Atomic decomposition of a weighted inductive limit
Transcripción
Atomic decomposition of a weighted inductive limit
RACSAM Rev. R. Acad. Cien. Serie A. Mat. VOL . 97 (2), 2003, pp. 325–337 Análisis Matemático / Mathematical Analysis Atomic decomposition of a weighted inductive limit J. Taskinen Dedicated to the memory of Klaus Floret Abstract. We study some structural questions concerning the locally convex space HV∞ , which consists of analytic functions on the open unit disc. We construct an atomic decomposition in this space, using a lattice of points of the unit disc which is more dense than a usual one. The coefficient space is a Köthe sequence space. We also prove that HV∞ is not nuclear. Descomposición atómica de un lı́mite inductivo ponderado Resumen. Estudiamos algunas cuestiones estructurales acerca del espacio localmente convexo HV∞ , que está formado por funciones analı́ticas en el disco unidad abierto. Construimos una descomposición atómica de este espacio, usando un retı́culo de puntos del disco unidad que es más denso que el usual. También demostramos que HV∞ no es nuclear. 1. Introduction In the paper [10] we introduced the space HV∞ := HV∞ (D) consisting of analytic functions on the open unit disc. This space is slightly larger than the classical Banach space H ∞ of bounded analytic functions. The motivation of HV∞ is the fact that it behaves much better than H ∞ with respect to Bergman and Szegö projections and the harmonic conjugation operator. The space HV∞ is aimed to be a substitute for H ∞ in situations where the continuity of the above mentioned mappings is critical. In [10] we also proved that this space is a some sense the smallest possible substitute. In this work we continue the study of HV∞ by showing that it admits an atomic decomposition: every analytic function in this space can be presented as a linear combination of “atoms” defined using the standard Bergman kernel. The coefficients form sequences belonging to a natural Köthe sequence space. For details, see Theorem 1. While atomic decompositions are interesting in itself in harmonic analysis, our results also give a representation of HV∞ as a complemented subspace of a Köthe–Schwartz coechelon space. The atom functions are defined using a lattice of points of the unit disc. Our choice of the lattice is not a typical one with essentially constant hyperbolic distances between the lattice points. In the present work the lattice is much more dense. This leads to many simplifications in proofs. Atomic decompositions were first studied by Coifman and Rochberg in the setting of Bergman spaces on the disc, see [3] and especially the book [12], Chapter 4, for a simple presentation. It is intuitively quite Presentado por José Bonet. Recibido: 27 de Noviembre de 2002. Aceptado: 25 de Julio de 2003. Palabras clave / Keywords: Weighted spaces of holomorphic functions, atomic decomposition, nuclear spaces. Mathematics Subject Classifications: Primary 46A13; Secondary 46E10, 46A11, 30H05 c 2003 Real Academia de Ciencias, España. 325 J. Taskinen clear that our method could be used also in the classical case of Bergman spaces to make many proofs shorter or even trivial. ∞ Before proceeding to the atomic decomposition let us recall from [10] the definition R of HV . By dA we denote the normalized 2–dimensional Lebesgue measure on D; moreover, m(B) := B dA for a measurable subset B of D. Definition 1 By V we denote the set of radial weight functions v : D → R+ i.e. positive bounded continuous functions, which satisfy for every n ∈ N sup v(z)| log(1 − |z|)|n ≤ Cn (1) z∈D and moreover p 1 − |z| ≤ v(z) ≤ 1 for all z ∈ D. (2) Notice that (2) is not explicitely required in [10]; nevertheless the spaces in the next definition are the same as those in [10]. Definition 2 We define the locally convex space n o HV∞ := f : D → C analytic kf kv := sup |f (z)|v(z) < ∞ for all v ∈ V . (3) z∈D The topology of HV∞ is determined by the uncountable family {k·kv |v ∈ V } of seminorms. It is a complete countable inductive limit of Banach spaces: HV∞ = indk→∞ Hv∞k , where vk (z) := min{1, | log(1−|z|)|−k } and n o Hv∞k := f : D → C analytic kf kvk < ∞ . (4) It is a Schwartz space; in particular, the bounded and precompact subsets coincide. It is not so surprising that the weight system is not “steep” enough to make the space a nuclear one. We consider this question in Section 4. The dual of HV∞ (with respect to the topology of uniform convergence on bounded sets) is the space n 1 HW := f : D → C analytic kf kk := Z o |f (z)|| log(1 − |z|)|k dA(z) < ∞ for all k ∈ N . (5) z∈D R 1 The dual pairing is the usual hf, gi := D f ḡdA. The spaces HV∞ and HW are reflexive, hence, the duality 1 0 ∞ 1 (HW )b = HV also holds. The dual HW is a Fréchet–Schwartz space, whose topology is determined by the sequence of seminorms (k · kk )∞ k=1 . For more details, especially for a proof of the duality, see [10]. We define K(z, w) := 1/(1 − z w̄)2 , where z, w ∈ D, and we denote by R the Bergman projection Z Rf (z) = K(z, w)f (w)dA(w). (6) D ∞ ∞ According to [10], R is a continuous projection from L∞ V onto HV , where LV is defined as in (3) but “measurable” replacing “analytic” and “ess sup” replacing “sup”. 326 Atomic decomposition To fix a constant later we remark that R also maps the space L∞ continuously into L∞ v1 (definition analogous to (4) ); see [10]. Moreover, R is self–dual with respect to the dual pairing h·, ·i. Let us thus show that R is also a bounded operator from Z o n L11 := f : D → C measurable kf k1 := |f (z)|(1 + | log(1 − |z|)|)dA(z) < ∞ (7) z∈D into L1 := L1 (D, dA): for f ∈ L11 and g ∈ L∞ , Z Z f (ζ)ḡ(z) dA(ζ)dA(z) |hRf, gi| = ¯2 (1 − z ξ) D D Z Z |f (ζ)| D D ≤ g(z) dA(z) dA(ζ) (1 − z̄ξ)2 Z |f (ζ)Rg(ζ)|dA(ζ) = D Z ≤ C |f (ζ)|(1 + | log(1 − |ζ|)|)dA(ζ) ≤ C 0 . D So, let us denote by C0 the positive constant such that (denoting the norm of L1 by k · k0 ) kRgk0 ≤ C0 kgk1 (8) for all g ∈ L11 . Concerning other notations and terminology, for analytic function spaces we refer to the book of Zhu, [12], and for locally convex spaces we mention [6], [7] and [8] and also the paper [10]. 2. Preliminaries for the atomic decomposition. We next turn to the construction of the atomic decomposition. Atomic decompositions for analytic function spaces on the unit disc were first introduced by Coifman and Rochberg, see [3]. A simple presentation of the classical result is given in [12], Chapter 4. By an “atomic decomposition” of some analytic function space we roughly mean a way to write an analytic function as an infinite linear combination of simple building blocks, “atoms”, in such a way that the complex coefficients for a sequence belong to a relevant sequence space. For our function space HV∞ the “atom” functions, i.e. the building blocks, will be defined using the Bergman kernel, as usual. Another essential ingredient is a suitable decomposition of the unit disc D. Our decomposition is finer than the usual one, which consists of sets with essentially constant hyperbolic area; see [3] or [12]. Our choice actually makes many preliminary lemmas nearly trivial to prove. It is very probable that this could be used to simplify the proofs also in the classical case of Bergman spaces Ap (D). We are aware of the fact that on the other hand our choice of the lattice is not an optimal, i.e. small, one. Hence we also do not particularly try to achieve optimal constants in various inequalities; the estimates may be quite crude in some cases. Definition 3 Let us fix the constant M = (C0 + 1)10000, where C0 is as in (8). For all m ∈ N, define first rm = 1 − m−1/3 ; then define for every m, for every m0 := 0, 1, 2, . . . , M rm,m0 := rm + m0 (rm+1 − rm ) M (9) 327 J. Taskinen and θ(m, k) := 2πk/[M m4/3 ] for k = 0, 1, . . . , [M m4/3 ]. (Here [a] stands for the largest integer strictly smaller than a + 1.) Define a bijection ρ from N onto the set {(m, m0 , k) | m ∈ N, m0 = 0, 1, 2, . . . , M − 1, k = 0, 1, 2, . . . , [M m4/3 ] − 1}. For every n ∈ N, pick up m, m0 and k such that n = ρ−1 (m, m0 , k) and define Dn := {z = reiθ ∈ D | rm,m0 ≤ r ≤ rm,m0 +1 and θ(m, k) ≤ θ ≤ θ(m, k + 1)}. (10) We clearly have [ Dn = D, (11) n and moreover, the set formed by the numbers z ∈ D belonging to more than one of the sets Dn has Lebesgue measure 0. Lemma 1 For every n the Euclidean diameter of Dn satisfies diamDn ≤ inf z∈Dn 20(1 − |z|)4 M (12) If n is given and m, m0 and k are such that n = ρ−1 (m, m0 , k), then m0 + 1 (rm+1 − rm ) ≥ 1 − rm+1 = (m + 1)−1/3 . (13) inf (1 − |z|) = 1 − rm,m0 +1 = 1 − rm + z∈Dn M P ROOF. On the other hand, the arguments of the points of Dn may vary at most from θ(m, k) to θ(m, k + 1), hence at most 2π/(M m4/3 ). The radii may vary from rm,m0 to rm,m0 +1 , which amounts to a change of (see (9) ) M −1 (rm+1 − rm ) = 1 1 m−4/3 1 − 1 + 1/3 ≤ . 1− 1/3 M M (m + 1) m (14) This implies the claim. Corollary 1 For every analytic function f on D, for all n, we have |f (z) − f (w)| ≤ for all z, w, λ ∈ Dn . 20 (1 − |λ|)4 sup |f 0 (ζ)| M ζ∈Dn (15) The proof is a direct application of the previous lemma and the mean value theorem. The crucial step for our main result is to strengthen the classical approximation lemma (e.g. [12], Lemma 4.3.4) using our new decomposition of D. Notice that the (relatively easy) proof is based on the density of our decomposition. p Lemma 2 Let v ∈ V , in particular, assume 1 − |z| ≤ v(z) ≤ 1. For all n ∈ N, let λn be an interior point of Dn . R 1 a) For all g ∈ HW with |g|dA ≤ 1 we have D ∞ Z X n=1 Dn |g(z) − g(λn )| 1 dA(z) ≤ 200M −1 v(z) (16) b) For all f ∈ HV∞ with |f (z)| ≤ 1/v(z) on D, and for all n ∈ N, we have sup |f (z) − f (λn )| ≤ 200M −1 sup (1 − |z|)1/4 . z∈Dn 328 z∈Dn (17) Atomic decomposition P ROOF. In the case a) we estimate using Corollary 1 (choose λ := z there) and (2) ∞ Z X 1 |g(z) − g(λn )|dA(z) v(z) n=1 Dn ∞ Z 20 X ≤ (1 − |z|)−1/2 (1 − |z|)4 sup |g 0 (ζ)|dA(z) M n=1 Dn ζ∈Dn (18) Since sup (1 − |ζ|) ≤ (1 + 1/10)(1 − |z|) for all z ∈ Dn (e.g.by Lemma 1), we can bound (18) by ζ∈Dn ∞ 40 X 40 m(Dn ) sup (1 − |z|)7/2 |g 0 (z)| ≤ sup (1 − |z|)7/2 |g 0 (z)| M n=1 M z∈D z∈Dn (19) g(ζ) dA(ζ) (1−z ζ̄)2 once under the integral sign to Z 80 g(ζ)ζ̄ 7/2 dA(ζ) sup (1 − |z|) 3 M z∈D (1 − z ζ̄) (20) Here we differentiate the reproducing formula g(z) = R D bound (19) by D 80 ≤ M Z |g(ζ)|dA(ζ) ≤ 80 . M (21) D As for b), let f be as indicated. Again by Corollary 1 20 sup (1 − |z|)4 |f 0 (z)| M z∈Dn z∈Dn 20 ≤ sup (1 − |z|)7/2 |f 0 (z)| sup (1 − |z|)1/4 , M z∈Dn z∈Dn sup |f (z) − f (λn )| ≤ and as in the previous case we see that the first factor is bounded by Z Z 40 40 200 |f (ζ)|dA(ζ) ≤ 1/v(ζ)dA(ζ) ≤ . M M M D (22) D The part b) of the above result immediately implies. Corollary 2 Let v ∈ V . If kf kv ≤ 1, we have |f (λ)| ≤ for every λ, z ∈ Dn . 3. 2 v(z) (23) Main result. In this section we construct an atomic decomposition for the space HV∞ . We shall show that every element f of HV∞ can be presented as a linear combination ∞ X α(n)m(Dn ) f (z) = , (1 − λ̄n z)2 n=1 (24) 329 J. Taskinen where λn ∈ Dn and the coefficient sequence (α(n))∞ n=1 belongs to a Köthe sequence space K∞ (A). Conversely, for every sequence (α(n))∞ n=1 ∈ K∞ (A), the analytic function defined by (24) belongs to HV∞ . However, given f ∈ HV∞ , the sequence (α(n))∞ n=1 is not unique. (The atoms do not form a Schauder basis.) We now consider the sequence space relevant to the atomic decomposition. Since HV∞ is a countable inductive limit of Banach spaces, the same is also true for the sequence space. Definition 4 Let us fix, for all n ∈ N, an interior point λn of Dn . For all k ∈ N, let ak be the sequence of positive real numbers with ak (n) := | log(1 − |λn |)|k , n ∈ N. (25) Denote A := {ak | k ∈ N}, define the Köthe co–echelon space K∞ (A) corresponding to A as in [1], Section 1. For a quick definition of K∞ (A), let us denote by VN the set of all bounded positive sequences ṽ satisfying ṽ(n) ≤ Ck /ak (n) for all k, n ∈ N. (26) Then a sequence α = (α(n))∞ n=1 of complex numbers belongs to K∞ (A) if and only if for every ṽ ∈ VN one can find a C > 0 such that sup |α(n)|ṽ(n) ≤ C. (27) n∈N The (non–metrizable) topology of K∞ (A) is defined by the seminorms kαkṽ := sup |α(n)|ṽ(n). (28) n∈N Let us next define three operators. Let the sequence (λn )∞ n=1 be as above. Let f be analytic on D and be a sequence of complex numbers. Set let α = (α(n))∞ n=1 Sf (z) := ∞ X m(Dn )f (λn ) , (1 − λ̄n z)2 n=1 (29) T (α) := ∞ X α(n)m(Dn ) , (1 − λ̄n z)2 n=1 (30) (Qf )(n) := f (λn ) for n ∈ N. (31) Hence, formally S = T Q. These operators have the following continuity properties. Lemma 3 The operators S : HV∞ → HV∞ , T : K∞ (A) → HV∞ and Q : HV∞ → K∞ (A) are continuous. 1 P ROOF. 1◦ Consider T . To prove the continuity it suffices to take an arbitrary bounded set B ∈ HW and ◦ show that a neighbourhood of 0 in K∞ (A) is mapped into the polar B of B by T , where B ◦ := {g ∈ HV∞ | |hf, gi| ≤ 1 for every f ∈ B}. (32) R 1 We first claim that B is contained in a set G := {f ∈ HW | |f |v(z)−1 ≤ 10} for a v ∈ V . To find v we use the boundedness of B to pick the numbers Ck ≥ 1 such that B ⊂ Ck Uk for every k; here Uk is the 1 closed unit ball of the continuous seminorm k · kk of HW . This means that every f ∈ B satisfies, for all k, Z |f || log(1 − |z|)|k dA ≤ Ck . (33) D 330 Atomic decomposition Define v by v(z) := P ∞ k=0 Since ∞ X 1 . (34) k!−1 4−k Ck−1 | log(1 − |z|)|k 1 k!−1 4−k | log(1 − |z|)|k = e− 4 log(1−|z|) = (1 − |z|)−1/4 , k=0 we get the estimate v(z) ≥ (1 − |z|)1/4 . (35) We clearly have sup v(z)| log(1 − |z|)|k ≤ k!4k Ck for every k, hence, v ∈ V . Moreover, it follows easily z from (33) and (34) that every function in B belongs to the set G: Z |f |v −1 dA ≤ ∞ X k=0 D Z k!−1 Ck−1 |f || log(1 − |z|)|k dA = ∞ X k!−1 ≤ 10. (36) k=0 D Set ṽ(n) := v(λn ) for n ∈ N; we then Rhave (ṽ(n))∞ n=1 ∈ VN , by definitions. For kαkṽ ≤ C we thus have, by the reproducing formula f (λn ) = f (ζ)(1 − λn ζ̄)−2 dA(ζ), ∞ X α(n)m(Dn )f¯(λn ) |hT α, f i| = n=1 ≤ ∞ X |α(n)|ṽ(n)v(λn )−1 m(Dn )|f (λn )| n=1 ≤ kαkṽ ∞ X v(λn )−1 m(Dn )|f (λn )| . (37) n=1 We prove in a moment that 1/v(λn ) < C/v(z) (38) for every n, for every z ∈ Dn . Using this, Lemma 2.b), the definition of the set G and (35), we can estimate (37) by a constant times XZ XZ |f (z)|v(z)−1 dA + (1 − |z|)4 v(z)−1 dA ≤ C. (39) n Dn n Dn The statement (38) follows from the smallness of diamDn . First, ∞ X ∂v(r)−1 = k!−1 k4−k Ck−1 | log(1 − r)|k−1 (1 − r)−1 ≤ C(1 − r)−1−1/4 , ∂r k=1 This, diamDn ≤ C(1 − |z|)4 (Lemma 1) and the mean value theorem imply |1/v(z) − 1/v(λn )| ≤ C(1 − |z|)2 ≤ C 0 ≤ C 0 /v(z). Hence, (38) follows. 2◦ Concerning Q, let the weight sequence ṽ ∈ VN be given. We define v ∈ V as follows. First let (ρj )∞ j=1 , 0 ≤ ρj < 1, be the unique strictly increasing sequence which satisfies: for every j there exists an n such that ρj = |λn |. Let us define the subset Nj of N by Nj = {n | |λn | = ρj }. For every j, define for all z ∈ D with |z| = ρj , v(z) = sup ṽ(n). (40) n∈Nj 331 J. Taskinen For every r ∈ [0, 1[, extend v(r) piecewise linearly, and then extend v radially to D. We show that v ∈ V . If z ∈ D satisfies |z| = ρj for some j, and k ∈ N is given, then v(z) = sup ṽ(n) ≤ sup Ck /ak (n) n∈Nj = n∈Nj sup Ck | log(1 − |λn |)|−k = Ck | log(1 − |z|)|−k . (41) n∈Nj Concerning other numbers z, Lemma 1 implies |ρj − ρj+1 | ≤ Cj (1 − ρj )4 , hence, for every k we have | log(1 − ρj )| ≤ | log(1 − ρj+1 )| ≤ ck | log(1 − ρj )| (42) for a constant ck > 0. So, if z satisfies ρj < |z| < ρj+1 , then the piecewise linearity of v and (41) and (42) imply v(z) ≤ max{v(ρj ), v(ρj+1 )} ≤ Ck0 | log(1 − ρj+1 )|−k ≤ Ck0 | log(1 − |z|)|−k . (43) Hence, v ∈ V . Now (40) and the definition of Q imply kQf kṽ ≤ Ckf kv , hence Q is continuous. 3◦ Finally, we have S = T Q, hence also S is continuous. The following fact can now be proved using a result of Garnir, De Wilde and Schmets to be found in [5], p. 346. Lemma 4 The operator S : HV∞ → HV∞ is invertible . P ROOF. We want to show that the operator A := I − S (i) is bounded (i.e. maps a neighbourhood of 0 of HV∞ into a bounded set), and (ii) for some neighbourhood U ⊂ HV∞ of 0, A(U ) ⊂ U/2. Then S is invertible by [5], p. 346. Let us take a weight as small as p w(z) := 1 − |z|. (44) Then, for sure, w ∈ V . Let U ⊂ HV∞ be the unit ball of k · kw : n o U := f ∈ HV∞ sup |f (z)|(1 − |z|)1/2 ≤ 1 ; (45) z∈D We want to show that 1 Z Af (z)ḡ(z)dA(z) ≤ 2 (46) D for every f ∈ U , g in the unit ball of L11 (see (8)), that is, Z n o g ∈ Ũ := h ∈ L1loc (D, dA) khk1 := |h(z)|(1 + | log(1 − |z|)|)dA(z) ≤ 1 . (47) z∈D Here L1loc (D, dA) is the space of locally integrable functions on D. Assuming (46) we show that (i) and (ii) hold. As for (i), let us pick up an arbitrary weight v ∈ V ; to 1 prove the boundedness of A it suffices to find a k such that for f ∈ HV∞ and g ∈ HW with kf kv ≤ 1 and kgkk ≤ 1 we have |hAf, gi| ≤ C. (48) p But v(z) ≥ 1 − |z|, see (2) in the definition of V . Hence, f ∈ U . Moreover, for every k ∈ N, k ≥ 2, 1 the subset {kgkk ≤ 1} ⊂ HW is smaller than the set Ũ above. Hence, (48) follows from (46). 332 Atomic decomposition Concerning (ii), the identification of the norm of L∞ (D, dA) as the dual norm of L1 (D, dA) implies the following: R An analytic function ϕ belongs to U/2, if | D ϕḡ| ≤ 1/2 for every Z n o g ∈ U ◦ := h ∈ L1loc (D) |h(z)|(1 − |z|)−1/2 dA(z) ≤ 1 . (49) D But U ◦ is a much smaller set than Ũ , hence (ii) follows from (46). So we want to prove (46). Let now f and g be given. Since Af is analytic and the Bergman projection is self–dual, the identities Z Z Af ḡ = (RAf )ḡ = hAf, Rgi (50) D hold formally. Moreover, by (8), kg̃k0 := kRgk0 ≤ C0 kgk1 ≤ C0 (51) We thus have Z Z ∞ X m(Dn )f (λn )g̃(λn ) Af ḡ = hAf, g̃i = f (z)g̃(z)dA(z) − D D = ∞ Z X n=1 n=1 ∞ X f (z) − f (λn ) g̃(z)dA(z) + Dn n=1 Z f (λn ) g̃(z) − g̃(λn ) dA(z). (52) Dn Using Lemma 2.b) and (51) we get the following bound for the first sum in (52): Z ∞ X |g̃(z)|dA(z) sup |f (z) − f (λn )| n=1 z∈Dn ≤ 200M −1 Dn Z |g̃(z)|dA(z) ≤ 2000C0 M −1 ≤ 1/5. (53) D To estimate the second term of (52) from above, we first use Corollary 2 to bound |f (λn )| by z ∈ Dn . Then Lemma 2 a) directly gives the bound ∞ Z X 2 400C0 |g̃(z) − g̃(λn )|dA(z) ≤ . w(z) M n=1 Dn 2 w(z) for any (54) This completes the proof. We now formulate the main result as follows. We denote m(n) := m(Dn ); this number can be estimated, if necessary, from the definition of the sets Dn after (9). Recall that λn was chosen in Lemma 2: it is an interior point of the set Dn . The space K∞ (A) was defined in the beginning of Section 3. Theorem 1 For every (α(n))∞ n=1 ∈ K∞ (A) the function f (z) = ∞ X α(n)m(n) (1 − λ̄n z)2 n=1 (55) belongs to HV∞ (D), and the mapping (α(n))∞ n=1 7→ f is continuous between these two spaces. For every analytic function f ∈ HV∞ (D) there exists a complex sequence (α(n))∞ n=1 ∈ K∞ (A) such that ∞ X α(n)m(n) f (z) = . (56) (1 − λ̄n z)2 n=1 ∞ The mapping f 7→ (α(n))∞ n=1 can be made linear and continuous from HV (D) into K∞ (A). 333 J. Taskinen P ROOF. The first statement follows from the continuity of T . For the second one, given f , choose g = S −1 f and then define the sequence α = Qg, i.e. α(n) = g(λn ). We then obtain (56) from (29) and (31). The continuity statement follows from the continuity of Q and Lemma 4. Finally, the operator P := QS −1 T is a continuous projection in the space K∞ (A); we have P 2 = P . This implies Proposition 1 The space HV∞ is isomorphic to a complemented subspace of the Köthe sequence space K∞ (A). 4. The space is not nuclear. Intuitively it is quite clear that the space HV∞ cannot be nuclear, since the weight system is not “steep” enough. Two arbitrary weights differ from each other at most by a factor which grows only logarithmically on the boundary of D. We give a proof for the nonnuclearity by extracting a subspace which is isomorphic to a nonnuclear Köthe coechelon space. The method of the proof may have some independent interest. It resembles the construction [2], Section 2. However, in [2] the weight system was non–radial, and the essential phenomena occurred on the boundary of the disc. In the present work the weights are radial; consequently, we have to play also with the radii of the points of D. More precisely, the interesting and crucial things happen in the interior points zn := eiθn (1 − (2n)−20 ) of the subdomains Dn (to be chosen below). Of course there is a lot of freedom in the construction; many other points could be used as well. Notice that the space K∞ (A) above is not nuclear; however, the results of Section 3 do not imply that HV∞ contains a subspace isomorphic to K∞ (A). The operator T need not be injection, since Q is not a surjection. Proposition 2 The space HV∞ is not nuclear, since it contains a subspace isomorphic to the Köthe coechelon space K∞ (M ), which is defined as K∞ (A) in the beginning of Section 3, but ak (n) replaced by µk (n) := (log n)k (57) So, a sequence α belongs to K∞ (M ), if and only if sup |αn |(log n)−k < ∞ for some k. It is well n known that K ∞ (M ) is not nuclear, see e.g. [9], Theorem 6.1.2. P ROOF. Let θn := 1/n and Jn := [θn − εn , θn + εn ] for all n ∈ N, where εn := 2−9 n−4 . Notice that the sets Jn are mutually disjoint. Choose the function ϕn : [0, 2π] → R+ such that ϕn (θ) = 1 for θ ∈ Jn , and ϕn (θ) = (2n)−4 for θ ∈ / Jn . Let en be an analytic function on the disc defined by en (z) := exp 1 Z2π eiθ + z log ϕn (θ)dθ . iθ 2π e −z 0 The radial limits (on the boundary of the disc) of the functions en exist a.e. (we denote them by the same symbol), and we have |en (eiθ )| = ϕn (θ) for a.e. θ. Compare with [2] and the references therein. Let us define Dn := {z ∈ D | |z − eiθn | < 1/n4 } , Cn := D \ Dn . We formulate a lemma containing some crucial estimates. Lemma 5 1◦ For all z ∈ D, all n, we have |en (z)| ≤ 1. 2◦ Fix a z ∈ D; if n is such that | arg z − θn | = min {| arg z − θm |}, then |en (z)| ≤ 4n−4 (1 − |z| + m∈N n−4 )−1 . Moreover, if m 6= n, then |em (z)| ≤ 2−3 m−2 . 3◦ For every n we have |en (zn )| ≥ 1/2, where zn := eiθn (1 − (2n)−20 ). 334 Atomic decomposition P ROOF. The statement 1◦ follows immediately from the maximum principle, since the moduli of the radial limits of every en are at most 1. Concerning 2◦ , if z and n are as in the assumption, the Jensen inequality implies (as in [2]) |en (z)| = Z2π iθ 1 e +z Re log ϕ (θ)dθ exp n 2π eiθ − z = 1 Z2π 1 − |z|2 log ϕ (θ)dθ exp n 2π |eiθ − z|2 0 0 ≤ 1 2π Z2π 1 − |z|2 ϕn (θ)dθ. |eiθ − z|2 (58) 0 We may assume |z| ≤ 1 − n−4 ; otherwise there is nothing to prove. We have |z − eiθ | ≥ n−4 , hence, |z − eiθ | ≥ max(|z − eiθ |, n−4 ) ≥ max(1 − |z|, n−4 ) ≥ (1 − |z| + n−4 )/2. So we have 1 − |z|2 1 − |z|2 4 ≤ ≤ . iθ 2 iθ |e − z| (1 − |z|)|e − z| 1 − |z| + n−4 Dividing the integration interval [0, 2π] in (58) into the parts Jn and [0, 2π] \ Jn , we thus find that the first part yields the estimate 2 , 4 n (1 − |z| + n−4 ) since the lenght of Jn is smaller than 2−1 n−4 . The second part has the smaller bound (2n)−4 , since ϕn = (2n)−4 on [0, 2π] \ Jn . In the case m 6= n we again have |em (z)| ≤ 1 2π Z2π 1 − |z|2 ϕm (θ)dθ. |eiθ − z|2 (59) 0 Dividing the integration interval into the parts Jm and its complement, on Jm we have |eiθ − z| ≥ 1 |n − m| 1 1 . − = 2 m n 2mn This is always larger than the number 1/(4m2 ), as seen by considering the cases n ≤ 2m and n > 2m separately: in the former case one uses |n − m| ≥ 1 and 2mn ≤ 4m2 , and in the latter |n − m| ≥ n/2 and a cancellation of n. Taking into account the length 2−9 m−4 of the interval Jm , this part yields an estimate 2−7 m−2 in (59). The rest of the integral is bounded by (2m)−4 , since ϕm = (2m)−4 on [0, 2π] \ Jm . For the lower bound 3◦ , we have log ϕn = 0 on Jn and log ϕn = −4 log(2n) < 0 on the complement of Jn , hence Z 1 1 − |zn |2 |en (zn )| = exp log ϕ (θ)dθ n 2π |eiθ − zn |2 [0,2π]\Jn ≥ exp − 4 2π Z (2n)−20 log(2n)dθ n−18 [0,2π]\Jn ≥ exp(−2−10 n−1/2 ) ≥ 1 . 2 (60) 335 J. Taskinen We return to the proof of the proposition. We define the mapping ψ from K∞ (M ) into HV∞ by ψ : ∞ P (αn )∞ αn en . We claim that the mapping is an isomorphism onto its image. Since the spaces are n=1 7→ n=1 strong duals of Fréchet–Schwartz spaces, it suffices to show that, for some constants ck , Ck > 0, ck sup |αn |(log n)−k ≤ k X αn en kvk ≤ Ck sup |αn |(log n)−k n∈N (61) n∈N −1/2 for all k. Since the expression µ((αn )∞ is obviously a continuous seminorm on n=1 ) := sup |αn |n n K∞ (M ), we may assume that µ((αn )) ≤ 1, i.e. |αn | ≤ n1/2 for all n in (61). We may moreover assume that |αn | ≥ 1 there, for all n. To prove the left hand side inequality, choose N such that |αN |(log(N ))−k ≥ 1 sup |αn |(log n)−k . 2 n∈N (62) We then have, by 3◦ of Lemma 5 above, and by |αn | ≥ 1, 1 −k 20 (log(2N ))−k , 2 |αN eN (zN )|vk (zN ) ≥ (63) and on the other hand X |αn en (zN )|vk (zN ) ≤ n6=N Here P X n1/2 2−3 n−2 20−k (log(2N ))−k (64) n6=N n−3/2 2−3 ≤ (1 + n R∞ x−3/2 )2−3 ≤ 3/8. Hence, combining (63) and (64) and using the triangle 1 inequality we get ∞ X X 1 αn en (zN )vk (zN ) ≥ |αN eN (zN )|vk (zN ) − αn en (zN )vk (zN ) ≥ |αN eN (zN )|vk (zN ), 8 n=1 n6=N hence, by (62) and (63), we get 20−k sup |αn |(log n)−k ≤ 2 · 20−k |αN |(log N )−k ≤ C|αN eN (zN )|vk (zN ) n ∞ X X ≤ C 0 αn en (zN )vk (zN ) ≤ C 0 αn en vk n=1 We finally prove the right hand side of (61). Let N still be as above, let z ∈ D and let n be such that | arg z − θn | = min {| arg z − θm |}. By 2◦ of Lemma 5 we have (since always vk (z) ≤ 1 ) m∈N X |αm em (z)|vk (z) ≤ m6=n ≤ X X |αm |m−2 m6=n −3/2 Ck m |αm |(log m)−k ≤ Ck0 |αN |(log N )−k . (65) m6=n Concerning the nth term, for |z| ≥ 1 − n−3 we have by 1◦ vk (z)|αn en (z)| ≤ vk (1 − n−3 )|αn en (z)| ≤ |αn |vk (1 − n−3 ) ≤ 3|αn |(log n)−k ≤ 3|αN |(log N )−k 336 Atomic decomposition For |z| ≤ 1 − n−3 we have vk (z)|αn en (z)| ≤ ≤ 4|αn | 4|αn | ≤ 4 −3 n4 (1 − |z| + n−4 ) n (n + n−4 ) 4|αn | ≤ Ck |αn |(log n)−k ≤ Ck |αN |(log N )−k . n This and (65) imply that the right hand side of (61) holds. Acknowledgement. The author is grateful for José Bonet for showing him the reference [5], see Lemma 4. The author also wishes to thank the referee for the careful reading of the paper and for some remarks and corrections. The research is partially supported by the Academy of Finland. References [1] Bierstedt, K.D., Meise, R., Summers, W. (1982). A projective description of weighted inductive limits, Trans. Amer. Math. Soc., 272 (1), 107–160. [2] Bonet, J., Taskinen, J. (1995). The subspace problem for weighted inductive limits of spaces of holomorphic functions, Michigan Math. J., 42, 259–268. [3] Coifman, R. (1980). Rochberg, R.:Representation theorems for Hardy spaces, Astérisque, 77, (1980), 11–66. [4] Garnett, J. (1981). Bounded Analytic Functions, Academic Press, New York. [5] Garnir, H.G, De Wilde, M., Schmets, J. (1968). Analyse Fonctionelle I, Birkhäuser Verlag, Basel Stuttgart., [6] Jarchow, H. (1981). Locally Convex Spaces, Teubner, Stuttgart. [7] Köthe, G. (1983). Topological Vector Spaces, Vol. 1. Second edition. Springer Verlag, Berlin–Heidelberg–New York. [8] Pérez Carreras, P., and Bonet, J. (1987). Barreled Locally Convex Spaces, North–Holland Mathematics Studies 131. North–Holland, Amsterdam. [9] Pietsch, A. (1972). Nuclear Locally Convex Spaces, Ergebnisse der Mathematik und ihrer Grenzgebiete 66, Springer. [10] Taskinen, J. On the continuity of the Bergman and Szegö projections, Houston J. Math., to appear. [11] Wojtaszczyk, P. (1991). Banach Spaces for Analysts, Cambridge University Press. [12] Zhu, K. (1995). Operator Theory in Function Spaces, Marcel Dekker, New York. J. Taskinen Department of Mathematics University of Joensuu P.O.Box 111 FIN–80101 Joensuu. Finland [email protected] 337