PJM 13 (2) 2024 8 To 17
PJM 13 (2) 2024 8 To 17
Keywords and phrases: Hypergroups, Gelfand pair, probability measure, positive definite measure, strongly positive
definite function.
1 Introduction
Hypergroups generalize locally compact groups. They appear when the Banach space of all
bounded Radon measures on a locally compact space carries a convolution having all proper-
ties of a group convolution apart from the fact that the convolution of two point measures is a
probability measure with compact support and not necessarily a point measure. The intention
was to unify harmonic analysis on duals of compact groups, double cosets spaces G//H (H a
compact subgroup of a locally compact group G), and commutative convolution algebras associ-
ated with product linearization formulas of special functions. The notion of hypergroup has been
sufficiently studied (see for example [2, 5, 8, 9]). Harmonic analysis and probability theory on
commutative hypergroups are well developed meanwhile where many results from group theory
remain valid (see [1]). When G is a commutative hypergroup, the convolution algebra Mc (G)
consisting of measures with compact support on G is commutative. The typical example of a
commutative hypergroup is the double coset G//K when G is a locally compact group, K is a
compact subgroup of G such that (G, K ) is a Gelfand pair. In [5], R. I. Jewett has shown the
existence of a positive measure called Plancherel measure on the dual space G b of a commutative
hypergroup G. In [6], R. Lasser relying on this result, has established a Bochner theorem on a
commutative hypergroup G. When the hypergroup G is not commutative, it is possible to involve
a compact subhypergroup K of G leading to a commutative subalgebra of Mc (G). In fact, if K
is a compact subhypergroup of a hypergroup G, the pair (G, K ) is said to be a Gelfand pair if
Mc (G//K ) the convolution algebra of measures with compact support on G//K is commuta-
tive. The notion of Gelfand pairs for hypergroups is well-known (see [3, 10, 11]). When (G, K )
is a Gelfand pair; it has been shown in [4] the existence of a Plancherel measure on G b . The
goal of this paper is to extend Lasser’s work by obtaining a Bochner theorem over Gelfand pair
associated with a noncommutative hypergroup. In the next section, we give notations and setup
useful for the remainder of this paper. In section 3, we define the inverse Fourier transform on
Mb (G b ), the algebra of bounded measures on G b and obtain a relationship between measures in
Mb (G) and measures in Mb (G b ). Then, we define the so-called positive (rep. strongly positive)
definite measure (resp. function) on Gb and obtain some of their characterizations. Finally, thanks
to these results, we prove that for any strongly positive definite measure, there exists a positive
measure in Mb (G) whose Fourier transform coincides with the function on the support of the
Plancherel measure.
A Bochner theorem on a noncommutative hypergroup 9
Definition 2.1. [5]. Let G be a locallly compact topological space. G is said to be a hypergroup
if the following assumptions are satisfied.
(H1) There is a binary operator ∗ named convolution on Mb (G) under which Mb (G) is an asso-
ciative algebra such that:
i) the mapping (µ, ν ) 7−→ µ ∗ ν is continuous from Mb (G) × Mb (G) in Mb (G).
ii) ∀x, y ∈ G, δx ∗ δy is a measure of probability with compact support.
iii) the mapping: (x, y ) 7−→ supp(δx ∗ δy ) is continuous from G × G in C(G).
(H2) There is a unique element e (called neutral element) in G such that δx ∗ δe = δe ∗ δx =
δx , ∀x ∈ G.
(H3) There is an involutive homeomorphism: x 7−→ x from G into G, named involution, such
that:
i) (δx ∗ δy )− = δy ∗ δx , ∀x, y ∈ G with µ− (f ) = µ(f − ) where f − (x) = f (x), ∀f ∈ C (G)
and µ ∈ M (G).
ii) ∀x, y, z ∈ G, z ∈ supp(δx ∗ δy ) if and only if x ∈ supp(δz ∗ δy ).
Let us now consider a hypergroup G provided with a left Haar measure µG and K a compact
10 Kouakou Germain BROU, Ibrahima TOURE and Kinvi KANGNI
subhypergroup of G with a normalized Haar measure ωK . Let us put MµG (G) the space of
measures in Mb (G) which are absolutely continuous with respect to µG . MµG (G) is a closed
self-adjoint ideal in Mb (G). For x ∈ G, the double coset of x with respect to K is K ∗ {x} ∗ K =
{k1S∗ x ∗ k2 ; k1 , k2 ∈ K}. We write simply KxK for a double coset and recall that KxK =
supp(δk1 ∗ δx ∗ δk2 ). All double cosets form a partition of G and the quotient topology
k1 ,k2 ∈K
with respect to the corresponding equivalence relation equips the double cosets space G//K
with a locally compact topology ([1], page 53). The natural mapping pK : G −→ G//K defined
by: pK (x) = KxK , x ∈ G is an open surjective continuous mapping. A function f ∈ C (G)
is said to be invariant by K or K − invariant if f (k1 ∗ x ∗ k2 ) = f (x) for all x ∈ G and
for all k1 , k2 ∈ K . We denote by C ♮ (G), (resp. K♮ (G)) the space of continuous functions
(resp. continuous functions with compact support) which are K−invariant. For f ∈ C ♮ (G),
one defines the function fe on G//K by fe(KxK ) = f (x) ∀x ∈ G. fe is well defined and
it is continuous on G//K . Conversely, for all continuous function φ on G//K , the function
f = φ ◦ pK ∈ C ♮ (G). One has the obvious consequence that the mapping f 7−→ fe sets up
a topological isomorphism between the topological vector spaces C ♮ (G) and C (G//K ) (see
[10, 11]). So, for any f in C ♮ (G), f = fe ◦ pK . Otherwise, we consider the K - projection
♮ ♮
fR 7− e♮
R→ f (by identifying f and f ) from C (G) into C (G//K ) where for x ∈ G, f ♮ (x) =
♮
K K
f (k1 ∗ x ∗ k2 )dωK (k1 )dωK (k2 ). If f ∈ K(G), then f ∈ K(G//K ). For a measure
♮ ♮
µ ∈ M (G), one defines µ by µ (f ) = µ(f ♮ ) for f ∈ K(G). µ is said to be K−invariant if
♮
µ = µ and we denote by M ♮ (G) the set of all those measures. Considering these properties,
R
one defines a hypergroup operation on G//K by: δKxK ∗ δKyK (fe) = K f (x ∗ k ∗ y )dωK (k ) (see
[2, p. 12] ). This defines uniquely the convolution (KxK ) ∗ (KyK ) on G//K.R The involution is
defined by: KxK = KxK and the neutral element is K. Let us put m = G δKxK dµG (x), m is
a left Haar measure on G//K . We say that (G, K ) is a Gelfand pair if the convolution algebra
Mc (G//K ) is commutative. Mc (G//K ) is topologically isomorphic to Mc♮ (G). Considering
the convolution product on K(G), K(G) is a convolution algebra and K♮ (G) is a subalgebra.
Thus (G, K ) is a Gelfand pair if and only if K♮ (G) is commutative ([3], theorem 3.2.2).
3 Bochner theorem
In this section, we give a Bochner theorem for a noncommutative hypergroup extending Lasser’s
works on commutative hypergroups. For that we give some characterizations of the inverse
Fourier transform as well as positive definite measures on the dual space of the hypergroup.
Let G be a hypergroup provided with a left Haar measure µG and K a compact subhypergroup
of G such that (G, K ) is a Gelfand pair.
Let G b be the space of continuous and bounded function ϕ on G satisfying the following condi-
tions:
(i) ϕ is K - invariant,
(ii) ϕR(e) = 1,
(iii) K ϕ(x ∗ k ∗ y )dwK (k ) = ϕ(x)ϕ(y ) ∀x, y ∈ G,
(iv ) ϕ(x) = ϕ(x) ∀x ∈ G.
Equipped with the topology of uniform convergence on compact subsets of G, G b is a locally
compact space (see [4]). G b is the dual space of the hypergroup G.
We can notice that the function 1 : x 7−→ 1 belongs to G b.
In the following results, we give some proporties of the inverse Fourier transform.
∨ ∨ ∨
b ). Then β belongs to C ♮ (G) and (β ∗ ) = β .
Theorem 3.2. Let β ∈ Mb (G b
Proof. For x ∈ G, we have
∨ R
β (x) = G
b ϕ(x)dβ (ϕ)
R R
= b( G
G
ϕ(y )dδx (y ))dβ (ϕ)
R R
= b( G
G
ϕ(y )dδx (y ))dβ (ϕ)
R
= b δx (ϕ))dβ (ϕ)
G
b
= β (δbx )
The mapping x 7−→ δx of G on Mb (G) is continuous (see [5] lemma 2.2B). Since the Fourier
∨ ∨
transform and β are continuous, then β is continuous. β is also bounded, then β is bounded. Let
k1 , k2 ∈ K and x ∈ G, we have
∨ R
β (k1 ∗ x ∗ k2 ) = Gb ϕ(k1 ∗ x ∗ k2 )dβ (ϕ)
R
= Gb ϕ(x)dβ (ϕ)
∨
= β (x).
∨
Thus, β is K−invariant.
Moreover for x ∈ G, we have
∨
(β ∗ )(x) = β ∗ (δbx )
b − (ϕ)dβ (ϕ)
R
= b δx
G
R R
= G
b G ϕ− (y )dδx (y )dβ (ϕ)
R
= G
b ϕ(x)dβ (ϕ)
∨
= β (x).
12 Kouakou Germain BROU, Ibrahima TOURE and Kinvi KANGNI
R
= G
(µ ∗ f )(x)g (x)duG (x)
R
= G
(µ ∗ f )(x)g ∗ (x)duG (x)
R
= G//K
∗ f )(KxK )ge∗ (KxK )dm(KxK )
(µ]
R
= G//K
∗ f )(KxK )ge∗ (KxK )dm(KxK )
(µ]
R [
= \ (µ
G//K
∗ f )(ϕe)b
] ge(ϕe)de
π (ϕe), (see[5] 12.1.C )
R ]
= \ (µ
G//K
∗ f )(ϕe)e
[ gb(ϕe)de
π (ϕe)
R
= b (µ
G
∗ f )(ϕ)gb(ϕ)dπ (ϕ)
[
R R
= G
b b(ϕ)fb(ϕ)gb(ϕ)dπ (ϕ) =
µ G
b b(ϕ)b
µ h(ϕ)dπ (ϕ).
n o
Theorem 3.5. (i) fb; f ∈ K(G) is a sup-norm dense subspace of C0 (G
b ).
∨
(ii) Let β ∈ Mb (Gb ). If β = 0, then β = 0.
∨
(iii) Let β ∈ Mb (Gb ) and µ ∈ M ♮ (G). Then µ = βµG if and only if β = µ bπ.
b
n o n o
Proof. (i) We know that fb; f ∈ K(G) ⊂ C0 (G b ). Let suppose that fb; f ∈ K(G) ̸= C0 (G b ).
By the Hahn-Banach theorem, there exists ψ in C0 (G b )′ such that ψ ̸= 0 and ψ (fb) = 0 ∀f ∈
K(G). Using the Riesz’s representation theorem, we can take µ in Mb (G b ) such that µ ̸= 0 and
R
ψ (φ) = Gb φ(ϕ)dµ(ϕ) ∀φ ∈ C0 (G). b
R
Let us consider f ∈ K(G) with sptf ⊂ K such that G f (x)dµG (x) = 1.
We have R
0 = Gb fb(ϕ)dµ(ϕ)
R R
= G
b G ϕ(x)f (x)dµG (x)dµ(ϕ)
R R
= G
b K f (x)dµG (x)dµ(ϕ) since ϕ(x) = 1 ∀x ∈ K .
= µ(G
b ).
A Bochner theorem on a noncommutative hypergroup 13
n o
Thus, µ = 0 and we have a contradiction. So fb; f ∈ K(G) = C0 (G b ).
n o
In addition, since fb; f ∈ K(G) ⊂ {b µ; µ ∈ MµG (G)} ⊂ C0 (Gb ), then we can
deduce that {bµ; µ ∈ MµG (G)} is a sup-norm dense subspace of C0 (G b ).
(ii) Let us suppose that β ̸= 0 and take µ ∈ MµG (G) such that β (µ
b) ̸= 0.
We have R R R
0 ̸= Gb µ
b(ϕ)dβ (ϕ) = Gb G ϕ(x)dµ(x)dβ (ϕ)
∨
β (x)dµ− (x).
R
= G
∨
Thus, β ̸= 0 and we have a contradiction.
(iii) Using (i) and ([1] th.1.4.30), it is sufficient to consider the functions of the form h = f ∗ g ,
with f, g ∈ K♮ (G) to ensure our proof.
∨
h(x)d(βµG )− (x) = b b
R R
Let us note that G
h(ϕ)dβ (ϕ). Indeed we have
G
∨
h(x)d(βµG )− (x) =
R R R
G G
h(x) G
b ϕ(x)dβ (ϕ)dµG (x)
R R
= G
b G h(x)ϕ(x)dµG (x)dβ (ϕ)
R
= G
b h(ϕ)dβ (ϕ).
b
h(x)dµ− (x) =
R R
Since G G
b h(ϕ)d(µ
b bπ )(ϕ), then we have the result.
R
= − (ϕ)φ(ϕ)dπ (ϕ)
µc
G
b
− π (φ) ∀φ ∈ K(G
= µc b ).
In the following proposition, we bring out some properties of the strongly positive definite func-
tions and the positive definite measures.
Proposition 3.10. Let φ ∈ SP (G b ). Then the following statements hold.
∨
(i) If µ ∈ P M (G b ), then (φµ) ≥ 0.
∨ R
(ii) If µ ∈ Mb (G) such that µ is real values, then µ(φ) = Gb φ(ϕ)dµ(ϕ) ∈ R.
b
(iii) ∀ϕ ∈ Gb , φ(ϕ) = φ(ϕ); φ and Re(φ) belong to SP (G b ).
(iv) If ψ ∈ SP (G), then φψ and c1 φ + c2 ψ belong to SP (G
b b ) for c1 , c2 ≥ 0.
R R R
= G
b G ϕ(z1 )dδy (z1 ) G
ϕ(z2 )dδx (z2 )dµ(ϕ)
R R R
= G
b G ϕ(z1 )dδy (z1 ) G
ϕ(z2 )dδx (z2 )dµ(ϕ)
R R R
= G
b G G ϕ(z1 )ϕ(z2 )dδy (z1 )dδx (z2 )dµ(ϕ)
R R R R
= G
b G G K ϕ(z1 ∗ k ∗ z2 )dωK (k )dδy (z1 )dδx (z2 )dµ(ϕ)
R R R R R
= G
b K G G
( G ϕ(z1 ∗ z3 ∗ z2 )dδk (z3 ))dδy (z1 )dδx (z2 )dωK (k )dµ(ϕ)
R R R R
= G
b K G G
ϕ(z ∗ z2 )d(δk ∗ δy )(z ))dδx (z2 )dωK (k )dµ(ϕ)
R R R
= G
b K G
ϕ(t)d {(δk ∗ δy ) ∗ δx } (t)dωK (k )dµ(ϕ)
R R ∨
= K G
µ(t)d {(δk ∗ δy ) ∗ δx } (t)dωK (k ) ≥ 0 since µ ∈ M P (G
b ).
R
= G
b φ(ϕ)d(δbx µ)(ϕ)
∨ ∨ ∨
(ii) If µ ∈ Mb (G
b ), then ( µ δ1 + µ)∨ ≥ 0, where µ is the sup-norm of µ.
∞ ∞
A Bochner theorem on a noncommutative hypergroup 15
∨ ∨ ∨ ∨
In fact,( µ δ1 + µ)∨ (x) = µ + µ(x) ≥ 0. Thus, ( µ δ1 + µ)(φ) ≥ 0.
∞ ∞ ∞
∨ ∨ ∨
Since ( µ δ1 + µ)(φ) = µ φ(1) + µ(φ), then µ φ(1) + µ(φ) ≥ 0.
∞ ∞ ∞
Thus, µ(φ) ∈ R.
(iii) Let ϕ ∈ G
b , then ϕ ∈ G
b and
(δϕ + δϕ )∨ (x) =
R R
G
b α(x)dδϕ (α) + G
b α(x)dδϕ (α)
= ϕ(x) + ϕ(x)
= 2Re(ϕ(x)).
Thus, (δϕ + δϕ )(φ) = φ(ϕ) + φ(ϕ) ∈ R and we deduce that Im(φ(ϕ)) = −Im(φ(ϕ)).
We have also, ((δϕ −δϕ )/i)∨ = 2Im(ϕ). So φ(ϕ)−φ(ϕ) ∈ iR and we have Re(φ(ϕ)) =Re(φ(ϕ)).
It comes that φ(ϕ) = φ(ϕ).
Let µ ∈ P M (Gb ). Then µ∗ ∈ P M (G
b ) and
R
µ(φ) = Gb φ(α)dµ(α)
R
= G
b φ(α)dµ(α)
R
= G
b φ(α)dµ∗ (α)
= µ∗ (φ) ≥ 0,
R
= G
b φ(ϕ)d(ψµ)(ϕ)
= (ψµ)(φ) ≥ 0.
Thus, φψ ∈ SP (G b ).
Let c1 , c2 ≥ 0. For µ ∈ M P (G b ), we have
R R R
b (c1 φ + c2 ψ )(ϕ)dµ(ϕ) = c1 G
G b φ(ϕ)dµ(ϕ) + c2 Gb ψ (ϕ)dµ(ϕ)
≥ 0.
Thus, c1 φ + c2 ψ ∈ SP (G
b ).
We have the following result which is the main theorem of this paper.
Theorem 3.11. Let φ in SP (G b ). Then there exists a unique positive measure ν ∈ M ♮ (G) such
b
that φ|spt(π) = νb|spt(π) . Conversely, νb is strongly positive for each positive measure ν ∈ Mb (G).
Proof. If ν ∈ Mb (G) such that ν ≥ 0, then νb ∈ SP (G
b ). In fact, for µ ∈ P M (G
b ), we have
R R R
G
b(ϕ)dµ(ϕ) = Gb G ϕ(x)dν (x)dµ(ϕ)
bν
R R
= G G
b ϕ(x)dµ(ϕ)dν (x)
R ∨
= G
µ(x)dν (x) ≥ 0.
16 Kouakou Germain BROU, Ibrahima TOURE and Kinvi KANGNI
∨ ∨ ∨
b ) such that µ is real-valued. Since µ(x) ≤ µ
Let µ ∈ Mb (G , then
∞
∨ ∨ ∨ ∨ ∨ ∨ ∨ ∨
− µ δ1 (x) ≤ µ(x) ≤ µ δ1 (x)∀x ∈ G and so ( µ δ1 ± µ) ≥ 0. This implies that
∞ ∞ ∞
∨ R R ∨
µ φ(1) ± Gb φ(ϕ)dµ(ϕ) ≥ 0, so Gb φ(ϕ)dµ(ϕ) ≤ µ φ(1).
∞ ∞
∨ ∨ ∨ ∨
b ), let us put µ1 = 1 (µ + µ∗ ) and µ2 =
For µ ∈ Mb (G 1 ∗
We have µ1 = 12 (µ + µ) = Re(µ)
2 2i (µ−µ ).
∨ ∨ ∨ ∨ ∨ ∨
and µ2 = 21i (µ − µ) = Im(µ). So µ1 and µ2 are in Mb (G
b ) such that µ1 and µ2 are real valued.
Since µ = µ1 + iµ2 , then
R R R
G
b φ(ϕ)dµ(ϕ) ≤ G
b φ(ϕ)dµ1 (ϕ) + G
b φ(ϕ)dµ2 (ϕ)
∨ ∨
≤ µ1 φ(1) + µ2 φ(1)
∞ ∞
∨
≤ 2φ(1) µ .
∞
The mapping
b ))∨
η : (Mb (G −→ C
∨ ∨ R
µ 7−→ η (µ) = G
b φ(ϕ)dµ(ϕ)
So ∀h ∈ K(G
b ),
R ∨ R R
G
b φ(ϕ)h(ϕ)dπ (ϕ) = η0 (h) = G G
b ϕ(x)h(ϕ)dπ (ϕ)dν (x)
R R
= G
b G ϕ(x)dν (x)h(ϕ)dπ (ϕ)
R
= − (ϕ)h(ϕ)dπ (ϕ).
νc
G
b
−
φ|spt(π) = νc [− ♮
|spt(π ) = (ν ) |spt(π ) .
Furthermore, ν is a positive measure. In fact let f ∈ C0 (G) such that f ≥ 0 and ε > 0. Since
∨
(K(Gb ))∨ is dense in C0 (G), then there exists h ∈ K(Gb ) such that f − h < ε. We may
∞
∨
assume that h is real valued. Define µ = εδ1 + hπ . µ ∈ Mb (G
b ) and
∨ ∨
µ(x) = ε + h(x) > f (x) ≥ 0 ∀x ∈ G.
∨ ∨ R
Thus, µ ≥ 0 and η (µ) = G
b φ(ϕ)dµ(ϕ) ≥ 0.
A Bochner theorem on a noncommutative hypergroup 17
In addition we have
R ∨ R ∨ R ∨
G
f (x)dν (x) − η (µ) = G
f (x)dν (x) − εη (δ1 ) − G
hdν (x)
∨ R ∨
≤ εη (δ1 ) + G
(f − h)(x) dν (x)
∨
≤ 2εη (δ1 )
≤ 4εφ(1).
R ∨
Thus, G
f (x)dν (x) ≥ η (µ) − 4εφ(1) and taking ε so small that we want, we can conclude that
f (x)dν (x) ≥ 0. So ν and (ν − )♮ are positive. Finally, the uniqueness follows from Theorem
R
G
3.5.
In this paper, we have proved a Bochner theorem for a noncommutative hypergoup G admitting
a compact subhypergroup K such that (G, K ) is a Gelfand pair. In fact, we have obtained a
one to one correspondence between the space of strong positive definite functions on G
b and the
space of K -invariant and positive measures on G. Our results generalize those established on
commutative locally compact hypergroups for in this case, (G, {e}) is a Gelfand pair with e the
neutral element of G.
References
[1] W. R. Bloom and H. Heyer, Harmonic analysis of probability measures on hypergroups. de Gruyter Stud-
ies in Mathematics. 20. Berlin: de Gruyter. vi, 601 p. 1995.
[2] C. F. Dunkl, The Measure Algebra of a Locally Compact Hypergroup, Trans. Amer. Math. Soc., Volume
179, 1973.
[3] B. K. Germain and K. Kangni, On Gelfand pair over Hypergroups, Far East J. Math, Vol. 132, Number 1,
63-76, 2021.
[4] B. K. Germain, I. Toure, K. Kangni, A Plancherel Theorem On a Noncommutatiove Hypergroup, Int. J.
Anal. Appl., 20 (2022), 32.
[5] R. I. Jewett, Spaces with an Abstarct Convolution of Measure, Adv. Math., Vol. 18, 1-101 1975.
[6] R. Lasser, Bochner Theorems for Hypergroups and Their Applications to Orthogonal Polynomial Expan-
sions, Journal Of Approximation Theory 37, 3 1 l-325 (1983)
[7] L. Nachbin, On the finite dimensionality of every irreducible representation of a group, Proc. Amer, Math.
Soc. 12, 11-12, 1961.
[8] K. A. Ross, Centers of Hypergroups, Trans. Amer. Math. Soc., Vol. 243, Septembre 1978.
[9] R. Spector, Mesures invariantes sur les Hypergroupes, Trans. Amer. Math. Soc., Vol. 239, 1978.
[10] L. Szekelyhidi, Spherical Spectral Synthetis On Hypergroups, Acta Math. Hungar.,
https://doi.org/10.1007/ s10474-020-01068-9 (2020)
[11] K. Vatti, Gelfand Pairs over Hypergroups Joins, Acta Math. Hungar., May 2019, Doi: 10.1007/s 10474-
019-00946-1.
Author information
Kouakou Germain BROU, Univerité de Man, Côte d’Ivoire.
E-mail: [email protected]
Ibrahima TOURE, UFR Mathématiques et Informatique, Université Félix Houphouët-Boigny, Côte d’Ivoire.
E-mail: [email protected]
Kinvi KANGNI, UFR Mathématiques et Informatique, Université Félix Houphouët-Boigny, Côte d’Ivoire.
E-mail: [email protected]
Received: 2023-01-11
Accepted: 2023-08-30