Home Continuity properties and Bargmann mappings of quasi-Banach Orlicz modulation spaces
Article Open Access

Continuity properties and Bargmann mappings of quasi-Banach Orlicz modulation spaces

  • Joachim Toft ORCID logo EMAIL logo , Rüya Üster , Elmira Nabizadeh Morsalfard and Serap Öztop
Published/Copyright: July 29, 2022

Abstract

We deduce continuity, compactness and invariance properties for quasi-Banach Orlicz modulation spaces on d . We characterize such spaces in terms of Gabor expansions and by their images under the Bargmann transform.

1 Introduction

In analysis and its applications in science and technology, it is common to perform investigations, e.g., on continuity, in the framework on functions and distribution spaces whose elements obey some kind of Lebesgue norm estimates. For example, for L p -spaces, such norm estimates are applied on the functions themselves, and for Sobolev, Besov and modulation spaces such norm estimates are applied on suitable Fourier multipliers or short-time Fourier transforms on the involved distributions. In analogous ways, discrete Lebesgue spaces like p , or weighted versions of p , are often used when estimating sequences. We refer to [26] and Section 2 for notations.

On the other hand, in several situations, e.g., when dealing with non-linear functionals, it might be beneficial to replace Lebesgue norm estimates with suitable Orlicz norm estimates. For example, in statistics or statistical physics, the entropy applied on probability density functions f on d is given by

𝖤 ( f ) = - d f ( x ) log f ( x ) 𝑑 x .

When investigating 𝖤 , it might be more efficient to replace the pair of Lebesgue spaces ( L 1 , L ) by the pair of Orlicz spaces ( L log ( L + 1 ) , L cosh - 1 ) . That is, it might be suitable to replace L 1 and L by the Orlicz spaces L Φ , where the Young functions are given by

Φ ( t ) = t log ( 1 + t ) and Φ ( t ) = cosh ( t ) - 1 ,

respectively (see [28, 29] and the references therein).

Orlicz spaces are important types of Banach function spaces that are considered in mathematical analysis. They naturally generalize L p spaces and contain certain Sobolev spaces as subspaces (see, e.g., [9]). Orlicz spaces appear in computation such as the Zygmund space L log + L , which is a Banach space related to Hardy–Littlewood maximal functions.

In this paper, we extend the analysis in [18, 21] concerning classical modulation spaces M ( ω ) p , q ( d ) , and in [40] concerning Banach Orlicz modulation spaces to quasi-Banach Orlicz modulation spaces (quasi-Orlicz modulation spaces) M ( ω ) Φ 1 , Φ 2 ( d ) . Here Φ 1 and Φ 2 are quasi-Young functions of certain degrees.

The design of Orlicz modulation spaces follows a path similar to classical modulation spaces, and they are defined by imposing a mixed L ( ω ) Φ 1 , Φ 2 (quasi-)norm condition on the short-time Fourier transforms of the involved distributions. For a convenient approach to mixed norm spaces of Banach type, see [5].

In the restricted case when Φ 1 and Φ 2 above are Young functions, corresponding Orlicz modulation spaces M ( ω ) Φ 1 , Φ 2 ( d ) were introduced and investigated in [40] by Schnackers and Führ. Here it is deduced that for such Φ 1 and Φ 2 , M ( ω ) Φ 1 , Φ 2 ( d ) is a Banach space and it admits, in similar ways as for classical modulation spaces, characterizations by Gabor expansions. In [40], it is also shown that M ( ω ) Φ 1 , Φ 2 ( d ) is completely determined by the behavior of Φ 1 and Φ 2 at the origin in the sense that if

(1.1) lim t 0 + Ψ 1 ( t ) Φ 1 ( t ) and lim t 0 + Ψ 2 ( t ) Φ 2 ( t )

exist, then

(1.2) M ( ω ) Φ 1 , Φ 2 ( d ) M ( ω ) Ψ 1 , Ψ 2 ( d )

with continuous embedding. We remark that several properties for (Banach) Orlicz modulation spaces in [40] hold true in more general contexts where the Orlicz norms are replaced by norms from solid translation invariant Banach function spaces (see, e.g., [4, 11, 15, 16, 13]).

In our situation, allowing, more generally, Φ 1 and Φ 2 to be quasi-Young functions, we show that these and several other continuity properties in [18, 21, 33, 43, 46], for classical modulation spaces, carry over to Orlicz modulation spaces.

More precisely, we show that M ( ω ) Φ 1 , Φ 2 ( d ) are quasi-Banach spaces, and deduce invariance properties concerning the choices of window functions in the quasi-norms of the short-time Fourier transforms. By our general continuity results and similar arguments to the ones for classical modulation spaces, it follows that the injection map

i : M ( ω 1 ) Φ 1 , Φ 2 ( d ) M ( ω 2 ) Φ 1 , Φ 2 ( d )

is compact if and only if ω 2 / ω 1 tends to zero at infinity. This extends results in [33] to the Orlicz modulation space case. One part of the analysis concerns investigations of mapping properties of Orlicz modulation spaces under the Bargmann transform, of independent interest, given in Section 3. These investigations lead to the observation that the Bargmann transform is isometric and bijective from M ( ω ) Φ 1 , Φ 2 ( d ) to certain weighted versions of

L ( ω ) Φ 1 , Φ 2 ( 2 d ) L ( ω ) Φ 1 , Φ 2 ( d )

of entire analytic functions on d .

Several of these properties follow from our characterizations of Orlicz modulation spaces in terms of Gabor expansions, given in Section 5. In fact, here it is proved that for a distribution f, some lattice Λ d and suitable (window) functions ϕ and ψ on d , the analysis and synthesis operators

( C ϕ f ) = { ( V ϕ f ) ( j , ι ) } j , ι Λ and ( D ψ c ) = j , ι Λ c ( j , ι ) e i , ι ψ ( - j )

are continuous between the spaces M ( ω ) Φ 1 , Φ 2 ( d ) and ( ω ) Φ 1 , Φ 2 ( Λ × Λ ) . These properties imply that Orlicz modulation spaces possess Gabor properties in similar ways as for classical modulation spaces, or more generally in the context of co-orbit space theory, given in [15, 16] in the Banach space situation and in [38, 37] in the more general quasi-Banach situation. For example, we show that for each suitable (ultra-)distribution f we have

f = j , ι Λ ( V ϕ f ) ( j , ι ) ψ ( - j ) e i , ι ,

and that

(1.3) f M ( ω ) Φ 1 , Φ 2 ( d ) if and only if { ( V ϕ f ) ( j , ι ) } j , ι Λ ( ω ) Φ 1 , Φ 2 ( Λ × Λ ) ,

provided that the lattice Λ is dense enough. Here the series converges in M ( ω ) ( d ) with respect to the weak * topology. If in addition test function spaces like the Schwartz space 𝒮 ( d ) or a suitable Gelfand–Shilov space Σ 1 ( d ) are dense in L ( ω ) Φ 1 , Φ 2 ( 2 d ) , then the series above converges unconditionally with respect to the M ( ω ) Φ 1 , Φ 2 -norm. In particular, the Gabor analysis for classical modulation spaces in [18, 21] is extended to quasi-Banach Orlicz modulation spaces. Here we remark that the theory on co-orbit spaces for quasi-Banach spaces, given in [38, 37] by Rauhut, does not host our situations for different reasons. For example, contrary to [38, 37], we permit modulation spaces which have no non-trivial compactly supported elements. It also seems that heavy deductions are needed to pass from the abstract results in [37] to those Orlicz modulation spaces which are reachable by such methods.

The paper is organized as follows. In Section 2, we recall some basic properties on Gelfand–Shilov spaces, weight functions, Pilipović spaces, Orlicz spaces, and introduce quasi-Orlicz modulation spaces. Here we also recall some properties for classical modulation spaces concerning Gabor analysis, images under pseudo-differential operators and under the Bargmann transform.

In Section 3, we deduce mapping properties of Orlicz modulation spaces under the Bargmann transform. At the same time, we prove that they are complete, and thereby quasi-Banach spaces. In Section 4, we obtain convolution estimates for quasi-Orlicz spaces.

In Section 5, we apply the convolution results in Section 4 to extend the Gabor analysis for classical modulation spaces to quasi-Orlicz modulation spaces. In particular, we deduce (1.3) for quasi-Young functions Φ 1 and Φ 2 . This extends [40, Theorem 9] by Schnackers and Führ. We also apply the analysis to deduce basic continuity properties of quasi-Orlicz modulation spaces. For example, we show invariance properties with respect to the choice of window function, and use the equivalence (1.3) to show that (1.1) leads to (1.2), also when Φ 1 and Φ 2 are quasi-Young functions.

2 Preliminaries

In this section, we provide a review of some basic facts. In the first part, we recall the definition and explain some well-known facts about Gelfand–Shilov and Pilipović spaces and their spaces of (ultra-)distributions. Thereafter, we consider (mixed) Orlicz and quasi-Orlicz spaces and explain some basic properties. Our family of quasi-Orlicz spaces contains the family of Orlicz spaces, but is less comprehensive than the families of quasi-Orlicz spaces given in, e.g., [25] by Harjulehto and Hästö. We refer to [36, 25, 32] for more facts about Orlicz spaces.

Then we introduce and discuss basic properties of quasi-Banach Orlicz modulation spaces, which are obtained by imposing quasi-Orlicz norm estimates on the short-time Fourier transforms of the involved functions and distributions. Finally, we recall some basic facts in Gabor frame theory, and for the Bargmann transform.

2.1 Gelfand–Shilov spaces

We start by discussing Gelfand–Shilov spaces and their properties. Let 0 < s be fixed. Then the Gelfand–Shilov space 𝒮 s ( d ) (resp. Σ s ( d ) ) of Roumieu type (resp. Beurling type) with parameter s consists of all f C ( d ) such that

(2.1) f 𝒮 s , h sup | x β α f ( x ) | h | α + β | ( α ! β ! ) s

is finite for some h > 0 (resp. for every h > 0 ). Here the supremum should be taken over all α , β d and x d . We equip 𝒮 s ( d ) (resp. Σ s ( d ) ) with the canonical inductive limit topology (resp. projective limit topology) with respect to h > 0 , induced by the semi-norms in (2.1).

Whenever s > s 0 1 2 , we have

(2.2) 𝒮 s 0 ( d ) Σ s ( d ) 𝒮 s ( d ) 𝒮 ( d ) 𝒮 ( d ) 𝒮 s ( d ) Σ s ( d ) 𝒮 s 0 ( d ) ,

with dense embeddings. Here A B means that the topological spaces A and B satisfy A B with continuous embeddings. The space Σ s ( d ) is a Fréchet space with semi-norms 𝒮 s , h , h > 0 . Moreover, Σ s ( d ) { 0 } if and only if s > 1 2 , and 𝒮 s ( d ) { 0 } if and only if s 1 2 .

The Gelfand–Shilov distribution spaces 𝒮 s ( d ) and Σ s ( d ) are the dual spaces of 𝒮 s ( d ) and Σ s ( d ) , respectively. As for the Gelfand–Shilov spaces, there is a canonical projective limit topology (inductive limit topology) for 𝒮 s ( d ) ( Σ s ( d ) ) (cf. [19, 34, 35]).

From now on, we let be the Fourier transform, which takes the form

( f ) ( ξ ) = f ^ ( ξ ) ( 2 π ) - d 2 d f ( x ) e - i x , ξ 𝑑 x

when f L 1 ( d ) . Here , denotes the usual scalar product on d . The map extends uniquely to homeomorphisms on 𝒮 ( d ) , 𝒮 s ( d ) and on Σ s ( d ) . Furthermore, restricts to homeomorphisms on 𝒮 ( d ) , 𝒮 s ( d ) and on Σ s ( d ) , and to a unitary operator on L 2 ( d ) .

Gelfand–Shilov spaces can in convenient ways be characterized in terms of estimates of the functions and their Fourier transforms. More precisely, in [8, 48] it is proved that if f 𝒮 ( d ) and s > 0 , then f 𝒮 s ( d ) (resp. f Σ s ( d ) ) if and only if

(2.3) | f ( x ) | e - r | x | 1 s and | f ^ ( ξ ) | e - r | ξ | 1 s ,

for some r > 0 (resp. for every r > 0 ). Here g 1 g 2 means that g 1 ( θ ) c g 2 ( θ ) holds uniformly for all θ in the intersection of the domains of g 1 and g 2 for some constant c > 0 , and we write g 1 g 2 when g 1 g 2 g 1 .

Gelfand–Shilov spaces and their distribution spaces can also be characterized by estimates of short-time Fourier transforms (see, e.g., [24, 46]). More precisely, let ϕ 𝒮 s ( d ) be fixed. Then the short-time Fourier transform V ϕ f of f 𝒮 s ( d ) with respect to the window function ϕ is the Gelfand–Shilov distribution on 2 d , defined by

(2.4) V ϕ f ( x , ξ ) = ( f ϕ ( - x ) ¯ ) ( ξ ) .

If f L 2 ( d ) , then it follows that

V ϕ f ( x , ξ ) = ( 2 π ) - d 2 d f ( y ) ϕ ( y - x ) ¯ e - i y , ξ 𝑑 y .

2.2 Weight functions

A weight or weight function on d is a positive function ω 0 L loc ( d ) such that 1 / ω 0 L loc ( d ) . The weight ω 0 is called moderate if there are a positive weight v 0 on d and a constant C 1 such that

(2.5) ω 0 ( x + y ) C ω 0 ( x ) v 0 ( y ) , x , y d .

If ω 0 and v 0 are weights on d such that (2.5) holds, then ω is also called v-moderate. We note that (2.5) implies that ω fulfills the estimates

(2.6) C - 1 v 0 ( - x ) - 1 ω 0 ( x ) C v 0 ( x ) , x d .

We let 𝒫 E ( d ) be the set of all moderate weights on d .

It can be proved that if ω 0 𝒫 E ( d ) , then ω 0 is v 0 -moderate for some v 0 ( x ) = e r | x | , provided the positive constant r is large enough (cf. [22]). That is, (2.5) implies

(2.7) ω 0 ( x + y ) ω 0 ( x ) e r | y | , x , y d ,

for some r > 0 . In particular, (2.6) shows that for any ω 0 𝒫 E ( d ) there is a constant r > 0 such that

e - r | x | ω 0 ( x ) e r | x | , x d .

We say that v 0 is submultiplicative if v 0 is even and (2.5) holds with ω 0 = v 0 and C = 1 . We observe that if v 𝒫 E ( d ) is even and satisfies v ( x + y ) v ( x ) v ( y ) , x , y d , then there is a submultiplicative weight v 0 𝒫 E ( d ) such that v v 0 (see, e.g., [12, 15, 21]). We also recall that if (2.5) holds, then there is a smallest positive even function v 0 such that (2.5) holds with C = 1 . We remark that this smallest function v 0 is submultiplicative (see, e.g., [15, 20, 43]). In the sequel, v and v j , for j 0 , always stand for submultiplicative weights if nothing else is stated. We also remark that in the literature it is common to define submultiplicative weights as (2.5) should hold with ω 0 = v 0 and C = 1 , without the condition that v 0 should be even (cf., e.g., [12, 15, 18, 21]). However, in the sequel it is convenient for us to include this property in the definition.

We let 𝒫 E 0 ( d ) be the set of all ω 𝒫 E ( d ) such that (2.7) holds for every r > 0 . We also let 𝒫 ( d ) be the set of all polynomially moderate weights, i.e. the set of all ω 𝒫 E ( d ) such that for some r > 0 it holds

ω ( x + y ) ω ( x ) ( 1 + | y | ) r , x , y d .

Evidently,

(2.8) 𝒫 ( d ) 𝒫 E 0 ( d ) 𝒫 E ( d ) .

In several situations, our weights are defined on the phase space 2 d instead of the configuration space d . Then the weight ω belongs to 𝒫 E ( 2 d ) if and only if there is a submultiplicative weight v 𝒫 E ( 2 d ) such that

(\ref{moderate})’ ω ( x + y , ξ + η ) C ω ( x , ξ ) v ( y , η ) , x , y , ξ , η d .

Remark 2.1.

The class 𝒫 E 0 ( d ) is important when dealing with spectral invariance for matrix or convolution operators on 2 ( d ) (see, e.g., [23]). If v 𝒫 E ( d ) is submultiplicative, then v 𝒫 E 0 ( d ) if and only if

(2.9) lim n v ( n x ) 1 n = 1

(see, e.g., [17]). The condition (2.9) is equivalent to

(\ref{Eq:GRSCond})’ lim n log ( v ( n x ) ) n = 0 ,

and is usually called the GRS condition, or Gelfand–Raikov–Shilov condition. A more restrictive condition on v is given by the Beurling–Domar condition

(2.10) n = 1 log ( v ( n x ) ) n 2 < ,

which is strongly linked to non-quasi-analytic classes that contain non-trivial compactly supported elements (see, e.g., [22]).

In our discussions, we go for maximal generality of the ingredients, both concerning Orlicz spaces and the weights. It turns out that we will be able to deduce the searched properties without the restrictions in (2.9), (\ref{Eq:GRSCond})’ and (2.10). Hence, for us the most relevant weight class will be 𝒫 E ( d ) .

We refer to [10, 21, 22] for more facts about weight functions.

2.3 Pilipović spaces

In the course of our work, we will also make use of the Pilipović space ( d ) and its dual ( d ) . We recall from [46] that the Pilipović space ( d ) = 1 ( d ) is the set of all Hermite series expansions

(2.11) f = α d c f ( α ) h α

such that

(2.12) | c f ( α ) | r | α | α ! - 1 2

for some r > 0 . Here h α is the Hermite function of order α > 0 given by

h α ( x ) = π - d 4 ( - 1 ) | α | ( 2 | α | α ! ) - 1 2 e 1 2 | x | 2 ( α e - | x | 2 ) , α d .

In the same way, ( d ) consists of all formal Hermite series expansion (2.11) such that

(2.13) | c f ( α ) | r | α | α ! 1 2

for every r > 0 . The topologies of ( d ) and ( d ) are given by suitable inductive and projective limit topologies, respectively, with respect to r in (2.12) and (2.13) (see [46] for details). By identifying elements in 𝒮 ( d ) , 𝒮 s ( d ) and Σ s ( d ) , we get the dense embeddings

(2.14) { ( d ) 𝒮 1 / 2 ( d ) Σ s ( d ) 𝒮 s ( d ) 𝒮 ( d ) , 𝒮 ( d ) 𝒮 s ( d ) Σ s ( d ) 𝒮 1 / 2 ( d ) ( d ) , s > 1 2 .

We also have

(2.15) ( f , g ) L 2 ( d ) = α d c f ( α ) c g ( α ) ¯ ,

when f , g L 2 ( d ) . By letting the L 2 -form ( f , g ) L 2 ( d ) be equal to the right-hand side of (2.15) when f ( d ) and g ( d ) , it follows that ( d ) is the dual of ( d ) through a unique extension of the L 2 -form on ( d ) × ( d ) , to ( d ) × ( d ) or to ( d ) × ( d ) .

For future references, we note that if

ϕ ( x ) = π - d 4 e - 1 2 | x | 2 and f ( d ) ,

then the short-time Fourier transform (2.4) makes sense as a smooth function in view of [46, (2.25) and Theorem 4.1].

2.4 Quasi-Banach spaces

We recall that a quasi-norm of order r 0 ( 0 , 1 ] on the vector-space over is a nonnegative functional on which satisfies

(2.16) f + g 2 1 r 0 - 1 ( f + g ) , f , g ,
(2.17) α f = | α | f , α , f ,

and

(2.18) f = 0 if and only if f = 0 .

The space is then called a quasi-normed space. A complete quasi-normed space is called a quasi-Banach space. If is a quasi-Banach space with quasi-norm satisfying (2.16), then, by [1, 39], there is an equivalent quasi-norm to which additionally satisfies

(2.19) f + g r 0 f r 0 + g r 0 , f , g .

From now on, we always assume that the quasi-norm of the quasi-Banach space is chosen in such a way that both (2.16) and (2.19) hold. The space is then also called an r 0 -Banach space.

2.5 Orlicz spaces

Next, we define and recall some basic facts for (quasi-) Orlicz spaces (see [36, 25]). First, we give the definition of Young functions and quasi-Young functions.

Definition 2.2.

A function Φ : { } is called convex if

Φ ( s 1 t 1 + s 2 t 2 ) s 1 Φ ( t 1 ) + s 2 Φ ( t 2 ) ,

for every s j , t j which satisfy s j 0 and s 1 + s 2 = 1 , j = 1 , 2 .

We observe that Φ might not be continuous, because we permit as a function value. For example,

Φ ( t ) = { c , t a , , t > a ,

is convex but discontinuous at t = a .

Definition 2.3.

Let r 0 ( 0 , 1 ] and let Φ 0 and Φ be functions from [ 0 , ) to [ 0 , ] . Then Φ 0 is called a Young function if the following conditions hold:

  1. Φ 0 is convex.

  2. Φ 0 ( 0 ) = 0 .

  3. lim t Φ 0 ( t ) = + .

The function Φ is called r 0 -Young function or quasi-Young function of order r 0 if Φ ( t ) = Φ 0 ( t r 0 ) , t 0 , for some Young function Φ 0 .

It is clear that Φ in Definition 2.3 is non-decreasing, because if 0 t 1 t 2 and s [ 0 , 1 ] is chosen such that t 1 = s t 2 , then

Φ ( t 1 ) = Φ ( s t 2 + ( 1 - s ) 0 ) s Φ ( t 2 ) + ( 1 - s ) Φ ( 0 ) Φ ( t 2 ) ,

since Φ ( 0 ) = 0 , Φ ( t 2 ) 0 and s [ 0 , 1 ] .

Orlicz spaces are defined as follows.

Definition 2.4.

Let Ω d , ( Ω , Σ , μ ) be a Borel measure space, Φ 0 be a Young function and let ω 0 𝒫 E ( d ) .

  1. L ( ω 0 ) Φ 0 ( μ ) consists of all μ-measurable functions f : Ω such that

    f L ( ω 0 ) Φ 0 = inf { λ > 0 : Ω Φ 0 ( | f ( x ) ω 0 ( x ) | λ ) 𝑑 μ ( x ) 1 }

    is finite.

  2. Let Φ be a quasi-Young function of order r 0 ( 0 , 1 ] , given by Φ ( t ) = Φ 0 ( t r 0 ) , t 0 , for some Young function Φ 0 . Then L ( ω 0 ) Φ ( μ ) consists of all μ-measurable functions f : Ω such that

    f L ( ω 0 ) Φ ( | f ω 0 | r 0 L Φ 0 ) 1 / r 0

    is finite. We let L ( ω 0 ) Φ ( μ ) be equipped by the topology, induced by the quasi-norm L ( ω 0 ) Φ .

Remark 2.5.

Let Φ, Φ 0 and ω 0 be the same as in Definition 2.3. Then it follows by straightforward computation that

f L ( ω 0 ) Φ = inf { λ > 0 : Ω Φ 0 ( | f ( x ) ω 0 ( x ) | r 0 λ r 0 ) 𝑑 μ ( x ) 1 } .

In several situations, we need to consider mixed quasi-normed Orlicz spaces (see, e.g., [2, 5, 6, 27, 30, 31, 40] for details in the Banach space case, i.e. on mixed norm Orlicz spaces.)

Definition 2.6.

Let ( Ω j , Σ j , μ j ) be Borel measure spaces with Ω j d j , r 0 ( 0 , 1 ] , Φ j be r 0 -Young functions, j = 1 , 2 , and let ω 𝒫 E ( d 1 + d 2 ) . Then the mixed quasi-normed Orlicz space L ( ω ) Φ 1 , Φ 2 = L ( ω ) Φ 1 , Φ 2 ( μ 1 μ 2 ) consists of all μ 1 μ 2 -measurable functions f : Ω 1 × Ω 2 such that

f L ( ω ) Φ 1 , Φ 2 f 1 , ω L Φ 2

is finite, where

f 1 , ω ( x 2 ) = f ( , x 2 ) ω ( , x 2 ) L Φ 1 .

We let L ( ω ) Φ 1 , Φ 2 ( μ 1 , μ 2 ) be equipped by the topology, induced by the quasi-norm L ( ω ) Φ 1 , Φ 2 .

If r 0 = 1 in Definition 2.6, then L ( ω ) Φ 1 , Φ 2 ( μ 1 μ 2 ) is a Banach space and is called a mixed norm Orlicz space.

Remark 2.7.

Suppose Φ j are quasi-Young functions of order q j ( 0 , 1 ] , j = 1 , 2 . Then both Φ 1 and Φ 2 are quasi-Young functions of order r 0 = min ( q 1 , q 2 ) .

Remark 2.8.

Let ω, μ 1 and μ 2 be as in Definition 2.6. For p ( 0 , ] , let Φ p ( t ) = t p p when p < , and set Φ ( t ) = 0 when 0 t 1 , and Φ ( t ) = when t > 1 . Then it is well known that

L ( ω ) Φ p , Φ q = L ( ω ) p , q ,

with equality in quasi-norms. Hence, the family of quasi-Orlicz spaces contains the usual Lebesgue spaces and mixed quasi-normed spaces of Lebesgue type.

Remark 2.9.

Let r 0 , Φ j , μ j and ω be as in Definition 2.6. Then

(2.20) f L ( ω ) Φ 1 , Φ 2 = ( | f | r 0 L ( ω ) Φ 0 , 1 , Φ 0 , 2 ) 1 r 0

for some Young functions Φ 0 , 1 and Φ 0 , 2 . It follows that L ( ω ) Φ 1 , Φ 2 ( μ 1 μ 2 ) is an r 0 -Banach space.

In fact, the completeness of L ( ω ) Φ 1 , Φ 2 ( μ 1 μ 2 ) follows from equation (2.20) and the completeness of L ( ω ) Φ 0 , 1 , Φ 0 , 2 ( μ 1 μ 2 ) . Furthermore, by (2.20) and the fact that r 0 ( 0 , 1 ] , we get for every f , g L ( ω ) Φ 1 , Φ 2 ( μ 1 μ 2 ) that

f + g L ( ω ) Φ 1 , Φ 2 r 0 = | f + g | r 0 L ( ω ) Φ 0 , 1 , Φ 0 , 2
| f | r 0 + | g | r 0 L ( ω ) Φ 0 , 1 , Φ 0 , 2
= | f | r 0 L ( ω ) Φ 0 , 1 , Φ 0 , 2 + | g | r 0 L ( ω ) Φ 0 , 1 , Φ 0 , 2
= f L ( ω ) Φ 1 , Φ 2 r 0 + g L ( ω ) Φ 1 , Φ 2 r 0 ,

which shows that L ( ω ) Φ 1 , Φ 2 is a quasi-norm of order r 0 , giving the assertion.

In what follows, let 0 ( Λ ) be the set of all formal sequences

{ a ( n ) } n Λ = { a ( n ) : n Λ } ,

and let 0 ( Λ ) be the set of all sequences { a ( n ) } n Λ such that a ( n ) 0 for at most finite numbers of n.

Remark 2.10.

Let Λ d be a lattice, Φ, Φ 1 and Φ 2 be r 0 -Young functions, ω 0 , v 0 𝒫 E ( d ) and let ω , v 𝒫 E ( 2 d ) be such that ω 0 and ω are v 0 -moderate and v-moderate, respectively (in the sequel it is understood that all lattices contain 0). Then we set

L ( ω 0 ) Φ ( d ) = L ( ω 0 ) Φ ( μ ) and L ( ω 0 ) Φ 1 , Φ 2 ( 2 d ) = L ( ω 0 ) Φ 1 , Φ 2 ( μ μ ) ,

when μ is the Lebesgue measure on d . If instead μ is the standard (Haar) measure on Λ, i.e. μ ( n ) = 1 , n Λ , then we let

( ω ) Φ ( Λ ) = ( ω ) Φ ( μ ) and ( ω ) Φ 1 , Φ 2 ( Λ × Λ ) = ( ω ) Φ 1 , Φ 2 ( μ μ ) .

Evidently, ( ω ) Φ 1 , Φ 2 ( Λ × Λ ) 0 ( Λ × Λ ) .

Lemma 2.11.

Let Φ, Φ j be quasi-Young functions, j = 1 , 2 , ω 0 , v 0 P E ( R d ) and let ω , v P E ( R 2 d ) be such that (2.5) and (\ref{moderate})’ hold for some constant C > 0 . Then L ( ω 0 ) Φ ( R d ) and L ( ω ) Φ 1 Φ 2 ( R 2 d ) are invariant under translations, and

f ( - x ) L ( ω 0 ) Φ C f L ( ω 0 ) Φ v 0 ( x ) , f L ( ω 0 ) Φ ( d ) , x d ,
f ( - ( x , ξ ) ) L ( ω ) Φ 1 , Φ 2 C f L ( ω ) Φ 1 , Φ 2 v ( x , ξ ) , f L ( ω ) Φ 1 , Φ 2 ( 2 d ) , ( x , ξ ) 2 d .

Specifically, we observe that the constant C in Lemma 2.11 does only depend on the involved weight functions. In particular, it is independent of the choice of quasi-Young functions.

Lemma 2.11 is essentially well known (see, e.g., [12, 15, 38, 37]). In order to be self-contained, we give a proof of the second statement.

Proof.

We have Φ j ( t ) = Φ 0 , j ( t r 0 ) , t 0 , for some r 0 ( 0 , 1 ] and Young functions Φ 0 , j , j = 1 , 2 . This gives

f ( - ( x , ξ ) L ( ω ) Φ 1 , Φ 2 C ( | f ( - ( x , ξ ) ) ω ( - ( x , ξ ) ) v ( x , ξ ) | r 0 L Φ 0 , 1 , Φ 0 , 2 ) 1 r 0
= C ( | f ω | r 0 L Φ 0 , 1 , Φ 0 , 2 ) 1 r 0 v ( x , ξ )
= C f L ( ω ) Φ 1 , Φ 2 v ( x , ξ ) .

Here the inequality follows from the fact that ω is v-moderate, and the last two relations follow from the definitions. ∎

2.6 Orlicz modulation spaces

Before considering Orlicz modulation spaces, we recall the definition of classical modulation spaces (cf. [11, 12]).

Definition 2.12.

Let ϕ ( x ) = π - d 4 e - 1 2 | x | 2 , x d , p , q ( 0 , ] and let ω be a weight on 2 d . Then the modulation space M ( ω ) p , q ( d ) is the set of all f 𝒮 1 / 2 ( d ) such that V ϕ f L ( ω ) p , q ( 2 d ) . Also, let

(2.21) f M ( ω ) p , q V ϕ f L ( ω ) p , q .

We observe that f f M ( ω ) p , q is a quasi-norm, and let the topology of M ( ω ) p , q ( d ) be induced by this quasi-norm (see, e.g., [18, 44]).

For convenience, we set M p , q ( d ) = M ( ω ) p , q ( d ) when ω ( x , ξ ) = 1 , and we set M p = M p , p and M ( ω ) p = M ( ω ) p , p .

In the following proposition, we have listed some basic properties for modulation spaces. We refer to [11, 15, 18, 21, 43, 46] for the proof.

Proposition 2.13.

Let r ( 0 , 1 ] , p , p j , q , q j ( 0 , ] , ω , ω j , v P E ( R 2 d ) , j = 1 , 2 , be such that r p , q , and let ω be v-moderate. Then the following is true:

  1. Σ 1 ( d ) M ( ω ) p , q ( d ) Σ 1 ( d ) with continuous inclusions. If in addition p , q < , then Σ 1 ( d ) is dense in M ( ω ) p , q ( d ) . If, more restrictedly, ω 𝒫 ( 2 d ) , then 𝒮 ( d ) M ( ω ) p , q ( d ) 𝒮 ( d ) .

  2. If ϕ M ( v ) r ( d ) 0 , then f M ( ω ) p , q ( d ) if and only if V ϕ f ω L p , q is finite. In particular, M ( ω ) p , q ( d ) is independent of the choice of ϕ M ( v ) r ( d ) 0 . Moreover, different choices of ϕ in ( 2.21 ) give rise to equivalent quasi-norms.

  3. M ( ω 1 ) p 1 , q 1 ( d ) M ( ω 2 ) p 2 , q 2 ( d ) when p 1 p 2 , q 1 q 2 and ω 2 ω 1 .

  4. If ω 0 ( ξ , x ) = ω ( - x , ξ ) , then is a homeomorphism from M ( ω ) p ( d ) to M ( ω 0 ) p ( d ) .

Let Φ , Φ 1 , Φ 2 be quasi-Young functions and let ω be a weight on 2 d . Then the Orlicz modulation spaces M ( ω ) Φ ( d ) and M ( ω ) Φ 1 , Φ 2 ( d ) are given by

(2.22) M ( ω ) Φ ( d ) = { f ( d ) : V ϕ f L ( ω ) Φ ( 2 d ) }

and

(2.23) M ( ω ) Φ 1 , Φ 2 ( d ) = { f ( d ) : V ϕ f L ( ω ) Φ 1 , Φ 2 ( 2 d ) } .

The quasi-norms on M ( ω ) Φ ( d ) and M ( ω ) Φ 1 , Φ 2 ( d ) are given by

(2.24) f M ( ω ) Φ = V ϕ f L ( ω ) Φ

and

(2.25) f M ( ω ) Φ 1 , Φ 2 = V ϕ f L ( ω ) Φ 1 , Φ 2 .

Remark 2.14.

Let Φ 1 and Φ 2 be quasi-Young functions and let ω 𝒫 E ( 2 d ) . We observe that L Φ 1 , Φ 2 ( 2 d ) is translation invariant, which leads to the fact that M ( ω ) Φ 1 , Φ 2 ( d ) is translation and modulation invariant. In particular, if in addition Φ 1 and Φ 2 are Young functions, then M ( ω ) Φ 1 , Φ 2 ( d ) is stable under continuous convolutions with elements in Beurling algebras, and it is stable under pointwise multiplication with elements in Fourier–Beurling algebras (see, e.g., the analysis in [15]).

We notice that (2.24) and (2.25) are norms when Φ , Φ 1 and Φ 2 are Young functions. If ω 𝒫 E ( 2 d ) as in Definition 2.12, then we prove later on that the conditions

V ϕ f L ( ω ) Φ 1 Φ 2 < and V ϕ f L ( ω ) Φ <

are independent of the choices of ϕ in Σ 1 ( d ) 0 and that different ϕ give rise to equivalent quasi-norms (see Theorem 5.1).

Later on, we need the following proposition.

Proposition 2.15.

Let Φ , Φ j be Young functions, j = 1 , 2 , and let ω 0 P E ( R d ) and ω P E ( R 2 d ) . Then

𝒮 ( d ) L Φ ( d ) 𝒮 ( d ) , 𝒮 ( 2 d ) L Φ 1 , Φ 2 ( 2 d ) 𝒮 ( 2 d ) ,
Σ 1 ( d ) L ( ω 0 ) Φ ( d ) Σ 1 ( d ) , Σ 1 ( 2 d ) L ( ω ) Φ 1 , Φ 2 ( 2 d ) Σ 1 ( 2 d ) .

Proof.

Let v 0 𝒫 E ( d ) and v 𝒫 E ( 2 d ) be chosen such that ω 0 is v 0 -moderate and ω is v-moderate. Since L ( ω 0 ) Φ ( d ) and L ( ω ) Φ 1 , Φ 2 ( 2 d ) are invariant under translation and modulation, we have (see [21, Theorem 12.1.8])

M ( v 0 ) 1 ( d ) L ( ω 0 ) Φ ( d ) M ( 1 / v 0 ) ( d )

and

M ( v ) 1 ( d ) L ( ω ) Φ 1 , Φ 2 ( 2 d ) M ( 1 / v ) ( d ) .

The result now follows from Proposition 2.13 (i). ∎

2.7 Gabor frames

Let { e 1 , , e d } be an ordered basis of d . The corresponding lattice is given by

(2.26) Λ = { n 1 e 1 + + n d e d : ( n 1 , , n d ) d } .

We put

Λ 2 = Λ × Λ = { ( m , n ) d × d : m , n Λ } .

Definition 2.16.

Let ω , v 𝒫 E ( 2 d ) be such that ω is v-moderate, ϕ , ψ M ( v ) 1 ( d ) and let Λ d be as in (2.26).

  1. The analysis operator C ϕ Λ is the operator from M ( ω ) ( d ) to ( ω ) ( Λ 2 ) , given by

    C ϕ Λ f { V ϕ f ( j , ι ) } j , ι Λ .

  2. The synthesis operator D ψ Λ is the operator from ( ω ) ( Λ 2 ) to M ( ω ) ( d ) , given by

    D ψ Λ c j , ι Λ c ( j , ι ) e i , ι ψ ( - j ) .

  3. The Gabor frame operator S ϕ , ψ Λ is the operator on M ( ω ) ( d ) , given by D ψ Λ C ϕ Λ , i.e.

    S ϕ , ψ Λ f j , ι Λ V ϕ f ( j , ι ) e i , ι ψ ( - j ) .

We set C ϕ ε = C ϕ Λ , D ψ ε = D ψ Λ and S ϕ , ψ ε = S ϕ , ψ Λ when Λ = ε d . It follows that

C ϕ ε f = { V ϕ f ( j , ι ) } j , ι ε d

and

D ψ ε c = j , ι ε d c ( j , ι ) e i , ι ψ ( - j ) .

Let ϕ L 2 ( d ) 0 be fixed and let Λ be as above. Then recall that

(2.27) { ϕ ( - j ) e i , ι } j , ι Λ

is called a Gabor frame if there are constants C 1 , C 2 > 0 such that

(2.28) C 1 f L 2 2 j , ι Λ | ( f , ϕ ( - j ) e i , ι ) L 2 | 2 C 2 f L 2 2 , f L 2 ( d ) .

We observe that, by slightly modifying the constants C 1 and C 2 , (2.28) is the same as

C 1 f L 2 2 j , ι Λ | V ϕ f ( j , ι ) | 2 C 2 f L 2 2 , f L 2 ( d ) .

The function ϕ is called the (frame) atom for the frame (2.27) (see, e.g., [20]).

If (2.27) is a Gabor frame, then there is a Gabor frame

(2.29) { ψ ( - j ) e i , ι } j , ι Λ

with ψ L 2 ( d ) 0 such that

f ( x ) = j , ι Λ V ϕ ( j , ι ) ψ ( x - j ) e i x , ι , f L 2 ( d ) ,

where the series converges in L 2 ( d ) . By duality, it follows that (see, e.g., [20])

f ( x ) = j , ι Λ V ψ ( j , ι ) ϕ ( x - j ) e i x , ι , f L 2 ( d ) .

Remark 2.17.

Gabor frames can be obtained for general lattices which are dense enough. Hence, the assumption above that the configuration variable j and momentum variable ι above should belong to separate lattices Λ d is not needed. It suffices to assume that ( j , ι ) should belong to a dense enough lattice in the phase space 2 d .

On the other hand, in some computations later on, it is convenient to restrict ourselves to separate lattices.

The next result shows that it is possible to find suitable ϕ and ψ in the previous definition. We omit the proof since the result is a special case of [20, Theorem S].

Lemma 2.18.

Let Λ R d be as in (2.26), v P E ( R 2 d ) and let ϕ M ( v ) 1 ( R d ) 0 . Then there is an ε 0 > 0 such that for every ε ( 0 , ε 0 ] there is a ψ M ( v ) 1 ( R d ) 0 such that

(2.30) { ϕ ( x - j ) e i x , ι } j , ι ε Λ 𝑎𝑛𝑑 { ψ ( x - j ) e i x , ι } j , ι ε Λ

are dual frames to each others.

Remark 2.19.

Let r ( 0 , 1 ] . Then there are several ways to achieve dual frames (2.30) such that ϕ and ψ belong to the subspace M ( v ) r ( d ) of M ( v ) 1 ( d ) . In fact, let v , v 0 𝒫 E ( 2 d ) be submultiplicative such that ω is v-moderate and

L ( v 0 ) 1 ( 2 d ) L r ( 2 d ) , r ( 0 , 1 ] .

Then Lemma 2.18 guarantees that for some choice of ϕ , ψ M ( v 0 v ) 1 ( d ) M ( v ) r ( d ) and lattice Λ, the sets in (2.30) are dual frames to each other, and that ψ = ( S ϕ , ϕ Λ ) - 1 ϕ (cf. [45, Proposition 1.5 and Remark 1.6]).

Lemma 2.20.

Let Λ R d be as in (2.26), v P E ( R 4 d ) be submultiplicative, ϕ 1 , ϕ 2 Σ 1 ( R d ) 0 and let

(2.31) φ ( x , ξ ) = ϕ 1 ( x ) ϕ ^ 2 ( ξ ) ¯ e - i x , ξ .

Then there is an ε > 0 such that

{ φ ( x - j , ξ - ι ) e i ( x , κ + k , ξ ) } j , k , ι , κ ε Λ

is a Gabor frame with canonical dual frame

{ ψ ( x - j , ξ - ι ) e i ( x , κ + k , ξ ) } j , k , ι , κ ε Λ ,

where ψ = ( S φ , φ ε , Λ 2 ) - 1 φ belongs to M ( v ) r ( R 2 d ) for every r > 0 .

We recall that φ in (2.31) is called the Rihaczek function of ϕ 1 and ϕ 2 (see, e.g., [7]).

The next result gives some information about the roles that Φ 1 and Φ 2 play for M Φ 1 , Φ 2 in the Banach space case. We omit the proof since it can be found in [40].

Proposition 2.21.

Let Λ R d be a lattice given by (2.26) and let Φ j , Ψ j be Young functions, j = 1 , 2 . Then the following conditions are equivalent:

  1. M Φ 1 , Φ 2 ( d ) M Ψ 1 , Ψ 2 ( d ) .

  2. Φ 1 , Φ 2 ( Λ × Λ ) Ψ 1 , Ψ 2 ( Λ × Λ ) .

  3. There is a constant t 0 > 0 such that Ψ j ( t ) Φ j ( t ) for all 0 t t 0 .

In Section 5, we extend Proposition 2.21 to the quasi-Banach case (see Theorem 5.10 and Proposition 5.11).

2.8 The Bargmann transform and modulation spaces

We finish this section by recalling the Bargmann transform and its mapping properties on modulation spaces.

The Bargmann kernel of dimension d is given by

𝔄 d ( z , y ) = π - d 4 exp ( - 1 2 ( z , z + | y | 2 ) + 2 1 2 z , y ) , y d , z d .

Here (see [3])

z , w = j = 1 d z j w j , z = ( z 1 , , z d ) d  and  w = ( w 1 , , w d ) d .

It follows that y 𝔄 d ( z , y ) belongs to 𝒮 1 / 2 ( d ) for every z d .

The Bargmann transform 𝔙 d f of f 𝒮 1 / 2 ( d ) is defined by

( 𝔙 d f ) ( z ) = f , 𝔄 d ( z , ) .

There are several results on Bargmann images of well-known function and distribution spaces. For example, it was already proved in [3] that 𝔙 d is an isometric bijection from L 2 ( d ) = M 2 ( d ) into L 2 ( d μ ) A ( d ) . Here

d μ ( z ) = π - d e - | z | 2 d λ ( z ) ,

where d λ ( z ) is the Lebesgue measure on d , and A ( d ) is the set of analytic functions on d . A more general result of the preceding result concerns Proposition 2.22 below, which is a special case of [46, Theorem 4.8] (see also [14, 41, 43] for sub-results.)

If F is a measurable function on d and p , q ( 0 , ] , then let

F B ( ω ) p , q F p , ω L q ( d ) ,

where

F p , ω ( ξ ) ( 2 π ) - d 2 e - 1 2 | ξ | 2 e - 1 4 | | 2 F ( 2 - 1 2 ( - i ξ ) ) ω ( , ξ ) L p ( d ) .

We let B ( ω ) p , q ( d ) be the set of all measurable functions F on d such that F B ( ω ) p , q is finite. We also let

A ( ω ) p , q ( d ) = B ( ω ) p , q ( d ) A ( d )

with topology inherited from the topology in B ( ω ) p , q ( d ) .

Proposition 2.22.

Let p , q ( 0 , ] and let ω be a weight on R 2 d C d . Then V d from S 1 / 2 ( R d ) to A ( C d ) is uniquely extendable to an isometric bijective map from M ( ω ) p , q ( R d ) to A ( ω ) p , q ( C d ) .

Apart from Proposition 2.22, there are several characterizations of well-known function and distribution spaces via their images under the Bargmann transform. For example, convenient characterization of ( d ) , 𝒮 s ( d ) , Σ s ( d ) , 𝒮 ( d ) and their duals can be found in [43, 46] for s > 0 . Specifically, we remark that the Bargmann transform on L 2 ( d ) is uniformly extendable to a bijective map from ( d ) to A ( d ) , and restricts to a bijective map ( d ) to the set of all entire functions F on d such that | F ( z ) | e r | z | for some r > 0 .

Let Φ 1 and Φ 2 be quasi-Young functions. In a similar way as for the definition of B ( ω ) p , q ( d ) above, we let B ( ω ) Φ 1 , Φ 2 ( d ) be the set of all measurable functions F on d such that F B ( ω ) Φ 1 , Φ 2 is finite, where

F B ( ω ) Φ 1 , Φ 2 F Φ 1 , ω L Φ 2 ( d ) ,

with

F Φ 1 , ω ( ξ ) ( 2 π ) - d 2 e - 1 2 | ξ | 2 e - 1 4 | | 2 F ( 2 - 1 2 ( - i ξ ) ) ω ( , ξ ) L Φ 1 ( d ) .

We also let

A ( ω ) Φ 1 , Φ 2 ( d ) = B ( ω ) Φ 1 , Φ 2 ( d ) A ( d )

with topology inherited from the topology in B ( ω ) Φ 1 , Φ 2 ( d ) . It follows that

B ( ω ) p , q = B ( ω ) Φ 1 , Φ 2 and A ( ω ) p , q = A ( ω ) Φ 1 , Φ 2

when Φ 1 and Φ 2 are chosen such that Φ 1 ( t ) = t p p and Φ 2 ( t ) = t q q , giving that L ( ω ) Φ 1 , Φ 2 = L ( ω ) p , q .

3 Continuity and Bargmann images of Orlicz modulation spaces

In this section, we extend Proposition 2.22 to more general weights and to the Orlicz case (see Theorem 3.1). At the same time, we prove that the Orlicz modulation spaces are quasi-Banach spaces, by deducing similar facts of their Bargmann images.

The extension of Proposition 2.22 is the following theorem. Here we let ϕ ( x ) = π - d 4 e - 1 2 | x | 2 in the modulation space norms.

Theorem 3.1.

Let ω be a weight on R 2 d C d and let Φ 1 , Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] . Then the following is true:

  1. A ( ω ) Φ 1 , Φ 2 ( d ) , B ( ω ) Φ 1 , Φ 2 ( d ) and M ( ω ) Φ 1 , Φ 2 ( d ) are quasi-Banach spaces of order r 0 .

  2. The Bargmann transform is isometric and bijective from M ( ω ) Φ 1 , Φ 2 ( d ) to A ( ω ) Φ 1 , Φ 2 ( d ) .

We need the following lemma for the proof. Here and in what follows, we let z = ( 1 + | z | 2 ) 1 2 when z d .

Lemma 3.2.

Let ω be a weight on R 2 d C d , and let Φ j be quasi-Young functions of order r 0 ( 0 , 1 ] , j = 1 , 2 . Then the following is true:

  1. If ρ ( 0 , 1 ] and ω r ( z ) = ω ( z ) e - r | z | ρ for some r > 0 , then A ( ω ) Φ 1 , Φ 2 ( d ) is continuously embedded in A ( ω r ) r 0 ( d ) .

  2. If r > 2 d r 0 and ω r ( z ) = ω ( z ) z - r , then A ( ω ) Φ 1 , Φ 2 ( d ) is continuously embedded in A ( ω r ) r 0 ( d ) .

Proof.

For every r 1 , r 2 > 0 , we have e - r 1 | z | ρ z - r 2 , which implies that it suffices to prove (ii). Since Φ 1 and Φ 2 are quasi-Young functions of order r 0 , we have

Φ j ( t ) { 0 , 0 t t 0 , C ( t - t 0 ) r 0 , t > t 0 ,

for some choices of t 0 > 0 and C > 0 . This implies that

L ( ω ) Φ 1 , Φ 2 L ( ω ) r 0 + L ( ω ) r 0 , + L ( ω ) , r 0 + L ( ω ) .

Since

L ( ω ) r 0 L ( ω r ) r 0 , L ( ω ) r 0 , L ( ω r ) r 0 , L ( ω ) , r 0 L ( ω r ) r 0 , L ( ω ) L ( ω r ) r 0 ,

we get

L ( ω ) Φ 1 , Φ 2 L ( ω r ) r 0 ,

which in turn leads to

A ( ω ) Φ 1 , Φ 2 ( d ) A ( ω r ) r 0 ( d ) ,

with continuous inclusion. ∎

Proof of Theorem 3.1.

Let ρ ( 0 , 1 ) and ω r = ω e - r | | ρ . Since B ( ω ) Φ 1 , Φ 2 ( d ) is essentially a weighted Orlicz space, the completeness of B ( ω ) Φ 1 , Φ 2 ( d ) follows from the completeness of L ( ω ) Φ 1 , Φ 2 ( 2 d ) . Suppose that { F j } j = 1 is a Cauchy sequence in A ( ω ) Φ 1 , Φ 2 ( d ) . Since A ( ω ) Φ 1 , Φ 2 ( d ) is continuously embedded in A ( ω r ) r 0 ( d ) for every r > 0 , { F j } j = 1 is a Cauchy sequence in A ( ω r ) r 0 ( d ) as well.

By the completeness of B ( ω ) Φ 1 , Φ 2 ( d ) and A ( ω r ) r 0 ( d ) , there are F B ( ω ) Φ 1 , Φ 2 ( d ) and F 0 A ( ω r ) r 0 ( d ) such that

F j F in  B ( ω ) Φ 1 , Φ 2 ( d )    and    F j F 0 in  A ( ω r ) r 0 ( d )     as j .

Because B ( ω ) Φ 1 , Φ 2 ( d ) and A ( ω r ) r 0 ( d ) are equipped with weighted Lebesgue norms, we have F = F 0 a.e. Since F 0 A ( d ) , we get

F B ( ω ) Φ 1 , Φ 2 ( d ) A ( d ) = A ( ω ) Φ 1 , Φ 2 ( d ) ,

giving the completeness of A ( ω ) Φ 1 , Φ 2 ( d ) .

Due to the completeness of A ( ω ) Φ 1 , Φ 2 ( d ) , the completeness of M ( ω ) Φ 1 , Φ 2 ( d ) follows if we prove (ii).

By the definitions, it follows that

(3.1) 𝔙 d f B ( ω ) Φ 1 , Φ 2 = f M ( ω ) Φ 1 , Φ 2 , f ( d ) .

Since 𝔙 d f A ( d ) when f ( d ) , it follows from (3.1) that

𝔙 d f B ( ω ) Φ 1 , Φ 2 A ( d ) = A ( ω ) Φ 1 , Φ 2 ( d )

when f M ( ω ) Φ 1 , Φ 2 ( d ) . This shows that 𝔙 d is an isometric injection from M ( ω ) Φ 1 , Φ 2 ( d ) to A ( ω ) Φ 1 , Φ 2 ( d ) . We need to prove the surjectivity of this map.

Suppose F A ( ω ) Φ 1 , Φ 2 ( d ) A ( d ) . Since any element in A ( d ) is a Bargmann transform of an element in ( d ) , we have

F = 𝔙 d f

for some f ( d ) . By (3.1), we get

f M ( ω ) Φ 1 , Φ 2 = 𝔙 d f A ( ω ) Φ 1 , Φ 2 < ,

giving that f M ( ω ) Φ 1 , Φ 2 ( d ) . This gives the asserted surjectivity, and thereby the result. ∎

Now we have the following inclusion relations between Orlicz modulation spaces, Gelfand–Shilov and Schwartz spaces and their duals. We recall Section 2.2 for notations on weight classes.

Proposition 3.3.

Let s 0 , σ 0 1 2 , s , σ > 1 2 , Φ 1 , Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] , and let

v r ( x , ξ ) = ( 1 + | x | + | ξ | ) r 𝑎𝑛𝑑 v s , σ , r ( x , ξ ) = e r ( | x | 1 s + | ξ | 1 σ ) .

Then

Σ s σ ( d ) = r > 0 M ( v s , σ , r ) Φ 1 , Φ 2 ( d ) , ( Σ s σ ) ( d ) = r > 0 M ( 1 / v s , σ , r ) Φ 1 , Φ 2 ( d ) ,
𝒮 s 0 σ 0 ( d ) = r > 0 M ( v s 0 , σ 0 , r ) Φ 1 , Φ 2 ( d ) , ( 𝒮 s 0 σ 0 ) ( d ) = r > 0 M ( 1 / v s 0 , σ 0 , r ) Φ 1 , Φ 2 ( d ) ,
𝒮 ( d ) = r > 0 M ( v r ) Φ 1 , Φ 2 ( d ) , 𝒮 ( d ) = r > 0 M ( 1 / v r ) Φ 1 , Φ 2 ( d ) .

Before giving the proof of Proposition 3.3, we give the following consequence.

Corollary 3.4.

Let Φ 1 , Φ 2 be the same as in Proposition 3.3. Then

Σ 1 ( d ) = ω 𝒫 E M ( ω ) Φ 1 , Φ 2 ( d ) , 𝒮 1 ( d ) = ω 𝒫 E 0 M ( ω ) Φ 1 , Φ 2 ( d ) , 𝒮 ( d ) = ω 𝒫 M ( ω ) Φ 1 , Φ 2 ( d )

and

Σ 1 ( d ) = ω 𝒫 E M ( ω ) Φ 1 , Φ 2 ( d ) , 𝒮 1 ( d ) = ω 𝒫 E 0 M ( ω ) Φ 1 , Φ 2 ( d ) , 𝒮 ( d ) = ω 𝒫 M ( ω ) Φ 1 , Φ 2 ( d ) .

Proof.

The result follows from the definitions, Proposition 3.3 and the fact that

M ( v ) Φ 1 , Φ 2 ( d ) M ( ω ) Φ 1 , Φ 2 ( d ) M ( 1 / v ) Φ 1 , Φ 2 ( d ) ,

where ω , v 𝒫 E ( 2 d ) are such that v is submultiplicative and ω is v-moderate. ∎

We need the following lemma for the proof of Proposition 3.3.

Lemma 3.5.

Let Φ 1 , Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] and let ω be a weight on R 2 d . Then

L ( ω ) r 0 ( 2 d ) L ( ω ) ( 2 d ) L ( ω ) Φ 1 , Φ 2 ( 2 d ) L ( ω ) r 0 ( 2 d ) + L ( ω ) ( 2 d ) .

A proof of Lemma 3.5 is essentially given in [40]. In order to be self-contained, here we give the arguments.

Proof.

By the mapping f | f ω | r 0 , we reduce ourselves to the case ω = 1 and r 0 = 1 . Since Φ 1 and Φ 2 are convex, there are constants t 1 , t 2 > 0 and C 1 , C 2 > 0 such that if

Ψ 1 ( t ) = { 0 , 0 t t 1 , C 1 ( t - t 1 ) , t > t 1 ,

and

Ψ 2 ( t ) = { C 2 t , 0 t t 2 , , t > t 2 ,

then

Ψ 1 ( t ) Φ j ( t ) Ψ 2 ( t ) , j = 1 , 2 .

This gives

L 1 ( 2 d ) L ( 2 d ) = L Ψ 2 ( 2 d ) L Φ 1 , Φ 2 ( 2 d ) L Ψ 1 ( 2 d ) = L 1 ( 2 d ) + L ( 2 d ) .

Corollary 3.6.

Let Φ 1 , Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] and let ω P E ( R 2 d ) . Then

M ( ω ) r 0 ( d ) M ( ω ) Φ 1 , Φ 2 ( d ) M ( ω ) ( d ) .

Proof.

Since ω 𝒫 E ( 2 d ) , it follows that M ( ω ) p , q ( d ) increases with p , q ( 0 , ] . Hence, Lemma 3.5 gives

M ( ω ) r 0 ( d ) = M ( ω ) r 0 ( d ) M ( ω ) ( d ) M ( ω ) Φ 1 , Φ 2 ( d ) M ( ω ) r 0 ( d ) + M ( ω ) ( d ) = M ( ω ) ( d ) .

Proof of Proposition 3.3.

If the Φ j are chosen such that

M ( ω ) Φ 1 , Φ 2 = M ( ω ) p , q

for some p , q ( 0 , ] , then the result follows by a combination of [42, Remark 1.3 (5)], [43, Theorem 3.9] and [46, Proposition 6.5]. For general Φ j , the result follows by combining [43, Theorem 3.2] with Lemma 3.5. ∎

4 Convolution estimates for quasi-Orlicz spaces

In this section, we extend the convolution estimates in [18] for Lebesgue spaces to the case of quasi-Orlicz spaces. In the first part, we deduce discrete convolution estimates between elements in discrete Orlicz and Lebesgue spaces. Thereafter, we focus on the semi-continuous convolution, and prove corresponding estimates for L ( ω ) Φ ( 2 d ) or in convolutions between elements in L ( ω ) Φ 1 , Φ 2 ( 2 d ) and ( v ) r 0 . Finally, at the end of this section, we also deduce similar estimates for continuous convolutions after L ( ω ) Φ ( 2 d ) , L ( ω ) Φ 1 , Φ 2 ( 2 d ) and ( v ) r 0 are replaced by the Wiener spaces W ( L ( ω ) Φ 1 , Φ 2 ) and W ( L 1 , L ( v ) r 0 ) .

Our investigations involve weighted Orlicz spaces where the weights should satisfy conditions of the form

(4.1) ω 0 ( x + y ) C ω 1 ( x ) ω 2 ( y ) , x , y d ,

or

(4.2) ω 0 ( x + y , ξ + η ) C ω 1 ( x , ξ ) ω 2 ( y , η ) , x , y , ξ , η d ,

for some constant C > 0 which is independent of x , y , ξ , η d .

4.1 Discrete convolution estimates on discrete Orlicz spaces

We recall that the discrete convolution between a 0 ( Λ ) and b 0 ( Λ ) is defined by

( a * b ) ( n ) = k Λ a ( k ) b ( n - k ) .

Lemma 4.1.

Let Λ be as in (2.26), let ω 0 P E ( R d ) , let B 1 0 ( Λ ) be a quasi-Banach space, and let B 2 0 ( Λ ) be a quasi-Banach space of order r 0 ( 0 , 1 ] such that k a ( k - n ) B 2 when a B 1 , n Λ and

a ( - n ) 2 C ω 0 ( n ) a 1 , a 1 , n Λ ,

for some constant C > 0 . Then the map ( a , b ) a * b from 0 ( Λ ) × B 1 to 0 ( Λ ) is uniquely extendable to a continuous map from ( ω 0 ) r 0 ( Λ ) × B 1 to B 2 , and

(4.3) a * b 2 C a ( ω 0 ) r 0 b 1 , a ( ω 0 ) r 0 ( Λ ) , b 1 .

Lemma 4.1 follows by similar arguments to the ones in the proof of [18, Lemma 2.5] (see also [15, 38, 37]). In order to be self-contained, here we show the arguments.

Proof.

Since 0 is dense in ( ω 0 ) r 0 , the result follows if we prove (4.3) for a 0 ( Λ ) .

Since j are r 0 -Banach spaces, j = 1 , 2 , we have

a * b 2 r 0 = k Λ a ( k ) b ( - k ) 2 r 0
k Λ a ( k ) b ( - k ) 2 r 0
C r 0 k Λ ( | a ( k ) ω 0 ( k ) ) r 0 | b 1 r 0
= ( C a ( ω 0 ) r 0 b 1 ) r 0 ,

which gives (4.3). ∎

By choosing j as ( ω j ) Φ ( d ) or ( ω j ) Φ 1 , Φ 2 ( 2 d ) in the previous lemma, for suitable ω j , and using Lemma 2.11, we get the following corollary.

Corollary 4.2.

Let Λ be as in (2.26) and let Φ, Φ 1 and Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] . Then the following is true:

  1. Suppose that ω j 𝒫 E ( d ) , j = 0 , 1 , 2 , satisfy ( 4.1 ) for some constant C > 0 . Then the map ( a , b ) a * b from 0 ( Λ ) × ( ω 2 ) Φ ( Λ ) to 0 ( Λ ) is uniquely extendable to a continuous map from ( ω 1 ) r 0 ( Λ ) × ( ω 2 ) Φ ( Λ ) to ( ω 0 ) Φ ( Λ ) , and

    a * b ( ω 0 ) Φ C a ( ω 1 ) r 0 b ( ω 2 ) Φ , a ( ω 1 ) r 0 ( Λ ) , b ( ω 2 ) Φ ( Λ ) .

  2. Let Λ 2 = Λ × Λ and suppose that ω j 𝒫 E ( 2 d ) , j = 0 , 1 , 2 , satisfy ( 4.2 ) for some constant C > 0 . Then the map ( a , b ) a * b from 0 ( Λ 2 ) × ( ω 2 ) Φ 1 , Φ 2 ( Λ 2 ) to 0 ( Λ 2 ) is uniquely extendable to a continuous map from ( ω 1 ) r 0 ( Λ 2 ) × ( ω 2 ) Φ 1 , Φ 2 ( Λ 2 ) to ( ω 0 ) Φ 1 , Φ 2 ( Λ 2 ) , and

    a * b ( ω 0 ) Φ 1 , Φ 2 C a ( ω 1 ) r 0 b ( ω 2 ) Φ 1 , Φ 2 , a ( ω 1 ) r 0 ( Λ 2 ) , b ( ω 2 ) Φ 1 , Φ 2 ( Λ 2 ) .

Next, we perform similar investigations for semi-discrete convolutions.

Definition 4.3.

Let Λ be as in (2.26). The semi-discrete convolution of a 0 ( Λ ) and f Σ 1 ( d ) with respect to Λ is given by

( a * Λ f ) ( x ) = k Λ a ( k ) f ( x - k ) .

For ε > 0 , we also set * ε = * Λ when Λ = ε d . Then

( a * ε f ) ( x ) = k ε d a ( k ) f ( x - k ) .

The following result corresponds to Lemma 4.1 in the framework of semi-discrete convolutions. Here and in what follows, we let ( d ) be the set of all (complex-valued) Borel measurable functions on d .

Lemma 4.4.

Let Λ be as in (2.26), ω 0 P E ( R d ) , B 1 M ( R d ) be a quasi-Banach space, and let B 2 M ( R d ) be a quasi-Banach space of order r 0 ( 0 , 1 ] such that y f ( y - x ) B 2 when f B 1 , x R d and

f ( - x ) 2 C ω 0 ( x ) f 1 , f 1 , x d ,

for some constant C > 0 . Then the map ( a , f ) a * Λ f from 0 ( Λ ) × B 1 to M ( R d ) is uniquely extendable to a continuous map from ( ω 0 ) r 0 ( Λ ) × B 1 to B 2 , and

(4.4) a * Λ f 2 C a ( ω 0 ) r 0 ( Λ ) f 1 , a ( ω 0 ) r 0 ( Λ ) , f 1 .

There are several results which are similar to Lemma 4.4 (see, e.g., [15, 18, 20, 38, 37]). In order to be self-contained, we here give the arguments.

Proof.

We shall argue as in the proof of Lemma 4.1. Since 0 is dense in ( ω 0 ) r 0 , the result follows if we prove (4.4) for a 0 ( Λ ) .

Since is an r 0 -Banach space, we have

a * Λ f 2 r 0 = k Λ a ( k ) f ( - k ) 2 r 0
k Λ a ( k ) f ( - k ) 2 r 0
C r 0 k Λ ( | a ( k ) ω 0 ( k ) ) r 0 | f 1 r 0
= ( C a ( ω 0 ) r 0 f 1 ) r 0 .

The proof is finished. ∎

By choosing j as L ( ω j ) Φ ( d ) or as L ( ω j ) Φ 1 , Φ 2 ( 2 d ) in the previous lemma, and using Lemma 2.11, we get the following corollary (cf. [15, 38, 37]).

Corollary 4.5.

Let Λ be as in (2.26), let Λ 2 = Λ × Λ , and let Φ, Φ 1 and Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] . Then the following is true:

  1. Suppose that ω j 𝒫 E ( d ) , j = 0 , 1 , 2 , satisfy ( 4.1 ) for some constant C > 0 . Then the map ( a , f ) a * Λ f from 0 ( Λ ) × L ( ω 2 ) Φ ( d ) to ( d ) is uniquely extendable to a continuous map from ( ω 1 ) r 0 ( Λ ) × L ( ω 2 ) Φ ( d ) to L ( ω 0 ) Φ ( d ) , and

    a * Λ f L ( ω 0 ) Φ C a ( ω 1 ) r 0 b L ( ω 2 ) Φ , a ( ω 1 ) r 0 ( Λ ) , f L ( ω 2 ) Φ ( d ) .

  2. Suppose that ω j 𝒫 E ( 2 d ) , j = 0 , 1 , 2 , satisfy ( 4.2 ) for some constant C > 0 . Then the map ( a , F ) a * Λ 2 F from

    0 ( Λ 2 ) × L ( ω 2 ) Φ 1 , Φ 2 ( 2 d ) 𝑡𝑜 ( 2 d )

    is uniquely extendable to a continuous map from

    ( ω 1 ) r 0 ( Λ 2 ) × L ( ω 2 ) Φ 1 , Φ 2 ( 2 d ) 𝑡𝑜 L ( ω 0 ) Φ 1 , Φ 2 ( 2 d ) ,

    and

    a * Λ 2 F L ( ω 0 ) Φ 1 , Φ 2 C a ( ω 1 ) r 0 F L ( ω 2 ) Φ 1 , Φ 2 , a ( ω 1 ) r 0 ( Λ 2 ) , F L ( ω 2 ) Φ 1 , Φ 2 ( 2 d ) .

In what follows, we set χ = χ [ 0 , 1 ] 2 d and Q d = [ 0 , 1 ] d .

Definition 4.6.

Let Φ , Φ 1 , Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] , and let ω 𝒫 E ( 2 d ) . Then the Wiener-amalgam space W ( L Φ , L ( ω ) Φ 1 , Φ 2 ) consists of all measurable functions F on 2 d such that

a F , ω , Φ ( k , κ ) = F ( + ( k , κ ) ) ω ( + ( k , κ ) ) L Φ ( Q 2 d ) = F ω T ( k , κ ) χ L Φ ( 2 d ) , k , κ d ,

belongs to Φ 1 , Φ 2 ( 2 d ) . The (quasi-) norm on W ( L Φ , L ( ω ) Φ 1 , Φ 2 ) is given by

F W ( L Φ , L ( ω ) Φ 1 , Φ 2 ) a F , ω , Φ Φ 1 , Φ 2 .

For convenience, we set W ( L Φ 1 , Φ 2 ) = W ( L , L Φ 1 , Φ 2 ) and a F , ω = a F , ω , Φ when L Φ = L . It is obvious that

(4.5) F L ( ω ) Φ 1 , Φ 2 F W ( L ( ω ) Φ 1 , Φ 2 )

for every measurable function F on 2 d .

Remark 4.7.

Let r 0 , Φ 1 , Φ 2 and let ω be the same as in Definition 4.6.

  1. It holds

    F W ( L ( ω ) Φ 1 , Φ 2 ) = G F , ω L Φ 1 , Φ 2 ,

    where

    G F , ω ( x , ξ ) = k , κ d a F , ω ( k , κ ) χ ( k , κ ) + Q 2 d ( x , ξ ) .

  2. It holds

    F W ( L ( ω ) Φ 1 , Φ 2 ) a F ( ω ) Φ 1 , Φ 2 = G F L ( ω ) Φ 1 , Φ 2 ,

    where

    a F = a F , 1 and G F = G F , 1 .

The next lemma corresponds to [18, Lemma 3.8] and gives suitable Orlicz estimates of samples in terms of Wiener norm estimates. Here and in what follows, we let [ t ] be the integer part of t .

Lemma 4.8.

Let Φ 1 , Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] , ω P E ( R 2 d ) , F W ( L ( ω ) Φ 1 , Φ 2 ) be continuous, c F ( k , κ ) = F ( α k , β κ ) for all α , β > 0 , and let C α = ( [ 1 α ] + 1 ) d . Then c F ( ω ) Φ 1 , Φ 2 ( Z 2 d ) , and for some constant C ω which only depends on ω, it holds

(4.6) c F ( ω ) Φ 1 , Φ 2 C ω ( C α C β ) 1 r 0 F W ( L ( ω ) Φ 1 , Φ 2 ) .

A more general result of Lemma 4.8 can be found in [37], and in the Banach space case in [15]. In order to be self-contained, here we give a proof.

Proof.

The map F F ω and the fact that ω is v-moderate for some v 𝒫 E ( 2 d ) , carry over the estimate (4.6) into the case ω = 1 . Hence, it suffices to prove the result for ω = 1 . Let a F be the same as in Definition 4.6. For ( α k , β κ ) ( j , ι ) + Q 2 d and ( j , ι ) 2 d , we have

| c F ( k , κ ) | F ( + ( j , ι ) ) L ( Q 2 d ) .

Since there are at most C α points α k j + Q d , the L Φ 1 -norm over k is bounded by

k d Φ 0 , 1 ( | c F ( k , κ ) | r 0 λ r 0 ) C α j d Φ 0 , 1 ( c F ( + ( j , ι ) L r 0 λ r 0 ) .

By the definition of the Φ 1 -norm, we obtain

c F ( , κ ) Φ 1 r 0 C α a F ( , ι ) Φ 1 r 0

for β κ ι + Q d .

Since there are at most C β = ( [ 1 β ] + 1 ) d of κ d such that β κ ι + Q d , we get

κ d Φ 0 , 2 ( c F ( , κ ) Φ 1 r 0 λ r 0 ) C β ι d Φ 0 , 2 ( C α a F ( , ι ) Φ 1 r 0 λ r 0 ) .

By the definition of the Φ 2 -norm, we get (4.6). ∎

Lemma 4.9.

Let Φ 1 , Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] , and let ω j P E ( R 2 d ) , j = 0 , 1 , 2 , be such that (4.2) holds for some constant C > 0 . Then the map ( F , G ) F * G from Σ 1 ( R 2 d ) × Σ 1 ( R 2 d ) to Σ 1 ( R 2 d ) extends uniquely to a continuous map from

W ( L 1 , L ( ω 1 ) r 0 ) × W ( L ( ω 2 ) Φ 1 , Φ 2 ) 𝑡𝑜 W ( L ( ω 0 ) Φ 1 , Φ 2 ) ,

and

(4.7) F * G W ( L ( ω 0 ) Φ 1 , Φ 2 ) C d C 3 F W ( L 1 , L ( ω 1 ) r 0 ) G W ( L ( ω 2 ) Φ 1 , Φ 2 ) , F W ( L 1 , L ( ω 1 ) r 0 ) , G W ( L ( ω 2 ) Φ 1 , Φ 2 ) ,

for some constant C d which only depends on the dimension d.

Lemma 4.9 is similar to [38, Theorem 5.1] when Y in [38] is an Orlicz space. Lemma 4.9 also essentially generalizes parts of [18, Lemma 2.9].

Proof.

We have

W ( L ( ω j ) Φ 1 , Φ 2 ) L ( ω j ) ( 2 d )

and we have that Σ 1 ( 2 d ) is dense in W ( L 1 , L ( ω 1 ) r 0 ) . Since F * G is uniquely defined when F Σ 1 ( 2 d ) and G L ( ω 2 ) ( 2 d ) , the result follows if we prove that (4.7) holds for F Σ 1 ( 2 d ) and G W ( L ( ω 2 ) Φ 1 , Φ 2 ) .

Let

a F * G ( k , κ ) = sup ( x , ξ ) ( k , κ ) + Q 2 d | ( F * G ) ( x , ξ ) ω 0 ( x , ξ ) | ,

F ω 1 = F ω 1 and G ω 2 = G ω 2 , where Q d = [ 0 , 1 ] d as usual. First, we estimate a F * G ( k , κ ) by

a F * G ( k , κ ) = sup | 2 d F ( y , η ) G ( x - y , ξ - η ) ω 0 ( x , ξ ) 𝑑 y 𝑑 η |
sup ( C d C 2 d | F ω 1 ( y , η ) | | G ω 2 ( x - y , ξ - η ) | 𝑑 y 𝑑 η )
= sup ( C d C j , ι d ( j , ι ) + Q 2 d | F ω 1 ( y , η ) G ω 2 ( x - y , ξ - η ) | d y d η )
C d C j , ι d ( j , ι ) + Q 2 d | F ω 1 ( y , η ) | ( sup | G ω 2 ( x - y , ξ - η ) | ) 𝑑 y 𝑑 η
C d C j , ι d F ω 1 L 1 ( ( j , ι ) + Q 2 d ) G ω 2 ( + ( k - j , κ - ι ) ) L ( [ - 1 , 1 ] 2 d )
(4.8) = C d C ( b * c ) ( k , κ ) ,

where

b ( j , ι ) = F ω 1 L 1 ( ( j , ι ) + Q 2 d ) and c ( j , ι ) = G ω 2 ( + ( j , ι ) ) L ( [ - 1 , 1 ] 2 d ) .

Here the suprema in (4.8) are taken with respect to ( x , ξ ) ( k , κ ) + Q 2 d .

By (4.8), Corollary 4.2 and the fact that

c Φ 1 , Φ 2 C 1 , d C G W ( L Φ 1 , Φ 2 )

for some constant C 1 , d which only depends on d, we get

F * G W ( L ( ω 0 ) Φ 1 , Φ 2 ) = a F * G Φ 1 , Φ 2 C b * c Φ 1 , Φ 2 C b r 0 c Φ 1 , Φ 2 C 3 F W ( L 1 , L ( ω 1 ) r 0 ) G W ( L ( ω 2 ) Φ 1 , Φ 2 ) ,

where all hidden constants only depend on d. This gives (4.7), and the result follows. ∎

By similar arguments, we get the following semi-discrete convolution relation. The result essentially generalizes [18, Lemma 2.10], while more general results are (implicitly) available in [37].

Lemma 4.10.

Let Φ j , ω j and r 0 be the same as in Lemma 4.9, and let ε > 0 . Then the map ( a , F ) a * ε F from

0 ( ε 2 d ) × ( L ( 2 d ) ( 2 d ) ) 𝑡𝑜 L loc ( 2 d )

extends uniquely to a continuous mapping from

( ω 1 ) Φ 1 , Φ 2 ( ε 2 d ) × W ( L ( ω 2 ) r 0 ) 𝑡𝑜 W ( L ( ω 2 ) Φ 1 , Φ 2 ) ,

and

a * ε F W ( L ( ω 0 ) Φ 1 , Φ 2 ) C d C 2 a ( ω 1 ) Φ 1 , Φ 2 F W ( L ( ω 2 ) r 0 ) , a ( ω 1 ) Φ 1 , Φ 2 ( ε 2 d ) , F W ( L ( ω 2 ) r 0 ) .

Proof.

Let v 𝒫 E ( 2 d ) be submultiplicative and such that ω 1 is v-moderate. If

a ( 1 / v ) ( ε 2 d ) and F L ( 2 d ) ( 2 d ) ,

then, for ( x , ξ ) 2 d belonging to a compact set, ( a * ε F ) ( x , ξ ) is given by a finite sum of locally bounded functions. This shows that a * ε F is uniquely defined as an element in L loc ( 2 d ) .

In particular, since

( ω 1 ) Φ 1 , Φ 2 ( ε 2 d ) ( 1 / v ) ( ε 2 d ) ,

a * ε F is uniquely defined as an element in ( 1 / v ) ( ε 2 d ) when a ( ω 1 ) Φ 1 , Φ 2 ( ε 2 d ) and F L ( 2 d ) ( 2 d ) .

The result now follows by similar arguments to the ones in the proof of Lemma 4.9, and the fact that

L ( 2 d ) ( 2 d )

is dense in W ( L ( v ) r 0 ) . The details are left to the reader. ∎

5 Gabor analysis of Orlicz modulation spaces

In this section, we extend the Gabor analysis in [18] to Orlicz modulation spaces. We show that the quasi-norms

f V ϕ 1 f L ( ω ) Φ 1 , Φ 2 and f V ϕ 2 f W ( L ( ω ) Φ 1 , Φ 2 )

are equivalent when ω 𝒫 E ( 2 d ) and ϕ 1 , ϕ 2 are suitable (cf. Proposition 5.3 below). This leads to the fact that the analysis operator C ϕ 1 is continuous from

M ( ω ) Φ 1 , Φ 2 ( d ) into ( ω ) Φ 1 , Φ 2 ( 2 d ) ,

and that the corresponding synthesis operator is continuous from

( ω ) Φ 1 , Φ 2 ( 2 d ) to M ( ω ) Φ 1 , Φ 2 ( d ) .

In the end, we are able to prove that an element belongs to M ( ω ) Φ 1 , Φ 2 ( d ) if and only if its Gabor coefficients belong to ( ω ) Φ 1 , Φ 2 ( 2 d ) (cf. Theorem 5.6).

We also remark that our investigations are related to those general results in [38, 37] by Rauhut on quasi-Banach co-orbit space theory, but note that Rauhut’s results do not cover our situation. For example, our weight functions are allowed to grow and decay exponentially, which is not the case in [38, 37].

5.1 Comparisons between V ϕ 1 f L ( ω ) Φ 1 , Φ 2 and V ϕ 2 f W ( L ( ω ) Φ 1 , Φ 2 )

The next extension of [18, Theorem 3.1] shows that the condition

V ϕ f L ( ω ) Φ 1 , Φ 2 <

is independent of the choice of window function ϕ, and that different ϕ give rise to equivalent norms.

Theorem 5.1.

Let ε > 0 , ϕ 0 ( x ) = π - d 4 e - 1 2 | x | 2 , ϕ Σ 1 ( R d ) 0 and let ψ and ψ 0 be canonical dual windows for ϕ and ϕ 0 , respectively, with respect to some lattice ε Z 2 d . Also, let Φ 1 , Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] , and let ω , v P E ( R 2 d ) be such that ω is v-moderate. Then

(5.1) V ϕ 0 f L ( ω ) Φ 1 , Φ 2 C V ϕ 0 ψ L ( v ) r 0 V ϕ f L ( ω ) Φ 1 , Φ 2

and

(5.2) V ϕ f L ( ω ) Φ 1 , Φ 2 C V ϕ ψ 0 L ( v ) r 0 V ϕ 0 f L ( ω ) Φ 1 , Φ 2 ,

where the constant C > 0 only depends on ϕ, ω, v and ε.

Proof.

Assume that V ϕ f L ( ω ) Φ 1 , Φ 2 ( 2 d ) and let ε > 0 be such that

{ e i , κ ϕ ( - k ) } k , κ ε d

is a Gabor frame for L 2 ( d ) . Let v 0 ( x , ξ ) = e | x | + | ξ | and b ( k , κ ) = ( V ψ ϕ 0 ) ( k , κ ) . As a consequence of [20, Theorem S] or the analysis in [21, Chapter 13], it follows that ψ M ( v ) r 0 ( d ) and

ϕ 0 = k , κ ε d b ( k , κ ) ϕ k , κ , ϕ k , κ ( x ) = e i x , κ ϕ ( x - k ) ,

with unconditional convergence in M ( v 0 v ) 1 ( d ) M ( v ) r 0 ( d ) (see also [45, Proposition 1.4] for details). We have

| V ϕ 0 f ( x , ξ ) | k , κ ε d | b ( k , κ ) V ϕ k , κ f ( x , ξ ) |
k , κ ε d | b ( k , κ ) | | V ϕ f ( x + k , ξ + κ ) |
= ( | b ˇ | * ε | V ϕ f | ) ( x , ξ ) ,

where b ˇ ( k , κ ) = b ( - k , - κ ) . By Corollary 4.5 and the fact that

b ( v ) r 0 V ϕ 0 ψ L ( v ) q 0 < ,

in view of [44, Theorem 3.7], we obtain

V ϕ 0 f L ( ω ) Φ 1 , Φ 2 | b ˇ | * ε | V ϕ f | L ( ω ) Φ 1 , Φ 2 C 1 b ˇ ( v ) r 0 V ϕ f L ( ω ) Φ 1 , Φ 2 C 2 V ϕ 0 ψ L ( v ) r 0 V ϕ f L ( ω ) Φ 1 , Φ 2

for some constants C 1 and C 2 which only depend on ϕ, ω, v and ε, where the last step follows from [18, Lemma 3.2] (see also Proposition 5.3 below). This gives (5.1). By interchanging the roles of ϕ and ϕ 0 , we obtain (5.2). ∎

Remark 5.2.

Let ϕ , ϕ 0 , ψ and ψ 0 be the same as in Theorem 5.1. By choosing the lattice dense enough, it follows that V ϕ 0 ψ L ( v ) r 0 ( d ) and V ϕ ψ 0 L ( v ) r 0 ( d ) . In fact, let v 0 ( x , ξ ) be subexponential. Then

ϕ 0 , ϕ M ( v 0 v ) 1 ( d ) M ( v ) r 0 ( d ) .

By [20, Theorem S],

ψ 0 , ψ M ( v 0 v ) 1 ( d ) M ( v ) r 0 ( d )

provided that the lattices of Gabor frames are dense enough. This implies that V ϕ 0 ψ L ( v ) r 0 and V ϕ ψ 0 L ( v ) r 0 in (5.1) and (5.2) are finite.

Proposition 5.3.

Let Φ 1 , Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] , ω P E ( R 2 d ) , f M ( ω ) Φ 1 , Φ 2 ( R d ) and let ϕ 0 ( x ) = π - d 4 e - 1 2 | x | 2 . Then V ϕ 0 f W ( L ( ω ) Φ 1 , Φ 2 ) and

(5.3) V ϕ 0 f W ( L ( ω ) Φ 1 , Φ 2 ) C V ϕ 0 f L ( ω ) Φ 1 , Φ 2

for some constant C which only depends on ω and d.

Proof.

Let F 0 = | V ϕ 0 f | and let

a F 0 ( k , κ ) = sup ( x , ξ ) Q 2 d F 0 ( x + k , ξ + κ ) , k , κ d .

For each k , κ d , choose

X k , κ = ( x k , κ , ξ k , κ ) ( k , κ ) + Q 2 d

such that

F 0 ( X k , κ ) = a F 0 ( k , κ ) .

We have

(5.4) f M ( ω ) Φ 1 , Φ 2 = F 0 r 0 L ( ω ) Φ 0 , 1 , Φ 0 , 2 1 / r 0 .

For any

X = ( x 1 , , x d , ξ 1 , , ξ d ) 2 d

and

X 0 = ( x 0 , 1 , , x 0 , d , ξ 0 , 1 , , ξ 0 , d ) 2 d ,

let X j = ( x j , ξ j ) 2 and X 0 , j = ( x 0 , j , ξ 0 , j ) 2 , j = 1 , , d , let D r ( X 0 ) be the polydisc

{ X d : | X j - X 0 , j | < r , j = 1 , , d } ,

and let U 1 , d = [ - r , 1 + r ] d and U 2 , d = [ - 2 - r , 2 + r ] d . By [18, Lemma 2.3], we get

F 0 ( X k , κ ) r 0 ω ( X k , κ ) r 0 C 1 D r ( X k , κ ) F 0 ( x , ξ ) r 0 ω ( X k , κ ) r 0 𝑑 x 𝑑 ξ C 2 X k , κ + U 1 , 2 d F 0 ( x , ξ ) r 0 ω ( x , ξ ) r 0 𝑑 x 𝑑 ξ

for some constants C 1 and C 2 which only depend on ω and d.

In order to estimate the left-hand side of (5.3), we apply the latter estimates on

k d Φ 0 , 1 ( F 0 ( X k , κ ) r 0 ω ( X k , κ ) r 0 λ r 0 ) k d Φ 0 , 1 ( C r 0 λ r 0 X k , κ + U 1 , 2 d F 0 ( x , ξ ) r 0 ω ( x , ξ ) r 0 𝑑 x 𝑑 ξ )
(5.5) k d Φ 0 , 1 ( C r 0 λ r 0 ( k , κ ) + U 2 , 2 d F 0 ( x , ξ ) r 0 ω ( x , ξ ) r 0 𝑑 x 𝑑 ξ ) ,

with C r 0 = C 2 > 0 . Since the volume of U 2 , d is equal to ( 4 + 2 r ) d and Φ 0 , 1 is convex, Jensen’s inequality gives

k d Φ 0 , 1 ( C r 0 λ r 0 ( k , κ ) + U 2 , 2 d F 0 ( x , ξ ) r 0 ω ( x , ξ ) r 0 𝑑 x 𝑑 ξ )
k d ( 4 + 2 r ) - d k + U 2 , d Φ 0 , 1 ( C r 0 ( 4 + 2 r ) d λ r 0 κ + U 2 , d F 0 ( x , ξ ) r 0 ω ( x , ξ ) r 0 𝑑 ξ ) 𝑑 x
(5.6) = 4 d ( 4 + 2 r ) - d d Φ 0 , 1 ( C r 0 ( 4 + 2 r ) d λ r 0 κ + U 2 , d F 0 ( x , ξ ) r 0 ω ( x , ξ ) r 0 𝑑 ξ ) 𝑑 x .

By (5.5), (5.6) and the definition of the L Φ 0 , 1 -norm, we get

a ( , κ ) ( ω ) Φ 1 C κ + U 2 , d F 0 ( , ξ ) r 0 ω ( , ξ ) r 0 𝑑 ξ L Φ 0 , 1 1 / r 0

for some constant C which only depends on ω and d.

Let

b ( κ ) = a ( , κ ) ( ω ) Φ 1 .

Then, by Minkowski’s inequality and again using Jensen’s inequality, for some suitable C > 0 we get

κ d Φ 0 , 2 ( b ( κ ) r 0 λ r 0 ) κ d Φ 0 , 2 ( C r 0 λ r 0 κ + U 2 , d F 0 ( , ξ ) r 0 ω ( , ξ ) r 0 𝑑 ξ L Φ 0 , 1 )
κ d Φ 0 , 2 ( C r 0 λ r 0 κ + U 2 , d F 0 ( , ξ ) r 0 ω ( , ξ ) r 0 L Φ 0 , 1 d ξ )
κ d ( 4 + 2 r ) - d κ + U 2 , d Φ 0 , 2 ( C r 0 ( 4 + 2 r ) d λ r 0 F 0 ( , ξ ) r 0 ω ( , ξ ) r 0 L Φ 0 , 1 ) 𝑑 ξ
= 4 d ( 4 + 2 r ) - d d Φ 0 , 2 ( C r 0 ( 4 + 2 r ) d λ r 0 F 0 ( , ξ ) r 0 ω ( , ξ ) r 0 L Φ 0 , 1 ) 𝑑 ξ .

By the definition of the L ( ω ) Φ 0 , 2 -norm, we get

(5.7) a ( ω ) Φ 1 , Φ 2 = b Φ 2 C F 0 L ( ω ) Φ 1 , Φ 2 ,

where C only depends on ω and d. Hence, (5.4) and (5.7) give

V ϕ 0 f W ( L ( ω ) Φ 1 , Φ 2 ) = a ( ω ) Φ 1 , Φ 2 C F 0 L ( ω ) Φ 1 , Φ 2 = C V ϕ 0 f L ( ω ) Φ 1 , Φ 2

for a suitable constant C. ∎

Now we have the following extension of Proposition 5.3, [18, Theorem 3.3] and [44, Proposition 3.4].

Theorem 5.4.

Let Φ 1 , Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] , ω , v P E ( R 2 d ) be such that ω is v-moderate, and let ϕ 1 , ϕ 2 M ( v ) r 0 ( R d ) with dual windows in M ( v ) r 0 ( R d ) with respect to some lattice in R 2 d . If V ϕ 1 f L ( ω ) Φ 1 , Φ 2 ( R 2 d ) , then V ϕ 2 f W ( L ( ω ) Φ 1 , Φ 2 ) and

V ϕ 2 f W ( L ( ω ) Φ 1 , Φ 2 ) C V ϕ 1 f L ( ω ) Φ 1 , Φ 2 ,

for some constant C > 0 which only depends on r 0 , ω, ϕ 1 and ϕ 2 .

Proof.

Let ϕ 0 ( x ) = π - d 4 e - 1 2 | x | 2 . Using the reproducing formula, we have (see, e.g., [21, Lemma 11.3.3])

| V ϕ 2 f ( x , ξ ) | ( 2 π ) - d ( | V ϕ 0 f | * | V ϕ 2 ϕ 0 | ) ( x , ξ ) .

By Lemma 4.9, Theorem 5.1 and Proposition 5.3, we obtain

V ϕ 2 f W ( L ( ω ) Φ 1 , Φ 2 ) C 1 V ϕ 0 f W ( L ( ω ) Φ 1 , Φ 2 ) V ϕ 2 ϕ 0 W ( L 1 , L ( v ) r 0 ) C 2 V ϕ 0 f L ( ω ) Φ 1 , Φ 2 V ϕ 2 ϕ 0 W ( L 1 , L ( v ) r 0 )

for some constants C 1 and C 2 which only depend on r 0 , ω and ϕ 2 . By [47, Proposition 1.15’], we get (see also [44])

V ϕ 2 ϕ 0 W ( L 1 , L ( v ) r 0 ) ϕ 2 M ( v ) r 0 .

Hence, if ψ 1 is the canonical dual window of ϕ 1 , then Theorem 5.1 gives

V ϕ 2 f W ( L ( ω ) Φ 1 , Φ 2 ) C 1 ϕ 2 M ( v ) r 0 V ϕ 0 ψ L ( v ) r 0 V ϕ 1 f L ( ω ) Φ 1 , Φ 2 = C 2 V ϕ 1 f L ( ω ) Φ 1 , Φ 2

for some constants C 1 and C 2 which only depend on r 0 , ω, ϕ 1 and ϕ 2 . This gives the result. ∎

Now we may deduce suitable continuity properties for analysis and synthesis operators, related to [37, Theorem 5.6] and to some results in [15], and which extend [18, Theorem 3.5].

Theorem 5.5.

Let ε > 0 , Φ 1 , Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] , ω , v P E ( R 2 d ) be such that ω is v-moderate and let ϕ , ψ M ( v ) r 0 ( R d ) be Gabor atoms with dual canonical windows in M ( ω ) r 0 ( R d ) with respect to ε Z 2 d . Then the following is true:

  1. The analysis operator C ϕ ε is continuous from M ( ω ) Φ 1 , Φ 2 ( d ) into ( ω ) Φ 1 , Φ 2 ( ε 2 d ) , and

    (5.8) C ϕ ε f ( ω ) Φ 1 , Φ 2 C f M ( ω ) Φ 1 , Φ 2 , f M ( ω ) Φ 1 , Φ 2 ( d ) ,

    for some constant C > 0 which only depends on ε, r 0 , ω and ϕ.

  2. The synthesis operator D ψ ε is continuous from ( ω ) Φ 1 , Φ 2 ( ε 2 d ) into M ( ω ) Φ 1 , Φ 2 ( d ) , and

    (5.9) D ψ ε c M ( ω ) Φ 1 , Φ 2 C c ( ω ) Φ 1 , Φ 2 , c ( ω ) Φ 1 , Φ 2 ( 2 d ) .

    for some constant C > 0 which only depends on ε, r 0 , ω and ψ.

Proof.

Since V ϕ f is continuous, by Lemma 4.8 with α = β = ε and Theorem 5.4, we have

C ϕ ε f ( ω ) Φ 1 , Φ 2 ( ε 2 d ) = V ϕ f ( ω ) Φ 1 , Φ 2 ( ε 2 d ) V ϕ f W ( L ( ω ) Φ 1 , Φ 2 ) V ϕ f L ( ω ) Φ 1 , Φ 2 f M ( ω ) Φ 1 , Φ 2 ,

which proves (i).

It remains to prove (ii). Let ϕ 0 be the standard Gaussian window. We have to show that

V ϕ 0 ( D ψ ε c ) L ( ω ) Φ 1 , Φ 2 ( 2 d )

when c ( ω ) Φ 1 , Φ 2 ( ε 2 d ) . Since

| ( V ϕ 0 ( e i , κ ψ ( - k ) ) ) ( x , ξ ) | = | V ϕ 0 ψ ( x - k , ξ - κ ) | ,

we get

| V ϕ 0 ( D ψ ε c ) ( x , ξ ) | = | V ϕ 0 ( k , κ ε d c ( k , κ ) ( e i , κ ψ ( - k ) ) ) ( x , ξ ) |
= | k , κ ε d c ( k , κ ) ( V ϕ 0 ( e i , κ ψ ( - k ) ) ) ( x , ξ ) |
= | k , κ ε d c ( k , κ ) ( V ϕ 0 ψ ) ( x - k , ξ - κ ) |
k , κ ε d | c ( k , κ ) ( V ϕ 0 ψ ) ( x - k , ξ - κ ) |
= ( | c | * ε | V ϕ 0 ψ | ) ( x , ξ ) .

Lemma 4.10 implies that

D ψ ε c M ( ω ) Φ 1 , Φ 2 = V ϕ 0 ( D ψ ε c ) L ( ω ) Φ 1 , Φ 2
V ϕ 0 ( D ψ ε c ) W ( L ( ω ) Φ 1 , Φ 2 )
| c | * ε | V ϕ 0 ψ | W ( L ( ω ) Φ 1 , Φ 2 )
C c ( ω ) Φ 1 , Φ 2 ( ε 2 d ) V ϕ 0 ψ W ( L ( v ) r 0 ) ,

and (ii) follows from Theorem 5.4. ∎

The next theorem is the main result of this section, and it shows that the Gabor analysis in [18] for modulation spaces also holds for quasi-Orlicz modulation spaces.

Theorem 5.6.

Let Φ 1 , Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] , ω , v P E ( R 2 d ) be such that ω is v-moderate, and choose ϕ , ψ M ( v ) r 0 ( R d ) and ε > 0 such that

(5.10) { e i , κ ϕ ( - k ) } k , κ ε d 𝑎𝑛𝑑 { e i , κ ψ ( - k ) } k , κ ε d

are dual frames to each others. Then the following is true:

  1. The Gabor frame operator S ϕ , ψ ε = D ψ ε C ϕ ε is the identity operator on M ( ω ) Φ 1 , Φ 2 ( d ) .

  2. If f M ( ω ) Φ 1 , Φ 2 ( d ) , then

    f = k , κ ε d ( V ψ f ) ( k , κ ) e i , κ ϕ ( - k )
    = k , κ ε d ( V ϕ f ) ( k , κ ) e i , κ ψ ( - k ) ,

    with unconditional convergence in M ( ω ) Φ 1 , Φ 2 ( d ) when 𝒮 ( 2 d ) is dense in L Φ 1 , Φ 2 ( 2 d ) , and with convergence in M ( ω ) ( d ) with respect to the weak * topology otherwise.

Furthermore, if f M ( ω ) Φ 1 , Φ 2 ( R d ) , then

(5.11) C - 1 f M ( ω ) Φ 1 , Φ 2 { ( V ϕ f ) ( k , κ ) } k , κ ε d ( ω ) Φ 1 , Φ 2 C f M ( ω ) Φ 1 , Φ 2

for some constant C > 0 which only depends on ε, r 0 , ω and ϕ. The same holds true with ψ in place of ϕ at each occurrence.

Again, we observe that the constant C in Theorem 5.6 does not depend on the quasi-Young functions Φ 1 and Φ 2 .

Proof.

Since (5.10) are dual frames, it follows that D ψ ε C ϕ ε is the identity operator on M ( ω ) ( d ) , in view of [21, Corollary 12.2.6]. A combination of this fact and

M ( ω ) Φ 1 , Φ 2 ( d ) M ( ω ) ( d )

shows that f = D ψ ε C ϕ ε holds for all f M ( ω ) Φ 1 , Φ 2 ( d ) . By Theorem 5.5, the norm equivalence between the first and last expressions in (5.11) follows from

f M ( ω ) Φ 1 , Φ 2 = ( D ψ ε C ϕ ε ) f M ( ω ) Φ 1 , Φ 2
D ψ ε ( ( ω ) Φ 1 , Φ 2 , M ( ω ) Φ 1 , Φ 2 ) C ϕ ε f ( ω ) Φ 1 , Φ 2
D ψ ε ( ( ω ) Φ 1 , Φ 2 , M ( ω ) Φ 1 , Φ 2 ) C ϕ ε ( M ( ω ) Φ 1 , Φ 2 , ( ω ) Φ 1 , Φ 2 ) f M ( ω ) Φ 1 , Φ 2 .

By interchanging the roles for ϕ and ψ, we deduce the other relations in (5.11). ∎

Remark 5.7.

Let v 𝒫 E ( 2 d ) be submultiplicative. We remark that [20, Theorem S] guarantees that we may always find ϕ , ψ M ( v ) 1 ( d ) and ε > 0 in Theorem 5.6 such that (5.10) are dual Gabor frames to each others. By playing with weights in suitable ways, we may then find other Gabor pairs with atoms belonging to the subset M ( v ) r 0 ( d ) of M ( v ) 1 ( d ) .

Here we note that the hidden constants in inequalities (5.8), (5.9) and (5.11) heavily depend on the weights ω, v, the parameter ε and the Young functions Φ 1 and Φ 2 . Specifically, we notice that, in view a numerical or computational approach, a lot of effort is needed to find suitable estimates of these hidden constants.

We also note in the spirit of [20, Theorem S], that, for ω, v fixed and ϕ as above, there is an ε 2 > 0 such that for every ε 1 ( 0 , ε 2 ] there are uniform bounds of the (hidden) constants in (5.8), (5.9) and (5.11) with respect to ε [ ε 1 , ε 2 ] (cf., e.g., [15, 16, 20, 38, 37]).

Remark 5.8.

Let ω 𝒫 E ( 2 d ) , Φ 0 , 1 , Φ 0 , 2 be Young functions, and let Φ 1 and Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] with respect to Φ 0 , 1 and Φ 0 , 2 , respectively.

Since 𝒮 ( 2 d ) is continuously embedded in L Φ 1 , Φ 2 ( 2 d ) , and Σ 1 ( 2 d ) is dense in 𝒮 ( 2 d ) , it follows that Σ 1 ( 2 d ) is dense in L Φ 1 , Φ 2 ( 2 d ) when 𝒮 ( 2 d ) is dense in L Φ 1 , Φ 2 ( 2 d ) , according to Theorem 5.6 (ii).

By straightforward computations, it follows that Σ 1 ( 2 d ) is dense in L ( ω ) Φ 1 , Φ 2 ( 2 d ) when 𝒮 ( 2 d ) is dense in L Φ 1 , Φ 2 ( 2 d ) .

A sufficient condition for 𝒮 ( 2 d ) to be dense in L Φ 0 , 1 , Φ 0 , 2 ( 2 d ) and in L Φ 1 , Φ 2 ( 2 d ) is that Φ 0 , 1 and Φ 0 , 2 fulfill the so-called Δ 2 -condition in [40]. In particular, this is true when Φ j ( t ) t θ , j = 1 , 2 , near the origin, for some θ > 0 .

5.2 Some consequences

Next we present some consequences of the previous results, and begin with showing the independence of M ( ω ) Φ 1 , Φ 2 ( d ) with respect to the choices of ϕ 1 and ϕ 2 in Theorem 5.4.

Theorem 5.9.

Let Φ 1 , Φ 2 be quasi-Young functions of order r 0 ( 0 , 1 ] , ω , v P E ( R 2 d ) be such that ω is v-moderate, and let ϕ M ( v ) r 0 ( R d ) with dual window in M ( v ) r 0 ( R d ) . Then

f V ϕ f L ( ω ) Φ 1 , Φ 2 𝑎𝑛𝑑 f V ϕ f W ( L ( ω ) Φ 1 , Φ 2 )

are quasi-norms on Σ 1 ( R d ) , which in turn are equivalent to the quasi-norm

f f M ( ω ) Φ 1 , Φ 2 .

Recall that

f M ( ω ) Φ 1 , Φ 2 = V ϕ 0 f L ( ω ) Φ 1 , Φ 2

when ϕ 0 ( x ) = π - d 4 e - 1 2 | x | 2 .

Proof.

The result is an immediate consequence of (4.5) and Theorem 5.4. ∎

We observe that ϕ in Theorem 5.9 exists in view of Remark 5.7.

Theorem 5.10.

Suppose that ω P E ( R 2 d ) and that Φ k and Ψ k are quasi-Young functions such that the

(5.12) lim t 0 + Ψ k ( t ) Φ k ( t )

exist and are finite, k = 1 , 2 . Then

(5.13) ( ω ) Φ 1 , Φ 2 ( 2 d ) ( ω ) Ψ 1 , Ψ 2 ( 2 d ) 𝑎𝑛𝑑 M ( ω ) Φ 1 , Φ 2 ( d ) M ( ω ) Ψ 1 , Ψ 2 ( d ) .

Proof.

By Theorem 5.6, it suffices to prove the first embedding in (5.13). Since a a ω is an isometric bijection from ( ω ) Φ 1 , Φ 2 ( 2 d ) to Φ 1 , Φ 2 ( 2 d ) , we may assume that ω = 1 . In view of (5.12), there is a t 0 > 0 such that

Ψ k ( t ) Φ k ( t ) , 0 t t 0 , k = 1 , 2 .

Let

Φ * , k ( t ) = { Φ k ( t ) , 0 t t 0 , , t > t 0 ,

and

Ψ * , k ( t ) = { Ψ k ( t ) , 0 t t 0 , , t > t 0 .

We claim

(5.14) Φ 1 , Φ 2 ( 2 d ) = Φ * , 1 , Φ * , 2 ( 2 d ) ,

also in the topological sense.

In fact, let

Φ * , k 0 ( t ) = Φ * , k ( t 1 r 0 ) and Φ k 0 ( t ) = Φ k ( t 1 r 0 ) , k = 1 , 2 .

Then Φ * , k 0 and Φ k 0 are Young functions, and, by Proposition 2.21, we have

a Φ * , 1 0 , Φ * , 2 0 a Φ 1 0 , Φ 2 0 .

This gives

(5.15) a Φ * , 1 , Φ * , 2 = | a | r 0 Φ * , 1 0 , Φ * , 2 0 1 r 0 | a | r 0 Φ 1 0 , Φ 2 0 1 r 0 a Φ 1 , Φ 2 ,

and (5.14) follows.

By (5.15) and the fact that

Ψ * , k ( t ) Φ * , k ( t ) , t + d ,

we get

a Ψ 1 , Ψ 2 a Ψ * , 1 , Ψ * , 2 a Φ * , 1 Φ * , 2 a Φ 1 , Φ 2 ,

and the result follows. ∎

By Theorem 5.10 and its proof, we may now extend Proposition 2.21 to the quasi-Banach case as follows. The details are left to the reader.

Proposition 5.11.

Let Φ j , Ψ j , j = 1 , 2 be quasi-Young functions and ω P E ( R 2 d ) . Then the following conditions are equivalent:

  1. M ( ω ) Φ 1 , Φ 2 ( d ) M ( ω ) Ψ 1 , Ψ 2 ( d ) .

  2. ( ω ) Φ 1 , Φ 2 ( 2 d ) ( ω ) Ψ 1 , Ψ 2 ( 2 d ) .

  3. There is a constant t 0 > 0 such that Ψ j ( t ) Φ j ( t ) for all 0 t t 0 .

We observe that in the Banach space case the equivalence between (i) and (ii) in Proposition 5.11 is a special case of [16, Theorem 8.4] (which, among others, hosts Orlicz modulation spaces with general moderate weights which are Banach spaces).

Next, we discuss compact embeddings of Orlicz modulation spaces. The next result follows by similar arguments to the ones in the proof of [33, Theorem 3.9], using the fact that M ( ω ) Φ 1 , Φ 2 ( d ) is continuously embedded in M ( ω ) ( d ) in view of Corollary 3.6. The details are left to the reader.

Theorem 5.12.

Let ω 1 , ω 2 P E ( R 2 d ) . Then the injection map

i : M ( ω 1 ) Φ 1 , Φ 2 ( d ) M ( ω 2 ) Φ 1 , Φ 2 ( d )

is compact if and only if

lim | X | ω 2 ( X ) ω 1 ( X ) = 0 .


Communicated by Christopher D. Sogge


Funding source: Vetenskapsrådet

Award Identifier / Grant number: 2019-04890

Funding statement: The first author was supported by Vetenskapsrådet (Swedish Science Council) within the project 2019-04890. The second author thanks Tinçel Kültür Vakfı and İstanbul University for the support during her stay at Linnæus University during Spring 2019.

Acknowledgements

We are very grateful to H. Feichtinger from Vienna university, for valuable advice which led to several improvements of the content and the style of the original manuscript.

References

[1] T. Aoki, Locally bounded linear topological spaces, Proc. Imp. Acad. Tokyo 18 (1942), 588–594. 10.3792/pia/1195573733Search in Google Scholar

[2] J. Appell, A. S. Kalitvin and P. P. Zabreĭko, Partial integral operators in Orlicz spaces with mixed norm, Colloq. Math. 78 (1998), no. 2, 293–306. 10.4064/cm-78-2-293-306Search in Google Scholar

[3] V. Bargmann, On a Hilbert space of analytic functions and an associated integral transform, Comm. Pure Appl. Math. 14 (1961), 187–214. 10.1002/cpa.3160140303Search in Google Scholar

[4] E. Berge, A primer on coorbit theory, J. Fourier Anal. Appl. 28 (2022), no. 1, Paper No. 2. 10.1007/s00041-021-09892-5Search in Google Scholar

[5] A. P. Blozinski, Multivariate rearrangements and Banach function spaces with mixed norms, Trans. Amer. Math. Soc. 263 (1981), no. 1, 149–167. 10.1090/S0002-9947-1981-0590417-XSearch in Google Scholar

[6] A. Boccuto, A. V. Bukhvalov and A. R. Sambucini, Some inequalities in classical spaces with mixed norms, Positivity 6 (2002), no. 4, 393–411. 10.1023/A:1021353215312Search in Google Scholar

[7] P. Boggiatto, G. De Donno and A. Oliaro, A class of quadratic time-frequency representations based on the short-time Fourier transform, Modern Trends in Pseudo-Differential Operators, Oper. Theory Adv. Appl. 172, Birkhäuser, Basel (2007), 235–249. 10.1007/978-3-7643-8116-5_13Search in Google Scholar

[8] J. Chung, S.-Y. Chung and D. Kim, Characterizations of the Gel’fand–Shilov spaces via Fourier transforms, Proc. Amer. Math. Soc. 124 (1996), no. 7, 2101–2108. 10.1090/S0002-9939-96-03291-1Search in Google Scholar

[9] A. Cianchi, L. Pick and L. Slavíková, Sobolev embeddings in Orlicz and Lorentz spaces with measures, J. Math. Anal. Appl. 485 (2020), no. 2, Article ID 123827. 10.1016/j.jmaa.2019.123827Search in Google Scholar

[10] H. G. Feichtinger, Gewichtsfunktionen auf lokalkompakten Gruppen, Österreich. Akad. Wiss. Math.-Natur. Kl. Sitzungsber. II 188 (1979), no. 8–10, 451–471. Search in Google Scholar

[11] H. G. Feichtinger, Modulation spaces on locally compact abelian groups, Technical Report, University of Vienna, Vienna, 1983. Search in Google Scholar

[12] H. G. Feichtinger, Modulation spaces: Looking back and ahead, Sampl. Theory Signal Image Process. 5 (2006), no. 2, 109–140. 10.1007/BF03549447Search in Google Scholar

[13] H. G. Feichtinger and K. Gröchenig, Multidimensional irregular sampling of band-limited functions in L p -spaces, Multivariate Approximation Theory. IV (Oberwolfach 1989), Internat. Ser. Numer. Math. 90, Birkhäuser, Basel (1989), 135–142. 10.1007/978-3-0348-7298-0_15Search in Google Scholar

[14] H. G. Feichtinger, K. Gröchenig and D. Walnut, Wilson bases and modulation spaces, Math. Nachr. 155 (1992), 7–17. 10.1002/mana.19921550102Search in Google Scholar

[15] H. G. Feichtinger and K. H. Gröchenig, Banach spaces related to integrable group representations and their atomic decompositions. I, J. Funct. Anal. 86 (1989), no. 2, 307–340. 10.1515/9781400827268.408Search in Google Scholar

[16] H. G. Feichtinger and K. H. Gröchenig, Banach spaces related to integrable group representations and their atomic decompositions. II, Monatsh. Math. 108 (1989), no. 2–3, 129–148. 10.1007/BF01308667Search in Google Scholar

[17] C. Fernández, A. Galbis and J. Toft, Characterizations of GRS-weights, and consequences in time-frequency analysis, J. Pseudo-Differ. Oper. Appl. 6 (2015), no. 3, 383–390. 10.1007/s11868-015-0122-zSearch in Google Scholar

[18] Y. V. Galperin and S. Samarah, Time-frequency analysis on modulation spaces M m p , q , 0 < p , q , Appl. Comput. Harmon. Anal. 16 (2004), no. 1, 1–18. 10.1016/j.acha.2003.09.001Search in Google Scholar

[19] I. M. Gel’fand and G. E. Shilov, Generalized Functions. Vol. II-III, Academic Press, New York, 1968. Search in Google Scholar

[20] K. Gröchenig, Describing functions: Aatomic decompositions versus frames, Monatsh. Math. 112 (1991), no. 1, 1–42. 10.1007/BF01321715Search in Google Scholar

[21] K. Gröchenig, Foundations of Time-Frequency Analysis, Birkhäuser, Boston, 2001. 10.1007/978-1-4612-0003-1Search in Google Scholar

[22] K. Gröchenig, Weight functions in time-frequency analysis, Pseudo-Differential Operators: Partial Differential Equations and Time-Frequency Analysis, Fields Inst. Commun. 52, American Mathematical Society, Providence (2007), 343–366. 10.1090/fic/052/16Search in Google Scholar

[23] K. Gröchenig, Wiener’s lemma: Theme and variations. An introduction to spectral invariance and its applications, Four Short Courses on Harmonic Analysis. Wavelets, Frames, Time-Frequency Methods, and Applications to Signal and Image Analysis, Appl. Numer. Harmon. Anal., Birkhauser, Basel (2010), 175–244. 10.1007/978-0-8176-4891-6_5Search in Google Scholar

[24] K. Gröchenig and G. Zimmermann, Spaces of test functions via the STFT, J. Funct. Spaces Appl. 2 (2004), no. 1, 25–53. 10.1155/2004/498627Search in Google Scholar

[25] P. Harjulehto and P. Hästö, Orlicz Spaces and Generalized Orlicz Spaces, Lecture Notes in Math. 2236, Springer, Cham, 2019. 10.1007/978-3-030-15100-3Search in Google Scholar

[26] L. Hörmander, The Analysis of Linear Partial Differential Operators. I–III, Class. Math., Springer, Berlin, 1983, 1985. Search in Google Scholar

[27] L. Huang and D. Yang, On function spaces with mixed norms—a survey, J. Math. Study 54 (2021), no. 3, 262–336. 10.4208/jms.v54n3.21.03Search in Google Scholar

[28] W. A. Majewski and L. E. Labuschagne, On applications of Orlicz spaces to statistical physics, Ann. Henri Poincaré 15 (2014), no. 6, 1197–1221. 10.1007/s00023-013-0267-3Search in Google Scholar

[29] W. A. Majewski and L. E. Labuschagne, On entropy for general quantum systems, Adv. Theor. Math. Phys. 24 (2020), no. 2, 491–526. 10.4310/ATMP.2020.v24.n2.a6Search in Google Scholar

[30] L. Maligranda, Calderón–Lozanovskiĭ construction for mixed norm spaces, Acta Math. Hungar. 103 (2004), no. 4, 279–302. 10.1023/B:AMHU.0000028829.15720.02Search in Google Scholar

[31] M. Milman, A note on L ( p , q ) spaces and Orlicz spaces with mixed norms, Proc. Amer. Math. Soc. 83 (1981), no. 4, 743–746. 10.2307/2044245Search in Google Scholar

[32] A. Osançlıol and S. Öztop, Weighted Orlicz algebras on locally compact groups, J. Aust. Math. Soc. 99 (2015), no. 3, 399–414. 10.1017/S1446788715000257Search in Google Scholar

[33] C. Pfeuffer and J. Toft, Compactness properties for modulation spaces, Complex Anal. Oper. Theory 13 (2019), no. 8, 3521–3548. 10.1007/s11785-019-00903-4Search in Google Scholar

[34] S. Pilipović, Generalization of Zemanian spaces of generalized functions which have orthonormal series expansions, SIAM J. Math. Anal. 17 (1986), no. 2, 477–484. 10.1137/0517037Search in Google Scholar

[35] S. Pilipović, Tempered ultradistributions, Boll. Un. Mat. Ital. B (7) 2 (1988), no. 2, 235–251. Search in Google Scholar

[36] M. M. Rao and Z. D. Ren, Theory of Orlicz Spaces, Monogr. Textb. Pure Appl. Math. 146, Marcel Dekker, New York, 1991. Search in Google Scholar

[37] H. Rauhut, Coorbit space theory for quasi-Banach spaces, Studia Math. 180 (2007), no. 3, 237–253. 10.4064/sm180-3-4Search in Google Scholar

[38] H. Rauhut, Wiener amalgam spaces with respect to quasi-Banach spaces, Colloq. Math. 109 (2007), no. 2, 345–362. 10.4064/cm109-2-13Search in Google Scholar

[39] S. Rolewicz, On a certain class of linear metric spaces, Bull. Acad. Polon. Sci. Cl. III. 5 (1957), 471–473. Search in Google Scholar

[40] C. Schnackers and H. Führ, Orlicz Modulation Spaces, Ph.D. thesis, RWTH Aachen University, 2014. Search in Google Scholar

[41] M. Signahl and J. Toft, Mapping properties for the Bargmann transform on modulation spaces, J. Pseudo-Differ. Oper. Appl. 3 (2012), no. 1, 1–30. 10.1007/s11868-011-0039-0Search in Google Scholar

[42] J. Toft, Continuity and Schatten properties for pseudo-differential operators on modulation spaces, Modern Trends in Pseudo-Differential Operators, Oper. Theory Adv. Appl. 172, Birkhäuser, Basel (2007), 173–206. 10.1007/978-3-7643-8116-5_11Search in Google Scholar

[43] J. Toft, The Bargmann transform on modulation and Gelfand–Shilov spaces, with applications to Toeplitz and pseudo-differential operators, J. Pseudo-Differ. Oper. Appl. 3 (2012), no. 2, 145–227. 10.1007/s11868-011-0044-3Search in Google Scholar

[44] J. Toft, Gabor analysis for a broad class of quasi-Banach modulation spaces, Pseudo-Differential Operators and Generalized Functions, Oper. Theory Adv. Appl. 245, Birkhäuser/Springer, Cham (2015), 255–284. 10.1007/978-3-319-14618-8_18Search in Google Scholar

[45] J. Toft, Continuity and compactness for pseudo-differential operators with symbols in quasi-Banach spaces or Hörmander classes, Anal. Appl. (Singap.) 15 (2017), no. 3, 353–389. 10.1142/S0219530516500159Search in Google Scholar

[46] J. Toft, Images of function and distribution spaces under the Bargmann transform, J. Pseudo-Differ. Oper. Appl. 8 (2017), no. 1, 83–139. 10.1007/s11868-016-0165-9Search in Google Scholar

[47] J. Toft, Wiener estimates on modulation spaces, Advances in Microlocal and Time-Frequency Analysis, Appl. Numer. Harmon. Anal., Birkhäuser/Springer, Cham (2020), 475–505. 10.1007/978-3-030-36138-9_26Search in Google Scholar

[48] S. J. L. van Eijndhoven, Functional analytic characterizations of the Gel’fand–Shilov spaces S α β , Nederl. Akad. Wetensch. Indag. Math. 49 (1987), no. 2, 133–144. 10.1016/S1385-7258(87)80035-5Search in Google Scholar

Received: 2021-11-03
Revised: 2022-06-01
Published Online: 2022-07-29
Published in Print: 2022-09-01

© 2022 Walter de Gruyter GmbH, Berlin/Boston

This work is licensed under the Creative Commons Attribution 4.0 International License.

Downloaded on 3.11.2025 from https://www.degruyterbrill.com/document/doi/10.1515/forum-2021-0279/html
Scroll to top button