Abstract
Inspired by Jaming’s characterization of the Fourier transform on specific groups via the convolution property, we provide a novel approach which characterizes the Fourier transform on any locally compact abelian group. In particular, our characterization encompasses Jaming’s results. Furthermore, we demonstrate that the cosine transform as well as the Laplace transform can also be characterized via a suitable convolution property.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
It is well-known that the Fourier transform satisfies the convolution property. More precisely, let G be a locally compact abelian group with a Haar measure \(\mu \) and a dual group \(\widehat{G}.\) The Fourier transform is a map \(\mathcal {F}: L^1(G)\longrightarrow C_0(\widehat{G})\) given by the formula
whereas the Fourier convolution \(\star _{\mathcal {F}}: L^1(G)\times L^1(G)\longrightarrow L^1(G)\) is given by
For the sake of convenience, from this point onwards we will write dx and du instead of “\(d\mu (x)\)” and “\(d\mu (u)\)”, respectively. The following Fourier convolution property (see Lemma 1.7.2 in [3], p. 30) holds for every \(f,g\in L^1(G)\):
Jaming (partly inspired by the works of Alesker, Artstein-Avidan, Faifman and Milman on product preserving maps, see [1]) proved that the convolution property characterizes (to a certain degree) the Fourier transform if \(G = \mathbb {R}, S^1, \mathbb {Z}\) or \(\mathbb {Z}_n\) (see [5]). His paper inspired Lavanya and Thangavelu to show that any continuous *-homomorphism of \(L^1(\mathbb {C}^d)\) (with twisted convolution as multiplication) into \(B(L^2(\mathbb {R}^d))\) is essentially a Weyl transform and deduce a similar characterization for the group Fourier transform on the Heisenberg group (see [8, 9]). Furthermore, Kumar and Sivananthan went on to demonstrate that the convolution property characterizes the Fourier transform on compact groups (see [7]).
Studying the topic we realized two things that prompted us to write this paper. Firstly, Jaming’s characterization of the Fourier transform need not be restricted to particular cases \(G = \mathbb {R}, S^1, \mathbb {Z}\) or \(\mathbb {Z}_n\). There is a unified approach for all locally compact abelian groups, which we demonstrate in the first part of Sect. 2.
Secondly, we discovered that the Fourier transform is not the only integral transform that can be characterized via the convolution property. In the second part of Sect. 2 and in Sect. 3 we explain that cosine and Laplace convolutions characterize cosine and Laplace transforms, respectively.
2 Fourier and Cosine Transform
As we have agreed in the Introduction, let G stand for a locally compact abelian group with Haar measure \(\mu \) and dual group \(\widehat{G}.\) The formula for the Fourier transform and Fourier convolution are given by (1) and (2), respectively. We start off with a well-known fact:
Lemma 1
For every function \(g\in L^1(G)\) and \(x,y\in G\) the following equality holds
where for every \(z\in G\) the operator \(L_z:L^1(G)\longrightarrow L^1(G)\) is given by
The following theorem generalizes Theorems 2.1 and 3.1 in Jaming’s paper (see [5]).
Theorem 1
Let \(T:L^1(G)\longrightarrow C^b(\widehat{G})\) be a linear and bounded operator. If it satisfies the Fourier convolution property
for every \(f,g\in L^1(G),\) then there exists an open set \(E \subset \widehat{G}\) and a continuous function \(\theta _{\mathcal {F}}: E \longrightarrow \widehat{G}\) such that
Furthermore, if we assume that T satisfies the additional property
for every \(f\in L^1(G),\ \phi \in \widehat{G}\) and \(x\in G\), then \(\theta _{\mathcal {F}}(\phi ) = \overline{\phi }.\)
Proof
We begin by defining
which is obviously a closed set. Next, we define
which is open. If \(\phi \in D = \widehat{G}\backslash E\) then the equality (3) is obvious. Hence, we choose \(\phi \in E\) and consider a nonzero, linear functional \(T_{\phi }: L^1(G)\longrightarrow \mathbb {C}\) given by \(T_{\phi }(f):= T(f)(\phi ).\) We pick \(g_* \in L^1(G)\) such that \(T_{\phi }(g_*) = 1\) and define a function \(\chi _{\phi }: G\longrightarrow \mathbb {C}\) by the formula
Let us remark, that the choice of \(g_*\) need not be unique, so there might be many functions \(\chi _{\phi }\) corresponding to \(\phi .\)
We will now focus on proving various properties of the function \(\chi _{\phi }.\) To begin with, we observe that for every \(x,y \in G\) we have
Taking the complex conjugate reveals the equation
which holds for every \(x,y\in G.\) Furthermore, since \(T_{\phi }\) is a continuous linear functional, then Lemma 1.4.2 in [3], p. 18 implies that \(\chi _{\phi }\) is continuous. It is also nonzero (as \(\chi _{\phi }(0) = 1\)) and bounded, since for every \(x\in G\) we have
Next, we argue that \(|\chi _{\phi }(x)| = 1\) for every \(x\in G.\) Indeed, suppose there exists \(\bar{x} \in G\) such that \(|\chi _{\phi }(\bar{x})| \ne 1.\) Since
then either \(|\chi _{\phi }(\bar{x})| > 1\) or \(|\chi _{\phi }(-\bar{x})| > 1.\) Without loss of generality, we may assume that the former is true. Consequently, we have
which contradicts boundedness of \(\chi _{\phi }.\) Hence, we conclude that \(|\chi _{\phi }(x)| = 1\) for every \(x\in G.\) This means that \(\chi _{\phi } \in \widehat{G}.\) We put \(\theta _{\mathcal {F}}(\phi ):= \chi _{\phi }\) and observe that by Lemma 11.45 in [2], p. 427 (or Proposition 7 in [4], p. 123) we have
for every \(f\in L^1(G).\) This is the equality (3).
In order to see that \(\theta _{\mathcal {F}}\) is continuous we invoke (again) Lemma 1.4.2 in [3], p. 18 and conclude that the map \(x \mapsto T(L_xg_*)\) is continuous. Consequently, the family \((T(L_xg_*))_{x\in K}\) is compact in \(C^b(\widehat{G})\) for every compact \(K\subset G.\) In particular, by the Arzelà–Ascoli theorem, the family \((T(L_xg_*))_{x\in K}\) is equicontinuous at every \(\phi _*\in E,\) i.e., for every \(\varepsilon >0\) there exists an open neighbourhood \(U_*\) of \(\phi _*\) such that
for every \(\phi \in U_*\) and \(x \in K.\) This implies that
for every \(\phi \in U_*\). In other words, as \(\phi \) converges to \(\phi _*\), \(\chi _{\phi }\) converges to \(\chi _{\phi _*}\), which proves continuity of \(\theta _{\mathcal {F}}\) and concludes the first part of the proof.
For the second part, we assume that the operator T satisfies (4). Hence, we compute that
for every \(\phi \in E\) and \(x \in G\), which concludes the proof. \(\square \)
We move on to show that the cosine transform admits a similar convolution characterization. The cosine transform is a map \(\mathcal {C}: L^1(G)\longrightarrow C_0(\mathcal {COS}(G))\) given by the formula
where
Functions in \(\mathcal {COS}(G)\) are called cosine functions on group G, since for \(G = \mathbb {R}\) we have
The cosine convolution \(\star _{\mathcal {C}}:L^1(G)\times L^1(G)\longrightarrow L^1(G)\) is given by
Let us recall a property of the cosine convolution which will be crucial in the sequel (see Theorem 2 in [6]):
Lemma 2
Let \(x,y\in G.\) If \(g\in L^1(G)\) is an even function, then
Before we present a characterization of the cosine transform we need one more technical result:
Lemma 3
Let \(T:L^1(G)\longrightarrow C^b(\mathcal {COS}(G))\) be a linear and bounded operator, which satisfies the cosine convolution property
for every \(f,g\in L^1(G)\). If \(\phi \in \mathcal {COS}(G)\) is such that \(T(f_*)(\phi ) \ne 0\) for some \(f_*\in L^1(G)\), then there exists an even function \(g_*\in L^1(G)\) such that \(T(g_*)(\phi )\ne 0.\)
Proof
Let \(\iota :G\longrightarrow G\) be the inverse function \(\iota (x):= -x\) and let \(T_{\phi }: L^1(G)\longrightarrow \mathbb {C}\) be a nonzero, linear functional given by \(T_{\phi }(f):= T(f)(\phi ).\) Then
for every \(f\in L^1(G).\) This leads to
for every \(f\in L^1(G).\) Consequently, we have \(T_{\phi }(f_*\circ \iota ) = T_{\phi }(f_*).\) Finally, we put \(g_*:= f_* + f_*\circ \iota \) and observe that
which concludes the proof. \(\square \)
We are now ready to demonstrate a counterpart of Theorem 1 for the cosine transform:
Theorem 2
Let \(T:L^1(G)\longrightarrow C^b(\mathcal {COS}(G))\) be a linear and bounded operator. If it satisfies the cosine convolution property (10), then there exists an open set \(E\subset \mathcal {COS}(G)\) and a continuous function \(\theta _{\mathcal {C}}: E \longrightarrow \mathcal {COS}(G)\) such that
Furthermore, if we assume that T satisfies the additional property
for every \(f\in L^1(G),\ \phi \in E\) and \(x\in G\), then \(\theta _{\mathcal {C}}(\phi ) = \phi .\)
Proof
The proof follows the same lines as the proof of Theorem 1. We define
and (as before) argue that it is an open set. For \(\phi \in \widehat{G}\backslash E\) we have \(T(f)(\phi ) = 0\) for every \(f\in L^1(G),\) i.e., (11) is satisfied. Thus, for the rest of the proof we choose \(\phi \in E.\)
Next, we consider a nonzero, linear functional \(T_{\phi }: L^1(G)\longrightarrow \mathbb {C}\) given by \(T_{\phi }(f):= T(f)(\phi ).\) By Lemma 3 there exists an even function \(g_* \in L^1(G)\) such that \(T_{\phi }(g_*) = 1,\) which we use to define a function \(\chi _{\phi }: G\longrightarrow \mathbb {C}\) by the formula
As in Theorem 1 we study the properties of \(\chi _{\phi }\) and see that
-
it is nonzero due to \(\chi _{\phi }(0) =1,\)
-
it is bounded due to (5),
-
it is continuous due to Lemma 1.4.2 in [3], p. 18.
Furthermore, we have
for every \(x,y\in G\). This can be written as
for every \(x,y\in G.\) We conclude that \(\chi _{\phi } \in \mathcal {COS}(G),\) which in particular means that
for every \(y\in G.\) We put \(\theta _{\mathcal {C}}(\phi ):= \chi _{\phi }\) and observe that by Lemma 11.45 in [2], p. 427 (or Proposition 7 in [4], p. 123) we have
for every \(f\in L^1(G)\). This demonstrates the equality (11). To conclude the first part of the theorem we observe that \(\theta _{\mathcal {C}}\) is continuous due to the same reasoning as in Theorem 1.
For the second part of the theorem we compute that
for every \(\phi \in E\) and \(x\in G.\) Dividing by \(T_{\phi }(g_*)\) we conclude the proof. \(\square \)
3 Laplace Transform
In the previous section we focused on Fourier and cosine transforms, characterizing them via suitable convolution properties. The purpose of the current section is to demonstrate that the Laplace transform enjoys a similar characterization. Let us recall that the Laplace transform is a map \(\mathcal {L}:L^1(\mathbb {R}_+)\longrightarrow C_0(\mathbb {R}_+)\) given by
Unlike in the previous section, when discussing the Laplace transform we assume that both spaces \(L^1(\mathbb {R}_+)\) and \(C^b(\mathbb {R}_+)\) comprise of real-valued (not complex-valued) functions.
If the Laplace convolution \(\star _{\mathcal {L}}:L^1(\mathbb {R}_+)\times L^1(\mathbb {R}_+)\longrightarrow L^1(\mathbb {R}_+)\) is given by
then it is well-known that following equality (see Theorem 2.39 in [10], p. 92) holds for every \(f,g\in L^1(\mathbb {R}_+)\):
Similarly to the previous section, our goal is to “reverse the implication” and ask whether there are any other operators satisfying (16). A (almost) negative answer will constitute a characterization of the Laplace transform in terms of the Laplace convolution (15). To this end, we need the following auxilary lemma:
Lemma 4
Let \(g\in L^{\infty }(\mathbb {R}_+)\) and let D be a dense subset in \(L^1(\mathbb {R}_+)\). If
for every \(f\in D,\) then \(g = 0.\)
At last, we are ready to characterize the Laplace transform in terms of the Laplace convolution property:
Theorem 3
Let \(T:L^1(\mathbb {R}_+)\longrightarrow C^b(\mathbb {R}_+)\) be a linear and bounded operator. If it satisfies the Laplace convolution property
for every \(f,g\in L^1(\mathbb {R}_+),\) then there exists an open set \(E\subset \mathbb {R}_+\) and a continuous function \(\theta _{\mathcal {L}}: E \longrightarrow \mathbb {R}_+\cup \{0\}\) such that
Furthermore, if we assume that T satisfies the additional property that there exists \(t_* \in \mathbb {R}_+\) such that
for every \(f\in L^1(\mathbb {R}_+)\) and \(y \in E\), where
then \(\theta _{\mathcal {L}}(y) = y.\)
Proof
Just as in Theorems 1 and 2 we commence the proof by defining
If \(y\in \mathbb {R}_+\backslash E\) then \(T(f)(y) = 0\) for every \(f\in L^1(\mathbb {R}_+)\) and thus the equality (18) holds. Hence, we focus on \(y \in E\) and define a nonzero, linear functional \(T_y: L^1(\mathbb {R}_+)\longrightarrow \mathbb {R}\) given by \(T_y(f):= T(f)(y).\) As the dual of \(L^1(\mathbb {R}_+)\) is \(L^{\infty }(\mathbb {R}_+)\) then there exists a nonzero function \(\chi _y\in L^{\infty }(\mathbb {R}_+)\) such that
for every \(f\in L^1(\mathbb {R}_+).\)
By Fubini’s theorem we have
for every \(f,g\in L^1(\mathbb {R}_+).\) Since \(T_y\) is nonzero we pick \(g_*\in C_c(\mathbb {R}_+)\) such that \(T_y(g_*) = 1.\) By the Laplace convolution property (17) we have
for every \(f\in L^1(\mathbb {R}_+)\). This implies that we have
or equivalently
for every \(f\in L^1(\mathbb {R}_+).\) Since
is a \(L^{\infty }-\)function then by Lemma 4 we conclude that
for every \(v\in \mathbb {R}_+\). We use this integral functional equation to establish continuity of \(\chi _y.\) For a fixed \(v_*\in \mathbb {R}_+\) and \(h\in (-v_*,v_*)\) we have
Since \(g_*\) is continuous and compactly supported and \(h \in (-v_*,v_*)\) then there exists some constant C such that the function
is continuous and takes the value 0 for \(h = 0.\) Consequently, we have
establishing the continuity of \(\chi _y\).
For the second time we use the Laplace convolution property (17) to obtain
which we rewrite as
for every \(f,g\in L^1(\mathbb {R}_+).\) Since \((u,v)\mapsto \chi _y(u+v) - \chi _y(v)\chi _y(u)\) is a \(L^{\infty }-\)function, then by Lemma 4 we have
for every \(u,v\in \mathbb {R}_+\).
Suppose, for the sake of contradiction, that there exists some \(\bar{u}\in \mathbb {R}_+\) such that \(\chi _y(\bar{u}) = 0.\) It immediately follows that \(\chi _y(v) = 0\) for every \(v > \bar{u}.\) Next, since
then \(\chi _y\left( \frac{\bar{u}}{2}\right) = 0.\) An inductive reasoning establishes that \(\chi _y\left( \frac{\bar{u}}{2^n}\right) = 0\) for every \(n\in \mathbb {N}\). This means that \(\chi _y = 0\) and consequently, \(T_y = 0.\) This contradicts our choice of \(y\in E\) (made at the beginning of the proof) so we assume that \(\chi _y(u) \ne 0\) for every \(u\in \mathbb {R}_+.\) By a standard (an easy) argument we establish that \(\chi _y(u) = e^{-zu}\) for some \(z\in \mathbb {R}_+\cup \{0\}.\) We set \(\theta _{\mathcal {L}}(y):= z\).
To conclude the first part of the proof we need to demonstrate that \(\theta _{\mathcal {L}}\) is continuous. We fix \(y_* \in E \) and observe that since \(T_{y_*}\) is a nonzero functional, then there exist \(a,b \in \mathbb {R}_+\) such that \(T(\mathbbm {1}_{(a,b)})(y_*)\ne 0.\) By continuity of \(T(\mathbbm {1}_{(a,b)})\) there exists an open neighbourhood \(U_*\subset E\) of \(y_*\) such that \(T(\mathbbm {1}_{(a,b)})(y) \ne 0\) for \(y\in U_*.\) Next, we define \(h:\mathbb {R}_+\cup \{0\}\longrightarrow \mathbb {R}\) with the formula
This function is clearly continuous on \(\mathbb {R}_+\) and (as strictly monotone) has a continuous inverse. Consequently, for every \(y \in U_*\) we have
This implies that \(\theta _{\mathcal {L}} = h^{-1}\circ T(\mathbbm {1}_{(a,b)})|_{U_*}\) is continuous at \(y_*\) as a composition of two continuous functions.
For the second part of the proof, we have
for every \(f\in L^1(\mathbb {R}_+)\) and \(y\in E\). For every \(y\in E\) we choose \(f\in L^1(\mathbb {R}_+)\) such that \(T(f)(y)\ne 0\) to conclude that \(\theta _{\mathcal {L}}(y) = y.\) \(\square \)
Data Availability Statement
Not applicable.
Code Availability
Not applicable.
References
Alesker, S., Artstein-Avidan, S., Faifman, D., Milman, V.: A characterization of product preserving maps with applications to a characterization of the Fourier transform. Ill. J. Math. 54(3), 1115–1132 (2010)
Aliprantis, C.D., Border, K.C.: Infinite Dimensional Analysis. A Hitchhiker’s Guide. Springer, Berlin (2006)
Deitmar, A., Echterhoff, S.: Principles of Harmonic Analysis. Springer, New York (2009)
Dinculeanu, N.: Vector Measures. Pergamon Press, Berlin (1967)
Jaming, P.: A characterization of Fourier transforms. Coll. Math. 118(2), 569–580 (2010)
Krukowski, M.: Cosine manifestations of the Gelfand transform. Results Math. 77, 83 (2022)
Kumar, N.S., Sivananthan, S.: Characterisation of the Fourier transform on compact groups. Bull. Aust. Math. Soc. 93, 467–472 (2016)
Lavanya, R.L., Thangavelu, S.: A characterization of the Fourier transform on the Heisenberg group. Ann. Funct. Anal. 3(1), 109–120 (2012)
Lavanya, R.L., Thangavelu, S.: Revisiting the Fourier transform on the Heisenberg group. Publ. Mat. 58, 47–63 (2014)
Schiff, J.L.: The Laplace Transform. Theory and Applications. Springer, New York (1999)
Acknowledgements
We would like to express our deepest gratitude for the anonymous Reviewer, whose insightful comments and instructive suggestions greatly improved the quality of the paper. We feel privileged to be provided with such a constructive criticism and invaluable assistance.
Funding
Not applicable.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The author declares that he has no conflict of interest.
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 licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Krukowski, M. Integral Transforms Characterized by Convolution. Results Math 78, 120 (2023). https://doi.org/10.1007/s00025-023-01897-4
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00025-023-01897-4