Arxiv:2010.02606V2 [Math.FA] 7 Oct 2020 Ohfo Nasrc on Fve Sfrcnrt Ucinad( Space and the Therein

Home , Ultrabornological space

arXiv:2010.02606v2 [math.FA] 7 Oct 2020 ohfo nasrc on fve sfrcnrt ucinad( space and the therein. function that references concrete proved the for and as [14] article view survey of the point see abstract spaces; an from both s oacmlmne usaeof subspace complemented a to futaonlgclsae sntncsaiyaanultrabornologic again ( necessarily projectiv when sp not the characterizing such is as o when matter spaces theory non-trivial determine ultrabornological a the to is of or this important that theorems is Note it graph functional bornological. [30], closed apply functor and to limit mapping able projective open be Wilde’s to De order as In functions. smooth creasing pcsaedfie sflos Let follows. as defined are spaces lrbrooia.Ltro adva[7 hwdta nfact in that showed [27] Valdivia on Later ultrabornological. n egnohuighmlgclmtos[23]. methods homological using Wengenroth and fdsrbtos h pc fra nltcfntosadtespac the and functions analytic real of space the distributions, of ( 12ZG921N. frames. Gabor transform; Fourier short-time LB P ⊗ b ntels ato i otrltei 1,Ca.2 Theorème 16, 2, Chap. [17, thesis doctoral his of part last the In nti ril esuywihe ( weighted study we article this In onal rjcielmt fcutbeidcielmt fBnc spa Banach of limits inductive countable of limits projective Countable .Ny rtflyakolde upr yFOVanee thro FWO-Vlaanderen by support acknowledges po the gratefully through Neyt FWO-Vlaanderen by L. supported was Debrouwere A. phrases. and words Key 2010 s ′ ieetpofo h atthat fact the of proof different A . -pcs rs aual nfntoa nlss lsia examples Classical analysis. functional in naturally arise )-spaces, WEIGHTED ahmtc ujc Classification. Subject Mathematics npriua,w hwta h utpirsaeo h pc fFour Fou of of space space the the of of one not. space the is multiplier ar whereas spaces the ultrabornological, such is that functions when show characterize we result, particular, Gelf above In of the spaces using of multiplier by context the and, determine the we to Furthermore, ultrabornological functions. is functions smooth increasing ytm hsgnrlzsGohnic’ lsia eutta h s the that result w classical a Grothendieck’s and Taylor) o generalizes terms and This in Meise ultrabornological system. Braun, are of spaces such sense when the characterize We (in function weight a via Abstract. UCIN N UTPIRSPACES MULTIPLIER AND FUNCTIONS esuywihe ( weighted study We NRA ERUEEADLNYNEYT LENNY AND DEBROUWERE ANDREAS LB P ( lrbrooia PB-pcs efn-hlvsae;multiplier spaces; Gelfand-Shilov (PLB)-spaces; Ultrabornological LB P O M -pcsaeutaonlgclhsbe xesvl studied, extensively been has ultrabornological are )-spaces sutaonlgcl esoe that showed He ultrabornological. is ) SAE FULTRADIFFERENTIABLE OF -SPACES 1. s ⊗ LB P b ω LB P Introduction s 61,4A8 61,46A63. 46A13, 46A08, 46E10, [0 : ′ -pcso lrdffrnibefntosdefined functions ultradifferentiable of )-spaces n eie ietyta h atrsaeis space latter the that directly verified and O , -pcso lrdffrnibefntos Our functions. ultradifferentiable of )-spaces M ∞ 1 sutaonlgclwsgvnb Larcher by given was ultrabornological is ) → [0 , ∞ eawih ucin(nthe (in function weight a be ) totrlgat12T0519N. grant stdoctoral g h ototrlgrant postdoctoral the ugh h enn weight defining the f ultrabornological. e irhyperfunctions rier pace ultradifferentiable n-hlvspaces and-Shilov O .11 Grothendieck 131] p. ii faspectrum a of limit e e ultrahyper- ier M e l h rbe of problem The al. nltctossuch tools analytic O O ih system. eight ultra)distribution O M sioopi to isomorphic is M M fslowly of csaeultra- are aces sisomorphic is r h space the are h derived the f fsol in- slowly of e,called ces, spaces; 2 A.DEBROUWEREANDL.NEYT sense of Braun, Meise and Taylor [6]) and set φ(x) = ω(ex). Denote by φ∗(y) = V supx≥0 xy φ(x) the Young conjugate of φ. Let = vλ λ (0, ) be a family of { − } d { | ∈ ∞ } continuous functions vλ : R (0, ) such that 1 vλ vµ for all µ λ. We call V a weight system. We then consider→ ∞ the weighted (≤P LB≤)-spaces of ultradifferentiable≤ functions of Beurling and Roumieu type (ω) := lim lim ω,h, {ω} := lim lim ω,h, Z(V ) Zvλ Z{V } Zvλ h←−→0+ λ−→→0+ λ←−→∞ h−→→∞ where ω,h denotes the Banach space consisting of all ϕ C∞(Rd) such that Zvλ ∈ (α) ϕ (x) 1 ∗ ϕ ω,h := sup sup | | exp φ (h α ) < . Zv k k λ α∈Nd x∈Rd vλ(x) −h | | ∞ We use [ω] as a common notation for (ω) and {ω} . The first main goal of this ar- Z[V ] Z(V ) Z{V } ticle is to characterize when [ω] is ultrabornological through conditions on V . These Z[V ] conditions will be closely related to the linear topological invariants (DN) and (Ω) for Fréchet spaces [29]. Following Grothendieck, the key idea in our proofs is to com- plement the space [ω] into a suitable weighted (P LB)-space of continuous functions Z[V ] and vice-versa, and then apply results from [1] concerning the ultrabornologicity of such spaces. To achieve the former, we use tools from time-frequency analysis [16], specifically, the short-time Fourier transform and Gabor frames. Such techniques have recently proved to be useful in the study of (generalized) function spaces; see e.g. [2, 8, 12, 28]. Schwartz [25] showed that M is equal to the multiplier space of the space of rapidly decreasing smooth functions,O i.e., S = f ′ ϕ f for all ϕ . OM { ∈ S | · ∈ S ∈ S} Moreover, the natural (P LB)-space topology of M coincides with the topology induced by the embedding O L ( , ), f (ϕ ϕ f). OM → b S S 7→ 7→ · The second main goal of this article is to obtain a similar result for a wide class of Gelfand-Shilov spaces [9]. Given a weight function ω and a weight system V , we define the Gelfand-Shilov spaces of Beurling and Roumieu type as (ω) ω,h {ω} ω,h V := lim , V := lim , S( ) Svh S{ } Svh h←−→0+ h−→→∞ where ω,h denotes the Banach space consisting of all ϕ C∞(Rd) such that Svh ∈ (α) 1 ∗ ϕ ω,h = sup sup ϕ (x) v (x) exp φ (h α ) < . Sv h k k h α∈Nd x∈Rd | | −h | | ∞ We shall show that [ω] is topologically equal to the multiplier space of [ω] . This Z[V ] S[V ] problem has been previously studied for Fourier (ultra)hyperfunctions [21, 24, 31] and for general Gelfand-Shilov spaces of non-quasianalytic type [13]. Our main improve- ment here is that we also consider the quasianalytic case and that, in contrast to the WEIGHTED (P LB)-SPACES OF ULTRADIFFERENTIABLE FUNCTIONS 3 aforementioned works, we obtain topological and not merely algebraic identities. Fur- thermore, by using the above results, we are able to determine when such multiplier spaces are ultrabornological. In particular, Theorem 5.6 below shows that the multiplier space of the space of the Fourier ultrahyperfunctions is ultrabornological, whereas the one of the space of Fourier hyperfunctions is not. We mention that analogous results for convolutor spaces of Gelfand-Shilov spaces have recently been obtained by Vindas and the first author [11] (see also [26]). The structure of this article is as follows. In the preliminary Sections 2 and 3 we define and study weight functions, weight sequences and weight systems. In Section 4 we introduce Gelfand-Shilov spaces and discuss the short-time Fourier transform and Gabor frames in the context of these function spaces. Our main results are stated and discussed in Section 5. In the auxiliary Section 6 we review some results from [1] about weighted (P LB)-spaces of continuous functions. Finally, the proofs of our main results are given in Section 7. For this we study the short-time Fourier transform and Gabor frame expansions on various function spaces.

2. Weight functions and weight sequences A non-decreasing continuous function ω : [0, ) [0, ) is called a weight function (in the sense of Braun, Meise and Taylor [6]) if∞ω(0)→ = 0∞ and ω satisfies the following properties: (α) ω(2t)= O(ω(t)); (γ) log t = o(ω(t)); (δ) φ : [0, ) [0, ), φ(x)= ω(ex), is convex. ∞ → ∞ We extend ω to Rd as the radial function ω(x)= ω( x ), x Rd. A weight function ω is called non-quasianalytic if | | ∈ ∞ ω(t) 2 dt< . Z0 1+ t ∞ We refer to [6] for more information on these conditions. The Young conjugate of φ is defined as φ∗ : [0, ) [0, ), φ∗(y) = sup xy φ(x) . ∞ → ∞ x≥0 { − } The function φ∗ is convex and increasing, (φ∗)∗ = φ and the function y φ∗(y)/y is increasing on [0, ) and tends to infinity as y . We shall often use7→ the following lemma. ∞ →∞ Lemma 2.1. [18, Lemma 2.6] Let ω be a weight function. Then, (i) For all h,k,l > 0 there are m,C > 0 such that 1 1 (2.1) φ∗(m(y + l)) + ky φ∗(hy) + log C, y 0. m ≤ h ≥ (ii) For all m,k,l > 0 there are h,C > 0 such that (2.1) holds.

A sequence M = (Mp)p∈N of positive numbers is called a weight sequence [22] if 1/p 2 Mp as p and M is log-convex, i.e., M M M for all p Z . → ∞ → ∞ p ≤ p−1 p+1 ∈ + 4 A.DEBROUWEREANDL.NEYT

We set mp = Mp+1/Mp, p Z+. We consider the following conditions on a weight sequence M: ∈ ′ p+1 (M.2) Mp+1 CH Mp, p N, for some C,H > 0; ≤ p+q ∈ (M.2) Mp+q CH MpMq, p, q N, for some C,H > 0; (M.2)∗ 2m ≤m , p p , for some∈ p , N Z . p ≤ Np ≥ 0 0 ∈ + Clearly, (M.2) implies (M.2)′. A weight sequence M is called non-quasianalytic if

∞ 1 < . m ∞ Xp=1 p Conditions (M.2)′ and (M.2) are due to Komatsu [22]. Condition (M.2)∗ was intro- duced by Bonet et al. [5] without a name; we use here the same notation as in [11]. The most important examples of weight sequences satisfying (M.2) and (M.2)∗ are the Gevrey sequences p!s, s> 0. The sequence p!s is non-quasianalytic if and only if s> 1. Given two weight sequences M and N, the relation M N means that there are p ⊂ C,H > 0 such that Mp CH Np for all p N. The stronger relation M N means that the latter inequality≤ is valid for every H∈ > 0 and suitable C > 0. ≺ The associated function of a weight sequence M is deﬁned as

p t M0 ωM (t) = sup log , t 0. p∈N Mp ≥ Given another weight sequence N, it holds that N M if and only if ⊂ ω (t) ω (Ht) + log C, t 0, M ≤ N ≥ for some C,H > 0 [22, Lemma 3.8]. Similarly, N M if and only if the latter inequality remains valid for every H > 0 and suitable C≺ > 0 [22, Lemma 3.10]. The next result explains when the weight sequence case can be reduced to the weight function case. Lemma 2.2. [5, Proposition 13 and its proof] Let M be a weight sequence satisfying ∗ (M.2). Then, ωM is a weight function if and only if M satisﬁes (M.2) . In such a x case, the following properties hold (with φM (x)= ωM (e )): (i) For all h> 0 there are k,C > 0 such that 1 (2.2) exp φ∗ (kp) ChpM , p N. k M ≤ p ∈ (ii) For all h> 0 there are k,C > 0 such that 1 (2.3) kpM C exp φ∗ (hp) , p N. p ≤ h M ∈ (iii) For all k > 0 there are h,C > 0 such that (2.2) holds. (iv) For all k > 0 there are h,C > 0 such that (2.3) holds. WEIGHTED (P LB)-SPACES OF ULTRADIFFERENTIABLE FUNCTIONS 5

3. Weight systems Let X be a topological space. A family V = v λ R of continuous functions { λ | ∈ +} vλ : X (0, ) is called a weight system [9] if 1 vλ(x) vµ(x) for all x X and µ λ.→ The following∞ two conditions play a crucial≤ role in this≤ article. ∈ ≤ Definition 3.1. A weight system V on X is said to satisfy condition (DN) if λ µ λ θ (0, 1) ν µ C > 0 x X : v (x) Cvθ (x)v1−θ(x). ∃ ∀ ≤ ∀ ∈ ∃ ≤ ∃ ∀ ∈ µ ≤ λ ν Definition 3.2. A weight system V on X is said to satisfy condition (Ω) if λ µ λ ν µ θ (0, 1) C > 0 x X : v (x) Cvθ(x)v1−θ(x). ∀ ∃ ≥ ∀ ≥ ∀ ∈ ∃ ∀ ∈ µ ≤ λ ν Remark 3.3. The previous conditions are inspired by and closely related to the linear topological invariants (DN) and (Ω) for Fréchet spaces [29]. Next, we consider weight systems on Rd. We write fˇ(t)= f( t) for reflection about d − the origin. Given a weight function system V on R , we write Vˇ = vˇλ λ (0, ) . We consider the following conditions on a weight system V on Rd: { | ∈ ∞ } Rd (wM) λ µ λ C > 0 x : sup|y|≤1 vλ(x + y) Cvµ(x); ∀ ∃ ≤ ∃ ∀ ∈ Rd ≤ wM µ λ µ C > 0 x : sup|y|≤1 vλ(x + y) Cvµ(x); { } ∀ ∃ ≥ ∃ ∀ ∈ d ≤ (M) λ µ, ν λ C > 0 x, y R : vλ(x + y) Cvµ(x)vν(y); ∀ ∃ ≤ ∃ ∀ ∈ d ≤ M µ, ν λ µ, ν C > 0 x, y R : vλ(x + y) Cvµ(x)vν (y); { } ∀ ∃ ≥ ∃ 1∀ ∈ ≤ (N) λ µ λ : vλ/vµ L ; ∀ ∃ ≤ ∈ 1 N µ λ µ : vλ/vµ L ; { } ∀ ∃ ≥ ∈ d (Sq) λ,µ ν λ,µ C > 0 x R : vλ(x)vµ(x) Cvν(x); Sq ∀ν λ,µ∃ ≤ ν C∃ > 0 x∀ R∈d : v (x)v (x) Cv≤ (x). { } ∀ ∃ ≥ ∃ ∀ ∈ λ µ ≤ ν Notation 3.4. We employ [wM] as a common notation for (wM) and wM . A similar convention will be used for other notations. In addition, we often first{ state} assertions for the Beurling case followed in parenthesis by the corresponding ones for the Roumieu case. Clearly, [M] implies [wM]. Note that [wM] yields that (3.1) a> 0 λ λ′ λ µ′ µ µ′ ( a> 0 λ′ λ λ′ µ µ′ µ) ∀ ∀ ∃ ≤ ∀ ∃ ≤ ∀ ∀ ∃ ≥ ∀ ∃ ≥ v (x + y) v ′ (x) C > 0 x Rd : sup λ C λ . ∃ ∀ ∈ |y|≤a vµ(x + y) ≤ vµ′ (x) Consequently, [wM] and [N] imply that (3.2) λ µ λ ( µ λ µ): v /v C , ∀ ∃ ≤ ∀ ∃ ≥ λ µ ∈ 0 and v (k) (3.3) a> 0 λ µ λ( a> 0 µ λ µ): λ < . ∀ ∀ ∃ ≤ ∀ ∀ ∃ ≥ vµ(k) ∞ kX∈aZd We refer to [9] for more information on these conditions. 6 A.DEBROUWEREANDL.NEYT

We end this section by discussing the above conditions for two classes of weight systems on Rd. Given a weight function ω, we deﬁne 1 ω V := e λ λ (0, ) . ω { | ∈ ∞ } Lemma 3.5. Let ω be a weight function. Then,

(i) Vω satisﬁes [M], [N] and [Sq]. (ii) Vω satisﬁes (DN).

(iii) Vω does not satisfy (Ω). Proof. The proof is straightforward and left to the reader. Given a weight sequence M, we deﬁne ω 1 · V := e M ( λ ) λ (0, ) . M { | ∈ ∞ } Lemma 3.6. Let M be a weight sequence. Then,

(i) VM satisfies [M]. ′ (ii) VM satisfies [N] if and only if M satisfies (M.2) . (iii) VM satisfies [Sq] if and only if M satisfies (M.2). (iv) VM satisfies (DN).

(v) VM satisﬁes (Ω) if and only if (3.4) C > 0 N Z p Z p p : m Cm . ∃ ∀ ∈ + ∃ 0 ∈ + ∀ ≥ 0 Np ≤ p Proof. (i) This follows from the fact that ωM is increasing. (ii) This is shown in [11, Lemma 3.3]. (iii) This follows from [22, Proposition 3.6]. (iv) For all H > 0 and θ (0, 1) it holds that ∈ p 1/(1−θ) p t M0 (H t) M0 ωM (Ht) = sup θ log + (1 θ) log p∈N Mp − Mp θω (t)+(1 θ)ω (H1/(1−θ)t), ≤ M − M for all t 0. This shows that V satisﬁes (DN). ≥ M (v) We denote by m the counting function of the sequence (mp)p∈Z+ , i.e., m(x)= 1, x 0. ≥ mXp≤x Then, [22, Equation (3.11)] t m(x) ωM (t)= dx, t 0. Z0 x ≥

Hence, VM satisﬁes (Ω) if and only if Kt m(x) Ht m(x) (3.5) H > 0 K 1 : m(x)= o(m(Cx)). ∃ WEIGHTED (P LB)-SPACES OF ULTRADIFFERENTIABLE FUNCTIONS 7

We now show that (3.5) and (3.6) are equivalent. First assume that (3.5) holds. Let ε > 0 be arbitrary. Condition (3.5) with H = 1 and L = K/e implies that for t large enough Kt m(x) t m(x) m(Kt/e) dx ε dx ε log(e/K)m(t), ≤ ZKt/e x ≤ ZKt/e x ≤ whence (3.6) holds (with C = e/K). Conversely, assume that (3.6) holds. Let H > 0 be arbitrary and set K = H/C. Fix L K. Let ε > 0 be arbitrary. Condition (3.6) implies that for all t large enough ≤ Kt m(x) Kt m(Cx) Ht m(x) Ht m(x) dx ε dx = ε dx ε dx, ZLt x ≤ ZLt x ZLCt x ≤ ZLt x whence (3.5) holds. Example 3.7. The weight sequences (log(p + e))sp, s > 0, satisfy (3.4). On the contrary, any weight sequence satisfying (M.2)∗ does not satisfy (3.4).

4. Gelfand-Shilov spaces and time-frequency analysis Let ω be a weight function. For h > 0 and a continuous function v : Rd (0, ) we define ω,h as the Banach space consisting of all ϕ C∞(Rd) such that → ∞ Sv ∈ (α) 1 ∗ ϕ ω,h := sup sup ϕ (x) v(x) exp φ (h α ) < . Sv k k α∈Nd x∈Rd | | −h | | ∞ Let V be a weight system (on Rd). We define the Gelfand-Shilov spaces of Beurling and Roumieu type as (ω) := lim ω,h, {ω} := lim ω,h. S(V ) Svh S{V } Svh h←−→0+ h−→→∞ Then, (ω) is a Fréchet space and {ω} is an (LB)-space. Following Notation 3.4, we S(V ) S{V } employ [ω] as a common notation for (ω) and {ω} . If V satisfies [wM], then [ω] is S[V ] S(V ) S{V } S[V ] V [ω] 1 ∞ p translation-invariant. If satisfies [N], then [V ] L L L for all p [1, ]. [ω] S ⊂ ∩ ⊂ ∈ ∞ We refer to [9] for more information on [V ]. Given another weight function η, we write [ω] [ω] S := V . S[η] S[ η] Let M and A be two weight sequences. For h > 0 we define M,h as the Banach SA,h space consisting of all ϕ C∞(Rd) such that ∈ xβϕ(α)(x) M,h ϕ S := sup sup |α| |+|β| | < . k k A,h α,β∈Nd x∈Rd h M|α|A|β| ∞ We define (M) := lim M,h, {M} := lim M,h. S(A) SA,h S{A} SA,h h←−→0+ h−→→∞ Then, (M) is a Fréchet space and {M} is an (LB)-space. S(A) S{A} 8 A.DEBROUWEREANDL.NEYT

Lemma 4.1. Let M and A be two weight sequences. Suppose that M satisfies (M.2) ∗ [M] [ωM ] and (M.2) . Then, = V as locally convex spaces. S[A] S[ A] Proof. This follows from Lemma 2.2 and the fact that for all h> 0 β 1 x 1 Rd ωA x sup log ||β| | ωA x , x . √dh| | ≤ β∈Nd h A|β| ≤ h| | ∈ Let r,s > 0. We write r (p!r) (t1/r ) r {p!r} {t1/r } Σ := s = , := s = , s S(p! ) S(t1/s) Ss S{p! } S{t1/s} 1 for the classical Gelfand-Shilov spaces [15]. In particular, Σ1 is the test function space 1 of the Fourier ultrahyperfunctions [31] and 1 is the test function space of the Fourier hyperfunctions [20]. S r r Remark 4.2. The space Σs ( s ) is non-trivial if and only if r + s > 1 (r + s 1) (cf. [15, Section 8]). Consequently,S given a weight function ω and a weight system≥ V , we have that [ω] = 0 if ω(t)= O(t1/r) and S[V ] 6 { } 1/s λ h ( h λ): v (x)= O(eh|x| ) ∀ ∃ ∀ ∃ λ for some r + s> 1 (r + s 1), as these conditions imply that Σr (ω) ( r {ω} ). ≥ s ⊆ S(V ) Ss ⊆ S{V } In particular, if η is another weight function, [ω] = 0 if ω(t) = O(t1/r) and η(t) = S[η] 6 { } O(t1/s) for some r + s > 1 (r + s 1). Similarly, given two weight sequences M ≥ and A, [M] = 0 if p!r M and p!s A for some r + s > 1 (r + s 1). In [10, S[A] 6 { } ⊂ ⊂ ≥ Proposition 2.7, Proposition 4.3 and Theorem 5.9] Vindas and the first author showed that [p!] = 0 if and only if (log(p + e))p A ((log(p + e))p A). In general, the S[A] 6 { } ≺ ⊂ characterization of the non-triviality of the spaces [ω] and [M] seems to be an open S[η] S[A] problem. Next, we introduce some tools from time-frequency analysis; see the book [16] for more information. The translation and modulation operators are denoted by Txf(t)= 2πiξ·t d f(t x) and Mξf(t) = e f(t), for x, ξ R . The short-time Fourier transform (STFT)− of f L2(Rd) with respect to a window∈ ψ L2(Rd) is defined as ∈ ∈ −2πiξ·t 2d Vψf(x, ξ)=(f, MξTxψ)L2 = f(t)ψ(t x)e dt, (x, ξ) R . ZRd − ∈ 2 d 2 2d We have that V f 2 = ψ 2 f 2 . In particular, V : L (R ) L (R ) is contin- k ψ kL k kL k kL ψ → uous. The adjoint of Vψ is given by the weak integral ∗ 2 R2d Vψ F = F (x, ξ)MξTxψdxdξ, F L ( ). Z ZR2d ∈ 2 d If γ L (R ) is such that (γ, ψ) 2 = 0, then ∈ L 6 1 ∗ (4.1) Vγ Vψ = idL2(Rd) . (γ, ψ)L2 ◦ WEIGHTED (P LB)-SPACES OF ULTRADIFFERENTIABLE FUNCTIONS 9

The above reconstruction formula is the basis for the proof of part (a) of Theorem 5.1 below. Next, we consider Gabor frames. Given a window ψ L2(Rd) and lattice parameters a,b > 0, the set of time-frequency shifts ∈ (ψ, a, b) := M T ψ :(k, n) aZd bZd G { n k ∈ × } is called a Gabor frame for L2(Rd) if there exist A,B > 0 such that

2 2 2 2 d A f 2 V f(k, n) B f 2 , f L (R ). k kL ≤ | ψ | ≤ k kL ∈ (k,n)∈XaZd×bZd We deﬁne the Wiener space W as the space consisting of all ψ L∞(Rd) such that ∈ T ψ ∞ d < . k n kL ([0,1] ) ∞ nX∈Zd Given a weight function ω and a weight system V satisfying [wM] and [N], we have that [ω] W , as follows from (3.1). Let ψ W . Then, the analysis operator S[V ] ⊂ ∈ a,b 2 d 2 d d C = C : L (R ) l (aZ bZ ), f (V f(k, n)) Zd Zd , ψ ψ → × 7→ ψ (k,n)∈a ×b and the synthesis operator a,b 2 d d 2 d D = D : l (aZ bZ ) L (R ), (c ) Zd Zd c M T ψ ψ ψ × → k,n (k,n)∈a ×b 7→ k,n n k (k,n)∈XaZd×bZd are continuous [16, Proposition 6.2.2 and Corollary 6.2.3]. Given another window γ W , we deﬁne ∈ 2 d 2 d Sψ,γ := Dγ Cψ : L (R ) L (R ). ◦ d →d We call (ψ,γ) a pair of dual windows (on aZ bZ ) if S = id 2 Rd . In such a case, × ψ,γ L ( ) also Sγ,ψ = idL2(Rd) and both (ψ, a, b) and (γ, a, b) are Gabor frames. Pairs of dual windows are characterized byG the Wexler-RazG biorthogonality relations: Theorem 4.3. [16, Theorem 7.3.1 and the subsequent remark] Let ψ,γ W and let a,b > 0. Then, (ψ,γ) is a pair of dual windows on aZd bZd if and only∈ if × d ′ ′ 1 d 1 d (M T ψ, M ′ T ′ γ) 2 =(ab) δ ′ δ ′ , (k, n), (k, n ) Z Z , n k n k L k,k n,n ∈ a × b or, equivalently,

1 1/a,1/b 1/a,1/b (4.2) Cψ Dγ = idl2 1 Zd× 1 Zd . (ab)d ◦ ( a b ) The proof of part (b) of Theorem 5.1 below is based on the formula (4.2). For it to be applicable in our context we need that, given a weight function ω and a weight system V , ψ [ω] and γ [ω] . Hence, we introduce the following general notion: ∈ S[V ] ∈ S[Vˇ] Deﬁnition 4.4. Let ω be a weight function and let V be a weight system. The space [ω] is called Gabor accessible if there exist ψ [ω] , γ [ω] and a,b > 0 such that S[V ] ∈ S[V ] ∈ S[Vˇ] (ψ,γ) is a pair of dual windows on aZd bZd. × 10 A.DEBROUWEREANDL.NEYT

The regularity and decay properties of pairs of dual windows is a well-studied topic in time-frequency analysis; see [16, Chapter 13] and the references therein. We now use such results to give growth conditions on ω and V which ensure that [ω] is Gabor S[V ] accessible. Proposition 4.5. Let ω be a weight function and let V be a weight system. Then, [ω] is Gabor accessible if one of the following two conditions is satisﬁed: S[V ] (i) ω is non-quasianalytic. (ii) ω(t)= o(t2) and λ h : v (x)= O(eh|x|2) ( h λ : v (x)= O(eh|x|2)). ∀ ∀ λ ∀ ∃ λ Proof. Theorem 4.3 implies that if (ψ,γ) W (R) is a pair of dual windows on aZ bZ, a,b > 0, then (ψ ψ,γ ⊂γ) W (Rd) is a pair of dual windows× on aZd bZd. Now assume⊗···⊗ that (i)⊗···⊗ holds. Then,⊂ there exists a function ψ : R R with supp×ψ [0, 2] such that → ⊆ 1 sup sup ψ(α)(x) exp φ∗(h α ) < α∈Nd x∈Rd | | −h | | ∞ for all h> 0 and

Tkψ =1. kX∈Zd Fix 0 < b 1/3. Define ≤ γ(x)= bψ(x)+2bψ(x + 1), x R. ∈ In [7, Theorem 2.2] it is shown that (ψ,γ) is a pair of dual windows on Z bZ. By the remark at the beginning of the proof, we obtain that that (ψ ψ,γ × γ) ⊗···⊗ ⊗···⊗ ⊂ [ω] (Rd) [ω] (Rd) is a pair of dual windows on Zd bZd. Next, assume that (ii) S[V ] ∩ S[Vˇ] × holds. This condition implies that 1/2(Rd) [ω] (Rd) [ω] (Rd). Hence, it suffices S1/2 ⊆ S[V ] ∩ S[Vˇ] to show that 1/2(Rd) is Gabor accessible. Moreover, by the remark at the beginning S1/2 of the proof, it is enough to consider the case d = 1. Set ψ(x) = e−πx2 , x R. Then, ∈ ψ 1/2(R) and Janssen [19, Proposition B and its proof] showed that for all a,b > 0 ∈ S1/2 with ab < 1 there exists γ 1/2(R) such that (ψ,γ) is a pair of dual windows on ∈ S1/2 aZ bZ (see also [4, p. 273]). × Next, we discuss the Gabor accessibility of the spaces [ω] and [M]. S[η] S[A] Proposition 4.6. Let ω and η be two weight functions. Then, [ω] is Gabor accessible S[η] if one of the following two conditions is satisfied: (i) ω or η is non-quasianalytic. (ii) ω(t)= o(t2) and η(t)= o(t2) (ω(t)= O(t2) and η(t)= O(t2)). Proof. If ω is non-quasianalytic or (ii) is satisfied, the result is a direct consequence of Proposition 4.5. Now assume that η is non-quasianalytic. Since the Fourier transform is an isomorphism from [ω] onto [η] and (ψ,γ) is a pair of dual windows on aZd S[η] S[ω] ⊆ S × WEIGHTED (P LB)-SPACES OF ULTRADIFFERENTIABLE FUNCTIONS 11 bZd, a,b > 0 if and only if (ψ, γ) is so (as follows from Theorem 4.3 and Plancherel’s theorem), the space [ω] is Gabor accessible because [η] is so. S[η] b b S[ω] Proposition 4.7. Let M and A two weight functions satisfying (M.2). Then, [M] is S[A] Gabor accessible if one of the following two conditions is satisfied: (i) M or A is non-quasianalytic. (ii) p!1/2 M and p!1/2 A (p!1/2 M and p!1/2 A ). ≺ ≺ ⊂ ⊂ Proof. If M is non-quasianalytic or (ii) is satisfied, the result can be shown in the same way as Proposition 4.5. Now assume that A is non-quasianalytic. Since the Fourier transform is an isomorphism from [M] onto [A] , the result can be shown by using S[A] S[M] the same argument as in the proof of Proposition 4.6.

r r Proposition 4.8. Let r,s > 0. Then, Σs ( s ) is Gabor accessible if max(r, s) > 1 or min(r, s) > 1/2 (max(r, s) > 1 or min(r, s) S 1/2). ≥ Proof. This follows from Proposition 4.7. Finally, we would like to point out the following open problem.

r Problem 4.9. Let r,s > 0. Is every non-trivial space s Gabor accessible? This r S would imply that every non-trivial space Σs is Gabor accessible. If not, characterize the Gabor accessibility of the spaces Σr and r in terms of r and s. s Ss 5. Statement of the main results Let ω be a weight function and let V be a weight system (on Rd). We define (ω) ω,h {ω} ω,h V := lim lim , V := lim lim . Z( ) S1/vλ Z( ) S1/vλ h←−→0+ λ−→→0+ λ←−→∞ h−→→∞ Then, [ω] is a (P LB)-space. The first main result of this article may now be formu- Z[V ] lated as follows. Theorem 5.1. Let ω be a weight function and let V be a weight system satisfying [M] and [N]. Consider the following statements: (i) V satisfies (DN) ((Ω)). (ii) [ω] is ultrabornological. Z[V ] (iii) [ω] is barrelled. Z[V ] Then, (a) If [ω] = 0 , then (i) (ii) (iii). S[V ] 6 { } ⇒ ⇒ (b) If [ω] is Gabor accessible, then also (iii) (i). S[V ] ⇒ The assumption that [ω] is non-trivial and Gabor accessible in part (a) and part (b) S[V ] of Theorem 5.1, respectively, should be interpreted as implicit growth conditions on ω and V under which these results are valid. We refer to Remark 4.2 and Proposition 12 A.DEBROUWEREANDL.NEYT

4.5 for explicit conditions on ω and V which ensure that [ω] is non-trivial and Gabor S[V ] accessible, respectively. Next, we discuss multiplier spaces. Let ω be a weight function and let V be a weight system satisfying [N]. Then, we may view [ω] (and thus also [ω] ) as a subspace of Z[V ] S[V ] ′[ω] by identifying f [ω] with the element of ′[ω] given by S[V ] ∈ Z[V ] S[V ] [ω] f,ϕ = f(x)ϕ(x)dx, ϕ [V ]. h i ZRd ∈ S The space [ω] is an algebra under pointwise multiplication and the mapping [ω] S[V ] S[V ] × [ω] [ω] , (ϕ, ψ) ϕ ψ is separately continuous. We denote by ′[ω] the strong S[V ] → S[V ] 7→ · S[V ] dual of [ω] . For f ′[ω] and ϕ [ω] we deﬁne ϕ f ′[ω] via transposition, i.e., S[V ] ∈ S[V ] ∈ S[V ] · ∈ S[V ] ϕ f, ψ := f,ϕ ψ for ψ [ω] . Then, the mapping [ω] ′[ω] ′[ω], (ϕ, f) ϕ f h · i h · i ∈ S[V ] S[V ] ×S[V ] → S[V ] 7→ · is separately continuous. We deﬁne the multiplier space of [ω] as S[V ] ( [ω] ) := f ′[ω] ϕ f [ω] for all ϕ [ω] . OM S[V ] { ∈ S[V ] | · ∈ S[V ] ∈ S[V ]} Fix f ( [ω] ). The closed graph theorem of De Wilde and the continuity of the ∈ OM S[V ] mapping [ω] ′[ω], ϕ ϕ f imply that the mapping [ω] [ω] , ϕ ϕ f is S[V ] → S[V ] 7→ · S[V ] → S[V ] 7→ · continuous. We endow ( [ω] ) with the topology induced by the embedding OM S[V ] ( [ω] ) L ( [ω] , [ω] ), f (ϕ ϕ f). OM S[V ] → b S[V ] S[V ] 7→ 7→ · We then have: Theorem 5.2. Let ω be a weight function and let V be a weight system satisfying [ω] [ω] [ω] [M], [N] and [Sq]. Suppose that [V ] = 0 . Then, M ( [V ]) = [V ] as locally convex spaces. S 6 { } O S Z We end this section by discussing the structural and topological properties of the [ω] [M] multiplier spaces of [η] and [A] . Given two weight functions ω and η, we write [ω] [ω] S S = V . Z[η] Z[ η] Theorem 5.3. Let ω and η be two weight functions. Suppose that [ω] = 0 . Then, S[η] 6 { } ( [ω])= [ω] as locally convex spaces. Moreover, OM S[η] Z[η] (i) ( (ω)) is ultrabornological. OM S(η) (ii) If {ω} is Gabor accessible, then ( {ω}) is not ultrabornological. S{η} OM S{η} Proof. This follows from Lemma 3.5, Theorem 5.1 and Theorem 5.2.

We refer to Proposition 4.6 for conditions on ω and η which ensure that {ω} is S{η} Gabor accessible. WEIGHTED (P LB)-SPACES OF ULTRADIFFERENTIABLE FUNCTIONS 13

Let M and A be two weight functions. For h,λ > 0 we define M,h as the Banach ZA,λ space consisting of all ϕ C∞(Rd) such that ∈ −ω 1 |x| ϕ(α)(x) e A( λ ) M,h ϕ Z := sup sup | |α|| < . k k A,λ α∈Nd x∈Rd h M|α| ∞ We define (M) := lim lim M,h, {M} := lim lim M,h. Z(A) ZA,λ Z{A} ZA,λ h←−→0+ λ−→→0+ λ←−→∞ h−→→∞ Then, [M] is a (P LB)-space. Z[A] Lemma 5.4. Let M and A be two weight sequences. Suppose that M satisfies (M.2) ∗ [M] [ωM ] and (M.2) . Then, = V as locally convex spaces. Z[A] Z[ A] Proof. This follows from Lemma 2.2. Theorem 5.5. Let M be a weight sequence satisfying (M.2) and (M.2)∗ and let A be weight function satisfying (M.2). Suppose that [M] = 0 . Then, ( [M]) = [M] S[A] 6 { } OM S[A] Z[A] as locally convex spaces. Moreover, (i) ( (M)) is ultrabornological. OM S(A) (ii) If A satisfies (3.4), then ( {M}) is ultrabornological. If {M} is Gabor ac- OM S{A} S{A} cessible, the converse holds true as well. Proof. In view of Lemma 4.1 and Lemma 5.4, this follows from Lemma 3.6, Theorem 5.1 and Theorem 5.2. We refer to Proposition 4.7 for conditions on M and A which ensure that {M} is S{A} Gabor accessible. r r (p! ) Theorem 5.6. Let r,s > 0 be such that r + s> 1 (r + s 1). Then, M (Σs)= (p!s) r ≥ O Z r {p! } ( ( )= s ) as locally convex spaces. Moreover, OM Ss Z{p! } r (i) M (Σs) is ultrabornological. (ii) IfO max(r, s) > 1 or min(r, s) 1/2, then ( r) is not ultrabornological. ≥ OM Ss Proof. This follows from Proposition 4.7 and Theorem 5.5. 6. Weighted (P LB)-spaces of continuous functions

Let X be a topological space. A double sequence = aN,n N, n N consisting of continuous functions a : X (0, ) is called aAweight{ grid| on X∈if a} (x) N,n → ∞ N,n+1 ≤ aN,n(x) aN+1,n(x) for all x X and N, n N. Following [1], we introduce the following≤ two conditions: ∈ ∈ Deﬁnition 6.1. A weight grid on X is said to satisfy condition (Q) if A N M N n K M m n ε> 0 k m C > 0 x X : ∀ ∃ ≥ ∃ ∀ ≥ ∀ ≥ ∀ ∃ ≥ ∃ ∀ ∈ 1 ε C + . aM,m(x) ≤ aN,n(x) aK,k(x) If “ ε> 0” is replaced by “ ε> 0”, then is said to satisfy condition (wQ). ∀ ∃ A 14 A.DEBROUWEREANDL.NEYT

For a continuous function v : X (0, ) we denote by C (X) the Banach space → ∞ v consisting of all f C(X) such that f v = supx∈X f(x) v(x) < . Given a weight grid on X, we define∈ the (P LB)-spacek k | | ∞ A C(X) = lim lim C (X). A aN,n N←−∈N n−→∈N We now give two results from [1] that will play an essential role in the proof of Theorem 5.1. Theorem 6.2. [1, Theorem 3.5] Let be a weight grid on X. If satisfies (Q), then C(X) is ultrabornological. A A A Theorem 6.3. [1, Theorem 3.8(2)] Let be a weight grid on X. If C(X) is barrelled, then satisfies (wQ). A A A Let X and Y be two topological spaces. Let V be a weight system on X and let W be a weight system on Y . We define the following weight grids on X Y × v1/N wn (V ,W ) := N, n N , {V ,W } := N, n N . A w1/n | ∈ A vN | ∈ The following result is inspired by [29, Theorem 4.2 and Theorem 4.3]. Lemma 6.4. Let ω be a weight function and let V be a weight system on a topological space X. Then, (a) The following statements are equivalent: Rd (i) (Vω,V ) on X satisfies (Q). A Rd × (ii) (Vω,V ) on X satisfies (wQ). (iii) AV satisfies (DN)× . (b) The following statements are equivalent: Rd (i) {V ,Vω} on X satisfies (Q). A × d (ii) V V on X R satisfies (wQ). A{ , ω} × (iii) V satisfies (Ω). Proof. We only show (a) as (b) can be shown similarly. The implication (i) (ii) is trivial. Next, we show (ii) (iii). Condition (wQ) implies that there exists⇒H > 0 such that ⇒ v (x) λ µ λ ν µ C > 0 x X t 0 : v (x) C v (x)eHω(t) + ν . ∃ ∀ ≤ ∃ ≤ ∃ ∀ ∈ ∀ ≥ µ ≤ λ eω(t) Since ω(0) = 0, ω is continuous and ω(t) as t , we obtain that →∞ →∞ v (x) λ µ λ ν µ C > 0 x X r > 0 : v (x) C v (x)rH + ν . ∃ ∀ ≤ ∃ ≤ ∃ ∀ ∈ ∀ µ ≤ λ r By calculating the minimum for r> 0 (with x X fixed) of the right-hand side of the above inequality, we find that ∈ θ (0, 1) λ µ λ ν µ C > 0 x X : v (x) Cvθ (x)v1−θ(x). ∃ ∈ ∃ ∀ ≤ ∃ ≤ ∃ ∀ ∈ µ ≤ λ ν WEIGHTED (P LB)-SPACES OF ULTRADIFFERENTIABLE FUNCTIONS 15

An induction argument now shows that V satisfies (DN). Finally, we show (iii) (i). Let N N be arbitrary and set M = N + 1. Since V satisfies (DN) there is n⇒ N such that∈ ∈ (6.1) m n θ (0, 1) k m C > 0 x X : v (x) Cvθ (x)v1−θ(x). ∀ ≥ ∀ ∈ ∃ ≥ ∃ ∀ ∈ 1/m ≤ 1/n 1/k Let K > M, m n and ε > 0 be arbitrary. Set θ = (K N 1)/(K N) (0, 1) and note that M≥= θN + (1 θ)K. Choose k and C as in− (6.1).− Then, − ∈ − θ −θ 1/(1−θ) 1−θ v1/m(x) εv1/n(x) (Cε ) v1/k(x) eMω(t) ≤ eNω(t) eKω(t) v (x) v (x) max ε 1/n , (Cε−θ)1/(1−θ) 1/k ≤ eNω(t) eKω(t) εv (x) (Cε−θ)1/(1−θ)v (x) 1/n + 1/k , ≤ eNω(t) eKω(t) for all t 0 and x X, whence V V satisfies (Q). ≥ ∈ A( ω, ) 7. Proof of the main results The proof of part (a) of Theorem 5.1 is based on the mapping properties of the STFT on [ω] . We start with the following three general results: Z[V ] d Lemma 7.1. Let ω be a weight function. Let vi : R (0, ), i = 1, 2, 3, 4, be continuous functions such that → ∞ v (x + t) C v (x)ˇv (t), x, t Rd, 2 ≤ 0 1 4 ∈ for some C > 0 and v /v L1. Let h > 0, i =1, 2, 3, be such that 0 4 3 ∈ i 1 1 φ∗(max h , h (y + 1)) + (log √d)y φ∗(h y) + log C , y 0, max h , h { 1 3} ≤ h 2 1 ≥ { 1 3} 2 for some C > 0. Let ψ ω,h3 . Then, the mapping 1 ∈ Sv3 ω,h1 R2d V : C 1 ω ( ) ψ v1 h x,ξ S → v2⊗e 2 is continuous.

ω,h1 R2d Proof. Let ϕ v1 be arbitrary. For all y 0 and (x, ξ) with ξ 1 it holds that ∈ S ≥ ∈ | | ≥ ξ y V ϕ(x, ξ) v (x) | | | ψ | 2 |α| α max √d ξ Vψϕ(x, ξ) v2(x) ≤ |α|=⌈y⌉ | |

|α| α (β) (α−β) max (√d/2π) v2(x) ϕ (t) ψ (t x) dt ≤ |α|=⌈y⌉ β ZRd | || − | Xβ≤α

|α| α (β) (α−β) C0 max (√d/2π) ϕ (t) v1(t) ψ (t x) v4(t x)dt ≤ |α|=⌈y⌉ β ZRd | | | − | − Xβ≤α 16 A.DEBROUWEREANDL.NEYT

ω,h ω,h 1 C0 ϕ S 1 ψ S 3 v4/v3 L ≤ k k v1 k k v3 k k × |α| α 1 ∗ 1 ∗ max (√d/2π) exp φ (h1 β )+ φ (h3 α β ) |α|=⌈y⌉ β h1 | | h3 | − | Xβ≤α

√ ω,h ω,h 1 dC0 ϕ S 1 ψ S 3 v4/v3 L ≤ k k v1 k k v3 k k × 1 exp φ∗(max h , h (y + 1)) + (log √d)y max h , h { 1 3} { 1 3} 1 ∗ √ ω,h ω,h 1 dC0C1 ϕ S 1 ψ S 3 v4/v3 L exp φ (h2y) . ≤ k k v1 k k v3 k k h2 Hence,

1 ∗ √ 1 Vψϕ(x, ξ) v2(x) dC0C1 ϕ Sω,h1 ψ Sω,h3 v4/v3 L inf exp φ (h2y) (log ξ )y | | ≤ k k v1 k k v3 k k y≥0 h2 − | | − 1 ω(ξ) √ 1 h2 = dC0C1 ϕ Sω,h1 ψ Sω,h3 v4/v3 L e . k k v1 k k v3 k k For all (x, ξ) R2d with ξ 1 it holds that ∈ | | ≤ 1 1 h ω(ξ) h ω(1) Vψϕ(x, ξ) v2(x)e 2 e 2 v2(x) ϕ(t) ψ(t x) dt | | ≤ ZRd | || − | 1 h ω(1) 2 ω,h ω,h 1 C0e ϕ S 1 ψ S 3 v4/v3 L . ≤ k k v1 k k v3 k k ω,h1 R2d This shows that V : C 1 ω ( ) is continuous. ψ v1 h x,ξ S → v2⊗e 2

Lemma 7.2. Let ω be a weight function. Choose C0,L> 0 such that (7.1) ω(2πt) Lω(t) + log C , t 0. ≤ 0 ≥ Let v : Rd (0, ), i =2, 3, 4, be continuous functions such that i → ∞ (7.2) v (x + t) C v (x)v (t), x, t Rd, 2 ≤ 1 3 4 ∈ for some C1 > 0. Let hi > 0, i =1, 2, 3, be such that 1 1 (7.3) φ∗(max h , h y) + (log 2)y φ∗(h y) + log C , y 0, max h , h { 1 3} ≤ h 2 2 ≥ { 1 3} 2 for some C2 > 0. Then, there is C > 0 such that L ω(ξ) 2d h1 R MξTxψ Sω,h2 C ψ Sω,h3 v4(x)e , (x, ξ) , k k v2 ≤ k k v3 ∈ for all ψ ω,h3 . ∈ Sv3 ω,h3 R2d Nd Rd Proof. Let ψ v3 and (x, ξ) be arbitrary. For all α and t it holds that ∈ S ∈ ∈ ∈ α (M T ψ)(α)(t) v (t) (2π ξ )|β| ψ(α−β)(t x) v (t) | ξ x | 2 ≤ β | | | − | 2 Xβ≤α α C v (x) (2π ξ )|β| ψ(α−β)(t x) v (t x) ≤ 1 4 β | | | − | 3 − Xβ≤α WEIGHTED (P LB)-SPACES OF ULTRADIFFERENTIABLE FUNCTIONS 17

α 1 ∗ ω,h C1 ψ S 3 v4(x) exp (log 2π ξ ) β φ (h1 β ) ≤ k k v3 β | | | | − h1 | | × Xβ≤α

1 ∗ 1 ∗ exp φ (h3 α β )+ φ (h1 β ) h3 | − | h1 | | 1 h ω(2πξ) ω,h 1 C1 ψ S 3 v4(x)e ≤ k k v3 × 1 exp φ∗(max h , h α ) + (log 2) α max h , h { 1 3}| | | | { 1 3} L 1 h ω(ξ) ∗ ω,h 1 C0C1C2 ψ S 3 v4(x)e exp φ (h2 α ) . ≤ k k v3 h2 | | This shows the result.

Lemma 7.3. Let ω be a weight function. Choose C0,L> 0 such that (7.1) holds. Let d vi : R (0, ), i =1, 2, 3, 4, be continuous functions such that (7.2) holds for some C > 0→and v∞/v L1. Let h > 0, i = 1, 2, 3, be such that (7.3) holds for some 1 4 1 ∈ i C > 0. Let γ ω,h3 . Then, the mapping 2 ∈ Sv3 ∗ R2d ω,h2 V : C 2L ω ( ) γ h x,ξ v2 v1⊗e 1 → S is continuous. Proof. This follows from Lemma 7.2. Given a weight function ω and a weight system V , we deﬁne 2d 2d R 1 R C (ω) ( x,ξ) := lim lim C 1 ω ( x,ξ), Z(V ) ⊗e h h←−→0+ λ−→→0+ vλ 2d 2d R 1 R C {ω} ( x,ξ) := lim lim C 1 ω ( x,ξ). Z{V } ⊗e h λ←−→∞ h−→→∞ vλ We then have: Proposition 7.4. Let ω be a weight function and let V be a weight system satisfying [M] and [N]. Let ψ [ω] and γ [ω] . Then, the mappings ∈ S[V ] ∈ S[Vˇ] [ω] R2d ∗ R2d [ω] Vψ : [V ] CZ[ω] ( x,ξ), Vγ : CZ[ω] ( x,ξ) [V ] Z → [V ] [V ] → Z are continuous. Moreover, if (γ, ψ) 2 =0, then L 6 1 ∗ (7.4) Vγ Vψ = idZ[ω] . (γ, ψ)L2 ◦ [V ] Proof. It suﬃces to show that ω,k R2d h k λ µ ( µ λ k h): Vψ : C 1 ω ( ) is continuous, 1/vλ 1 ⊗e h x,ξ ∀ ∃ ∀ ∃ ∀ ∃ ∀ ∃ S → vµ and ∗ R2d ω,h h k λ µ ( µ λ k h): V : C 1 ω ( ) is continuous. γ 1 ⊗e k x,ξ 1/vµ ∀ ∃ ∀ ∃ ∀ ∃ ∀ ∃ vλ → S 18 A.DEBROUWEREANDL.NEYT

By Lemma 2.1 and the fact that V satisfies [M] and [N], this follows from Lemma 7.1 and Lemma 7.3, respectively. We now show (7.4). Let ϕ [ω] be arbitrary. Since ∈ Z[V ] V satisfies [N], the continuous functions ϕT ψ and V ϕ(x, ), with x Rd fixed, both x ψ · ∈ belong to L1. As V ϕ(x, ξ)= (ϕT ψ)(ξ), we obtain that ψ F x 2πiξ·t Vψϕ(x, ξ)MξTxγ(t)dxdξ = Vψϕ(x, ξ)e dξ Txγ(t)dx Z ZR2d ZRd ZRd

= ϕ(t) Txψ(t)Txγ(t)dx =(γ, ψ)L2 ϕ(t) ZRd for all t Rd. ∈ Lemma 7.5. Let ω be a weight function and let V be a weight system satisfying [wM]. If [ω] =0, then also [ω] [ω] =0. S[V ] 6 S[V ] ∩ S[Vˇ] 6 Proof. Since the space [ω] is translation-invariant and non-trivial, there exists ϕ [ω] S[V ] ∈ S[V ] with ϕ(0) = 0. Then, ψ = ϕϕˇ [ω] [ω] and ψ(0) = ϕ2(0) = 0. 6 ∈ S[V ] ∩ S[Vˇ] 6 Proof of part (a) of Theorem 5.1. The implication (ii) (iii) holds for any locally ⇒ convex space. We now show (i) (ii). By Lemma 7.5, there exists ψ [ω] [ω] 0 ⇒ ∈ S[V ] ∩S[Vˇ]\{ } [ω] (and thus ψ 2 = 0). Proposition 7.4 (with γ = ψ) implies that is isomorphic to k kL 6 Z[V ] R2d a complemented subspace of C [ω] ( x,ξ). Note that Z[V ] 2d 2d 2d 2d (ω) R V V R {ω} R V V R CZ ( x,ξ)= ( ω, )C( ξ,x), CZ ( x,ξ)= { , ω}C( x,ξ). (V ) A {V } A R2d Hence, C [ω] ( x,ξ) is ultrabornological by Theorem 6.2 and Lemma 6.4. The result Z[V ] now follows from the fact that a complemented subspace of an ultrabornological space is again ultrabornological. The proof of part (b) of Theorem 5.1 is based on the on the mapping properties of the analysis and synthesis operator on [ω] . Given a weight function ω, a weight Z[V ] system V and a,b > 0, we deﬁne d d d d Z Z 1 Z Z C (ω) (a b ) := lim lim C 1 ω (a b ), Z V x ξ ⊗e h x ξ ( ) × v × h←−→0+ λ−→→0+ λ d d d d Z Z 1 Z Z C {ω} (a x b ξ ) := lim lim C 1 ω (a x b ξ ). Z V ⊗e h { } × v × λ←−→∞ h−→→∞ λ Proposition 7.6. Let ω be a weight function, let V be a weight system satisfying [M] and [N] and let a,b > 0. Let ψ [ω] and γ [ω] . Then, the mappings ∈ S[V ] ∈ S[Vˇ] a,b [ω] Zd Zd a,b Zd Zd [ω] Cψ : [V ] CZ[ω] (a x b ξ ),Dγ : CZ[ω] (a x b ξ ) [V ] Z → [V ] × [V ] × → Z are continuous. Moreover, if (ψ,γ) is a dual pair of windows on aZd bZd, then × 1 1/a,1/b 1/a,1/b (7.5) C D = id 1 Zd 1 Zd . d ψ γ C [ω] ( x× ) (ab) Z a b ξ ◦ [V ] WEIGHTED (P LB)-SPACES OF ULTRADIFFERENTIABLE FUNCTIONS 19

Proof. It suffices to show that a,b ω,k Zd Zd h k λ µ ( µ λ k h): C : C 1 ω (a b ) is continuous, ψ 1/vλ 1 ⊗e h x ξ ∀ ∃ ∀ ∃ ∀ ∃ ∀ ∃ S → vµ × and a,b Zd Zd ω,h h k λ µ ( µ λ k h): D : C 1 ω (a b ) is continuous. γ 1 ⊗e k x ξ 1/vµ ∀ ∃ ∀ ∃ ∀ ∃ ∀ ∃ vλ × → S By Lemma 2.1 and the fact that V satisfies [M] and [N], this follows from Lemma 7.1 and Lemma 7.2 together with (3.3) (cf. Lemma 7.2), respectively. Since, by (3.2), the Zd Zd Zd Zd set of finite sequences on a x b ξ is dense in CZ[ω] (a x b ξ ), the identity (7.5) × [V ] × follows from Lemma 4.3. Proof of part (b) of Theorem 5.1. As [ω] is Gabor accessible, Proposition 7.6 implies S[V ] 1 Zd 1 Zd [ω] that CZ[ω] a x b ξ is isomorphic to a complemented subspace of [V ]. Since a [V ] × Z complemented subspace of a barrelled space is again barrelled, we may conclude that 1 Zd 1 Zd CZ[ω] a x b ξ is barrelled. Note that [V ] × 1 d 1 d 1 d 1 d C (ω) Z Z = 1 d 1 d C Z Z , Z x ξ (Vω | Z ,V | Z ) ξ x (V ) a × b A b a b × a

1 d 1 d 1 d 1 d C {ω} Z Z = 1 d 1 d C Z Z . Z x ξ {V | Z ,Vω| Z } x ξ {V } a × b A a b a × b

Hence, (V | 1 Zd,V | 1 Zd) ( {V | 1 Zd,V | 1 Zd}) satisﬁes (wQ) by Theorem 6.3. Property (3.1) A ω b a A a ω b implies that then also (Vω,V ) ( {V ,Vω}) satisﬁes (wQ). The result now follows from Lemma 6.4. A A Finally, we show Theorem 5.2. We need various results in preparation. d Lemma 7.7. Let ω be a weight function. Let vi : R (0, ), i =1, 2, 3, be continuous functions such that → ∞ v (x) C v (x)v (x), x Rd, 2 ≤ 0 1 3 ∈ for some C0 > 0. Let hi > 0, i =1, 2, 3, be such that 1 1 φ∗(max h , h y) + (log 2)y φ∗(h y) + log C , y 0, max h , h { 1 3} ≤ h 2 1 ≥ { 1 3} 2 for some C1 > 0. Then, the mapping ω,h1 ω,h3 ω,h2 , (ϕ, ψ) ϕ ψ Sv1 × Sv3 → Sv2 7→ · is continuous.

ω,h1 ω,h3 Nd Rd Proof. Let ϕ v1 and ψ v3 be arbitrary. For all α and x it holds that ∈ S ∈ S ∈ ∈ (ϕψ)(α)(x) v (x) | | 2 α C ϕ(β)(x) v (x) ψ(α−β)(x) v (x) ≤ 0 β| | 1 | | 3 Xβ≤α 20 A.DEBROUWEREANDL.NEYT

α 1 ∗ 1 ∗ ω,h ω,h C0 ϕ S 1 ψ S 3 exp φ (h1 β )+ φ (h3 α β ) ≤ k k v1 k k v3 β h1 | | h3 | − | Xβ≤α

1 ∗ ω,h ω,h C0 ϕ S 1 ψ S 3 exp φ (max h1, h3 α ) + (log 2) α ≤ k k v1 k k v3 max h , h { }| | | | { 1 3} 1 ∗ ω,h ω,h C0C1 ϕ S 1 ψ S 3 exp φ (h2 α ) . ≤ k k v1 k k v3 h2 | | This shows that the mapping ω,h1 ω,h3 ω,h2 , (ϕ, ψ) ϕ ψ is continuous. Sv1 × Sv3 → Sv2 7→ · Next, we extend the STFT and its adjoint to ′[ω]. Given a weight function ω and S[V ] a weight system V satisfying [M] and [N], we define 2d 2d R 1 R C ′(ω) ( x,ξ) := lim C 1 − ω ( x,ξ), S(V ) ⊗e h h−→→0+ vh 2d 2d R 1 R C ′{ω} ( x,ξ) := lim C 1 − ω ( x,ξ). S{V } ⊗e h h←−→∞ vh The STFT of an element f ′[ω] with respect to ψ [ω] is defined as ∈ S[V ] ∈ S[V ] V f(x, ξ) := f, M T ψ , (x, ξ) R2d. ψ h ξ x i ∈ 2d [ω] ′[ω] R We define the adjoint STFT of F CS ( x,ξ) with respect to γ [Vˇ] as ∈ [V ] ∈ S

∗ [ω] Vγ F,ϕ := F, Vγϕ = F (x, ξ)Vγϕ(x, ξ)dxdξ, ϕ [V ]. h i h i Z ZR2d − ∈ S Proposition 7.8. Let ω be a weight function and let V be a weight system satisfying [M] and [N]. Let ψ [ω] and γ [ω] . Then, the mappings ∈ S[V ] ∈ S[Vˇ] ′[ω] R2d ∗ R2d ′[ω] Vψ : [V ] CS′[ω]( x,ξ), Vγ : CS′[ω]( x,ξ) [V ] S → [V ] [V ] → S are continuous. Moreover, if (γ, ψ) 2 =0, then L 6 1 ∗ (7.6) Vγ Vψ = idS′[ω] . (γ, ψ)L2 ◦ [V ]

Proof. By Lemma 2.1 and the fact that V satisfies [M], Lemma 7.2 implies that Vψ : ′{ω} R2d ′(ω) R2d {V } CS′{ω} ( x,ξ) is continuous and that Vψ : (V ) CS′(ω) ( x,ξ) maps bounded S → {V } S → (V ) sets into bounded sets. Since ′(ω) is bornological, we obtain that also V : ′(ω) S(V ) ψ S(V ) → R2d ∗ C ′(ω) ( x,ξ) is continuous. Next, we treat Vψ . We define S(V ) R2d R2d R2d R2d C (ω) ( x,ξ) := lim C 1 ω ( x,ξ), C {ω} ( x,ξ) := lim C 1 ω ( x,ξ). S(V ) vh⊗e h S{V } vh⊗e h h←−→0+ h−→→∞ By Lemma 2.1 and the fact that V satisfies [M] and [N], Lemma 7.1 implies that the mapping [ω] R2d [V ] CS[ω] ( x,ξ), ϕ Vγϕ(x, ξ), S → [V ] 7→ − WEIGHTED (P LB)-SPACES OF ULTRADIFFERENTIABLE FUNCTIONS 21

∗ R2d ′[ω] is continuous. Hence, the continuity of Vψ : CS′[ω]( x,ξ) [V ] follows from the [V ] → S continuity of the mapping

R2d R2d ′ CS′[ω]( x,ξ) (CS[ω] ( x,ξ))b, F f F (x, ξ)f(x, ξ)dxdξ . [V ] → [V ] 7→ 7→ Z ZR2d ∗ ′[ω] ′[ω] We now verify (7.6). Define A = (1/(γ, ψ)L2 )Vγ Vψ, then A : [V ] [V ] is ◦ S → S 2 continuous. By (4.1) we have that A 2 = id 2 . Then the result would follow if L is |L L dense in ′[ω]. The latter is true, as [ω] is reflexive (it is nuclear [9, Theorem 5.1]), so S[V ] S[V ] the transpose of the inclusion mapping ι : L2 ′[ω] : f (ϕ fϕ) is exactly the → S[V ] 7→ 7→ trivial inclusion [ω] L2 and thus injective. R S[V ] → R2d We also need the following result about the projective description of C [ω] ( x,ξ). Z[V ] Lemma 7.9. (i) Let V be a weight system satisfying [wM] and [N]. Set d V ( ) := v : R (0, ) continuous sup v(x)vλ(x) < for all λ . V { → ∞ | x∈Rd ∞ } Then, R2d R2d C (ω) ( x,ξ) = lim lim C 1 ω ( x,ξ). Z(V ) v⊗e h h←−→0+ v∈←−V (V) (ii) Let ω be a weight function. Set V := σ : [0, ) [0, ) weight function σ = o(ω) . ω { ∞ → ∞ | } Then, 2d 2d R 1 R CZ{ω} ( x,ξ) = lim lim C ⊗eσ ( x,ξ). {V } vλ λ←−→∞ σ←−∈V ω Proof. In view of [3, Theorem 1.3], (i) follows from (3.2) and [3, Proposition, p. 112, and its proof] and (ii) follows from [6, Lemma 1.7 and Remark 1.8]. Proof of Theorem 5.2. We first show that [ω] is continuously included in ( [ω] ). Z[V ] OM S[V ] To this end, it suffices to show that h k λ µ ( µ λ k h): ω,µ ω,k ω,h, (ϕ, ψ) ϕ ψ is continuous. ∀ ∃ ∀ ∃ ∀ ∃ ∀ ∃ Svµ × S1/vλ → Svh 7→ · By Lemma 2.1 and the fact that V satisfies [Sq], this follows from Lemma 7.7. Next, [ω] [ω] we show that M ( [V ]) is continuously included in [V ]. By Lemma 7.5, there exists [ω] [ω]O S Z ψ 0 (and thus ψ 2 = 0). Proposition 7.4 and Proposition 7.8 (with ∈ S[V ] ∩ S[Vˇ]\{ } k kL 6 ψ = γ) imply that it suffices to show that the mapping [ω] R2d Vψ : M ( [V ]) CZ[ω] ( x,ξ) O S → [V ] is continuous. We start by showing that

−2πiξ·t 2d Vψf(x, ξ)= ((Txψ) f)(t)e dt, (x, ξ) R , ZRd · ∈ 22 A.DEBROUWEREANDL.NEYT for all f ( [ω] ). Choose χ [ω] such that χ(0) = 1 and set χ (x)= χ(x/n) for ∈ OM S[V ] ∈ S[V ] n n Z . Then, χ [ω] by [M] and [Sq]. Property (3.2) implies that χ ϕ ϕ in ∈ + n ∈ S[V ] n · → [ω] as n for all ϕ [ω] . Hence, S[V ] →∞ ∈ S[V ]

Vψf(x, ξ)= f, MξTxψ = lim f, χn (MξTxψ) = lim (Txψ) f, M−ξχn h i n→∞h · i n→∞h · i −2πiξ·t −2πiξ·t = lim ((Txψ) f)(t)χn(t)e dt = ((Txψ) f)(t)e dt n→∞ ZRd · ZRd · for all (x, ξ) R2d. In the rest of the proof we treat the Beurling and Roumieu case separately. We∈ ﬁrst consider the Beurling case. By Lemma 7.9(i), it suﬃces to show that for all h > 0 and v V ( ) there is a bounded subset B (ω) , µ > 0 and ∈ V ⊂ S(V ) k,C > 0 such that (ω) (7.7) V f 1 ω C sup ϕ f ω,k , f ( ). ψ h Sv M (V ) k kv⊗e ≤ ϕ∈B k · k µ ∈ O S Set B = T ψv(x) x Rd { x | ∈ } and note that B is a bounded subset of (ω) . Next, choose µ> 0 such that 1/v L1. S(V ) µ ∈ By Lemma 2.1(i), there are k,C > 0 such that 1 1 φ∗(k(y + 1)) + (log √d)y φ∗(hy) + log C, y 0. k ≤ h ≥ For all y 0 and (x, ξ) R2d with ξ 1 it holds that ≥ ∈ | | ≥ ξ y V f(x, ξ) v(x) | | | ψ | |α| α −2πiξ·t max √d v(x) ξ ((Txψ) f)(t)e dt ≤ |α|=⌈y⌉ ZRd ·

|α| (α) max (√d/(2π)) v(x) ((Txψ) f) (t) d t ≤ |α|=⌈y⌉ ZRd | · |

|α| 1 ∗ 1/v 1 sup ϕ f ω,k max (√d/(2π)) exp φ (k α ) µ L Sv ≤k k ϕ∈B k · k µ |α|=⌈y⌉ k | |

1 ∗ √d 1/v 1 sup ϕ f ω,k exp φ (k(y + 1)) + (log √d)y µ L Sv ≤ k k ϕ∈B k · k µ k

1 ∗ √dC 1/v 1 sup ϕ f ω,k exp φ (hy) . µ L Sv ≤ k k ϕ∈B k · k µ h Hence,

1 ∗ V f(x, ξ) v(x) √dC 1/v 1 sup ϕ f ω,k inf exp φ (hy) (log ξ )y ψ µ L Sv | | ≤ k k ϕ∈B k · k µ y≥0 h − | | − 1 ω(ξ) = √dC 1/v 1 sup ϕ f ω,k e h . µ L Sv k k ϕ∈B k · k µ WEIGHTED (P LB)-SPACES OF ULTRADIFFERENTIABLE FUNCTIONS 23

For all (x, ξ) R2d with ξ 1 it holds that ∈ | | ≤ 1 ω(ξ) 1 ω(1) Vψf(x, ξ) v(x)e h e h v(x) ((Txψ) f)(t) dt | | ≤ ZRd | · | 1 ω(1) e h 1/v 1 sup ϕ f ω,k . µ L Sv ≤ k k k ϕ∈B k · k µ This shows (7.7). Next, we consider the Roumieu case. By Lemma 7.9(ii), it suﬃces to show that for all λ> 0 and σ V there is a bounded subset B {ω} , a continuous ∈ ω ⊂ S{V } seminorm p on {ω} and C > 0 such that S{V } {ω} (7.8) Vψf 1 ⊗eσ C sup p(ϕ f), f M ( {V }). k k vλ ≤ ϕ∈B · ∈ O S Set T ψ B = x x Rd vλ(x) | ∈ and note that B is a bounded subset of {ω} . By Lemma 2.1(i), there are k,C > 0 S{V } such that 1 φ∗ (k(y + 1)) + (log √d)y φ∗ (y) + log C, y 0. k σ ≤ σ ≥ ∗ x where φσ denotes the Young conjugate of the function φσ(x)= σ(e ). Deﬁne

(α) 1 ∗ {ω} p(ϕ) = sup ϕ L1 exp φσ(k α ) , ϕ {V }. α∈Nd k k −k | | ∈ S Then, p is a continuous seminorm on {ω} . One can now show (7.8) in the same way S{V } as (7.7) was shown in the Beurling case.

References [1] S. Agethen, K. D. Bierstedt, J. Bonet, Projective limits of weighted (LB)-spaces of continuous functions, Arch. Math. 92 (2009), 384–398. [2] C. Bargetz, N. Ortner, Characterization of L. Schwartz’ convolutor and multiplier spaces ′ C and M by the short-time Fourier transform, Rev. R. Acad. Cienc. Exactas F´ıs. Nat. Ser. OA. Math.O RACSAM 108 (2014), 833–847. [3] K. D. Bierstedt, R. Meise, W. H. Summers, A projective description of weighted inductive limits, Trans. Amer. Math. Soc. 272 (1982), 107–160. [4] H. Bolcskei,¨ A. J. E. M. Janssen, Gabor frames, unimodularity, and window decay,J. Fourier Anal . Appl. 6(3) (2000), 255–276. [5] J. Bonet, R. Meise, S. N. Melikhov, A comparison of two different ways to define classes of ultradifferentiable functions, Bull. Belg. Math. Soc. Simon Stevin 14 (2007), 425–444. [6] R. W. Braun, R. Meise, B. A. Taylor, Ultradifferentiable functions and Fourier analysis, Results Math. 17 (1990), 206–237. [7] O. Christensen, Pairs of dual Gabor frame generators with compact support and desired frequency localization, Appl. Comput. Harmon. Anal., 20 (2006), 403–410. [8] A. Debrouwere, L. Neyt, J. Vindas, On the space of ultradistributions vanishing at infinity, Banach J. Math. Anal., 14 (2020), 915–934. [9] A. Debrouwere, L. Neyt, J. Vindas, The nuclearity of Gelfand-Shilov spaces and kernel theorems, Collect. Math. (2020), in press (DOI: 10.1007/s13348-020-00286-2). 24 A.DEBROUWEREANDL.NEYT

[10] A. Debrouwere, J. Vindas, On the non-triviality of certain spaces of analytic functions. Hyper- functions and ultrahyperfunctions of fast growth, Rev. R. Acad. Cienc. Exactas F´ıs. Nat. Ser. A. Math. RACSAM 112 (2018), 473–508. [11] A. Debrouwere, J. Vindas, On weighted inductive limits of spaces of ultradifferentiable functions and their duals, Math. Nachr. 292 (2019), 573–602. [12] A. Debrouwere, J. Vindas, Topological properties of convolutor spaces via the short-time Fourier transform, Trans. Amer. Math. Soc., in press (arXiv:1801.09246). [13] P. Dimovski, B. Prangoski, D. Velinov, Multipliers and convolutors in the space of tem- pered ultradistributions, Novi Sad J. Math. 44 (2014), 1–18. [14] P. Domanski´ , Classical (PLS)-spaces: spaces of distributions, real analytic functions and their relatives, pp. 51–70, in: Orlicz Centenary Volume, Banach Center Publications, Warszawa, 2004. [15] I. M. Gel’fand, G. E. Shilov, Generalized functions. Vol. 2: Spaces of fundamental and generalized functions, Academic Press, New York-London, 1968. [16] K. Grochenig¨ , Foundations of time-frequency analysis, Birkhäuser Boston, Inc., Boston, MA, 2001. [17] A. Grothendieck, Produits tensoriels topologiques et espaces nucléaires, Mem. Amer. Math. Soc. 16, 1955. [18] T. Heinrich, R. Meise, A support theorem for quasianalytic functionals, Math. Nachr. 280 (2007), 364–387. [19] A. J. E. M. Janssen, Signal analytic proofs of two basic results on lattice expansions, Appl. Comp. Harmonic Anal. 1 (1994), 350–354. [20] T. Kawai, On the theory of Fourier hyperfunctions and its applications to partial differential equations with constant coefficients, J. Fac. Sci. Univ. Tokyo, Sec. IA 17 (1970), 467–517. [21] D. Kim, K. W. Kim, E. L. Lee, Convolution and multiplication operators in Fourier hyperfunctions, Integral Transforms Spec. Funct. 17 (2006), 53–63. [22] H. Komatsu, Ultradistributions I. Structure theorems and a characterization, J. Fac. Sci. Univ. Tokyo Sect. IA Math. 20 (1973), 25–105. [23] J. Larcher, J. Wengenroth, A new proof for the bornologicity of the space of slowly increasing functions, Bull. Belg. Math. Soc. Simon Stevin 5 (2014), 887–894. [24] M. Morimoto, Convolutors for ultrahyperfunctions, Internat. Sympos. Math. Problems in The- oret. Phys. (Kyoto, 1975), Lecture Notes in Phys., vol. 39, Springer-Verlag, 1975, pp. 49–54. [25] L. Schwartz, Théorie des distributions, Hermann, Paris, 1966. [26] M. A. Soloviev, Inclusion theorems for the Moyal multiplier algebras of generalized Gelfand- Shilov spaces, preprint (arXiv: 2007.13627). [27] M. Valdivia, A representation of the space M , Math. Z. 77 (1981), 463–478. [28] S. Kostadinova, S. Pilipović, K. Saneva andO J. Vindas, The short-time Fourier transform of distributions of exponential type and Tauberian theorems for S-asymptotics, Filomat 30 (2016), 3047–3061. 1 [29] D. Vogt, On the functors Ext (E, F ) for Fréchet spaces, Studia Math. 85 (1987), 163–197. [30] J. Wengenroth, Derived functors in functional analysis, Springer-Verlag, Berlin, 2003. [31] V. V. Zharinov, Fourier ultrahyperfunctions, (Russian) Izv. Akad. Nauk SSSR Ser. Mat. 44 (1980), 533–570; English translation: Math. USSR-Izv.16 (1981), 479–511.

Department of Mathematics: Analysis, Logic and Discrete Mathematics, Ghent Uni- versity, Krijgslaan 281, 9000 Gent, Belgium Email address: [email protected] Email address: [email protected]