We introduce a family of Sobolev-type spaces on Gelfand pairs, built from the Stockwell transform: a function belongs to the space of a given smoothness when its Stockwell transform, multiplied by a weight that grows with the smoothness, is square integrable. We prove that these spaces are Hilbert spaces, that they form a decreasing scale under continuous embeddings, and that an interpolation inequality controls the norm at an intermediate smoothness by the norms at the two endpoints. Under integrability conditions on the weight and on the family of analysing windows, they embed continuously into Lebesgue spaces. Under a local domination condition on the window, we show that their elements have bounded continuous representatives. On the Euclidean motion group, where the Stockwell transform reduces to a windowed Hankel transform, this gives boundedness and continuity as soon as the smoothness exceeds half the dimension, which is the classical threshold for Euclidean spaces.
The Stockwell transform, introduced by Stockwell, Mansinha and Lowe [12], can be viewed either as a short-time Fourier transform with a frequency-dependent window, or as a phase-corrected wavelet transform, and it has since become a standard tool for analysing non-stationary signals in seismology, biomedical imaging, and mechanical fault diagnosis. While the Stockwell transform was first developed on the real line and on Euclidean space, subsequent work has sought to place it within the broader framework of abstract harmonic analysis, in the same way the short-time Fourier and wavelet transforms have already been extended to locally compact groups and homogeneous spaces. Esmaeelzadeh carried out the first such extension [5, 6], defining the Stockwell transform, together with its associated localization operators, on locally compact abelian groups. More recently, this construction was pushed further, to the setting of Gelfand pairs [4], a class that strictly contains locally compact abelian groups and that includes many spaces of independent geometric interest, such as spheres, hyperbolic spaces, and other symmetric spaces. Gelfand pairs occupy a distinguished position in noncommutative harmonic analysis; although the ambient group need not be commutative, the algebra of bi-invariant functions still is, so that a spherical Fourier analysis remains available. Working at the level of Gelfand pairs therefore makes it possible to transport time-frequency methods, originally designed for commutative groups, to a much wider and geometrically richer class of settings.
On the other hand, Sobolev spaces have long played a central role in the study of partial differential equations and integral equations, providing the natural framework in which questions of regularity, boundedness, and embedding are addressed. The interplay between Sobolev-type smoothness and time-frequency transforms has proved fruitful. Classically, Sobolev spaces of Bessel potentiel type are built from the Fourier transform on Euclidean spaces [1] and on locally compact groups/hypergroups [7, 9] and homogeneous spaces [2, 8, 10].
The purpose of the present paper is to construct a family of Sobolev-type spaces associated with the Stockwell transform on Gelfand pairs, and to study some of their major properties. We show that these spaces carry a natural Hilbert space structure, we establish a scale of continuous embeddings relating spaces of different smoothness to one another and to the underlying square-integrable functions, and we obtain an embedding into Lebesgue spaces under a suitable integrability condition. We further show, under a local domination hypothesis, that sufficiently smooth elements of these spaces are in fact bounded and continuous. Throughout, the theory is illustrated on the Euclidean motion group, a noncommutative Gelfand pair on which it recovers, as a special case, the classical smoothness threshold for the continuous embedding into bounded continuous functions on Euclidean spaces.
The remainder of the paper is organized as follows. Section 2 collects the background material on harmonic analysis on Gelfand pairs and on the Stockwell transform needed in the sequel. Section 3 contains the main results.
Preliminaries
In this section, we provide the tools needed in the sequel.
Harmonic analysis on Gelfand pairs
For this subsection we refer to [3, 13]. Let G be a locally compact group (assumed to be Hausdorff once for all) with a left Haar measure and let K be a compact subgroup of G endowed with its normalized Haar measure. A function f:G→C is said to be K-bi-invariant if it satisfies the property:
∀x∈G,∀k,k′∈K,
f(kxk′)=f(x).
Let Cc(G//K) and Lp(G//K), 1≤p<∞, denote respectively the space of complex-valued continuous compactly supported K-bi-invariant functions on G and the space of complex-valued K-bi-invariant Haar p-integrable functions on G. For 1≤p<∞, Lp(G//K) is a complex Banach space with respect to the norm
∥f∥Lp=(∫G∣f(x)∣p)p1.
The convolution of f,g∈L1(G//K) is defined by
f∗g(x)=∫Gf(y)g(y−1x)dy,x∈G.
The pair (G,K) is called a Gelfand pair if Cc(G//K) is a commutative convolution algebra; by density of Cc(G//K) in L1(G//K), this is equivalent to say that L1(G//K) is a commutative convolution algebra. It is well known that if (G,K) is a Gelfand pair, then G is unimodular [3 Proposition 6.1.2].
A continuous K-bi-invariant function φ is called a spherical function for the Gelfand pair (G,K) if the map χφ:Cc(G//K)→C, f↦∫Gf(x)φ(x−1)dx is a nontrivial character of the convolution algebra Cc(G//K), that is
χ(f∗g)=χ(f)χ(g),f,g∈Cc(G//K).
The set of bounded spherical functions for the Gelfand pair (G,K) is denoted by Sb(G,K). The spherical Fourier transform for f∈L1(G//K) is defined by
f(φ)=∫Gf(x)φ(x−1)dx,φ∈Sb(G,K).
A function φ:G→C is said to be positive-definite function if
Theorem 2.1.[13 Proposition 8.4.2] Let φ be a positive-definite function. Then,
∀x∈G, ∣φ(x)∣≤φ(e).
∀x∈G, φ(x−1)=φ(x).
If φ is a spherical function, then φ(e)=1[3 Proposition 6.1.5].
Let S+ denote the set of positive definite spherical functions for the Gelfand pair (G,K). It carries a Plancherel measure which ensures the inversion formula for the spherical Fourier transform. Indeed, the spherical Fourier transform extends to an isometric isomorphism L2(G//K)⟶L2(S+) and
f(x)=∫S+f(φ)φ(x)dφ,f∈L2(G//K),x∈G.
The Stockwell transform on Gelfand pairs
We recall from [4] the definition of the Stockwell transform on a Gelfand pair together with the two properties, the inversion formula and the Plancherel-type isometry, that will actually be used in the sequel.
Let (G,K) be a Gelfand pair and α a topological automorphism of G. There exists a positive constant δα such that d(α(x))=δαdx. It follows,
∫Gf(α(x))dx=δα−1∫Gf(x)dx,f∈Cc(G//K).
Definition 2.2 ([4 Definition 1]). Let (G,K) be a Gelfand pair, let α be a topological automorphism of G and let θ∈L1(G//K)∩L2(G//K). The Stockwell transform of f∈L2(G//K) with respect to α and θ is
In particular, if moreover ∥θ∥L2=1, then Sθ,α is an isometry:
∥Sθ,αf∥L2(G×S+)=∥f∥L2,∀f∈L2(G//K).
Main results
We state the main definition of this article.
Definition 3.1. Let γ:G×S+→R+ be a positive measurable function and s∈R+. The Sobolev space associated with the Stockwell transform on the Gelfand pair (G,K) is the set
Thus, (fn)n is also Cauchy in L2(G//K). Since L2(G//K) is complete, (fn)n converges to some f in L2(G//K).
On the other hand, (Γfn)n=((1+γ2)2sSθ,αfn)n is Cauchy in L2(G×S+) as Γ is an isometry, hence converges to some g∈L2(G×S+). That is, (Sθ,αfn)n converges to (1+γ2)−2sg.
Since Sθ,α is an isometry L2(G//K)→L2(G×S+), it is in particular continuous, so Sθ,αfn converges to Sθ,αf in L2(G×S+). By uniqueness of the limit, Sθ,αf=(1+γ2)−2sg a.e., that is g=Γf.
as n goes to ∞. Hence, every Cauchy sequence in Hγ,θ,αs(G//K) converges in Hγ,θ,αs(G//K), which is therefore a Hilbert space. □
Next, we show that increasing the smoothness parameter yields a decreasing scale of spaces, in the sense of continuous embeddings.
Theorem 3.3. Let (G,K) be a Gelfand pair and let θ∈L1(G//K)∩L2(G//K) be such that ∥θ∥L2=1. If τ>σ≥0, then the space Hγ,θ,ατ(G//K) is continuously embedded into Hγ,θ,ασ(G//K) with
∥f∥Hγ,θ,ασ≤∥f∥Hγ,θ,ατ,∀f∈Hγ,θ,ατ(G//K).
Proof. Since (1+γ(t,φ)2)≥1, then for τ>σ≥0, we have (1+γ(t,φ)2)τ≥(1+γ(t,φ)2)σ. The assertion follows. □
We next establish an interpolation inequality, showing that the norm at an intermediate smoothness level is controlled by the norms at the two endpoint levels.
Theorem 3.4. Let s1,s2 be nonnegative real numbers satisfying s1<s2. Let ζ∈[0,1], put s=(1−ζ)s1+ζs2. For all f∈Hγ,θ,αs(G//K), we have
In order to obtain embeddings into other Lebesgue spaces, we first establish the following auxiliary estimate, which bounds the Lebesgue norm of a function by a Lebesgue-type norm of its Stockwell transform.
Lemma 3.5. Let (G,K) be a Gelfand pair and let θ∈L1(G//K)∩L2(G//K) with ∥θ∥L2=1. Let 1≤p<∞ and 1≤q≤∞, with conjugate exponent q′, and set
Np,q=(∫S+∫G∥θα,φ,t∥Lp(G)q′dtdφ)q′1
(an essential supremum over (t,φ) if q′=∞). If Np,q<∞, then for every f∈L2(G//K),
∥f∥Lp≤δα21Np,q∥Sθ,αf∥Lq(G×S+).
Proof. Let g∈Cc(G//K)⊂L2(G//K). Since Sθ,αg(t,φ)=δα1/2⟨g,θα,φ,t⟩, Theorem 2.4 (with η=θ, $\lVert\theta\rVert_{L^2}=1) gives
Building on the previous lemma, we now show that a suitable condition on the weight yields a continuous embedding of the Sobolev-type space into a Lebesgue space.
Theorem 3.6. Let (G,K) be a Gelfand pair and let θ∈L1(G//K)∩L2(G//K) with ∥θ∥L2=1. Let s>0, β>2 and set p=β+22β∈(1,2), so that p1=21+β1. Assume that (1+γ2)−2s∈Lβ(G×S+) and Np:=Np,p<∞ (with Np,p as in Lemma 3.5, i.e. with q=p). Then Hγ,θ,αs(G//K) is continuously embedded in Lp(G//K) and
Lemma 3.5 with q=p concludes ∥f∥Lp≤δα21Np∥Sθ,αf∥Lp.
□
We now give a second embedding into Lebesgue spaces, based on the endpoint case q=1 of Lemma 3.5. It requires only that the window belongs to Lp and that the inverse of the weight is square integrable, and it involves no condition on the family of analysing windows.
Proposition 3.7. Let θ∈L1(G//K)∩L2(G//K)∩Lp(G//K) with ∥θ∥L2=1, 1≤p<∞, s>0, and assume (1+γ2)−s/2∈L2(G×S+). Then Hγ,θ,αs(G//K)↪Lp(G//K) with
(left invariance of the Haar measure and the change of variables x↦α(x)), we get Np,1≤δα−1/p∥θ∥Lp. Moreover, by Cauchy–Schwarz,
∥Sθ,αf∥L1≤∥(1+γ2)s/2Sθ,αf∥L2∥(1+γ2)−s/2∥L2.
Finally, apply Lemma 3.5 with q=1, q′=∞ yields the result. □
We finally turn to pointwise regularity. The following theorem states that, under a local domination condition on the window, the elements of the Sobolev-type space are continuous, and that the space embeds continuously into the bounded continuous functions.
Theorem 3.8. Let (G,K) be a Gelfand pair with G first countable, and let α be a topological automorphism of G with α(K)=K. Let θ∈L1(G//K)∩L2(G//K) be continuous with ∥θ∥L2=1, and let s≥0. Consider the following hypotheses:
(H) There exist a K-bi-invariant open neighbourhood V of x0∈G and h∈L1(G×S+) such that, for every y∈V and for a.e. (t,φ),
(using ∣φ(x)∣≤1 and the Cauchy-Schwarz inequality)
≤∥f∥Hγ,θ,αs∥h∥L121<∞
(using (H)).
Let x0∈V and let (xn) be a sequence which converges to x0. Since V is open, xn∈V for n large. Put In(t,φ)=Sθ,αf(t,φ)φ(xn)θ(α(t−1xn)). Since there are countably many xn, for a.e. (t,φ) and all n, by (H), we have
We have G∈L1(G×S+) because (1+γ2)s/2Sθ,αf∈L2(G×S+) and h1/2∈L2(G×S+). Moreover, for every (t,φ), In(t,φ)→I0(t,φ), because φ and θ are continuous and so are α and the group operations. The Lebesgue dominated convergence theorem implies that F(xn) converges to F(x0). Since G is first countable, F is continuous at x0.
Now let us show that F=f a.e. on V. First, F is K-bi-invariant on V: left K-invariance follows from the change of variables t↦kt together with Sθ,αf(kt,φ)=Sθ,αf(t,φ) (because f and φ are K-bi-invariant); right K-invariance follows from φ(xk)=φ(x), α(K)=K and the right K-invariance of θ. Now let g∈Cc(G//K) with suppg⊂V. We have
and the following continuous embedding follows: Hγ,θ,αs(G//K)↪Cb(G//K).
Example 3.9 (The Euclidean motion group). Let n≥2 and (G,K)=(E(n),SO(n)), where E(n)=Rn⋊SO(n) with product and inverse
(y,k)(y′,k′)=(y+k⋅y′,kk′),(y,k)−1=(−k−1y,k−1).
The Haar measure of G is dydk, where dy is the Lebesgue measure on Rn and dk is the normalized Haar measure of K. A function on G is K-bi-invariant if and only if it is of the form (y,k)↦f(∣y∣), i.e. it is a radial function on Rn[3 Subsection 7.1]. We identify Lp(G//K) with the radial functions of Lp(Rn).
The bounded spherical functions are indexed by λ≥0:
where Jν is the Bessel function of the first kind. They are real-valued, S+≅[0,∞), and the spherical Fourier transform f(λ)=∫Rnf(y)φλ(y)dy is the Hankel transform. With the Haar measure dy, the Plancherel measure is dφλ=cnλn−1dλ, where cn>0 depends only on n and on the normalization of the Haar measure.
For a>0 let αa(y,k)=(ay,k). Then αa is a topological automorphism of E(n) with αa(K)=K (it fixes K pointwise), and d(αa(x))=andx, i.e. δαa=an. Let
θ(y,k)=θ(y)=π−4ne−2∣y∣2,
which is radial, continuous, belongs to L1∩L2, and satisfies ∥θ∥L2=1. For t=(u,h) and x=(y,k) in G,
αa(t−1x)=(ah−1(y−u),h−1k),
hence
θ(αa(t−1x))=π−4ne−2a2∣y−u∣2,
which depends neither on h nor on k. Since φλ and θ are real and the integral over K is trivial, for a radial f we get
Take the weight γ(t,λ)=λ, which depends only on the spherical variable. Definition 3.1 gives
Hγ,θ,αas(E(n)//SO(n))={f radial on Rn:cn∫0∞∫Rn(1+λ2)s∣Sθ,αaf(u,λ)∣2duλn−1dλ<∞}.
We check the hypotheses of Theorem 3.8. Fix ρ≥0, r>0 and the K-bi-invariant open set V={(y,k):∣∣y∣−ρ∣<r}; every point of G has such a neighbourhood. For y∈V and u∈Rn we have ∣y−u∣≥∣∣u∣−ρ∣−r whenever ∣∣u∣−ρ∣≥r, hence, with t=(u,h),
The function κ is integrable on Rn (a bounded annulus and a Gaussian tail). Put h(t,λ)=π−n/2κ(u)(1+λ2)−s: it satisfies (H) on V, and it is integrable on G×S+ as soon as
∫0∞(1+λ2)−sλn−1dλ<∞.
Moreover, for every y∈G, the change of variables w=t−1y and the unimodularity of G give ∫G∣θ(αa(t−1y))∣2dt=δαa−1=a−n, so that
M=a−ncn∫0∞(1+λ2)−sλn−1dλ,
which is finite if and only if s>n/2. Theorem 3.8 therefore shows that, for s>n/2, every element of Hγ,θ,αas(E(n)//SO(n)) has a representative F which is a bounded continuous radial function on Rn, with
∥F∥∞≤(cn∫0∞(1+λ2)−sλn−1dλ)21∥f∥Hγ,θ,αas.
The condition s>n/2 is the classical threshold for the embedding of Hs(Rn) into the bounded continuous functions [11] where
Hs(Rn)={u∈S′(Rn):(1+∣ξ∣2)s/2u(ξ)∈L2(Rn)},
S′(Rn) being the space of tempered distributions and u is the Fourier transform of u.
Conclusion
In this paper we have defined Sobolev-type spaces on a Gelfand pair through the Stockwell transform, with the smoothness measured by a weight acting on the time-frequency variables rather than on the frequency alone. The Plancherel-type identity for the Stockwell transform makes these spaces Hilbert spaces, and the usual structural properties of a Sobolev scale hold: continuous embeddings between different smoothness levels and an interpolation inequality. We also obtained embeddings into Lebesgue spaces and into the bounded continuous functions. The embedding into bounded continuous functions rests on a domination condition on the window and requires no equicontinuity of the family of spherical functions. On the Euclidean motion group, the construction leads to a windowed Hankel transform and to the classical smoothness threshold for continuity. Finally, the same construction could be attempted for the Stockwell transform in related frameworks, such as hypergroups or commutative triples, and applied to the regularity of localization operators.
References
Department of Mathematics, University of Lomé, Togo
[1]Adams, R.A., Fournier, J.J.F.: Sobolev Spaces (2nd ed.), Vol. 140, Pure and Applied Mathematics, Academic Press, 2003.
[2]Bataka, K.T., Egwe, M. E., Mensah, Y.: Sobolev spaces on hypergroup Gelfand pairs, Far East J. Math. Sci. 142(1), (2025), 57-69.DOI
[3]van Dijk, G.: Introduction to Harmonic Analysis and Generalized Gelfand Pairs, Walter de Gruyter GmbH & Co. KG, Berlin, 2009.
[4]Dosseh, C.G., Todjro, M., Mensah, Y.: The Stockwell transform on Gelfand pairs and localization operators, J. Pseudo-Differ. Oper. Appl. 17(3), (2026), article 55.
[5]Esmaeelzadeh, F.: Localization operators related to Stockwell transforms on locally compact abelian groups, J. Pseudo-Differ. Oper. Appl. 14(4), (2023), article 67.DOI
[6]F. Esmaeelzadeh, The Stockwell transform on locally compact abelian groups, Infinite Dimensional Analysis, Quantum Probability and Related Topics, 2024, vol. 27, no. 1, p.2350002.DOI
[7]Górka, P., Reyes, E. G.: Sobolev spaces on locally compact abelian groups and the bosonic string equation, J. Aust. Math. Soc. 97(1), (2014), 101-114.DOI
[9]Kumar, M., Kumar, N.S.: Sobolev spaces on compact groups, Forum Math. 35(4), 2023, 901-911.DOI
[10]Mensah, Y.: Sobolev spaces arising from a generalized spherical Fourier transform, Adv. Math. Sci. 10(7) (2021), 2947-2955.DOI
[11]Stein, E.M., Weiss, G.: Introduction to Fourier Analysis on Euclidean Spaces, Princeton University Press, 1971.
[12]Stockwell, R.G., Mansinha, L., Lowe, R.P.: Localization of the complex spectrum: the S-transform, IEEE Trans. Signal Proc. 44 (4), (1996) 998-1001.DOI
[13]Wolf, J.A.: Harmonic Analysis on Commutative spaces, Amer. Math. Soc., Mathematical Surveys and Monographs 142, 2007.DOI