Abstract
We show basic properties on tensor products for Gelfand–Shilov distributions and Pilipović distributions. This also includes the Fubini’s property of such tensor products. We also apply the Fubini property to deduce some properties for short-time Fourier transforms of Gelfand–Shilov and Pilipović distributions.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
An important issue in mathematics concerns tensor products. When considering the functions \(f_j\) defined on \(\Omega _j\subseteq {\mathbf {R}}^{d_j}\), \(j=1,2\), and with values in \({\mathbf {C}}\), their tensor product \(f_1\otimes f_2\) is the function from \(\Omega _1\times \Omega _2\) to \({\mathbf {C}}\) given by the formula
Let \(f_j,\varphi _j\in {\mathscr {S}}({\mathbf {R}}^{d_j})\), \(f=f_1\otimes f_2\), \(\varphi \in {\mathscr {S}}({\mathbf {R}}^{d_1+d_2})\), and let \(\psi _1\) and \(\psi _2\) be given by
(For notations, see. [9] and Sect. 2.) Then it follows that
and that the Fubini’s property
holds.
The formula (1.2) and (1.3) are essential when searching for extensions of tensor products to distributions. By the analysis in [9, Chapter V and VII], we have the following.
Theorem 1.1
Let \(f_j\in {\mathscr {S}}'({\mathbf {R}}^{d_j})\), \(\varphi \in {\mathscr {S}}({\mathbf {R}}^{d_1+d_2})\) and let \(\psi _j\) be given by (1.1), \(j=1,2\). Then \(\psi _j\in {\mathscr {S}}({\mathbf {R}}^{d_2})\), \(j=1,2\), and there is a unique \(f\in {\mathscr {S}}'({\mathbf {R}}^{d_1+d_2})\) such that for every \(\varphi _1\in {\mathscr {S}}({\mathbf {R}}^{d_1})\) and \(\varphi _2\in {\mathscr {S}}({\mathbf {R}}^{d_2})\), (1.2) and (1.3) hold.
The existence of a distribution f in the previous theorem which satisfies (1.2) can also be deduced by a general and abstract result on tensor products for nuclear spaces (see, [17, Chapter 50]). On the other hand, in order to reach the Fubini property (1.3), it seems that more structures are needed.
A more specific approach in the lines of the ideas in [17] is indicated in [10, 14], where \({\mathscr {S}}({\mathbf {R}}^{d})\) and \({\mathscr {S}}'({\mathbf {R}}^{d})\) are described by suitable series expansions of Hermite functions. By following such approaches, the situations are essentially reduced to questions on tensor products of weighted \(\ell ^2\) spaces, and both properties (1.2) and (1.3) follows from such approach.
In Sects. 3 and 4 we show that Theorem 1.1 holds in the context of Gelfand–Shilov spaces, Pilipović spaces and their distribution (dual) spaces. In particular, we prove that the following results hold true.
Theorem 1.2
Let \(s_j,\sigma _j>0\), \(f_j\in ({\mathcal {S}}_{s_j}^{\sigma _j})'({\mathbf {R}}^{d_j})\), \(\varphi \in {\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2}({\mathbf {R}}^{d_1+d_2})\) and let \(\psi _j\) be given by (1.1), \(j=1,2\). Then \(\psi _j\in {\mathcal {S}}_{s_j}^{\sigma _j}({\mathbf {R}}^{d_2})\), \(j=1,2\), and there is a unique \(f\in ({\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2})'({\mathbf {R}}^{d_1+d_2})\) such that for every \(\varphi _1\in {\mathcal {S}}_{s_1}^{\sigma _1}({\mathbf {R}}^{d_1})\) and \(\varphi _2\in {\mathcal {S}}_{s_2}^{\sigma _2}({\mathbf {R}}^{d_2})\), (1.2) and (1.3) hold.
The same holds true with \(\Sigma _{s_j}^{\sigma _j}\), \((\Sigma _{s_j}^{\sigma _j})'\), \(\Sigma _{s_1,s_2}^{\sigma _1,\sigma _2}\) and \((\Sigma _{s_1,s_2}^{\sigma _1,\sigma _2})'\) in place of \(\Sigma _{s_j}^{\sigma _j}\), \(({\mathcal {S}}_{s_j}^{\sigma _j})'\), \({\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2}\) and \(({\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2})'\), respectively, at each occurrence.
Theorem 1.3
Let \(s\in \overline{{\mathbf {R}}_\flat }\), \(f_j\in {\mathcal {H}}_s'({\mathbf {R}}^{d_j})\), \(\varphi \in {\mathcal {H}}_s({\mathbf {R}}^{d_1+d_2})\) and let \(\psi _j\) be given by (1.1), \(j=1,2\). Then \(\psi _j\in {\mathcal {H}}_s({\mathbf {R}}^{d_2})\), \(j=1,2\), and there is a unique \(f\in {\mathcal {H}}_s'({\mathbf {R}}^{d_1+d_2})\) such that for every \(\varphi _1\in {\mathcal {H}}_s({\mathbf {R}}^{d_1})\) and \(\varphi _2\in {\mathcal {H}}_s({\mathbf {R}}^{d_2})\), (1.2) and (1.3) hold.
The same holds true with \({\mathcal {H}}_{0,s}\) and \({\mathcal {H}}_{0,s}'\) in place of \({\mathcal {H}}_s\) and \({\mathcal {H}}_s'\), respectively, at each occurrence.
The distribution f in Theorems 1.1, 1.2 or in Theorem 1.3 is called the tensor product of \(f_1\) and \(f_2\) and is denoted by \(f_1\otimes f_2\) as before. We notice that in some cases, Theorem 1.2 is deduced in [4] (cf. [4, Appendix]).
We remark that Gelfand–Shilov spaces of functions and distributions appear naturally when discussing analyticity and well-posedness of solutions to partial differential equations (cf. [2, 3]). Pilipović spaces of functions and distributions often agree with Fourier-invariant Gelfand–Shilov spaces, and possess convenient mapping properties with respect to the Bargmann transform. They therefore seems to be suitable to have in background on problems in partial differential equations which have been transformed by the Bargmann transform (see, [6, 16] for more details).
Since the spaces in Theorems 1.2 and 1.3 are unions and intersections of nuclear spaces, the existence of f satisfying (1.2) may be deduced by the abstract analogous results in [17]. Some parts of Theorem 1.2 are also proved in [10].
In Sect. 3 we give a proof of Theorem 1.2, by using the framework in [9] for the proof of Theorem 1.1. In Sect. 4 we use that Pilipović spaces and their distribution spaces can be described by unions and intersections of Hilbert spaces of Hermite series expansions. In similar ways as in [14], this essentially reduce the situation to deal with questions on tensor products of weighted \(\ell ^2\) spaces.
In the end of Sect. 3 we also give examples on how to apply the Fubini property (1.3) to deduce certain relations for short-time Fourier transforms (which is often called coherent state trasnforms in physics) of Gelfand–Shilov distributions (see, Example 3.7). In Sect. 4 we also discuss such questions for Pilipović spaces which are not Gelfand–Shilov distributions (cf. Remark 4.4).
2 Preliminaries
In this section we recall some basic facts. We start by giving the definition of Gelfand–Shilov spaces. Thereafter we recall the definition of Pilipović spaces and some of their properties.
2.1 Gelfand–Shilov spaces
We start by recalling some facts about Gelfand–Shilov spaces (cf. [5, 8]). Let \(0<h,s_j,\sigma _j\in {\mathbf {R}}\), \(j=1,\dots ,n\), be fixed, \(d=d_1+\cdots +d_n\), where \(d_j\ge 0\) are integers, and let
For multi-indices of multi-indices we let
when
For any \(f\in C^\infty ({\mathbf {R}}^{d})\), we let
where the supremum is taken over all \(\alpha _j,\beta _j\in {\mathbf {N}}^{d_j}\), \(j=1,\ldots ,n\). Then \(f\mapsto \Vert f\Vert _{{\mathcal {S}}_{{{\varvec{s}}};h}^{{\varvec{\sigma }}}}\) defines a norm on \(C^\infty ({\mathbf {R}}^{d})\) which might attend infinity. The space \({\mathcal {S}}_{{{\varvec{s}}};h}^{{\varvec{\sigma }}}({\mathbf {R}}^{d})\) is the Banach space which consist of all \(f\in C^\infty ({\mathbf {R}}^{d})\) such that \(\Vert f\Vert _{{\mathcal {S}}_{{{\varvec{s}}};h}^{{\varvec{\sigma }}}}\) is finite. In the case \(d_1=d\ge 1\), \(d_2=\cdots =d_n=0\), \(s=s_1\), \(\sigma =\sigma _1\) and \(x_1=x\), (2.1) is interpreted as
The Gelfand–Shilov spaces\({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\) and \(\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\) are defined as the inductive and projective limits, respectively of \({\mathcal {S}}_{{{\varvec{s}}};h}^{{\varvec{\sigma }}}({\mathbf {R}}^{d})\). This implies that
and that the topology for \({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\) is the strongest possible one such that the inclusion map from \({\mathcal {S}}_{{{\varvec{s}}};h}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\) to \({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\) is continuous, for every choice of \(h>0\). The space \(\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\) is a Fréchet space with semi-norms \(\Vert \, \cdot \, \Vert _{{\mathcal {S}}_{{{\varvec{s}}};h}^{{\varvec{\sigma }}}}\), \(h>0\). Moreover,
and
There are various kinds of characterizations of the spaces \({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\) and \(\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\), e. g., in terms of the exponential decay of their elements. Later on it will be useful that \(f \in {\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\) (respectively, \(f \in \Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\)), if and only if
for some \(h,r>0\) (respectively, for every \(h>0,r>0\)).
If \({\varvec{1}}=(1,\ldots ,1)\in {\mathbf {R}}^{n}\) and \({{\varvec{s}}},{\varvec{\sigma }}\in {\mathbf {R}}^{n}_+\), then
for every \(\varepsilon >0\). If in addition \(s_j+\sigma _j\ge 1\) for every j, then the last two inclusions in (2.3) are dense, and if \(s_j+\sigma _j\ge 1 \, \text {and}\,(s_j,\sigma _j)\ne (\frac{1}{2},\frac{1}{2})\) for every j, then the first inclusion in (2.3) is dense.
In order for discuss duality properties of Gelfand–Shilov spaces we first recall the definition of Gelfand tripples.
Definition 2.1
Let \({\mathcal {V}},{\mathcal {H}},{\mathcal {V}}'\) be topological vector spaces. Then \(({\mathcal {V}},{\mathcal {H}},{\mathcal {V}}')\) is called a Gelfand tripple, if the following conditions are fulfilled:
- (1)
\({\mathcal {V}}\subseteq {\mathcal {H}}\subseteq {\mathcal {V}}'\);
- (2)
\({\mathcal {H}}\) is a Hilbert space and the restriction of the \({\mathcal {H}}\)-scalar product \((\, \cdot \, ,\, \cdot \, )_{{\mathcal {H}}}\) to \(\mathcal V\times {\mathcal {V}}\) is uniquely extendable to continuous mappings from \({\mathcal {V}}'\times {\mathcal {V}}\) to \({\mathbf {C}}\) and from \({\mathcal {V}}\times {\mathcal {V}}'\) to \({\mathbf {C}}\);
- (3)
the dual space of \({\mathcal {V}}\) can be identified with \({\mathcal {V}}'\) through the form \((\, \cdot \, ,\, \cdot \, )_{{\mathcal {H}}}\) on \(\mathcal V'\times {\mathcal {V}}\) in (2).
The Gelfand–Shilov distribution spaces\(({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}})' ({\mathbf {R}}^{d})\) and \((\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}})' ({\mathbf {R}}^{d})\) are the projective and inductive limits respectively of \(({\mathcal {S}}_{{{\varvec{s}}};h}^{{\varvec{\sigma }}})'({\mathbf {R}}^{d})\) with respect to \(h>0\). Here \(({\mathcal {S}}_{{{\varvec{s}}};h}^{{\varvec{\sigma }}})'({\mathbf {R}}^{d})\) is the dual of \({\mathcal {S}}_{{{\varvec{s}}};h}^{{\varvec{\sigma }}}({\mathbf {R}}^{d})\). This implies that
If in addition \(d_1=d\ge 1\), \(d_2=\cdots =d_n=0\), \(s=s_1\) and \(\sigma _1=\sigma \), then we set \(({\mathcal {S}}_s^\sigma )'({\mathbf {R}}^{d}) =({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}})' ({\mathbf {R}}^{d})\) and \((\Sigma _s^\sigma )'({\mathbf {R}}^{d}) =(\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}} )' ({\mathbf {R}}^{d})\). We remark that the analysis in [12] shows that
is a Gelfand tripple when \(s_j+\sigma _j\ge 1\), \(j=1,\dots ,n\), and that
is a Gelfand tripple when \(s_j+\sigma _j\ge 1\) and \((s_j,\sigma _j) \ne (\frac{1}{2},\frac{1}{2})\), \(j=1,\dots ,n\). In particular, \(({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}} )'({\mathbf {R}}^{d})\) and \((\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}} )'({\mathbf {R}}^{d})\) are the topological duals of \({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\) respectively, \(\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\) through unique extensions of the \(L^2\)-scalar product on \({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\times {\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\) respectively, on \(\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\times \Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\) to continuous mappings from \(({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}})' ({\mathbf {R}}^{d})\times {\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\) respectively, \((\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}})' ({\mathbf {R}}^{d})\times \Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}} ({\mathbf {R}}^{d})\) to \({\mathbf {C}}\).
By the inequalities \(n!k! \le (n+k)! \le 2^{n+k}n!k!\) it follows that
Corresponding relations to (2.3) for Gelfand–Shilov distributions are
when \(s_j+\sigma _j \ge 1\), \(j=1,\ldots ,n\), and
when \(s_j+\sigma _j \ge 1\) and \((s_j,\sigma _j)\ne (\frac{1}{2},\frac{1}{2})\), \(j=1,\ldots ,n\).
The Gelfand–Shilov spaces and their distribution spaces possess several convenient properties. For example they are complete, invariant under translations, dilations, and to some extent (partial) Fourier transformations. For any \(f\in L^1({\mathbf {R}}^{d})\), its Fourier transform is defined by
If instead \(f\in L^1({\mathbf {R}}^{d_1+\cdots +d_n})\), then the partial Fourier transform of f with respect to \(k\in \{ 1,\ldots ,n\}\) is given by
Remark 2.2
Let \(d=d_1+\cdots +d_n\), \(j\in \{ 1,\ldots ,n\}\), \({{\varvec{s}}},{\varvec{\sigma }}\in {\mathbf {R}}^{n}_+\), \(\tau _{k,j}({{\varvec{s}}},{\varvec{\sigma }}) = (r_{k,j,1},\ldots ,r_{k,j,n})\), \(k=1,2\), where
Then the following assertions follow from the general theory of Schwartz functions and Gelfand–Shilov functions and their distributions (see, e.g. [5, 9]):
- (1)
the definition of \({\mathscr {F}}_j\) extends to a homeomorphism on \({\mathscr {S}}'({\mathbf {R}}^{d})\) and restricts to a homeomorphism on \({\mathscr {S}}({\mathbf {R}}^{d})\);
- (2)
the definition of \({\mathscr {F}}_j\) extends uniquely to a homeomorphism from \(({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}})'({\mathbf {R}}^{d})\) to \(({\mathcal {S}}_{\tau _{1,j}({{\varvec{s}}},{\varvec{\sigma }})}^{\tau _{2,j}({{\varvec{s}}},{\varvec{\sigma }})})'({\mathbf {R}}^{d})\), and from \((\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}})'({\mathbf {R}}^{d})\) to \((\Sigma _{\tau _{1,j}({{\varvec{s}}},{\varvec{\sigma }})}^{\tau _{2,j}({{\varvec{s}}},{\varvec{\sigma }})})'({\mathbf {R}}^{d})\);
- (3)
\({\mathscr {F}}_j\) restricts to homeomorphisms from \({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}}({\mathbf {R}}^{d})\) to \({\mathcal {S}}_{\tau _{1,j}({{\varvec{s}}},{\varvec{\sigma }})}^{\tau _{2,j}({{\varvec{s}}},{\varvec{\sigma }})}({\mathbf {R}}^{d})\), and from \(\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}}({\mathbf {R}}^{d})\) to \(\Sigma _{\tau _{1,j}({{\varvec{s}}},{\varvec{\sigma }})}^{\tau _{2,j}({{\varvec{s}}},{\varvec{\sigma }})}({\mathbf {R}}^{d})\).
Remark 2.3
There are several characterizations of Gelfand–Shilov spaces in the literature. For example, let \(s,\sigma \in {\mathbf {R}}_+\) and \(p\in [1,\infty ]\). Then it is proved in [5] that \(\psi \in {\mathscr {S}}'({\mathbf {R}}^{d})\) belongs to \({\mathcal {S}}_s^\sigma ({\mathbf {R}}^{d})\) (\(\Sigma _s^\sigma ({\mathbf {R}}^{d})\)), if and only if
is finite for some \(r>0\) (for every \(r>0\)). Furthermore, the topology of \({\mathcal {S}}_s^\sigma ({\mathbf {R}}^{d})\) (\(\Sigma _s^\sigma ({\mathbf {R}}^{d})\)) is obtained by the inductive (projective) limit topologies, supplied by the semi-norms \(\Vert \, \cdot \, \Vert _{(r,p)}\) with respect to \(r>0\).
2.2 Pilipović spaces
Next we make a review of Pilipović spaces. These spaces can be defined in terms of Hermite series expansions. We recall that the Hermite function of order \(\alpha \in {\mathbf {N}}^{d}\) is defined by
It follows that
for some polynomial \(p_\alpha \) on \({\mathbf {R}}^{d}\), which is called the Hermite polynomial of order \(\alpha \). The Hermite functions are eigenfunctions to the Fourier transform, partial Fourier transforms and to the Harmonic oscillator \(H_d\equiv |x|^2-\Delta \) which acts on functions and (ultra-)distributions defined on \({\mathbf {R}}^{d}\). For example, we have
It is well-known that the set of Hermite functions is a basis for \({\mathscr {S}}({\mathbf {R}}^{d})\) and an orthonormal basis for \(L^2({\mathbf {R}}^{d})\) (cf. [14]). In particular, if \(f\in L^2({\mathbf {R}}^{d})\), then
where
is the Hermite seriers expansion of f, and
is the Hermite coefficient of f of order \(\alpha \in {\mathbf {R}}^{d}\).
In order to define the full scale of Pilipović spaces, their order s should belong to the extended set
of \({\mathbf {R}}_+\), with extended inequality relations as
when \(s_1<\frac{1}{2}\le s_2\) and \(\sigma _1<\sigma _2\). (Cf. [16].)
For \(r>0\) and \(s\in {\mathbf {R}}_\flat \) we set
and
Definition 2.4
Let \(s\in \overline{{\mathbf {R}}_\flat } = {\mathbf {R}}_\flat \cup \{ 0\}\), and let \(\vartheta _{r,s}\) and \(\vartheta _{r,s}'\) be as in (2.7) and (2.8).
- (1)
\({\mathcal {H}}_0({\mathbf {R}}^{d})\) consists of all finite expansions in (2.5), and \({\mathcal {H}}_0'({\mathbf {R}}^{d})\) consists of all formal Hermite series expansions in (2.5);
- (2)
if \(s\in {\mathbf {R}}_\flat \), then \({\mathcal {H}}_s({\mathbf {R}}^{d})\) (\({\mathcal {H}}_{0,s}({\mathbf {R}}^{d})\)) consists of all \(f\in L^2({\mathbf {R}}^{d})\) such that
$$\begin{aligned} |c_h(f,h_\alpha )| \,\,\lesssim\,\, \vartheta _{r,s}(\alpha ) \end{aligned}$$holds true for some \(r\in {\mathbf {R}}_+\) (for every \(r\in \mathbf R_+\));
- (3)
if \(s\in {\mathbf {R}}_\flat \), then \({\mathcal {H}}_s'({\mathbf {R}}^{d})\) (\({\mathcal {H}}_{0,s}'({\mathbf {R}}^{d})\)) consists of all formal Hermite series expansions in (2.5) such that
$$\begin{aligned} |c_h(f,h_\alpha )| \,\lesssim\, \vartheta _{r,s}'(\alpha ) \end{aligned}$$holds true for every \(r\in {\mathbf {R}}_+\) (for some \(r\in {\mathbf {R}}_+\)).
The spaces \({\mathcal {H}}_s({\mathbf {R}}^{d})\) and \({\mathcal {H}}_{0,s}({\mathbf {R}}^{d})\) are called Pilipović spaces of Roumieu respectively Beurling types of order s, and \({\mathcal {H}}_s'({\mathbf {R}}^{d})\) and \({\mathcal {H}}_{0,s}'({\mathbf {R}}^{d})\) are called Pilipović distribution spaces of Roumieu respectively Beurling types of order s.
Remark 2.5
Let \({\mathcal {S}}_s({\mathbf {R}}^{d})\equiv {\mathcal {S}}_s^s({\mathbf {R}}^{d})\) and \(\Sigma _s({\mathbf {R}}^{d})\equiv \Sigma _s^s({\mathbf {R}}^{d})\) be the Fourier invariant Gelfand–Shilov spaces of order \(s\in {\mathbf {R}}_+\) and of Roumieu and Beurling types, respectively. Then it is proved in [11, 12] that
and
Next we recall the topologies for Pilipović spaces. Let \(s\in {\mathbf {R}}_\flat \), \(r>0\), and let \(\Vert f\Vert _{{\mathcal {H}}_{s;r}}\) and \(\Vert f\Vert _{{\mathcal {H}}_{s;r}'}\) be given by
and
when f is a formal expansion in (2.5). Then \({\mathcal {H}}_{s;r}({\mathbf {R}}^{d})\) consists of all expansions (2.5) such that \(\Vert f\Vert _{{\mathcal {H}}_{s;r}}\) is finite, and \({\mathcal {H}}_{s;r}'({\mathbf {R}}^{d})\) consists of all expansions (2.5) such that \(\Vert f\Vert _{{\mathcal {H}}_{s;r}'}\) is finite. It follows that both \({\mathcal {H}}_{s;r}({\mathbf {R}}^{d})\) and \({\mathcal {H}}_{s;r}'({\mathbf {R}}^{d})\) are Banach spaces under the norms \(f\mapsto \Vert f\Vert _{{\mathcal {H}}_{s;r}}\) and \(f\mapsto \Vert f\Vert _{{\mathcal {H}}_{s;r}'}\), respectively.
We let the topologies of \({\mathcal {H}}_s({\mathbf {R}}^{d})\) and \({\mathcal {H}}_{0,s}({\mathbf {R}}^{d})\) be the inductive respectively, projective limit topology of \({\mathcal {H}}_{s;r}({\mathbf {R}}^{d})\) with respect to \(r>0\). In the same way, the topologies of \({\mathcal {H}}_s'({\mathbf {R}}^{d})\) and \({\mathcal {H}}_{0,s}'({\mathbf {R}}^{d})\) are the projective respectively, inductive limit topology of \({\mathcal {H}}_{s;r}'({\mathbf {R}}^{d})\) with respect to \(r>0\).
Suppose instead \(s=0\). For any integer \(N\ge 0\), we set
The topology for \({\mathcal {H}}_0'({\mathbf {R}}^{d})\) is defined by the semi-norms \(\Vert \, \cdot \, \Vert _{(N)}\).
We also let \({\mathcal {H}}_{0}^{(N)}({\mathbf {R}}^{d})\) be the vector space which consists of all \(f\in {\mathcal {H}}_0'({\mathbf {R}}^{d})\) such that \(c_h(f,\alpha )=0\) when \(|\alpha |>N\), and equip this space with the topology, defined by the norm \(\Vert \, \cdot \, \Vert _{(N)}\). The topology of \({\mathcal {H}}_0({\mathbf {R}}^{d})\) is then defined as the inductive limit topology of \({\mathcal {H}}_{0}^{(N)}({\mathbf {R}}^{d})\) with respect to \(N\ge 0\).
It follows that the spaces in Definition 2.4 are complete, and that \({\mathcal {H}}_{0,s}({\mathbf {R}}^{d})\) and \({\mathcal {H}}_s'({\mathbf {R}}^{d})\) are Fréchet space with semi-norms \(f\mapsto \Vert f\Vert _{{\mathcal {H}}_{s;r}}\) and \(f\mapsto \Vert f\Vert _{{\mathcal {H}}_{s;r}'}\), respectively. By [16] it follows that
is a Gelfand tripple when \(s\in {\overline{{\mathbf {R}}}}_\flat \) and that
is a Gelfand tripple when \(s\in {\mathbf {R}}_\flat \). Obviously, if \(\langle \, \cdot \, ,\, \cdot \, \rangle \) denotes the action between (ultra-)distributions and their corresponding (ultra-)test functions, then \(\langle f,\varphi \rangle = (f,\overline{\varphi })=(f,\overline{\varphi })_{L^2}\), when \(f\in {\mathcal {H}}_s'({\mathbf {R}}^{d})\) and \(\varphi \in {\mathcal {H}}_s({\mathbf {R}}^{d})\). The same holds true with \({\mathcal {H}}_{0,s}\), \({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}}\) or \(\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}}\) in place of \({\mathcal {H}}_s\) at each occurrence.
The following characterizations of Pilipović spaces can be found in [16]. The proof is therefore omitted.
Proposition 2.6
Let \(s\in {\mathbf {R}}_+\cup \{ 0\}\) and let \(f\in {\mathcal {H}}_0'({\mathbf {R}}^{d})\). Then \(f\in {\mathcal {H}}_{0,s}({\mathbf {R}}^{d})\) (\(f\in {\mathcal {H}}_s({\mathbf {R}}^{d})\)), if and only if \(f\in C^\infty ({\mathbf {R}}^{d})\) and satisfies \(|H_d^N f(x)|\,\lesssim\, h^N N!^{2s}\) for every \(h>0\) (for some \(h>0\)).
Finally we remark that the Pilipović spaces of functions and distributions possess convenient mapping properties under the Bargmann transform (cf. [16] and Remarks 4.4 and 4.5 in Sect. 4).
3 Tensor product for Gelfand–Shilov spaces
In this section we start by proving Theorem 1.2. Thereafter we deduce a multi-linear version of this result.
For the proof of Theorem 1.2 we first need the following analogy of Lemma 4.1.3 in [9].
Lemma 3.1
Let \(s_1,s_2,\sigma _1,\sigma _2>0\), \(\varphi ,\psi \in {\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2}({\mathbf {R}}^{d_1+d_2})\). Then the Riemann sum
converges to \((\varphi *\psi )(x)\) in \({\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2}({\mathbf {R}}^{d_1+d_2})\) as \(\varepsilon \rightarrow 0\).
The same holds true if each \({\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2}\) are replaced by \(\Sigma _{s_1,s_2}^{\sigma _1,\sigma _2}\).
Proof
We may assume that \(\varepsilon >0\), and consider first the case when \(\varphi \) and \(\psi \) are real-valued. Set
and \(Q_{d,r}= [0,r]^d\). By the mean-value theorem we have for some \(\rho _k=\rho _k(x,y)\in Q_{d,1}\), \(k\in {\mathbf {Z}}^{d}\) that
where
and
Since \((m+1)!\le 2^mm!\) when \(m\ge 0\) is an integer, and \(\varphi ,\psi \in {\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2} ({\mathbf {R}}^{d_1+d_2})\), we get
\(j=1,2\), for some positive constants h and r. This implies that for some \(h>0\) we have
for some positive constants C and h which are independent of \(\varepsilon \).
Since the right-hand side tends to zero when \(\varepsilon >0\) tends to zero, the stated convergence follows in this case.
The general case follows from the previous case, after writing \(\varphi = \varphi _1+i\varphi _2\) and \(\psi = \psi _1+i\psi _2\) with \(\varphi _{j}\) and \(\psi _{j}\) being real-valued, \(j=1,2\), giving that \(\varphi *\psi \) is a superposition of \(\varphi _{j_1}*\psi _{j_2}\), \(j_1,j_2\in \{ 1,2\}\), and using the fact that \(\varphi _j\in {\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2}({\mathbf {R}}^{d_1+d_2})\) when \(\varphi \in {\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2}({\mathbf {R}}^{d_1+d_2})\). \(\square \)
We may now prove the following result related to [9, Theorem 4.1.2]
Lemma 3.2
Let \(s_1,s_2,\sigma _1,\sigma _2>0\), \(\varphi ,\psi \in {\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2}({\mathbf {R}}^{d_1+d_2})\) and let \(f\in ({\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2})'({\mathbf {R}}^{d_1+d_2})\). Then
The same holds true if each \({\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2}\) and \(({\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2})'\) are replaced by \(\Sigma _{s_1,s_2}^{\sigma _1,\sigma _2}\) and \((\Sigma _{s_1,s_2}^{\sigma _1,\sigma _2})'\), respectively.
Proof
We use the same notations as in the previous proof. Since the Riemann sum in Lemma 3.1 converges to \(\varphi *\psi \) in \({\mathcal {S}}_{s_1,s_2} ^{\sigma _1,\sigma _2}\), we get
Here the second equality follows by the fact that
converges in \({\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2}\) for every fixed \(x\in {\mathbf {R}}^{d}\) and \(\varepsilon \in {\mathbf {R}}\).
We have that \(f*\varphi \) is smooth, and for some \(r_0>0\) we have
for every \(r>0\). This gives
for some constant \(C_x\) which only depends on x and \(r_0\). It follows that
is a Riemann sum which converges to
as \(\varepsilon >0\) tends to zero. Hence (3.2) holds and the result follows. \(\square \)
The next lemma shows that similar limit properties for convolutions of distributions with mollifiers also work for Gelfand–Shilov distributions and Gelfand–Shilov mollifiers.
Lemma 3.3
Let \(s,\sigma >0\), \(\phi ,\psi \in {\mathcal {S}}_s^\sigma ({\mathbf {R}}^{d})\) (\(\phi ,\psi \in \Sigma _s^\sigma ({\mathbf {R}}^{d})\)) be such that
\(\phi _\varepsilon (x) = \varepsilon ^{-d}\phi (\varepsilon ^{-1}x)\), \(\varepsilon >0\), and let \(f\in ({\mathcal {S}}_s^\sigma )'({\mathbf {R}}^{d})\) (\(f\in (\Sigma _s^\sigma )'({\mathbf {R}}^{d})\)). Then the following is true:
- (1)
\(\psi *\phi _\varepsilon \rightarrow \psi \) in \({\mathcal {S}}_s^\sigma \) (in \(\Sigma _s^\sigma \)) as \(\varepsilon \rightarrow 0+\);
- (2)
\(f*\phi _\varepsilon \rightarrow f\) in \(({\mathcal {S}}_s^\sigma )'\) (in \((\Sigma _s^\sigma )'\)) as \(\varepsilon \rightarrow 0+\).
We notice that under the additional assumption
(1) in Lemma 3.3 is deduced in [7, Proposition 4.1]. By straight-forward modifications of the arguments in the proof of the latter result it also follows that the condition (3.4) may be removed.
In order to be self-contained, we here present a different proof of Lemma 3.3 compared to [7, Proposition 4.1], based on the characterizations of Gelfand–Shilov spaces given in Remark 2.3 and without requiring that (3.4) should hold.
Proof of Lemma 3.3
By duality it suffices to prove (1). We only prove the result in the Roumieu case, i. e. that \(\psi *\phi _\varepsilon \rightarrow \psi \) in \({\mathcal {S}}_s^\sigma \) when \(\phi ,\psi \in {\mathcal {S}}_s^\sigma ({\mathbf {R}}^{d})\). The other (Beurling) case follows by similar arguments and is left for the reader.
Let \(c\ge 1\) be chosen such that
and let \(r>0\) be chosen such that \(\Vert \phi \Vert _{(rc,1)}<\infty \) and \(\Vert \psi \Vert _{(rc,1)}<\infty \). (Cf. Remark 2.3 for notations.) By Remark 2.3 and the fact that \({\mathscr {F}}(f*\psi ) = (2\pi )^{\frac{d}{2}}{\widehat{f}} \cdot \widehat{\psi }\), the result follows if we prove
and
For any \(\varepsilon \in (0,1]\), let
and
Then
Here the last inequality follows from the fact that \(\varepsilon \le 1\).
Since \(\Vert \phi \Vert _{(cr,1)}+\Vert \psi \Vert _{(cr,1)}<\infty \), it follows that
Hence, Lebesgue’s theorem gives
and (3.5) follows.
Since \(\int \phi (x)\, dx = 1\) and
it follows that \(\widehat{\phi }(0) = (2\pi )^{-\frac{d}{2}}\) and that
by Lebesgue’s theorem. This gives (3.6) and thereby the result. \(\square \)
Remark 3.4
By the previous lemma and Fourier transformations it follows that if \(f\in ({\mathcal {S}}_s^\sigma )'({\mathbf {R}}^{d})\) and \(\phi ,\psi \in {\mathcal {S}}_s^\sigma ({\mathbf {R}}^{d})\) fullfil \(\phi (0)=1\), then \(f\cdot \phi (\varepsilon \, \cdot \, )\rightarrow f\) in \(({\mathcal {S}}_s^\sigma )'\) and \(\psi \cdot \phi (\varepsilon \, \cdot \, )\rightarrow \psi \) in \({\mathcal {S}}_s^\sigma \) when \(\varepsilon \rightarrow 0\). The same holds true with \(\Sigma _s^\sigma \) and \((\Sigma _s^\sigma )'\) in place of \({\mathcal {S}}_s^\sigma \) and \(({\mathcal {S}}_s^\sigma )'\) respectively, at each occurrence.
By the previous results it is now straight-forward to prove the following.
Lemma 3.5
Let \({{\varvec{s}}},{\varvec{\sigma }}\in {\mathbf {R}}^{n}_+\), \(d=d_1+\cdots +d_n\) and suppose \(f\in ({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}})'({\mathbf {R}}^{d})\) satisfies \(\langle f,\varphi _1\otimes \cdots \otimes \varphi _n\rangle =0\) for every \(\varphi _j\in {\mathcal {S}}_{s_j}^{\sigma _j}({\mathbf {R}}^{d_j})\), \(j=1,\dots ,n\). Then \(f=0\).
The same holds true if each \({\mathcal {S}}_{s_j}^{\sigma _j}\), \(({\mathcal {S}}_{s_j}^{\sigma _j})'\), \({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}}\) and \(({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}})'\) are replaced by \(\Sigma _{s_j}^{\sigma _j}\), \((\Sigma _{s_j}^{\sigma _j})'\), \(\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}}\) and \((\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}})'\), respectively, at each occurrence.
Proof
We only prove the result in the Roumieu case. The Beurling case follows by similar arguments and is left for the reader. We use the same notations as in the previous proofs.
First suppose that \(n=2\). Let \(\varphi \in {\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2}({\mathbf {R}}^{d_1+d_2})\), \(\phi _j\in {\mathcal {S}}_{s_j}^{\sigma _j}({\mathbf {R}}^{d_j})\) be such that
and let
when \(\varepsilon >0\) is real. Then the assumptions imply that \(\check{f} *\phi _\varepsilon =0\) for every \(\varepsilon \). Here \(\check{f}\) is defined by \(\check{f}(x)= f(-x)\). By Lemmas 3.2 and 3.3 we get
and the result follows for \(n=2\).
For general \(n\ge 2\), the result follows from the case \(n=2\) and induction. The details are left for the reader. \(\square \)
Proof of Theorem 1.2
We only prove the result in the Roumieu cases. The Beurling cases follow by similar arguments and are left for the reader.
By straight-forward computations it follows that
and
define continuous linear forms \(g_1\) and \(g_2\) on \({\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2}({\mathbf {R}}^{d_1+d_2})\). Hence \(g_1,g_2 \in ({\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2})'({\mathbf {R}}^{d_1+d_2})\). It is obvious that both \(g_1\) and \(g_2\) in place of f satisfy (1.2), and the existence of f follows.
If \(f\in ({\mathcal {S}}_{s_1,s_2}^{\sigma _1,\sigma _2})'({\mathbf {R}}^{d_1+d_2})\) is arbitrary such that (1.2) holds, then
and Lemma 3.5 shows that \(f=g_1=g_2\). This gives the uniqueness of f, as well as (1.3). \(\square \)
In order to consider corresponding multi-linear situation of Theorem 1.2, we let \(S_n\) be the permutation group of \(\{ 1,\dots ,n\}\), and let inductively
and
when \(f_j\) for \(j=1,\dots ,n\) are suitable (ultra-)distributions and \(\varphi \) is a suitable function. Also set
when \(j=1,\dots ,n\) and \({{\varvec{s}}},{\varvec{\sigma }}\in {\mathbf {R}}^{n}_+\). Then Theorem 1.2 can be reformulated as follows.
Theorem 3.6
Let \(\tau \in S_2\), \(d=d_1+d_2\), \({{\varvec{s}}},{\varvec{\sigma }}\in {\mathbf {R}}^{2}_+\), \(d_{j,\tau }\), \({{\varvec{s}}}_{j,\tau }\) and \({\varvec{\sigma }}_{j,\tau }\) be as in (3.9), \(f_j\in ({\mathcal {S}}_{s_j}^{\sigma _j})'({\mathbf {R}}^{d_j})\), \(\varphi \in {\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}}({\mathbf {R}}^{d})\) and let \(\varphi _{j,\tau }\) be given by (3.7) and (3.8), \(j=1,2\). Then \(\varphi _{j,\tau }\in {\mathcal {S}}_{{{\varvec{s}}}_{j\tau }} ^{{\varvec{\sigma }}_{j,\tau }} ({\mathbf {R}}^{d_{j,\tau }})\), and there is a unique distribution f in \(({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}})'({\mathbf {R}}^{d})\) such that for every \(\varphi _j\in {\mathcal {S}}_{s_j}^{\sigma _j}({\mathbf {R}}^{d_j})\), \(j=1,2\),
hold.
The same holds true with \(\Sigma _{s_j}^{\sigma _j}\), \((\Sigma _{s_j}^{\sigma _j})'\), \(\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}}\) and \((\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}})'\) in place of \({\mathcal {S}}_{s_j}^{\sigma _j}\), \(({\mathcal {S}}_{s_j}^{\sigma _j})'\), \({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}}\) and \(({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}})'\), respectively, at each occurrence.
Here the second equality in (3.10) is the same as the Fubini property (1.3). The multi-linear version of the previous theorem is the following, and follows by similar arguments as for the proof of Theorem 3.6. The details are left for the reader.
Theorem 3.6′
Let \(\tau \in S_n\), \(d=d_1+\dots +d_n\), \({{\varvec{s}}},{\varvec{\sigma }}\in {\mathbf {R}}^{n}_+\), \(d_{j,\tau }\), \({{\varvec{s}}}_{j,\tau }\) and \({\varvec{\sigma }}_{j,\tau }\) be as in (3.9), \(f_j\in ({\mathcal {S}}_{s_j}^{\sigma _j})'({\mathbf {R}}^{d_j})\), \(\varphi \in {\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}}({\mathbf {R}}^{d})\) and let \(\varphi _{j,\tau }\) be given by (3.7) and (3.8), \(j=1,\dots ,n\). Then \(\varphi _{j,\tau }\in {\mathcal {S}}_{{{\varvec{s}}}_{j\tau }} ^{{\varvec{\sigma }}_{j,\tau }} ({\mathbf {R}}^{d_{j,\tau }})\), and there is a unique distribution f in \(({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}})'({\mathbf {R}}^{d})\) such that for every \(\varphi _j\in {\mathcal {S}}_{s_j}^{\sigma _j}({\mathbf {R}}^{d_j})\), \(j=1,\dots ,n\),
hold.
The same holds true with \(\Sigma _{s_j}^{\sigma _j}\), \((\Sigma _{s_j}^{\sigma _j})'\), \(\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}}\) and \((\Sigma _{{{\varvec{s}}}}^{{\varvec{\sigma }}})'\) in place of \({\mathcal {S}}_{s_j}^{\sigma _j}\), \(({\mathcal {S}}_{s_j}^{\sigma _j})'\), \({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}}\) and \(({\mathcal {S}}_{{{\varvec{s}}}}^{{\varvec{\sigma }}})'\), respectively, at each occurrence.
Example 3.7
Let \(s,\sigma >0\). An important object in time-frequency and micro-local analysis concerns the short-time Fourier transform (cf. e.g., [15]). If \(\phi \in {\mathcal {S}}_s^\sigma ({\mathbf {R}}^{d}){\setminus } \{ 0\}\) is fixed, then the short-time Fourier transform of \(f\in {\mathcal {S}}_s^\sigma ({\mathbf {R}}^{d})\) is defined by
It follows that
and
for such choices of \(\phi \) and f.
We notice that the right-hand side of (3.11) also makes sense as a smooth function on \({\mathbf {R}}^{2d}\) if the assumption on f is relaxed into \(f\in ({\mathcal {S}}_s^\sigma )'({\mathbf {R}}^{d})\). For such f we therefore let (3.11) define the short-time Fourier transform of f with respect to \(\phi \). Since the map which takes \(\phi \) into \(y\mapsto \phi (y-x)e^{i\langle y,\xi \rangle }\) is continuous and smooth with respect to \((x,\xi )\) from \({\mathcal {S}}_s^\sigma ({\mathbf {R}}^{d})\) to itself it follows that \(V_\phi\, f\) is smooth. By [16, Proposition 2.2] it follows that \(V_\phi \,f\) belongs to \(({\mathcal {S}}_{s,\sigma }^{\sigma ,s})'({\mathbf {R}}^{2d})\). Consequently,
Let U be the operator which takes any F(x, y) into \(F(y,y-x)\) and recall that \({\mathscr {F}}_2F\) is the partial Fourier transform of F(x, y) with respect to the y variable. Then the right-hand side of (3.12) equals
We notice that the right-hand side makes sense as an element in \(({\mathcal {S}}_{s,\sigma }^{\sigma ,s})'({\mathbf {R}}^{2d})\) for any \(f,\phi \in ({\mathcal {S}}_s^\sigma )'({\mathbf {R}}^{d})\) in view of Remark 2.2, which may be used to extend the definition of the short-time Fourier transform to even more general situations.
We claim that the right-hand sides of (3.11) and (3.12) agree when \(f \in ({\mathcal {S}}_s^\sigma )'({\mathbf {R}}^{d})\) and \(\phi \in {\mathcal {S}}_s^\sigma ({\mathbf {R}}^{d})\). In particular, by letting \(s=\sigma \), we recover (A.1) in [4, Appendix].
In fact, let \(\psi \in {\mathcal {S}}_{s,\sigma }^{\sigma ,s}({\mathbf {R}}^{2d})\) and set
and let g be the right-hand side of (3.12). By the Fubini property at the right-hand of (3.10)\('\) we get
and
Since \(\psi \) was arbitrarily chosen, it follows that \(g=V_\phi \, f\) in \(({\mathcal {S}}_{s,\sigma }^{\sigma ,s})'({\mathbf {R}}^{2d})\), and the claim follows.
Remark 3.8
By straight-forward modifications of the arguments in the previous example it also follows that the right-hand sides of (3.11) and (3.12) agree when \(f\in (\Sigma _s^\sigma )'({\mathbf {R}}^{d})\) and \(\phi \in \Sigma _s^\sigma ({\mathbf {R}}^{d})\).
4 Tensor product of Pilipović spaces
In this section we discuss the tensor products of Pilipović spaces. Especially we prove Theorem 1.2. Thereafter we deduce a multi-linear version of this result.
First we show that the tensor products take Pilipović spaces into Pilipović spaces.
Proposition 4.1
Let \(s\in \overline{{\mathbf {R}}_\flat }\). Then the following is true:
- (1)
the map \((f_1,f_2)\mapsto f_1\otimes f_2\) from \({\mathscr {S}}({\mathbf {R}}^{d_1}) \times {\mathscr {S}}({\mathbf {R}}^{d_2})\) to \({\mathscr {S}}({\mathbf {R}}^{d_1+d_2})\), restricts to a continuous map from \({\mathcal {H}}_s({\mathbf {R}}^{d_1}) \times {\mathcal {H}}_s ({\mathbf {R}}^{d_2})\) to \({\mathcal {H}}_s ({\mathbf {R}}^{d_1+d_2})\);
- (2)
the map \((f_1,f_2)\mapsto f_1\otimes f_2\) from \({\mathscr {S}}({\mathbf {R}}^{d_1}) \times {\mathscr {S}}({\mathbf {R}}^{d_2})\) to \({\mathscr {S}}({\mathbf {R}}^{d_1+d_2})\), restricts to a continuous map from \({\mathcal {H}}_{0,s}({\mathbf {R}}^{d_1}) \times {\mathcal {H}}_{0,s} ({\mathbf {R}}^{d_2})\) to \({\mathcal {H}}_{0,s} ({\mathbf {R}}^{d_1+d_2})\).
Proof
We only prove (1) and in the case \(s>0\). The assertion (2) and the case \(s=0\) in (1) follow by similar arguments and are left for the reader. If
and \(f=f_1\otimes f_2\), then
If \(s\in {\mathbf {R}}_+\), then
for some \(r>0\). This gives
and it follows that \(f\in {\mathcal {H}}_s({\mathbf {R}}^{d})\).
If instead \(s=\flat _\sigma \), for some \(\sigma >0\), then
for some \(r>0\). Hence, if \(\alpha =(\alpha _1,\alpha _2)\), we get
for some \(r>0\), and it follows that \(f\in {\mathcal {H}}_s({\mathbf {R}}^{d})\) in this case as well. This shows that the map \((f_1,f_2)\rightarrow f_1\otimes f_2\) is continuous from \({\mathcal {H}}_s({\mathbf {R}}^{d_1}) \times {\mathcal {H}}_s ({\mathbf {R}}^{d_2})\) to \({\mathcal {H}}_s ({\mathbf {R}}^{d_1+d_2})\), and the result follows. \(\square \)
The uniqueness assertions of Theorem 1.3 follow from the following lemma.
Lemma 4.2
Let \(s\in \overline{{\mathbf {R}}_\flat }\), \(d=d_1+\cdots +d_n\) and suppose \(f\in {\mathcal {H}}_s'({\mathbf {R}}^{d})\) satisfies \(\langle f,\varphi _1\otimes \cdots \otimes \varphi _n\rangle =0\) for every \(\varphi _j\in {\mathcal {H}}_s({\mathbf {R}}^{d_j})\), \(j=1,\dots ,n\). Then \(f=0\).
The same holds true if each \({\mathcal {H}}_s\) and \({\mathcal {H}}_s'\) are replaced by \({\mathcal {H}}_{0,s}\) and \({\mathcal {H}}_{0,s}'\), respectively.
Proof
Let \(c_h(f,\alpha )\) be the Hermite coefficient of f of order \(\alpha =(\alpha _1,\dots ,\alpha _n)\), where \(\alpha _j\in {\mathbf {N}}^{d_j}\). By choosing \(\varphi _j=h_{\alpha _j}\), we get
giving that \(f=0\). \(\square \)
Proof of Theorem 1.3
Let \(d=d_1+d_2\). We shall deal with the Hermite sequence representations of the elements in the Pilipović spaces. Such approach is performed in [14], when deducing tensor product and kernel results for tempered distributions. We only prove the results when \(f_j\in {\mathcal {H}}_s'({\mathbf {R}}^{d_j})\) and \(s>0\). The cases when \(f_j\in {\mathcal {H}}_{0,s}'({\mathbf {R}}^{d_j})\) or \(s=0\) follow by similar arguments and are left for the reader.
The uniqueness follows from Lemma 4.2. We need to prove the existence of f which fullfils the asserted properties.
We have
where \(c_h(f_j,\alpha _j)\) for every \(\alpha _j\in {\mathbf {N}}^{d_j}\) are unique and equal to \((f_j,h_{\alpha _j})\), \(j=1,2\).
Now let f be the element in \({\mathcal {H}}_0'({\mathbf {R}}^{d})\), \(d=d_1+d_2\) with expansion
where
We claim that \(f\in {\mathcal {H}}_s'({\mathbf {R}}^{d})\).
In fact, if \(s\in {\mathbf {R}}_+\), then
for every \(\varepsilon >0\), and it follows that
for every \(\varepsilon >0\). This is the same as \(f\in {\mathcal {H}}_s'({\mathbf {R}}^{d})\).
If instead \(s=\flat _\sigma \), for some \(\sigma >0\), then
for every \(r>0\). Hence, if \(\alpha =(\alpha _1,\alpha _2)\), we get
for every \(r>0\), and it follows that \(f\in {\mathcal {H}}_s'({\mathbf {R}}^{d})\) in this case as well.
If \(\varphi _j\in {\mathcal {H}}_0({\mathbf {R}}^{d_j})\) and \(\varphi \in {\mathcal {H}}_0({\mathbf {R}}^{d})\), \(j=1,2\), then (1.2) and (1.3) follow by straight-forward computations, using the fact that the set of Hermite functions is an orthonormal basis of \(L^2\). For general \(\varphi _j\in {\mathcal {H}}_s({\mathbf {R}}^{d_j})\) and \(\varphi \in {\mathcal {H}}_s({\mathbf {R}}^{d})\), \(j=1,2\), the result now follows from dominating convergence, using the fact that \({\mathcal {H}}_0({\mathbf {R}}^{d})\) is dense in \({\mathcal {H}}_s({\mathbf {R}}^{d})\).
In order to formulate a multi-linear version of Theorem 1.3 we first reformulate the result as follows.\(\square \)
Theorem 4.3
Let \(\tau \in S_2\), \(d=d_1+d_2\), \(s\in \overline{{\mathbf {R}}_\flat }\), \(d_{j,\tau }\) be as in (3.9), \(f_j\in ({\mathcal {H}}_s)'({\mathbf {R}}^{d_j})\), \(\varphi \in {\mathcal {H}}_s({\mathbf {R}}^{d})\) and let \(\varphi _{j,\tau }\) be given by (3.7) and (3.8), \(j=1,2\). Then \(\varphi _{j,\tau }\in {\mathcal {H}}_s ({\mathbf {R}}^{d_{j,\tau }})\), and there is a unique distribution f in \({\mathcal {H}}_s'({\mathbf {R}}^{d})\) such that for every \(\varphi _{j}\in {\mathcal {H}}_s({\mathbf {R}}^{d_{j}})\), \(j=1,2\),
hold.
The same holds true with \({\mathcal {H}}_{0,s}\) and \({\mathcal {H}}_{0,s}'\) in place of \({\mathcal {H}}_{s}\) and \({\mathcal {H}}_{s}'\), respectively, at each occurrence.
The multi-linear version of the previous theorem is the following, and follows from the previous result and induction. The details are left for the reader.
Theorem 4.3′
Let \(\tau \in S_n\), \(d=d_1+\cdots +d_n\), \(s\in \overline{{\mathbf {R}}_\flat }\), \(d_{j,\tau }\) be as in (3.9), \(f_j\in ({\mathcal {H}}_s)'({\mathbf {R}}^{d_j})\), \(\varphi \in {\mathcal {H}}_s({\mathbf {R}}^{d})\) and let \(\varphi _{j,\tau }\) be given by (3.7) and (3.8), \(j=1,\dots ,n\). Then \(\varphi _{j,\tau }\in {\mathcal {H}}_s ({\mathbf {R}}^{d_{j,\tau }})\), and there is a unique distribution f in \({\mathcal {H}}_s'({\mathbf {R}}^{d})\) such that for every \(\varphi _{j}\in {\mathcal {H}}_s({\mathbf {R}}^{d_{j}})\), \(j=1,\dots ,n\),
hold.
The same holds true with \({\mathcal {H}}_{0,s}\) and \({\mathcal {H}}_{0,s}'\) in place of \({\mathcal {H}}_{s}\) and \({\mathcal {H}}_{s}'\), respectively, at each occurrence.
Remark 4.4
Only certain parts of the properties in Example 3.7 carry over to Pilipović spaces of functions and distributions, in the case when these spaces do not agree with Gelfand–Shilov spaces of functions and distributions. (See Remark 2.5). In order to deal with such questions, it it convenient to consider the image of such spaces under the Bargmann transform, which is defined by
when f is a suitable (ultra-)distribution on \({\mathbf {R}}^{d}\) (cf. [1, 16]). Here
In fact, let \({\mathcal {A}}_s({\mathbf {C}}^{d})\) (\({\mathcal {A}}_{0,s}({\mathbf {C}}^{d})\)) be the set of all F in \(A({\mathbf {C}}^{d})\), the set of entire functions on \({\mathbf {C}}^{d}\), which satisfies
when \(s<\frac{1}{2}\) and
when \(s=\flat _\sigma \), for some \(r>0\) (for every \(r>0\)). Here \(\langle z\rangle = (1+|z|^2)^{\frac{1}{2}}\) when \(z\in {\mathbf {Z}}^{d}\). Also let \({\mathcal {A}}_{0,1/2}({\mathbf {C}}^{d})\) be the set of all \(F\in A({\mathbf {C}}^{d})\) such that \(|F(z)|\,\lesssim\, e^{r|z|^2}\) for all \(r>0\). Then it is proved in [6, 16] that \({\mathfrak {V}}_d\) is bijective from \({\mathcal {H}}_s({\mathbf {R}}^{d})\) to \({\mathcal {A}}_s({\mathbf {C}}^{d})\) when \(s\in {\mathbf {R}}_\flat \) and \(s<\frac{1}{2}\), and from \({\mathcal {H}}_{0,s}({\mathbf {R}}^{d})\) to \({\mathcal {A}}_{0,s}({\mathbf {C}}^{d})\) when \(s\in {\mathbf {R}}_\flat \) and \(s\le \frac{1}{2}\).
By straight-forward computations we have for every \(x_0\in {\mathbf {R}}^{d}\) that
and
Consequently, by Remark 2.5 and the mapping properties of the Pilipović spaces above under the Bargmann transform, it follows that the following is true:
- (1)
\({\mathcal {H}}_s({\mathbf {R}}^{d})\) and \({\mathcal {H}}_s'({\mathbf {R}}^{d})\) are invariant under translations and modulations, if and only if \(s\ge \flat _1\);
- (2)
\({\mathcal {H}}_{0,s}({\mathbf {R}}^{d})\) and \({\mathcal {H}}_{0,s}'({\mathbf {R}}^{d})\) are invariant under translations and modulations, if and only if \(s> \flat _1\).
In particular, the short-time Fourier transform
makes sense as a smooth function when \(s\ge \flat _1\), \(f\in {\mathcal {H}}_s'({\mathbf {R}}^{d})\) and \(\phi \in {\mathcal {H}}_s({\mathbf {R}}^{d})\), or when \(s> \flat _1\), \(f\in {\mathcal {H}}_{0,s}'({\mathbf {R}}^{d})\) and \(\phi \in {\mathcal {H}}_{0,s}({\mathbf {R}}^{d})\).
On the other hand, for \(s<\frac{1}{2}\), it seems to be difficult to guarantee that (3.12) is true in general, since the map U in Example 3.7 seems not to be well-defined on Pilipović spaces which fail to be Gelfand–Shilov spaces.
Remark 4.5
It is observed already in [1] that if
then \({\mathfrak {V}}_dh_\alpha =e_\alpha \). For any expansion f in (2.5), we define the Bargmann transform of f as the formal power series expansion
with
For any \(s\in {\overline{{\mathbf {R}}}}_\flat \), we also let
be the images of
respectively, under the Bargmann transform. We also let the topologies of the former spaces be inherited from the latter spaces. Let \(d\lambda (z)\) be the Lebesgue measure on \({\mathbf {C}}^{d}\), \(d\mu (z)\) be the Gauss measure given by \(d\mu (z)=\pi ^{-d}e^{-|z|^2}\, d\lambda (z)\) and let \(A^2({\mathbf {C}}^{d})\) be the Hilbert space of all \(F\in A({\mathbf {C}}^{d})\) such that
is finite. Then the scalar product of \(A^2({\mathbf {C}}^{d})\) is given by
It follows that the Bargmann transform is a homeomorphism from the Gelfand tripples in (2.9) and (2.10) to the Gelfand tripples
respectively. Furthermore, the definition of the spaces in Remark 4.4 agrees with corresponding spaces in (4.2) (cf. [16]).
It now follows that Theorems 1.3, 4.3 and 4.3\('\) remains valid after the spaces in (4.3) are replaced by corresponding spaces in (4.2), where the \(L^2\) products are replaced by the \(A^2\) products.
References
Bargmann, V. 1961. On a Hilbert space of analytic functions and an associated integral transform. Communications on Pure and Applied Mathematics 14: 187–214.
Cappiello, M., and L. Rodino. 2006. \(\operatorname{SG}\)-pseudodifferential operators and Gelfand-Shilov spaces. Rocky Mountain Journal of Mathematics 36: 1117–1148.
Cappiello, M., and J. Toft. 2017. Pseudo-differential operators in a Gelfand–Shilov setting. Mathematische Nachrichten 290: 738–755.
Chen, Y., J. Toft, and P. Wahlberg. 2019. The Weyl product on quasi-Banach modulation spaces. Bulletin of Mathematical Sciences 9: 1950018-1.
Chung, J., S.-Y. Chung, and D. Kim. 1996. Characterizations of the Gelfand-Shilov spaces via Fourier transforms. Proceedings of the American Mathematical Society 124: 2101–2108.
Fernandez, C., A. Galbis, and J. Toft. 2017. The Bargmann transform and powers of harmonic oscillator on Gelfand–Shilov subspaces. RACSAM 111: 1–13.
Garetto, C., and M. Ruzhansky. 2015. Hyperbolic second order equations with non-regular time dependent coefficients. Archive for Rational Mechanics and Analysis 217: 113–154.
Gelfand, I.M., and G.E. Shilov. 1968. Generalized functions, II–III. NewYork, London: Academic Press.
Hörmander, L. 1983, 1985. The analysis of linear partial differential operators, vol. I–III. Berlin, Heidelberg, NewYork: Springer.
Lozanov Crvenković, Z., and D. Perišić. 2007. Hermite expansions of elements of Gelfand Shilov spaces in quasianalytic and non quasianalytic case. Novi Sad Journal of Mathematics 37: 129–147.
Pilipović, S. 1986. Generalization of Zemanian spaces of generalized functions which have orthonormal series expansions. SIAM Journal on Mathematical Analysis 17: 477–484.
Pilipović, S. 1986. Structural theorems for periodic ultradistributions. Proceedings of the American Mathematical Society 98: 261–266.
Pilipović, S. 1988. Tempered ultradistributions. Bollettino dell’UMI - Unione Matematica Italiana 7: 235–251.
Reed, M., and B. Simon. 1979. Methods of modern mathematical physics, vol. I, II. London, New York: Academic Press.
Toft, J. 2012. The Bargmann transform on modulation and Gelfand–Shilov spaces, with applications to Toeplitz and pseudo-differential operators. Journal of Pseudo-Differential Operators and Applications 3: 145–227.
Toft, J. 2017. Images of function and distribution spaces under the Bargmann transform. J. Pseudo-Differ. Oper. Appl. 8: 83–139.
Treves, F. 1967. Topological vector spaces, distributions and kernels. New York-London: Academic Press.
Acknowledgements
Open access funding provided by Linnaeus University.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The author declares that he has no conflict of interest.
Human participants
This article does not contain any studies with human participants or animals performed by the author.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
About this article
Cite this article
Toft, J. Tensor products for Gelfand–Shilov and Pilipović distribution spaces. J Anal 28, 591–613 (2020). https://doi.org/10.1007/s41478-019-00205-0
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s41478-019-00205-0