], a function f ∗ ∈ L2 (∂Ω, µ eΩ ) such that: Z |f (sz) − f ∗ (z)|2 de µΩ (z) −→ 0 . s→1

∂Ω

It ensues that the map f ∈ H 2 (Ω) 7→ f ∗ ∈ L2 (∂Ω, µ eΩ ) is an isometric embedding (in fact, f ∗ is the radial limit of f : see [21, Lemma 2]). Therefore, we can consider H 2 (Ω) as a complemented subspace of L2 (∂Ω, µ eΩ ), and we call P the orthogonal projection of L2 (∂Ω, µ eΩ ) onto H 2 (Ω). Every holomorphic function f in a Reinhardt domain Ω containing 0 (in particular if Ω is a complete Reinhardt domain) has a power series expansion about 0: X f (z) = bα z α α

which converges normally on compact subsets of Ω ([32, Proposition 2.3.14]). αN Recall that if z = (z1 , . . . , zN ) and α = (α1 , . . . , αN ), then z α = z1α1 · · · zN , |α| = α1 + · · · + αN , and α! = α1 ! · · · αN !. We have: 15

Proposition 4.5. Let Ω be a bounded hyperconvex and complete Reinhardt domain, and set eα (z) = z α . Then the system (eα )α is orthogonal in H 2 (Ω). Proof. We use the fact that the level sets S(r) and the Demailly-Monge-Ampère N are pluri-rotation invariant. For α 6= β, we choose measures µr = ddc (gΩ )r θ1 , . . . , θN ∈ R such that 1, (θ1 /2π), . . . , (θN /2π) are rationally independent. PN 6= 1. Hence, as in [25, p. 78], we have, making Then exp i j=1 (αj − βj )θj iθ1 the change of variables z = (e w1 , . . . , eiθN wN ): Z X Z N α β z z dµr (z) = exp i wα wβ dµr (w) , (αj − βj )θj S(r)

S(r)

j=1

which implies that:

Z

and hence:

z α z β dµr (z) = 0 , S(r)

(z α | z β ) := lim

r→0

Z

z α z β dµr (z) = 0 .

S(r)

For the polydisk, we have keα kH 2 (DN ) = 1, and for the ball (see [48, Proposition 1.4.9]): (N − 1)! α! · keα k2H 2 (BN ) = (N − 1 + |α|)!

Definition 4.6. We say that Ω is a good complete Reinhardt domain if, for some positive constant CN and some positive integer c, we have, for all p ≥ 0: X |z α |2 ≤ CN pcN [jΩ (z)]2p , keα k2H 2 (Ω) |α|=p

where jΩ is the Minkowski functional of Ω.

Examples 1. The polydisk DN is a good Reinhardt domain because keα kH 2 (DN ) = 1, |α| |z α | ≤ kzk∞ , and the number of indices α such that |α| = p is N −1+p ≤ CN pN p (see [35, p. 498] or [37, pp. 213–214]). 2. The ball BN is a good Reinhardt domain. In fact, observe that: (p + 1)(p + 2) · · · (p + N − 1) (N − 1 + p)! = p! ≤ p! (p + 1)N −1 ≤ p! (p + 1)N ; (N − 1)! 1 × 2 × · · · × (N − 1)

hence:

X

|α|=p

X |z α |2 (N − 1 + |α|)! |z α |2 = 2 keα kH 2 (BN ) (N − 1)! α! |α|=p

≤ (p + 1)N

X |α|! |z1 |2α1 · · · |zN |2αN α!

|α|=p

N

= (p + 1) (|z1 |2 + · · · + |zN |2 )p , 16

by the multinomial formula, so: X

|α|=p

|z α |2 2p N N ≤ (p + 1)N kzk2p 2 ≤ 2 p kzk2 . keα k2H 2 (BN )

3. More generally, if Ω = Bl1 × · · · × Blm , l1 + · · · + lm = N , is a product of balls, we have, writing α = (β1 , . . . , βm ), where each βj is an lj -tuple: Z keα k2H 2 (Ω) = |uβ1 1 |2 . . . |uβmm |2 dσl1 (u1 ) . . . dσlm (um ) Sl1 ×···×Sl2

=

m Y

j=1

(lj − 1)! βj ! , (lj − 1 + |βj |)!

and, writing z = (z1 , . . . , zm ), with zj ∈ Blj : X

|α|=p

|z α |2 ≤ keα k2H 2 (Ω) p

X

1 +···+pm =p

m Y

j=1

2pj

(pj + 1)lj kzj k2

≤ Cm p (p + 1)l1 +···+lm [jΩ (z)]2(p1 +···+pm ) , m

since jΩ (z) = max{kz1 k2 , . . . , kzm k2 }. Hence: X

|α|=p

4.2.2

|z α |2 ≤ CN p2N [jΩ (z)]2p . keα k2H 2 (Ω)

The result

Theorem 4.7. Let Ω be a bounded hyperconvex domain which is a good complete Reinhardt domain in CN , and let ϕ : Ω → Ω be an analytic function such that ϕ(Ω) ⊆ Ω. Then, for every compact subset K ⊇ ϕ(Ω) of Ω with non void interior, we have: (4.3)

+ βN (Cϕ ) ≤ ΓN (K) .

In particular, if ϕ is moreover non-degenerate, we have: + (4.4) βN (Cϕ ) ≤ ΓN ϕ(Ω) .

The last assertion holds because ϕ(Ω) is open if ϕ is non-degenerate.

Corollary 4.8. Let Ω be a good complete bounded symmetric domain in CN , and ϕ : Ω → Ω a non-degenerate analytic map such that ϕ(Ω) ⊆ Ω. Then: − + (Cϕ ) ≤ βN (Cϕ ) ≤ ΓN ϕ(Ω) . ΓN ϕ(Ω) ≤ βN

For the proof of Theorem 4.7, we will use the following result ([56, Proposition 6.1]), which do not need any regularity condition on the compact set (because it may be written as a decreasing sequence of regular compact sets). 17

Proposition 4.9 (Zakharyuta). If K is any compact subset of a bounded hyperconvex domain Ω of CN with non-empty interior, we have: 1/N log dn (AK ) N! lim sup . ≤ − τN (K) n1/N n→∞ Proof of Theorem 4.7. In the sequel we write k . kH 2 for k . kH 2 (Ω) . We set: ΛN = lim sup[dn (AK )]n

−1/N

.

n→∞

Changing n into nN , Proposition 4.9 means that for every ε > 0, there exists, for n large enough, an (nN − 1)-dimensional subspace F of C(K) such that, for any g ∈ H ∞ (Ω), there exists h ∈ F such that: (4.5)

kg − hkC(K) ≤ (1 + ε)n ΛnN kgk∞ .

Let l be an integer to be adjusted later, and X f (z) = bα z α ∈ H 2 (Ω) with kf kH 2 ≤ 1 . α

By Proposition 4.5, we have: kf k2H 2 =

X α

We set: g(z) =

|bα |2 keα k2H 2 .

X

bα z α .

|α|≤l

By the Cauchy-Schwarz inequality: X X X |z α |2 |z α |2 2 2 2 ≤ · |bα | keα kH 2 |g(z)| ≤ 2 keα kH 2 keα k2H 2 |α|≤l

|α|≤l

|α|≤l

Since Ω is a good complete Reinhardt domain and since jΩ (z) < 1 for z ∈ Ω, we have: l X 2 |g(z)| ≤ pcN [jΩ (z)]2p ≤ (l + 1)cN +1 . p=0

It follows from (4.5) that there exists h ∈ F such that:

kg − hkC(K) ≤ (1 + ε)n ΛnN (l + 1)(cN +1)/2 . Since Cϕ f (z) − Cϕ g(z) = f ϕ(z) − g ϕ(z) and ϕ(Ω) ⊆ K, we have kCϕ f − Cϕ gk∞ ≤ kf − gkC(K) ; therefore: (4.6)

kg ◦ ϕ − h ◦ ϕkH 2 ≤ kg ◦ ϕ − h ◦ ϕk∞ ≤ kg − hkC(K) ≤ (1 + ε)n ΛnN (l + 1)(cN +1)/2 . 18

Now, the subspace Fe formed by functions v ◦ ϕ, for v ∈ F , can be viewed as a subspace of L∞ (∂Ω, µ eΩ ) ⊆ L2 (∂Ω, µ eΩ ) (indeed, since v is continuous, we can ∗ ∗ write (v ◦ ϕ) = v ◦ ϕ , where ϕ∗ denotes the almost everywhere existing radial limits of ϕ(rz), which belong to K). Let ﬁnally E = P (Fe ) ⊆ H 2 (Ω) where P : L2 (∂Ω, µ eΩ ) → H 2 (Ω) is the orthogonal projection. This is a subspace of H 2 (Ω) with dimension < nN , and we have dist (Cϕ g, E) ≤ kg ◦ ϕ − P (h ◦ ϕ)kH 2 ; hence, by (4.6): (4.7)

dist (Cϕ g, E) ≤ (1 + ε)n ΛnN (l + 1)(cN +1)/2 .

Now, the same calculations give that: X |f (z) − g(z)|2 ≤ pcN [jΩ (z)]2p ; p>l

hence, for some positive constant MN : |f (z) − g(z)| ≤ MN (l + 1)(cN +1)/2

[jΩ (z)]l , (1 − [jΩ (z)]2 )(cN +1)/2

by using the following lemma, whose proof is postponed. Lemma 4.10. For every non-negative integer m, there exists a positive constant Am such that, for all integers l ≥ 0 and all 0 < x < 1, we have: X p≥l

pm xp ≤ Am lm

xl . (1 − x)m+1

Since K is a compact subset of Ω, there is a positive number r0 < 1 such that jΩ (z) ≤ r0 for z ∈ K. Since Cϕ f (z) − Cϕ g(z) = f ϕ(z) − g ϕ(z) and ϕ(Ω) ⊆ K, we have kCϕ f − Cϕ gk∞ ≤ kf − gkC(K) , and we get: (4.8) kCϕ f −Cϕ gkH 2 ≤ kCϕ f −Cϕ gk∞ ≤ MN (l+1)(cN +1)/2

r0l · (1 − r02 )(cN +1)/2

Now, (4.7) and (4.8) give: dist (Cϕ f, E) ≤ MN (l + 1)(cN +1)/2

r0l + (1 + ε)n ΛnN 2 (1 − r0 )(cN +1)/2

.

It ensues, thanks to (2.7), that:

1/n anN (Cϕ ) ≤ [MN (l + 1)(cN +1)/2 ]1/n

l/n r0 + (1 + ε) ΛN . (1 − r02 )(cN +1)/2n

Taking now for l the integer part of n log n, and passing to the upper limit as n → ∞, we obtain (since l/n → ∞ and (log l)/n → 0): + βN (Cϕ ) ≤ (1 + ε) ΛN ,

19

and therefore, since ε > 0 is arbitrary: + βN (Cϕ ) ≤ ΛN .

That ends the proof, by using Proposition 4.9. Proof of Lemma 4.10. We make the proof by induction on m. We set: X Sm = pm xp p≥l

P xl · The result is obvious for m = 0, with A0 = 1, since then S0 = p≥l xp = 1−x Let us assume that it holds till m − 1 and prove it for m. We observe that, since pm − (p − 1)m ≤ mpm−1 , we have: X X X X (1 − x)Sm = pm xp − pm xp+1 = pm xp − (p − 1)m xp p≥l

X

=

p≥l+1

≤

X p≥l

p≥l

m

p≥l

m

p

m l

(p − (p − 1) )x + l x ≤

p≥l+1

X

p≥l+1

mpm−1 xp + lm xl ≤ mAm−1 lm−1

≤ (mAm−1 + 1) lm

mpm−1 xp + lm xl

xl + lm xl (1 − x)m

xl , (1 − x)m

giving the result, with Am = mAm−1 + 1.

4.3

Equality

Proposition 4.11. Let Ω be a bounded hyperconvex domain and ω a relatively compact open subset of Ω. Assume that: (4.9)

For every a ∈ ∂ω, except on a pluripolar set E ⊆ ∂ω, there exists z0 ∈ ω such that the open segment (z0 , a) is contained in ω.

Then: Cap (ω) = Cap (ω) . In particular, if ϕ : Ω → Ω a non-degenerate holomorphic map such that ϕ(Ω) ⊆ Ω and ω = ϕ(Ω) satisfies (4.9), we have: Cap ϕ(Ω) = Cap ϕ(Ω) . Before proving Proposition 4.11, let us give an example of such a situation.

Proposition 4.12. Let Ω be a bounded hyperconvex domain with C 1 boundary. Let U be an open neighbourhood of Ω and ϕ : U → CN be a non-degenerate holomorphic function such that ϕ(Ω) ⊆ Ω. Then the condition (4.9) is satisfied. 20

Proof. Let ω = ϕ(Ω). We may assume that U is connected, hyperconvex and bounded. Let Bϕ be the set of points z ∈ U such that the complex Jacobian Jϕ is null. Since Jϕ is holomorphic in Ω, we have log |Jϕ | ∈ PSH(U ) and hence (see [31, proof of Lemma 10.2]): Bϕ = {z ∈ U ; Jϕ (z) = 0} = {z ∈ U ; log |Jϕ (z)| = −∞} is pluripolar . Therefore (see [5, Theorem 6.9]), Cap (Bϕ , U ) = 0. It follows (see [5, page 2, line -8]) that Cap [ϕ(Bϕ )] := Cap [ϕ(Bϕ ), Ω] = 0. Now, for every a ∈ ∂ω ∩ [ϕ(U \ Bϕ )], there is a tangent hyperplane Ha to ω, and hence an inward normal to ∂ω (note that ∂ω ⊆ ϕ(∂Ω) ⊆ ϕ(U )). It follows that there is z0 ∈ ω such that the open interval (z0 , a) is contained in ω. Proof of Proposition 4.11. Let a ∈ ∂ω and L be a complex line containing (z0 , a); we have a ∈ ω ∩ L. Assume now that this point a is a fine (“effilé”) point of ω, i.e. that there exists u ∈ PSH(V ), for V a neighbourhood of a, such that: lim sup u(z) < u(a) . z→a ,z∈ω

By deﬁnition, the restriction u e of u to ω ∩ L is subharmonic and we keep the inequality: lim sup u e(z) < u e(a) = u(a) . z→a ,z∈ω∩L

That means that a is a ﬁne point of ω ∩L. But a ∈ ω ∩ L and ω ∩L is connected, so this is not possible, by [40, Lemma 2.4]. Hence no point of ∂ω \ E is ﬁne. Let now ω f be the closure of ω for the ﬁne topology (i.e. the coarsest topology on U for which all the functions in PSH(U ) are continuous; it is known: see [6, comment after Theorem 2.3], that it is the trace on U of the ﬁne topology on CN ). It is also known (see [30, Corollary 4.8.10]) that ω f is the set of points of ω which are not ﬁne. By the above reasoning, we thus have: ω \ ωf ⊆ E . Since Cap (E) = 0, we have: Cap (ω \ ω f ) = 0 , and it follows that: Cap (ω) = Cap [ω f ∪ (ω \ ω f )] ≤ Cap (ω f ) + Cap (ω \ ω f ) = Cap (ω f ) ,

and hence Cap (ω f ) = Cap (ω). But, since, by deﬁnition, the psh functions are continuous for the ﬁne topology, it is clear, that the relative extremal functions uω,Ω and uωf ,Ω are equal; hence we have, by [30, Proposition 4.7.2]: Z Z c ∗ N Cap (ω) = (dd uω,Ω ) = (ddc u∗ωf ,Ω )N = Cap (ω f ) . Ω

Ω

Hence Cap (ω) = Cap (ω).

21

4.4

Consequences of the spectral radius type formula

Theorem 4.3 has the following consequence. Proposition 4.13. Let Ω be a regular bounded symmetric domain in CN , and let ϕ : Ω → Ω be a non-degenerate analytic function inducing a bounded composition operator Cϕ on H 2 (Ω). Then, if Cap [ϕ(Ω)] = ∞, we have βN (Cϕ ) = 1. 1/N In other words, if, for some constants C, c > 0, we have an (Cϕ ) ≤ C e−cn for all n ≥ 1, then Cap [ϕ(Ω)] < ∞. As a corollary, we can give a new proof of [41, Theorem 3.1]. Corollary 4.14. Let τ : D → D be an analytic map such that kτ k∞ = 1 and ψ : DN −1 → DN −1 such that the map ϕ : DN → DN , defined as: ϕ(z1 , z2 , . . . , zN ) = τ (z1 ), ψ(z2 , . . . , zN ) , is non-degenerate. Then βN (Cϕ ) = 1.

Proof. Since the map ϕ is non-degenerate, ψ is also non-degenerate. Hence (see [44, Proposition 2] ψ(DN −1 ) is not pluripolar, i.e. CapN −1 [ψ(DN −1 )] > 0. On the other hand, it follows from [40, Theorem 3.13 and Theorem 3.14] that Cap1 [τ (D)] = +∞. Then, by [8, Theorem 3], we have: CapN [ϕ(DN )] = CapN [τ (D) × ψ(DN −1 )]

= Cap1 [τ (D)] × CapN −1 [ψ(DN −1 )] = +∞ .

It follows from Proposition 4.13 that βN (Cϕ ) = 1. Proof of Proposition 4.13. If R : H 2 (Ω) → H 2 (Ω) is a ﬁnite-rank operator, we set, for t < 0: (Rt f )(w) = (Rf )(et w) ,

f ∈ H 2 (Ω) .

Then the rank of the operator Rt is less or equal to that of R. Recall that if k . k is the norm whose unit ball is Ω, then the pluricomplex Green function of Ω is gΩ (z) = log kzk, and hence the level set S(r) is the sphere S(0, er ) = er ∂Ω for this norm. Since: Z Z t t 2 |f [ϕ(z)] − (Rf )(z)|2 dµr+t (z) , |f [ϕ(e w)] − (Rf )(e w)| dµr (w) = S(r+t)

S(r)

we have, setting ϕt (w) = ϕ(et w): kCϕt (f ) − Rt (f )kH 2 ≤ kCϕ (f ) − R(f )kH 2 . − (Cϕt ) ≤ It follows that an (Cϕt ) ≤ an (Cϕ ) for every n ≥ 1. Therefore βN − βN (Cϕ ).

22

By Theorem 4.3, we have: " exp − 2π

N! Cap [ϕt (Ω)]

1/N #

− ≤ βN (Cϕt ) .

Since ϕt (Ω) = ϕ(et Ω) increases to ϕ(Ω) as t ↑ 0, we have (see [30, Corollary 4.7.11]): Cap [ϕ(Ω)] = lim Cap [ϕt (Ω)] . t→0

As Cap [ϕ(Ω)] = ∞, we get: − − βN (Cϕ ) ≥ lim sup βN (Cϕt ) = 1 . t→0

Remark 1. In [41, Theorem 5.12], we construct a non-degenerate analytic function ϕ : D2 → D2 such that ϕ(D2 ) ∩ ∂D2 6= ∅ and for which β2+ (Cϕ ) < 1. We hence have Cap [ϕ(D2 )] < ∞. Remark 2. The capacity cannot tend to inﬁnity too fast when the compact set approaches the boundary of Ω; in fact, we have the following result, that we state for the ball, but which holds more generally. Proposition 4.15. For every compact set K of BN , we have, for some constant CN : CN · Cap (K) ≤ [dist (K, SN )]N Proof. We know that: Cap (K) =

Z

BN

(ddc u∗K )N .

Let ρ(z) = |z|2 − 1 and aK := minz∈K [−ρ(z)] = − maxz∈K ρ(z). Then ρ is in PSH and is non-positive. Since aK > 0, the function: v(z) =

ρ(z) aK

is in PSH, non-positive on BN , and v ≤ −1 on K. Hence v ≤ uK ≤ u∗K . Since v(w) = 0 for all w ∈ SN and (see [5, Proposition 6.2 (iv)], or [30, Proposition 4.5.2]): lim u∗K (z) = 0 , z→w

for all w ∈ SN , the comparison theorem of Bedford and Taylor ([5, Theorem 4.1]; [30, Theorem 3.7.1] gives, since v ≤ u∗K and v, u∗K ∈ PSH: Z Z Z 1 c N c ∗ N (dd v) = N (dd uK ) ≤ (ddc ρ)N . a BN BN K BN As (ddc ρ)N = 4N N ! dλ2N , we get, with CN := 4N N ! λ2N (BN ): Cap (K) ≤ 23

CN · aN K

That ends the proof since: aK = min(1 − |z|2 ) ≥ min(1 − |z|) = dist (K, SN ) z∈K

z∈K

We have assumed that the symbol ϕ is non-degenerate. For a degenerate symbol ϕ, we have: Proposition 4.16. Let Ω be a bounded hyperconvex and good complete Reinhardt domain in CN , and let ϕ : Ω → Ω be an analytic function such that ϕ(Ω) ⊆ Ω is pluripolar. Then βN (Cϕ ) = 0. Recall that ϕ(Ω) is pluripolar when ϕ is degenerate (see [44, Proposition 2]); its closure is also pluripolar if it satisﬁes the condition (4.9). Proof. Let K = ϕ(Ω). By hypothesis, we have Cap (K) = 0. For every ε > 0, + let Kε = {z ∈ Ω ; dist (z, K) ≤ ε}. By Theorem 4.7, we have βN (Cϕ ) ≤ ΓN (Kε ). As limε→0 Cap (Kε ) = Cap (K) = 0 ([30, Proposition 4.7.1(iv)]), we get βN (Cϕ ) = 0. Remark 1. In [41, Section 4], we construct a degenerate symbol ϕ on the bi-disk D2 , deﬁned by ϕ(z1 , z2 ) = λθ (z1 ), λθ (z1 ) , where λθ is a lens map, for which β − (Cϕ ) > 0. For this function ϕ(D2 ) ∩ ∂D2 6= ∅ and hence ϕ(D2 ) is not a compact subset of D2 . Remark 2. In the one dimensional case, for any (non constant) analytic map ϕ : D → D, the parameter β(Cϕ ) = β1 (Cϕ ) is determined by its range ϕ(D), as shown by the formula: β(Cϕ ) = e−1/Cap [ϕ(D)] proved in [40]. This is no longer true in dimension N ≥ 2. In [42], we construct pairs of (degenerate) symbols ϕ1 , ϕ2 : D2 → D2 , such that ϕ1 (D2 ) = ϕ2 (D2 ) and: 1) Cϕ1 is not bounded, but Cϕ2 is compact, and even β2 (Cϕ2 ) = 0; 2) Cϕ1 is bounded but not compact, so β2 (Cϕ1 ) = 1, and Cϕ2 is compact, with β2 (Cϕ2 ) = 0; 3) Cϕ1 is compact, with 0 < β2 (Cϕ1 ) < 1, and Cϕ2 is compact, with β2 (Cϕ2 ) = 0. Acknowledgements. We thank S. Nivoche and A. Zeriahi for useful discussions and informations, and Y. Tiba, who send us his paper [51]. We than specially S. Nivoche, who carefully read a preliminary version of this paper. The third-named author is partially supported by the project MTM201563699-P (Spanish MINECO and FEDER funds).

24

References [1] L. Aizenberg and E. Liﬂyand, Hardy spaces in Reinhardt domains, and Hausdorff operators, Illinois J. Math. 53, no. 4 (2009), 1033–1049. [2] M. A. Alan, Hardy spaces on hyperconvex domains, Master thesis, Middle East Technical University, Ankara, Turkey (2003), https://etd.lib.metu.tr/upload/1046101/index.pdf [3] F. Bayart, Composition operators on the polydisc induced by affine maps, J. Funct. Anal. 260, no. 7 (2011), 1969–2003. [4] F. Bayart, D. Li, H. Queﬀélec and L. Rodríguez-Piazza, Approximation numbers of composition operators on the Hardy and Bergman spaces of the ball and of the polydisk, Math. Proc. Cambridge Philos. Soc., to appear (DOI: https://doi.org/10.1017/S0305004117000263). [5] E. Bedford and B. A. Taylor, A new capacity for plurisubharmonic functions, Acta Mathematica 149 (1982), 1–40. [6] E. Bedford and B. A. Taylor, Fine topology, Šilov boundary and (ddc )n , J. Funct. Anal. 72, no. 2 (1987), 225–251. [7] Z. Błocki, The complex Monge-Ampère operator in pluripotential theory, Course given at the Jagiellonian University in 1997/98, http://gamma.im.uj.edu.pl/ blocki/publ/ln/wykl.pdf [8] Z. Błocki, Equilibrium measure of a product subset of Cn , Proc. Amer. Math. Soc. 128, no. 12 (2000), 3595–3599. [9] S. Bochner, Classes of holomorphic functions of several variables in circular domains, Proc. Nat. Acad. Sci. U.S.A. 46, no.5 (1960), 721–723. [10] B. Carl and I. Stephani, Entropy, compactness and the approximation of operators, Cambridge Tracts in Mathematics 98, Cambridge University Press, Cambridge (1990). [11] J. A. Cima, C. S. Stanton and W. R. Wogen, On boundedness of composition operators on H 2 (B2 ), Proc. Amer. Math. Soc. 91, no. 2 (1984), 217–222. [12] J. A. Cima and W. R. Wogen, Unbounded composition operators on H 2 (B2 ), Proc. Amer. Math. Soc. 99, no. 3 (1987), 477–483. [13] J.-L. Clerc, Geometry of the Shilov boundary of a bounded symmetric domain, J. Geom. Symmetry Phys. 13 (2009), 25–74. [14] J.-L. Clerc, Geometry of the Shilov boundary of a bounded symmetric domain, Geometry, integrability and quantization, 11–55, Avangard Prima, Soﬁa (2009).

25

[15] C. C. Cowen and B. D. MacCluer, Composition operators on spaces of analytic functions, Studies in Advanced Mathematics, CRC Press, Boca Raton, FL (1995). [16] J.-P. Demailly, Mesures de Monge-Ampère et mesures pluriharmoniques, Math. Z. 194, no. 4 (1987), 519–564. [17] J.-P. Demailly, Potential theory in several complex variables, Course given at the CIMPA Summer School in Complex Analysis, Nice, France (1989), https://www-fourier.ujf-grenoble.fr/ demailly/manuscripts/ nice_cimpa.pdf. [18] J.-P. Demailly, Monge-Ampère operators, Lelong numbers and intersection theory, Complex analysis and geometry, 115–193, Univ. Ser. Math., Plenum, New York (1993), available at https://www-fourier.ujf-grenoble.fr/ demailly/manuscripts/ trento.pdf. [19] V. D. Erokhin, Best linear approximations of functions analytically continuable from a given continuum into a given region, with an appendix by A. L. Levin and V. M. Tikhomirov, Russ. Math. Surv. 23 (1) (1968), 93–135. [20] A. Guichardet. Leçons sur certaines algèbres topologiques, Gordon & Breach, Paris-London-New York, distributed by Dunod Editeur (1967). [21] K. T. Hahn, Properties of holomorphic functions of bounded characteristic on star-shaped circular domains, J. Reine Angew. Math. 254 (1972), 33–40. [22] K. T. Hahn and J. Mitchell. H p spaces on bounded symmetric domains, Trans. Amer. Math. Soc. 146 (1969), 521–531. [23] K. T. Hahn and J. Mitchell, H p spaces on bounded symmetric domains, Ann. Polon. Math. 28 (1973), 89–95. [24] L. A. Harris, Bounded symmetric homogeneous domains in infinite dimensional spaces, Proceedings on Inﬁnite Dimensional Holomorphy (Internat. Conf., Univ. Kentucky, Lexington, Ky., 1973), pp. 13–40, Lecture Notes in Math., Vol. 364, Springer, Berlin (1974). [25] L. K. Hua Harmonic analysis of functions of several complex variables in the classical domains, American Mathematical Society, Providence, R.I. (1963). [26] M. Jarnicki and P. Pﬂug, First steps in several complex variables: Reinhardt domains, EMS Textbooks in Mathematics, European Mathematical Society (EMS), Z¨rich (2008). [27] W. Kaup, A Riemann mapping theorem for bounded symmetric domains in complex Banach spaces, Math. Z. 183, no. 4 (1983), 503–529. 26

[28] M. Klimek, Extremal plurisubharmonic functions and L-regularity in Cn , Proc. Roy. Irish Acad. Sect. A 82, no. 2 (1982), 217–230. [29] M. Klimek, On the invariance of the L-regularity under holomorphic mappings, Zeszyty Nauk. Uniw. Jagiellon, Prace Mat. No. 23 (1982), 27–38. [30] M. Klimek, Pluripotential theory, London Mathematical Society Monographs, New Series 6, Oxford Science Publications, The Clarendon Press, Oxford University Press, New York (1991). [31] M. Koskenoja, Pluripotential theory and capacity inequalities, Ann. Acad. Sci. Fenn. Math. Diss. No. 127 (2002). [32] S. G. Krantz, Function theory of several complex variables, Pure and Applied Mathematics. A Wiley-Interscience Publication, John Wiley & Sons, Inc., New York (1982). [33] C. Laurent-Thiébaut, Théorie des fonctions holomorphes de plusieurs variables, Savoirs actuels, InterEditions/ CNRS Éditions (1997). [34] J.-S. Lee, J. J. Kim, C.-Y. Oh and S.-H. Park, Hyperconvexity and neighborhood basis of Reinhardt domains, Kyushu J. Math. 63, no. 1 (2009), 103–111. [35] D. Li and H. Queﬀélec, Introduction à l’étude des espaces de Banach Ű Analyse et probabilités, Cours Spécialisés 12, Société Mathématique de France (2004). [36] D. Li and H. Queﬀélec, Introduction to Banach Spaces: Analysis and Probability, Vol. 1, Cambridge studies in advanced mathematics 166, Cambridge University Press (2017). [37] D. Li and H. Queﬀélec, Introduction to Banach Spaces: Analysis and Probability, Vol. 2, Cambridge studies in advanced mathematics 167, Cambridge University Press (2017). [38] D. Li, H. Queﬀélec and L. Rodríguez-Piazza, On approximation numbers of composition operators, J. Approx. Theory 164 (4) (2012), 431–459. [39] D. Li, H. Queﬀélec and L. Rodríguez-Piazza, Estimates for approximation numbers of some classes of composition operators on the Hardy space, Ann. Acad. Scient. Fennicae 38 (2013), 547–564. [40] D. Li, H. Queﬀélec and L. Rodríguez-Piazza, A spectral radius formula for approximation numbers of composition operators, J. Funct. Anal. 160 (12) (2015), 430–454. [41] D. Li, H. Queﬀélec and L. Rodríguez-Piazza, Some examples of composition operators and their approximation numbers on the Hardy space of the bidisk, Trans. Amer. Math. Soc., to appear. 27

[42] D. Li, H. Queﬀélec and L. Rodríguez-Piazza, Composition operators with surjective symbol and small approximation numbers, in preparation. [43] B. MacCluer, Spectra of compact composition operators on H p (BN ), Analysis 4 (1984), 87–103. [44] Nguyen Thanh Van and W. Pleśniak, Invariance of L-regularity and Leja’s polynomial condition under holomorphic mappings, Proc. R. Ir. Acad. 84A, No. 2 (1984), 111–115. [45] S. Nivoche, Proof of a conjecture of Zahariuta concerning a problem of Kolmogorov on the ε-entropy, Invent. Math. 158, no. 2 (2004), 413–450. [46] W. Pleśniak, Siciak’s extremal function in complex and real analysis, Ann. Polon. Math. 80 (2003), 37–46. [47] E. A. Poletsky and M. I. Stessin, Hardy and Bergman spaces on hyperconvex domains and their composition operators, Indiana Univ. Math. J. 57, no. 5 (2008), 2153–2201. [48] W. Rudin, Function theory in the unit ball of Cn , Reprint of the 1980 edition, Classics in Mathematics, Springer-Verlag, Berlin (2008). [49] S. Şahin, Monge-Ampère measures and Poletsky-Stessin Hardy spaces on bounded hyperconvex domains, Constructive approximation of functions, 205–214, Banach Center Publ., 107, Polish Acad. Sci. Inst. Math., Warsaw (2015). [50] J. Siciak, Extremal plurisubharmonic functions in Cn , Ann. Polon. Math. 39 (1981), 175–211. [51] Y. Tiba, Shilov boundaries of the pluricomplex Green function’s level sets, Int. Math. Res. Not. IMRN 2015, no. 24 ( 2015), 13717Ű13727. [52] J.-P. Vigué, Le groupe des automorphismes analytiques d’un domaine borné d’un espace de Banach complexe. Application aux domaines bornés symétriques, Ann. Sci. École Norm. Sup. (4) 9, no. 2 (1976), 203–281. [53] H. Widom, Rational approximation and n-dimensional diameter, J. Approx. Theory 5 (1972), 342–361. [54] Ö. Yazici, Kolmogorov problem on widths asymptotics and pluripotential theory, Master thesis, SabancıUniversity, Istanbul, Turkey (2008), https://oyazici.expressions.syr.edu/wp-content/uploads/2014/ 07/tez-4-haz-2.pdf [55] V. Zakharyuta, Kolmogorov problem on widths asymptotics and pluripotential theory, Functional analysis and complex analysis, 171–196, Contemp. Math. 481, Amer. Math. Soc., Providence, RI (2009).

28

[56] V. Zakharyuta, Extendible bases and Kolmogorov problem on asymptotics of entropy and widths of some classes of analytic functions, Annales de la Faculté des Sciences de Toulouse Vol. XX, numéro spécial (2011), 211–239. [57] A. Zeriahi, Private Communication. Daniel Li Univ. Artois, Laboratoire de Mathématiques de Lens (LML) EA 2462, & Fédération CNRS Nord-Pas-de-Calais FR 2956, Faculté Jean Perrin, Rue Jean Souvraz, S.P. 18 F-62 300 LENS, FRANCE [email protected] Hervé Queffélec Univ. Lille Nord de France, USTL, Laboratoire Paul Painlevé U.M.R. CNRS 8524 & Fédération CNRS Nord-Pas-de-Calais FR 2956 F-59 655 VILLENEUVE D’ASCQ Cedex, FRANCE [email protected] Luis Rodríguez-Piazza Universidad de Sevilla, Facultad de Matemáticas, Departamento de Análisis Matemático & IMUS, Calle Tarfia s/n 41 012 SEVILLA, SPAIN [email protected]

29