A SCALAR VALUED FOURIER TRANSFORM

FOR THE HEISENBERG GROUP

SUNDARAM THANGAVELU
arXiv:2206.00883v1 [math.FA] 2 Jun 2022

Abstract. We define a scalar valued Fourier transform for functions on the Heisenberg
group and establish some of its basic properties like inversion formula, Plancherel theorem
and Riemann-Lebesgue lemma. We also restate certain well known theorems for the group
Fourier transform in terms of the new transform which we would like to call Strichartz
Fourier transform.

Dedicated to the memory of Robert Strichartz

I think it was during the summer of 1987 when my friend Der-Chen Chang brought a preprint
of Strichartz to my attention. In this work, which was published in 1989 [17], Strichartz
developed harmonic analysis as spectral theory of Laplacians. In the context of the Heisenberg
group Hn , the relevant operator was the sublaplacian L, and Strichartz developed the joint
spectral theory of L and T. I read this paper with great zeal and learnt for the first time
about special Hermite functions and their importance in the harmonic analysis on Hn . This
point of view was further developed in another influential paper [18] and both works have
played a vital role in all my works related to the Heisenberg group. During the academic
year 1991-92, as a visitor to Cornell University, I was fortunate to have the opportunity to
discuss mathematics with Strichartz. Needless to say, I greatly benefitted by interacting with
him and I am eternally grateful for his kindness and the genuine interest he showed in my
works. As one can see, this article draws quite a lot from the two papers mentioned above.

1. Introduction
Fourier transform of a function f on a locally compact Lie group G is defined as a function
on the unitary dual Gb consisting of the equivalence classes of irreducible unitary represen-
tations of G. Given a representative π of an element from G b we define
Z
fˆ(π) = f (g)π(g)dg (1.1)
G

where dg is the left invariant Haar measure on G. It is also customary to use the notation
π(f ) instead of fˆ(π) especially when the group is non-abelian. Note that for f ∈ L1 (G),
fˆ(π) is a bounded linear operator on the Hilbert space Hπ on which the representation
π is realised. When G is abelian, all the irreducible unitary representations of G are one
dimensional and hence we can think of them as homomorphisms of G into the circle group

2010 Mathematics Subject Classification. Primary: 43A85, 42C05. Secondary: 33C45, 35P10.
Key words and phrases. Heisenberg group, unitary representations, Fourier transform, Gelfand transform,
spectral theory, special Hermite expansions, Hecke-Bochner formula, Hardy and Ingham theorems.
1
2 S. THANGAVELU

S 1 consisting of all complex numbers of absolute value one. They are known as unitary
b the Fourier transform
characters and for any given χ ∈ G
Z
fˆ(χ) = f (g)χ(g)dg (1.2)
G
b into a multiplicative group and hence fˆ becomes a function
is a scalar. We can also turn G
on this dual group. This allows us to treat Fourier transform on locally compact abelian
groups on par with the well understood Euclidean Fourier transform on Rn .

When the group G is non abelian, then it is no longer true that all the elements of G b
are one dimensional. If the group is compact, then G b consists only of finite dimensional
representations and hence Fourier transform becomes a matrix valued function which is still
manageable. When the group is non compact and non abelian, we need to deal with infinite
dimensional representations and hence the Fourier transform becomes operator-valued. As
we are mainly interested in studying the Fourier transform on the Heisenberg group, let us
leave the generality and specialise to the case in hand.

The simplest example of a non abelian nilpotent Lie group is provided by the Heisenberg
group Hn whose underlying manifold is Cn × R. The representation theory of Hn is very
simple and its unitary dual can be described explicitly. There are certain one dimensional
representations of Hn which are characters, parametrised by Cn , but we discard them as
they do not play any role in the Plancherel theorem. We let H cn to stand for the set of all
infinite dimensional irreducible unitary representations of H . By a theorem of Stone and
n

von Neumann, the members of H cn are parametrised by non zero real numbers R∗ . Thus to
each λ ∈ R∗ we have an irreducible unitary representation πλ realised on the same Hilbert
space L2 (Rn ). Thus the Fourier transform of f ∈ L1 (Hn ) is the operator valued function
Z
ˆ
f (λ) = f (g)πλ(g)dg
Hn
defined on R . This operator valued Fourier transform satisfies all the standard properties
∗

of a Fourier transform. Hence for suitable functions we have an inversion formula and for
f ∈ L1 ∩ L2 (Hn ) we have the Plancherel formula:
Z Z ∞
2
|f (g)| dg = (2π) −n−1
kfˆ(λ)k2HS |λ|n dλ. (1.3)
Hn −∞

This allows us to extend the Fourier transform to the whole of L2 (Hn ) as a unitary operator
onto L2 (R∗ , S2 , dµ), the L2 space of Hilbert-Schmidt operator valued functions on R∗ taken
with respect to the measure dµ(λ) = (2π)−n−1 |λ|n dλ.

Though the Fourier transform defined above shares many features with its abelian coun-
terparts, it is unwieldy and not suitable for studying several standard problems in harmonic
analysis. For example, there are no simple descriptions of the images either of the Schwartz
space S(Hn ) or C0∞ (Hn ) under the Fourier transform. Consequently, it is not clear how to
define Fourier transforms of distributions on the Heisenberg group. Though there are ver-
sions of Paley-Wiener theorems for the so called Fourier-Weyl transform, the results are not
very satisfactory, see [22], [23].
 FOURIER TRANSFORM ON THE HEISENBERG GROUP 3

Recently the authors of the paper [6] have attempted with certain degree of success to de-
fine a scalar valued Fourier transform on Hn . The idea is very simple: as fˆ(λ) for f ∈ L1 (Hn )
is a bounded linear operator on L2 (Rn ), it is natural to fix an orthonormal basis Φλα , α ∈ Nn
for L2 (Rn ) and consider the map f → hfˆ(λ)Φλα , Φλβ i as a candidate for a scalar valued Fourier
transform. In [6] the authors have introduced a metric d on the set H e n = Nn × Nn × R∗ so
that the above map is continuous from L1 (Hn ) into H b n , the completion of H
e n in the metric d.

In this article we propose the following definition of a scalar valued Fourier transform on
Hn that shares several properties with the Helgason Fourier transform on noncompact rank
one Riemannian symmetric spaces. In order to define this Fourier transform, which we call
Strichartz Fourier transform, we need to set up some notations. For any δ ≥ −1/2 we let
Lδk (t), k ∈ N, t ≥ 0 stand for the Laguerre polynomials of type δ. For any λ ∈ R∗ , we define
the Laguerre functions by
n−1 1 1 2
ϕk,λ (z) = Lkn−1 ( |λ||z|2 )e− 4 |λ||z| , z ∈ Cn .
2
n−1 iλt n−1 ∂
Then ek,λ (z, t) = e ϕk,λ (z) are joint eigenfunctions of the sublaplacian L and T = ∂t :
n−1 n−1 n−1 n−1
Lek,λ (z, t) = (2k + n)|λ| ek,λ (z, t), −iT ek,λ (z, t) = λ ek,λ (z, t).
Let Ω stand for the Heisenberg fan which is the union of the rays Rk = {(λ, τ ) ∈ R2 :
τ = (2k + n)|λ|} for k = 0, 1, 2, ... and the limiting ray R∞ = {(0, τ ) : τ ≥ 0}. For
n−1
each a = (λ, (2k + n)|λ|) ∈ Rk we use the notation ea (z, t) in place of ek,λ (z, t). For any
2 n b
f ∈ L ∩ L (H ), we define its Strichartz Fourier transform f (a, z) on Ω × C as follows. For
1 n

a ∈ Rk , z ∈ Cn , λ 6= 0 we define
Z
b
f (a, z) = f (w, s)ea ((w, s)−1(z, 0))dw ds. (1.4)
Hn
Here (w, s) = (−w, −s) and the group law in Hn is given in Section 2. For a = (0, τ )
−1

coming from the limiting ray R∞ we set

Z √
J ( τ |z − w|)
fb(0, τ, z) = (n − 1)!2 n−1 n−1
f (w, s) √ dw ds (1.5)
Hn ( τ |z − w|)n−1
where Jn−1 is the Bessel function of order (n − 1). As a subset of R2 , Ω inherits the Euclidean
metric and topology. We define the normalised Strichartz Fourier transform fe(a, z) by
k!(n − 1)! b
fe(a, z) = f (a, z), a ∈ Rk , fe(0, τ, z) = fb(0, τ, z).
(k + n − 1)!
For this transform we can prove the following analogue of Riemann-Lebesgue lemma.

Theorem 1.1. For any f ∈ L1 (Hn ), the normalised Strichartz Fourier transform fe(a, z) is
uniformly continuous on Ω for any z ∈ Cn fixed. Moreover, fe(a, z) vanishes at infinity, i.e.
fe(a, z) → 0 as |a| → ∞ for each z ∈ Cn fixed.
On the Heisenberg fan Ω we consider the measure ν which is defined by
Z Z ∞ X∞ 
−2n−1
ϕ(a)dν(a) = (2π) ϕ(λ, (2k + n)|λ|) |λ|2n dλ.
Ω −∞ k=0
4 S. THANGAVELU

We can now state the inversion formula and the Plancherel theorem for our Strichartz Fourier
transform.
Theorem 1.2. For any Schwartz class function f on Hn we have the following inversion
formula for the Strichartz Fourier transform:
Z Z
f (z, t) = fb(a, w)ea ((−w, 0)(z, t))dw dν(a). (1.6)
Ω Cn

Moreover, for any f ∈ L1 ∩ L2 (Hn ) we have the Plancherel formula

Z Z Z
2
|f (z, t)| dz dt = |fb(a, w)|2 dw dν(a). (1.7)
Hn Ω Cn

It is not true that the Strichartz Fourier transform takes L2 (Hn ) onto L2 (Ω × Cn , dν dw).
This is because fb(a, z) has to satisfy a necessary condition (see (2.12)) which we write as
(2π)−n |λ|n ϕk,λ
n−1
∗λ fb(a, ·)(z) = fb(a, z), a ∈ Rk .
If we let L20 (Ω×Cn , dν dw) stand for the subspace of L2 (Ω×Cn , dν dw) consisting of functions
F (a, z) satisfying the above condition, we can prove the following result.
Theorem 1.3. The Strichartz Fourier transform initially defined on L1 ∩ L2 (Hn ) can be
extended as a unitary operator from L2 (Hn ) onto L20 (Ω × Cn , dν dw).
Remark 1.4. For f ∈ L2 (Hn ) the function f λ is defined only for almost every λ ∈ R∗ and
hence we may not be able to define fb(a, z) at every a ∈ Ω. In particular, fb(0, τ, z) need not
be defined. As the limiting ray R∞ has zero Plancherel measure, this does not create any
problem.
n−1
As a consequence of the estimates satisfied by the Laguerre functions ϕk,λ (z) and the
1 n
Bessel function Jn−1 (t), the Strichartz Fourier transform of any f ∈ L (H ) satisfies the
estimate
sup |fe(a, z)| ≤ cn,1 kf k1 .
(a,z)∈Ω×Cn

Restating the Plancherel theorem theorem in terms of fe and interpolating with change of
measures we can prove the following Hausdorff-Young inequality.
Theorem 1.5. For 1 ≤ p ≤ 2 there is a measure νp on Ω such that for all f ∈ Lp (Hn ) we
have
Z Z 1/p′ Z 1/p
e p′
|f (a, w)| dw dνp (a) ≤ cn,p |f (z, t)|p dz dt .
Ω Cn Hn

Let us compare the definition (1.4) with the Euclidean Fourier transform and the Helgason
Fourier transform on noncompact rank one Riemannian symmetric spaces X = G/K. For
functions f on Rn we can write the Fourier transform in the form
Z
b
f (λ, ω) = (2π) −n/2
f (x)e−iλx·ω dx, λ ∈ R, ω ∈ S n−1 . (1.8)
Rn
 FOURIER TRANSFORM ON THE HEISENBERG GROUP 5

Note that the functions eλ,ω (x) = e−iλx·ω are eigenfunctions of the Laplacian ∆ and Rn and
the inversion formula which reads as
Z ∞Z
f (x) = (2π) −n/2
fb(λ, ω)eλ,ω (x)|λ|n−1 dω dλ (1.9)
−∞ S n−1

expresses f as a superposition of the eigenfunctions eλ,ω . As ea (z, t) are eigenfunctions of L

the similarity between (1.6) and (1.9) is clear. It is even more illuminating to compare our
Fourier transform with the Helgason Fourier transform.

Let X = G/K be a rank one symmetric space of noncompact type. The Helgason Fourier
transform fe(λ, b) of a function f on G/K is given by
Z
e
f (λ, b) = f (x)e(−iλ+ρ)A(x,b) dx (1.10)
G/K

where λ ∈ C and b ∈ K/M, see Helgason [11, 12] for the notations. The inversion formula
is given by Z ∞Z
f (x) = c fe(λ, b)e(iλ+ρ)A(x,b) |c(λ)|−2db dλ (1.11)
−∞ K/M

where db is the normalised measure on K/M and c(λ) is the Harish-Chandra c-function.
Here again, the functions eλ,b (x) = e(iλ+ρ)A(x,b) are eigenfunctions of the Laplace-Beltrami
operator ∆X on the symmetric space X bringing out the analogy between (1.4) and (1.10).

Since functions f on G/K can be considered as right K-invariant functions on the Lie
group G we can consider the group Fourier transform of f. For any λ ∈ C if we let πλ
stand for the spherical principal series representations realized on L2 (K/M), each of these
representations has a unique K-fixed vector Y0 ∈ L2 (K/M) which is the constant function
Y0 (b) = 1. The Helgason Fourier transform of f is then related to the group Fourier transform
via the equation Z
fe(λ, b) = πλ (f )Y0 (b), πλ (f ) = f (g)πλ(g)dg. (1.12)
G
We will show that there is a similar relation in the case of Strichartz Fourier transform. The
Heisenberg group Hn can be considered as a subgroup of a bigger group Gn := Hn ⋉ U(n)
known as the Heisenberg motion group, so that functions on Hn are precisely the right
U(n)-invariant functions on Gn . For each a = (λ, (2k + n)|λ|) ∈ Rk there is a class one
representation ρλk of Gn so that
Z
fb(a, z) = ρk (f )ea (z, 0), ρk (f ) =
λ λ
f (g)ρλk (g)dg. (1.13)
Gn

We will demonstrate that the Strichartz Fourier transform shares several other properties
with the Helgason Fourier transform.

We conclude this introduction with a brief description of the plan of the paper. In Section
2, after recalling the representation theory of Hn and the Heisenberg group motion group
and the spectral theory of the sublaplacian, we define the Strichartz Fourier transform and
prove all its basic properties. In Section 3 we explore the connection between the Strichartz
Fourier transform and the Gelfand transform on the commutative Banach algebra consisting
6 S. THANGAVELU

of L1 (Hn ) functions which are radial in the z variable. We prove an analogue of Hecke-
Bochner formula for the Strichartz Fourier transform and use it to characterize the image
of the Schwartz space under the Strichartz Fourier transform. In Section 4 we describe
the connection between the operator valued group Fourier transform and the scalar valued
Strichartz Fourier transform and restate the well known Hardy and Ingham theorems which
are originally stated in terms of the former, in terms of the new transform.

2. Fourier transforms on the Heisenberg group

2.1. Schrödinger representations and the group Fourier transform. Let Hn := Cn ×
R be the (2n + 1)- dimensional Heisenberg group with the group law

1

(z, t)(w, s) := z + w, t + s + Im(z.w̄) , ∀(z, t), (w, s) ∈ Hn .
2
This is a step two nilpotent Lie group where the Lebesgue measure dzdt on Cn × R serves as
the Haar measure. The representation theory of Hn is well-studied in the literature, see the
monographs [8], [20] and [23]. In order to define Fourier transform, we use the Schrödinger
representations as described below.
For each non zero real number λ we have an infinite dimensional representation πλ realised
on the Hilbert space L2 (Rn ). These are explicitly given by
1
πλ (z, t)ϕ(ξ) = eiλt eiλ(x·ξ+ 2 x·y) ϕ(ξ + y),
where z = x + iy and ϕ ∈ L2 (Rn ). These representations are known to be unitary and
irreducible. Moreover, by a theorem of Stone and Von-Neumann, (see e.g., [8]) up to unitary
equivalence these account for all the infinite dimensional irreducible unitary representations
of Hn which act as eiλt I on the center. Also there is another class of finite dimensional
irreducible representations. As they do not contribute to the Plancherel measure we will not
describe them here.

The Fourier transform of a function f ∈ L1 (Hn ) is the operator valued function on the set
of all nonzero reals R∗ given by
Z
ˆ
f (λ) = f (z, t)πλ (z, t)dzdt.
Hn

Note that fˆ(λ) is a bounded linear operator on L2 (Rn ). It is known that when f ∈ L1 ∩L2 (Hn )
its Fourier transform is actually a Hilbert-Schmidt operator and one has
Z Z ∞
2
|f (z, t)| dzdt = (2π) −(n+1)
kfb(λ)k2HS |λ|n dλ
Hn −∞

where k.kHS denotes the Hilbert-Schmidt norm. The above allows us to extend the Fourier
transform as a unitary operator between L2 (Hn ) and the Hilbert space of Hilbert-Schmidt
operator valued functions on R which are square integrable with respect to the Plancherel
measure dµ(λ) = (2π)−n−1|λ|n dλ.

Let S2 stand for the Hilbert space of Hilbert-Schmidt operators on L2 (Rn ) equipped with
the inner product (T, S) = tr(S ∗ T ). We then have the following Plancherel theorem:
Theorem 2.1. The group Fourier transform is a unitary operator from L2 (Hn ) onto L2 (R∗ , S2 , dµ).
 FOURIER TRANSFORM ON THE HEISENBERG GROUP 7

By polarizing the Plancherel formula we obtain the Parseval identity: for f, g ∈ L2 (Hn )
Z Z ∞
f (z, t)g(z, t)dzdt = tr(fb(λ)b
g (λ)∗ ) dµ(λ).
Hn −∞
Also for suitable functions f on Hn we have the following inversion formula
Z ∞
f (z, t) = tr(πλ (z, t)∗ fb(λ))dµ(λ).
−∞
R
Now from the definition of πλ it is easy to see that fb(λ) = Cn f λ (z)πλ (z, 0)dz where f λ
stands for the inverse Fourier transform of f in the central variable:
Z ∞
λ
f (z) := eiλt f (z, t)dt.
−∞
1 n
This suggests that for any g ∈ L (C ), we consider the following operator valued function
Z
Wλ (g) := g(z)πλ (z, 0)dz.
Cn

With these notations we note that fˆ(λ) = Wλ (f λ ). These transforms are called the Weyl
transforms. We have the following Plancherel formula for Weyl transform (See [24, (2.2.9),
p. 49])
kWλ (g)k2HS |λ|n = (2π)n kgk22, g ∈ L2 (Cn ). (2.1)
Many results for the group Fourier transform are proved by studying the analogues for the
Weyl transform.
2.2. Joint spectral theory of L and T . The Heisenberg Lie algebra hn has a basis con-
sisting of the left invariant vector fields
∂ 1 ∂ ∂ 1 ∂ ∂
Xj = + y j , Xj = − xj , T = , j = 1, 2, ..., n.
∂xj 2 ∂t ∂yj 2 ∂t ∂t
Pn

The sublaplacian L = − j=1 Xj2 + Yj2 plays the role of the Laplacian on Hn . As L
commutes with T there is a well defined joint spectral theory of these two operators. The
n−1
functions ek,λ (z, t) in the introduction are joint eigenfunctions of L and T. The operators Lλ
defined by the relation L(f (z)eiλt ) = eiλt Lλ f (z) are known as a special Hermite operators
and they have explicit spectral decomposition which we describe now.

The convolution on Hn gives rise to the so called λ-twisted convolutions by the relation
Z
i
λ
(f ∗ g) (z) = f λ (z − w)g λ(w)e 2 λ Im(z·w̄) dw =: f λ ∗λ g λ (z).
Cn
n−1
The Laguerre functions ϕk,λ are eigenfunctions of Lλ with eigenvalues (2k + n)|λ| and every
g ∈ L2 (Cn ) has the L2 convergent expansion
X∞
−n n n−1
g(z) = (2π) |λ| g ∗λ ϕk,λ (z). (2.2)
k=0

The eigenspaces of Lλ corresponding to the eigenvalues (2k + n)|λ| are infinite dimensional
and an orthonormal basis is provided by
Φλα,β (z) = (2π)−n/2 (πλ (z, 0)Φλα , Φλβ ), α, β ∈ Nn , |α| = k.
8 S. THANGAVELU

Here Φλα are the scaled Hermite functions which are eigenfunctions of the scaled Hermite op-
erator H(λ) = −∆ + λ2 |x|2 with eigenvalues (2k + n)|λ|. We refer to [24] for the details. The
functions Φλα,β , known as special Hermite functions, form an orthonormal basis for L2 (Cn )
and (2.2) is the compact form of the special Hermite expansion of a function g.
n−1
The projections (2π)−n |λ|n g ∗λ ϕk,λ are mutually orthogonal and the Plancherel theorem
associated to the expansion (2.2) reads as
Z ∞ Z
X
2 −2n 2n n−1
|g(z)| dz = (2π) |λ| |g ∗λ ϕk,λ (z)|2 dz. (2.3)
Cn k=0 Cn

Weyl transform converts the twisted convolution into products and we have the formula
n−1
(2π)−n |λ|n Wλ (g ∗λ ϕk,λ ) = Wλ (g)Pk (λ). (2.4)
Here Pk (λ) is the orthogonal projection of L2 (Rn ) onto the k-th eigenspace of the Hermite
operator H(λ) spanned by Φλα , |α| = k.

2.3. Heisenberg motion group and some class one representations. Let U(n) denote
the group of all unitary matrices of order n. This acts on Hn by the automorphisms
σ.(z, t) = (σz, t), σ ∈ U(n).
We consider the semi-direct product of Hn and U(n), Gn := Hn ⋉ U(n) which acts on Hn by
1

(z, t, σ).(w, s) = z + σw, t + s + Im(z · σw)
2
whence the group law in Gn is given by
1

(z, t, σ).(w, s, τ ) = z + σw, t + s +Im(z · σw), στ .
2
The group Gn is called the Heisenberg motion group which contains Hn and U(n) as sub-
groups. Also Hn can be identified with the quotient group Gn /U(n). As a matter of fact,
functions on Hn can be viewed as right U(n) invariant functions on Gn . The Haar measure
on Gn is given by dσ dz dt where dσ denotes the normalised Haar measure on U(n). To
bring out the connection between the group Fourier transform on Hn and the Heisenberg
motion group, we need to describe a family of class one representations of Gn . We start with
recalling the definition of such representations.

Let G be a locally compact topological group and K be a compact subgroup of G. Suppose

π is a representation of G realised on the Hilbert space H. Let HK denote the set of all
K-fixed vectors given by
HK := {v ∈ H : π(k)v = v, ∀k ∈ K}.
It can be easily checked that HK is a subspace of H. We say that π is a class one repre-
sentation of the pair (G, K) if HK 6= {0}. Moreover, when (G, K) is a Gelfand pair it is
well-known that dim HK = 1. In the following, we describe certain family of class one rep-
resentation for the Gelfand pair (Gn , U(n)). For that we need to set up some more notations.
 FOURIER TRANSFORM ON THE HEISENBERG GROUP 9
p
For α ∈ Nn and λ 6= 0 let Φλα (x) := |λ|n/4Φα ( |λ|x), x ∈ Rn where Φα denote the
normalised Hermite functions on Rn . We know that for each λ 6= 0, {Φλα : α ∈ Nn } forms an
orthonormal basis for L2 (Rn ). Suppose
λ
Eα,β (z, t) := (πλ (z, t)Φλα , Φλβ ), (z, t) ∈ Hn
denotes the matrix coefficients of the Schrödinger representation πλ of Hn . For each λ = 6 0
and k ∈ N, we consider the Hilbert space Hkλ spanned by {Eα,β λ
: α, β ∈ Nn , |β| = k} and
equipped with the inner product
Z
−n n
(f, g)Hλk := (2π) |λ| f (z, 0)g(z, 0)dz.
Cn

We define a representation ρλk of Gn realised on Hkλ by the prescription

ρλk (z, t, σ)ϕ(w, s) := ϕ((z, t, σ)−1 (w, s)), (w, s) ∈ Hn .
It is well-known that ρλk is an irreducible unitary representation of Gn for all λ 6= 0 and
n−1
k ∈ N. Also for λ 6= 0 and k ∈ N we consider the function ek,λ on Hn defined by
X
n−1
ek,λ (z, t) = (πλ (z, t)Φλα , Φλα ).
|α|=k

It is known that the above function can be expressed in terms of Laguerre functions as follows
(See [24, p. 52])
n−1 n−1
ek,λ (z, t) = eiλt ϕk,λ (z).
n−1
It can be checked that ek,λ is a U(n)-fixed vector corresponding to the representation ρλk and
hence ρλk is a class one representation of the pair (Gn , U(n)). Moreover, (Gn , U(n)) being a
n−1
Gelfand pair, ek,λ is unique up to scalar multiple.

The representations ρλk when restricted to Hn are not irreducible but split into finitely
many irreducible unitary representations each one being equivalent to πλ . Given f ∈ L1 (Hn ),
considering it as an U(n)-invariant function on Gn , we associate an operator valued function
ρλk (f ) acting on Hkλ defined by
Z
λ
ρk (f ) := f (z, t)ρλk (z, t, σ)dσ dz dt.
Gn

Now since ρλk is unitary, it can be easily checked that ρλk (f ) is a bounded operator and the
operator norm is bounded above by kf k1 . From the definition of ρλk the following can be
easily checked:
ρλk (f )ek,λ
n−1
(z, t) = eiλt f −λ ∗−λ ϕk,λ n−1
(z). (2.5)
This leads to the Plancherel formula, proved in [13, Proposition 2.1], for the representations
ρλk . Z
(k + n − 1) λ 2 −n n n−1
kρ (f )kHS = (2π) |λ| |f −λ ∗−λ ϕk,λ (z)|2 dz. (2.6)
k!(n − 1)! k Cn
n−1
It is easy to see that ρλk (f )ek,λ (z, t) are eigenfunctions of the sublaplacian L with eigenvalues
(2k + n)|λ| and f can be recovered by the formula
Z ∞ X∞
−n−1 iλt


f (z, t) = (2π) e ρλk (f )ek,λ
n−1
(z, 0) |λ|n dλ. (2.7)
−∞ k=0
10 S. THANGAVELU

This is a consequence of the special Hermite expansion (2.2) applied to f λ . We can thus view
the above as the spectral decomposition of the sublaplacian.

2.4. Strichartz Fourier transform on the Heisenberg group. We propose the following
definition as a scalar valued Fourier transform for functions on the Heisenberg group. Let Ω
stand for the Heisenberg fan which is the union of the rays Rk = {(λ, τ ) ∈ R2 : τ = (2k +
n)|λ|} for k = 0, 1, 2, ... and the limiting ray R∞ = {(0, τ ) : τ ≥ 0}. For any f ∈ L1 ∩ L2 (Hn ),
we define its Strichartz Fourier transform fb(a, z) for a ∈ Rk , z ∈ Cn by the relation
fb(a, z) = f −λ ∗−λ ϕn−1 (z), a = (λ, (2k + n)|λ|).
k,λ
n−1 n−1
In view of the relation (2.5), namely ρλk (f )ek,λ (z, t) = eiλt f −λ ∗−λ ϕk,λ (z) we see that
fb(a, z) = e−iλt ρλk (f )ek,λ
n−1
(z, t) = ρλk (f )ek,λ
n−1
(z, 0).
For a = (0, τ ) coming from the limiting ray R∞ we set
Z √
b n−1 0 Jn−1( τ |w|)
f (0, τ, z) = (n − 1)!2 f (z − w) √ dw
Cn ( τ |w|)n−1
R
where Jn−1 is the Bessel function of order (n−1). At (0, 0, z) we let fb(0, 0, z) = Hn f (w, t)dwdt.
As a subset of R2 , Ω inherits the Euclidean metric and topology.

We define the normalised Strichartz Fourier transform by fe(a, z) = (k+n−1)! k!(n−1)! b

f (a, z) for
a ∈ Rk . On the limiting ray we simply set fe(0, τ, z) = fb(0, τ, z). It then follows that fe is a
continuous function on Ω for f ∈ L1 (Hn ).

Proof of Theorem 1.1: The only nontrivial part which requires a proof is the continuity
at (0, τ, z) when a sequence aj ∈ Ω converges to (0, τ ) running through different rays Rk .
Thus we have (λj , (2kj + n)|λj |) with λj → 0 and (2kj + n)|λj | → τ. We need to show that
Z √
kj !(n − 1)! −λj n−1 n−1 0 Jn−1( τ |w|)
lim f ∗−λj ϕkj ,λj (z) = (n − 1)!2 f (z − w) √ dw.
j→∞ (kj + n − 1)! Cn ( τ |w|)n−1
n−1
This can be proved by using asymptotic properties of the Laguerre functions ϕk,λ (z). From
[19, (8.22.4), p.193] we have
p
n−1 J n−1 ( (2k + n)|λ| |z|) p
ϕk,λ (z) = (n − 1)!2n−1 p + mk ( |λ||z|) (2.8)
( (2k + n)|λ| |z|)n−1
where the error term satisfies the uniform estimates
|mk (t)| ≤ C (2k + n)−(n−1)/2−3/4 , 0 < t ≤ b. (2.9)
Thus we see that fe(aj , z) is a sum of two terms of which the main term is given by
Z p
n−1 λj Jn−1 ( (2kj + n)|λj | |w|) i λj Im(z·w̄)
(n − 1)!2 f (z − w) p e2 dw.
Cn ( (2kj + n)|λj | |w|)n−1
As λj → 0 and (2kj + n)|λj | → τ, it is clear that the above converges to
Z √
n−1 0 Jn−1 ( τ |w|)
(n − 1)!2 f (z − w) √ dw.
Cn ( τ |w|)n−1
 FOURIER TRANSFORM ON THE HEISENBERG GROUP 11
p
As λj remains bounded, the estimates on mk ( |λ| |z − w|) shows that the error term goes
to zero proving our claim.

In order to prove that fe(a, z) vanishes at infinity, we first observe that the Strichartz
Fourier transform of any f ∈ L1 (Hn ) satisfies the estimate
sup |fe(a, z)| ≤ cn,1 kf k1 . (2.10)
(a,z)∈Ω×Cn

n−1
This is a consequence of the well known fact that the Laguerre functions ϕk,λ (z) and the
Bessel function Jn−1 (t) satisfy the uniform estimates (see [19, (8.22.4) p. 193, (1.71.6) p. 15]
)
k!(n − 1)!
|ϕn−1 (z)| ≤ 1, |Jn−1 (t)| ≤ cn tn−1 . (2.11)
(k + n − 1)! k,λ
Therefore, it is enough to prove that fe(a, z) vanishes at infinity whenever f is compactly
supported. The case a = (0, τ ) → ∞ is easy to handle. In this case
Z √
b n−1 Jn−1( τ |w|)
f (0, τ, z) = (n − 1)! 2 f (z − w) √ dw
Cn ( τ |w|)n−1
is a constant multiple of the Hankel transform of the radial function
Z
fz (w) = f (z − σw)dσ.
U (n)

Therefore, by appealing to Riemann-Lebesgue lemma for Hankel transforms we can conclude

that fb(0, τ, z) vanishes at infinity.
Consider the case when a = (λ, (2k + n)|λ|) ∈ Rk goes to infinity. In case |λ| → ∞ we
have Z
k!(n − 1)! b
|f (a, z)| ≤ |f λ (z)| dz
(k + n − 1) C n

in view of (2.11). The above certainly vanishes as |λ| → ∞ in view of Riemann-Lebesgue

lemma for the Euclidean Fourier transform. In case (2kj + n)|λj → ∞ but λj remains
bounded, the main term in fe(aj , z) is bounded by a contant times
Z p
|J n−1 ( (2kj + n)|λj | |w|)|
|f λj (z − w)| p dw.
Cn ( (2kj + n)|λj | |w|)n−1
Since |f λj (w)| ≤ |f 0(w)| is integrable, the estimate |Jn−1 (t)| ≤ c tn−1 (1 + t)−1/2 shows that
the main term goes to zero as j goes to infinity. The same is true of the error term in view
of (2.9) since f 0 (w) is compactly supported. This completes the proof.

Proof of Theorem 1.2: In view of the relation (see Corollary 2.3.4 in [24])
n−1 n−1
ϕk,λ ∗λ ϕk,λ (z) = (2π)n |λ|−n ϕk,λ
n−1
(z)

we see that the Fourier transform fb(a, z) satisfies

Z
fb(a, w)ϕk,λ (z) = fb(a, z).
i
−n n n−1
(2π) |λ| (z − w)e 2 λ Im(z·w̄) dw = f −λ ∗−λ ϕk,λ
n−1
(2.12)
Cn
12 S. THANGAVELU

n−1
Recalling the definition of ek,λ (z, t) we can rewrite the above as
Z
(2π)−n |λ|n fb(a, w)ek,λ
n−1
((w, 0)−1(z, t))dw = eiλt f −λ ∗−λ ϕk,λ
n−1
(z).
Cn

Integrating the above over Ω with respect to dν and recalling (2.7) we obtain the inversion
formula Z Z
f (z, t) = fb(a, w)ea ((−w, 0)(z, t))dw dν(a). (2.13)
Ω Cn
The Plancherel theorems for the special Hermite expansions (2.3) and the Euclidean Fourier
transform gives us the identity
Z Z Z
2
|f (z, t)| dz dt = cn |fb(a, w)|2dw dν(a). (2.14)
Hn Ω Cn
This completes the proof of Theorem 1.2.

Proof of Theorem 1.3: For a given F ∈ L20 (Ω × Cn , dν dw), we define

Z Z
f (z, t) = F (a, w)ea ((−w, 0)(z, t))dw dν(a). (2.15)
Ω Cn

As F ∈ L20 (Ω × Cn , dν dw), the function f defined by

Z ∞ X∞
−n−1


f (z, t) = (2π) eiλt F (a, z) |λ|n dλ
−∞ k=0
P∞

belongs to L (H ). But then f (z) = (2π) |λ|n
2 n −λ −n
k=0 F (a, z) , which gives, in view of
n−1 n−1
the condition (2.12) and the orthogonality relation ϕk,λ ∗−λ ϕj,λ = 0, j 6= k we obtain
fb(a, z) = f −λ ∗−λ ϕk,λ
n−1
(z) = F (a, z)
which completes the proof.

Proof of Theorem 1.5: Recall that in the course of the proof of Theorem 1.1 we have
verified the estimate (2.10), namely
sup |fe(a, z)| ≤ cn,1 kf k1 .
(a,z)∈Ω×Cn

This simply means that f → fe is bounded from L1 (Hn ) into L∞ (Ω × Cn , dν1 (a) dz) where
Z Z ∞ X
∞ 
−2n−1
ϕ(a)dν1 (a) = (2π) ϕ(λ, (2k + n)|λ|) dλ.
Ω −∞ k=0

Restating the Plancherel theorem for fe we also have

Z Z 1/2  Z 1/2
e 2
|f (a, w)| dw dν2 (a) = |f (z, t)|2 dz dt
Ω Cn Hn

with another measure dν2 (a) on Ω defined by

Z Z ∞ X∞ 
k + n − 1)! 2 
ϕ(a)dν2 (a) = ϕ(λ, (2k + n)|λ|) |λ|2n dλ.
Ω −∞ k=0
k!(n − 1)!
 FOURIER TRANSFORM ON THE HEISENBERG GROUP 13

Using interpolation theorem with change of measures [15] we obtain the Hausdorff-Young
inequality
Z Z 1/p′ Z 1/p
e p′
|f (a, w)| dw dνp (a) ≤ cn,p |f (z, t)|p dz dt
Ω Cn Hn
p n
for all f ∈ L (H ), 1 ≤ p ≤ 2 for some measure νp (a) on Ω. The measure νp (a) can be
constructed explicitly.

Remark 2.2. Since f → fe is also bounded from L1 (Hn ) into L∞ (Ω × Cn , dν2 (a) dz), interpo-
lation without change of measures gives the inequality
Z Z 1/p′ Z 1/p
|fe(a, w)| dw dν2 (a)
p ′ ′
≤ cn,p |f (z, t)|p dz dt .
Ω Cn Hn
However, the result proved above with change of measures is sharper than this inequality.

3. Strichartz Fourier transform vs. Gelfand transform

We now investigate further properties of the Fourier transform f → fb which justifies our
claim that this transform is the analogue of the Helgason Fourier transform on Riemannian
symmetric spaces of noncompact type. If f is a K-biinvariant function on X = G/K it is
known that the Helgason Fourier transform fe(λ, b) is independent of b and reduces to the
Jacobi transform. In fact if f (g) = f (kar k ′ ) = f0 (r) then fe(λ, b) = Jα,β f0 (λ)Y0 (b) where
Y0 (b) = 1 is the unique K-fixed vector associated to the representation πλ and
Z ∞
Jα,β f0 (λ) = f0 (r)ϕα,β
λ (r)wα,β (r)dr
0

where ϕα,β
λ is the Jacobi function of type (α, β) which depends on the symmetric space. For


the Helgason Fourier transform, it is not true that ^ f ∗ g (λ, b) = fe(λ, b)e
g (λ, b). However if
g is K-biinvariant, then it is indeed true and we have
^

f ∗ g (λ, b) = Jα,β g0 (λ) fe(λ, b), g(kar k ′ ) = g0 (r).
We have a similar situation for the Strichartz Fourier transform on the Heisenberg group.
3.1. Gelfand transform on the Heisenberg group. In the case of Hn , the role of K-
biinvariant functions is played by U(n)-invariant functions of Hn , also known as radial func-
tions. Note that when f is such a function, then f (z, t) = f0 (|z|, t) for a unique function on
R+ × R and we know that the special Hermite expansion of the radial function f λ (z) reduces
to the Laguerre expansion, see [24, (2.4.5),p. 61] . Thus
n−1
f λ ∗λ ϕk,λ (z) = Rkn−1 (λ, f )ϕk,λ
n−1
(z)
where Rkn−1 (λ, f ) is the k-th Laguerre coefficient of f λ given by
Z ∞
n−1 n −n+1 k!
Rk (λ, f ) = (2π) 2 f0λ (r)ϕk,λ
n−1
(r) r 2n−1 dr.
(k + n − 1)! 0
π n
Since the surface measure of S 2n−1 is 2 Γ(n) we can write the above in the form
Z
n−1 k!(n − 1)!
Rk (λ, f ) = f λ (z)ϕk,λ
n−1
(z) dz.
(k + n − 1)! Cn
14 S. THANGAVELU

Thus we see that if f is a radial function on Hn , then

fb(a, z) = Rn−1 (−λ, f )en−1 (z, 0)
k k,λ (3.1)
so that fb(a, z) is proportional to the unique U(n)-fixed vector for the representation ρλk . For
f, g ∈ L1 (Hn ) we also have
f[
∗ g(a, z) = Rkn−1 (−λ, g0 ) fb(a, z).
As in the case of the Helgason Fourier transform, a more general result, known as the Hecke-
Bochner formula is true.

Though the algebra L1 (Hn ) under convolution is noncommutative, the subalgebra L1 (Hn /U(n))
consisting of radial functions is commutative. This follows from the fact that for any radial
function f the special Hermite expansion of f λ reduces to the Laguerre expansion as proved
above:
X∞
λ −n n
f (z) = (2π) |λ| Rkn−1 (λ, f ) ϕk,λ
n−1
(z). (3.2)
k=0
1 n
As elements of L (H /U(n)) are precisely the U(n)-biinvariant functions on the Heisenberg
motion group, this simply means that (Gn , U(n)) is a Gelfand pair. The multiplicative
linear functionals on this algebra are given by bounded spherical functions which come in
two families as shown in [7]. These are given by
√
k!(n − 1)! n−1 n−1 Jn−1 ( τ |z|)
e
ea (z, t) = e (z, t), χτ (z, t) = (n − 1)!2 √ , τ ≥ 0. (3.3)
(k + n − 1)! k,λ ( τ |z|)n−1
Thus we see that the Gelfand spectrum of the algebra L1 (Hn /U(n)) is precisely the Heisen-
berg fan Ω. The Gelfand transform on L1 (Hn /U(n)) is defined as the map which takes f
into Gf given by
Z Z
Gf (a) = f (z, t) eea (z, t) dz dt, Gf (0, τ ) = f (z, t) χτ (z, t) dz dt.
Hn Hn
Thus the relation (3.1) reads as
fb(a, z) = Gf (a) ek,λ
n−1
(z, 0), f ∈ L1 (Hn /U(n)). (3.4)
The same relation holds also for a = (0, τ ) as can be easily verified.

From the expansion (3.2) for radial functions we see that their Fourier transforms are
functions of the Hermite operator:
∞
X ∞
X
fb(λ) = Rk (λ, f ) Pk (λ) = Gf (a) Pk (λ).
k=0 k=0

This relation allows us to study the Fourier transform of radial functions in terms of their
Gelfand transforms, also called spherical Fourier transforms for obvious reasons. As the
Gelfand transform is scalar valued, it has been used as an alternate for the group Fourier
transform by several authors. For example, in the work [1] the authors have studied the
image of radial Schwartz functions under the spherical Fourier transform. In what follows,
we will further explore the connection between the Strichartz Fourier transform and Gelfand
transforms.
 FOURIER TRANSFORM ON THE HEISENBERG GROUP 15

3.2. Hecke-Bochner formula for the Strichartz Fourier transform. We begin by

recalling briefly some basic facts about bigraded spherical harmonics, referring to [24, Section
2.5] for details. Given a pair (p, q) of non-negative integers, we say that a polynomial P (z, z̄)
on Cn is a bigraded solid harmonic if it is harmonic and has the form
X X
P (z, z̄) = cα,β z α z̄ β .
|α|=p |β|=q

They are uniquely determined by their restrictions to the unit sphere S 2n−1 . We denote by
Sp,q the space of all bigraded spherical harmonics,
L i.e. restrictions of bigraded solid har-
2 2n−1
monics. It is then known that L (S ) = p,q Sp,q . Each Sp,q is finite dimensional and by
j
choosing an orthonormal basis Sp,q , j = 1, 2, ..., d(p, q) for each (p, q) we obtain an orthonor-
mals basis for L2 (S 2n−1 ).

Let us return to the Euclidean Fourier transform (1.8) for a moment and consider the
integral Z
fb(λ, ω) Ym(ω) dω
S n−1
where Ym is a (standard) spherical harmonic of degree m. Then it is known that (see for
example [16])
Z Z ∞
−m b Jn/2+m−1 (λr) n+2m−1
λ f (λ, ω) Ym(ω) dω = cn,m r −m fm (r) r dr (3.5)
S n−1 0 (λr)n/2+m−1
where fm (r) is the spherical harmonic coefficient (f (r·), Ym)L2 (S n−1 ) . Suppose g is a radial
function on Rn and f (x) = g(x)Pm (x) where Pm is a solid harmonic of degree m. Then from
the above formula, we easily see that the spherical harmonic expansion of fb(λ, ω) has only
one term and hence
Z ∞
b Jn/2+m−1 (λr) n+2m−1
f (λ, ω) = cn,m Pm (λω) g(r) r dr. (3.6)
0 (λr)n/2+m−1
We also observe that if we consider g as a radial function on Rn+2m then the above can be
rewritten as Fn (gPm ) = cn,m Pm Fn+2m (g) where Fk stands for the Fourier transform on Rk .
These formulas are known as Hecke-Bochner identities in the literature.

There is a Hecke-Bochner formula for the Helgason Fourier transform too, but we do not
intend to state it here as it requires quite a bit of preparation, see [11, 12]. Given below is
the Hecke-Bochner formula for the Strichartz Fourier transform which is the exact analogue
of (3.5). In what follows we use the following convention. A radial function g on Hn will
be simultaneously considered as a radial function on any Hm . When k is a negative integer
we set ϕmk,λ (z) = 0 for any m and z ∈ C . Moreover, the Gelfand tranform for the algebra
m

L1 (Hm /U(m)) will be denoted by Gm . With these notations we have

Theorem 3.1. For any f ∈ L1 (Hn ) and Sp,q ∈ Sp,q , let fp,q (r, t) = (f (r·, t), Sp,q )L2 (S 2n−1 )
stand for the spherical harmonic coefficient of f (rω, t). For any a ∈ Rk , λ > 0 we have the
following:
Z
fb(a, rω) Sp,q (ω) dω = (2π)−p−q (|λ|r)p+q Gn+p+q (gp,q )(a(p, q)) ek−q,λ
n+p+q−1
(r, 0)
S 2n−1
16 S. THANGAVELU

where gp,q (z, t) = |z|−p−q fp,q (|z|, t) and a(p, q) = (λ, (2k + p − q + n)|λ|). When λ < 0 we
have Z
fb(a, rω) Sp,q (ω) dω = (2π)−p−q (|λ|r)p+q Gn+p+q (gp,q )(a(q, p)) ek−p,λ
n+p+q−1
(r, 0).
S 2n−1

Proof. The proof is indeed a rewriting of [24, Theorem 2.6.1] where the following result is
proved. Let f ∈ L1 (Hn ) be of the form f (z, t) = g(z, t)P (z) where g is radial and P is a
bigraded solid harmonic of degree (p, q). Then for any λ > 0 we have
n+p+q−1
f λ ∗λ ϕk,λ
n−1
(z) = (2π)−p−q λp+q P (z) g λ ∗λ ϕk−p,λ (z) (3.7)
where the convolution on the right hand side is taken over Cn+p+q . There is a similar formula
when λ < 0 where the roles of p and q are interchanged. This result, stated in terms of Weyl
j
transform, is due to Geller [10]. It is enough to prove the result when Sp,q = Sp,q is a member
of the orthonormal basis for Sp,q which we have described earlier.
Expanding f λ (z) in terms of Sp,q
j
and recalling the definitions of fp,q and gp,q we have
X d(p,q)
X
f λ (z) = j
Pp,q j λ
(z) (gp,q ) (z).
p,q j=1

As fb(a, z) = f −λ ∗−λ ϕk,λ

n−1
(z), assuming λ < 0 (so that −λ > 0), the above formula gives
X  d(p,q)
X 
fb(a, z) = (2π)−p−q |λ|p+q j
Pp,q j |λ|
(z) (gp,q n+p+q−1
) ∗|λ| ϕk−p,λ (z) .
p,q j=1

As j
gp,q (z, t) is radial on H n+p+q
using (3.4) we can rewrite the above expansion as
X  d(p,q)
X 
fb(a, z) = (2π) −p−q
|λ| p+q j
Pp,q j
(z) Gn+p+q (gp,q n+p+q−1
)(a(q, p)) ek−p,λ (z, 0).
p,q j=1

By calculating the spherical harmonic coefficients of fb(a, rω) we obtain the stated formula.

The identity (3.7) which was used in the proof of the above theorem can be restated as
the following Hecke-Bochner formula for the Strichartz Fourier transfiorm.
Corollary 3.2. Suppose f ∈ L1 (Hn ) is of the form f (z, t) = P (z)g(z, t) where g is radial
and P is a solid harmonic of degree (p, q). Then for any λ > 0,
fb(a, z) = (2π)−p−q |λ|p+q P (z) Gn+p+q g(a(p, q)) en+p+q−1(z, 0). k−q,λ
A similar formula holds for λ < 0.

We also have a representation theoretic interpretation of the Hecke-Bochner formula. The

spherical harmonic expansion of a function f (z, t) on Hn can be written as
X d(p,q)
X
j
f (z, t) = g(p,q),j (r, t) Pp,q (z), z = rω (3.8)
(p,q)∈N2 j=1
j j
where Pp,q is the solid harmonic corresponding to Sp,q . For each pair (p, q) the space Sp,q
supports an irreducible unitary representation R(p,q) of the unitary group K = U(n). The
 FOURIER TRANSFORM ON THE HEISENBERG GROUP 17

action of R(p,q) on Sp,q is given by R(p,q) (σ)S(ω) = S(σ −1 ω), σ ∈ K. Let M be the isotropic
subgroup of K which fixes the coordinate vector e1 ∈ Cn which can be indentified with
U(n−1). It is known that each R(p,q) has a unique M-fixed vector in Sp,q . Such representations
are called class one representations and it is known that the any such irreducible unitary
representation of K is unitarily equivalent to one and only one of R(p,q) . Let K b 0 stand for
the set of all equivalence classes of such representations of K. We can rewrite (3.8) as
d(p,q)
X X
j
f (z, t) = fδ (z, t), fδ (z, t) = g(p,q),j (r, t) Pp,q (z), δ = (p, q). (3.9)
b0
δ∈K j=1

We can view the functions fδ as radial functions on Hn taking values in the finite dimensional
Hilbert space Sδ .

Let L2 (Hn /U(n), Hδ ) stand for the space of all such radial functions taking values in Hδ
so that M
L2 (Hn ) = L2 (Hn /U(n), Sδ ). (3.10)
b0
δ∈K

It then follows that each of the spaces L (Hn /U(n), Hδ ) is invariant under the Strichartz
2

Fourier transform. Indeed, by the Hecke-Bochner formula it is clear that, for λ > 0
 d(p,q)
X 
fbδ (a, z) = (2π)−p−q |λ|p+q j
Pp,q n+p+q−1
(z) Gn+p+q (gδ,j )(a(p, q)) ϕk−q,λ (z). (3.11)
j=1

Thus we can consider the Strichartz Fourier transform as a family of Gelfand transforms Gδ
indexed by the class one representations δ of the pair (U(n), U(n−1)). If fδ is thus identified
with the vector (gδ,j )j=1 , then fbδ (a, z) is given by the vector with components
d(δ)

n+p+q−1
(2π)−p−q |λ|p+q r p+q Gn+p+q (gδ,j )(a(p, q)) ϕk−q,λ (r). (3.12)
We will make use of this view in the next section where we study the image of the Schwartz
space under the Strichartz Fourier transform.

3.3. The image of S(Hn ) under the Strichartz Fourier transform. Let S(Hn ) stand
for the space of all Schwartz class functions on the Heisenberg group. As the underlying
manifold of Hn is just R2n+1 this space is nothing but S(R2n+1 ). In his pioneering work
Geller [10] has proved a characterisation of the image of S(Hn ) under the group Fourier
transform in terms of certain asymptotic series. In 1998, Benson et al. [7] studied the image
of Schwartz functions on the Heisenberg group under the spherical Fourier transform asso-
ciated to Gelfand pairs. In particular, for radial functions on Hn they have described the
image as a space of rapidly decreasing functions on the Gelfand spectrum identified with the
Heisenberg fan Ω. In their work, rapidly decreasing functions on Ω are defined in terms of
certain derivatives and finite difference operators.

In a series of papers [1, 2, 3, 4], Astengo et al have studied the problem of characterizing
the image of S(Hn ) under the Fourier transform. By using multiple Fourier series, they have
reduced the problem to the characterization of the image of polyradial Schwartz functions
under the Gelfand transform (spherical Fourier transform). See the survey [14] for a readable
18 S. THANGAVELU

account of these works with connections to spectral multipliers. In particular, for the class
SK (Hn ) of radial functions (recall that K = U(n)) they have proved the following elegant
result.

Theorem 3.3. [1, Astengo et al.] Let S(Ω) stand for the space of restrictions of Schwartz
functions on R2 to Ω equipped with the quotient topology S(R2 )/{f : f |Ω = 0}. The Gelfand
transform Gn is a topological isomorphism between SK (Hn ) and S(Ω).

Recall that the components gδ,j of fδ are defined by r −p−q fδ,j (r, t) where
Z
fδ,j (r, t) = f (rω, t)Sδ,j (ω) dω.
S 2n−1
It is clear that when f ∈ S(H ) these functions gδ,j considered as radial functions on Hn+p+q
n

are Schwartz functions. For the sake of brevity let us use the following notations: we write
Hδ in place of Hn+p+q and U(δ) in place of U(n + p + q) so that Hn and U(n) correspond
to the trivial representation of U(n). Following the previous authors, let SU (δ) (Hδ , Sδ ) stand
for the Schwartz space of U(δ) invariant functions on Hδ taking values in Sδ . The Gelfand
transform for the pair (Hδ , U(δ)) will be denoted by Gδ . We then observe that for Schwartz
class functions the decomposition (3.10) takes the form
M
S(Hn ) = SU (δ) (Hδ , Sδ ).
b0
δ∈K

We also use the notation S(R2 , Sδ ) for the space of Sδ valued Schwartz functions on R2 . We
define S(Ω, Sδ ) as in the scalar valued case and take their orthogonal sum to define
M
b
S(Ω) = S(Ω, Sδ ).
b0
δ∈K

With these notations we can restate Theorem 3.3 in the following form.

Theorem 3.4. The Strichartz Fourier transform is an isomorphism between S(Hn ) and
b
S(Ω).
Even though this result is a consequence of Theorem 3.3, due to the various identifications
we have made, the following explanations are in order. Given f ∈ S(Hn ) and fδ defined as
in (3.9) the function
d(δ)
X
e
fδ (z, t) = gδ,j (z, t) Sδ,j , (z, t) ∈ Hδ
j=1

is an element of the space SU (δ) (Hδ , Sδ ). The Gelfand transform Gδ feδ is a function on the
Heisenberg fan Ωδ for the pair (Hδ , U(δ)) which is a proper subset of Ω. By Theorem 3.3 there
exists mδ ∈ S(Ω, Sδ ) such that Gδ feδ is the restriction of mδ to Ωδ . If mδ,j are the components
P
of mδ then we have the relation Gδ feδ = d(δ) j=1 mδ,j Sδ,j and the Strichartz Fourier transform
of fδ considered as a function on H is given by
n

X
d(δ) 
fbδ (a, z) = (2π)−p−q |λ|p+q Pδj (z) mδ,j (a(p, q)) n+p+q−1
ϕk−q,λ (z)
j=1
 FOURIER TRANSFORM ON THE HEISENBERG GROUP 19

for λ > 0 with a similar expression for λ < 0. Thus the correspondence alluded to in the
statement of the theorem is the one given by fbδ → mδ . Conversely, given mδ ∈ S(Ω, Sδ ) we
can define gδ,j by applying Gδ−1 and the function fδ is then defined as in (3.9). It is a routine
matter to check that the resulting f is Schwartz.

4. Group Fourier transform vs. Strichartz Fourier transform

In this section we investigate the relation between group Fourier transform and the
Strichartz Fourier transform on Hn . Though the former is defined in terms of the Schrödinger
representations πλ and the latter in terms of ρλk there is a direct connection between fb(λ)
and fb(a, z). We will show that this relation allows us to recast some theorems for the group
Fourier transform in terms of the Strichartz Fourier transform.

Combining (2.4) and (2.1) we obtain the following relation between the Strichartz and the
group Fourier transforms:
Z
|fb(a, z)|2 dz = (2π)n |λ|−n kfb(λ)Pk (λ)k2HS .
Cn
As an application of this relation, let us rewrite theorems of Hardy and Ingham for the
Heisenberg group in terms of the Strichartz Fourier transform.

Let pa stand for the heat kernel associated to the sublaplacian L on Hn . The Heisenberg
analogue of the classical Hardy’s uncertainty principle reads as follows (see [24, Theorem
2.9.2, p. 89]):

For a nontrivial function f ∈ L1 (Hn ) the condtions

|f (z, t)| ≤ c pa (z, t), fb(λ)∗ fb(λ) ≤ c pc
2b (λ)

cannot hold simultaneously unless a ≥ b. We can now restate this result in the following
form.
Theorem 4.1. Suppose f ∈ L1 (Hn ) satisfies the conditions
Z
(k + n − 1)! −2b(2k+n)|λ|
|f (z, t)| ≤ c pa (z, t), |λ| n
|fb(a, z)|2 dz ≤ c e
Cn k!(n − 1)!
for every a ∈ Rk . Then f = 0 whenever a < b.
The decay condition on fb(a, z) comes from the Hardy condition fb(λ)∗ fb(λ) ≤ c pc 2b (λ).
−bH(λ)
In fact, it is well known that pbb (λ) = e is the semigroup generated by the Hermite
operator and an easy calculation using the Hermite basis shows that
(k + n − 1)! −2b(2k+n)|λ|
kfb(λ)Pk (λ)k2HS ≤ c e (4.1)
k!(n − 1)!
under the Hardy condition on fb(λ). An examination of the proof of Hardy’s theorem pre-
sented in [24, Theorem 2.9.2], shows that the result holds under the weaker assumption (4.1).
We remark that other more refined versions of Hardy’s theorem can also be stated in terms
of fb(a, z).
20 S. THANGAVELU

Ingham’s uncertainty principle is another theorem that has received considerable attention
in recent years. For functions on R Ingham proved in 1934 the following result:

Let θ be a nonnegative even function on R which decreases to 0 at infinity. There exists a

compactly supported function f on R whose Fourier transform satisfies the decay condition
|fb(y)| ≤ c e−|y| θ(y)
R∞
if and only if 1 θ(t)t−1 dt < ∞.

Recently the following analogue of Ingham’s theorem for the group Fourier transform has
been proved in [5]:

Let Θ be a nonnegative even function on R which decreases to 0 at infinity. There exists a

compactly supported function f on Hn whose Fourier transform satisfies the decay condition
√ √
|fb(λ)| ≤ c e− H(λ) Θ( H(λ))
R∞
if and only if 1 Θ(t)t−1 dt < ∞.

As before using the relation between fb(λ) and fb(a, z) we can restate the Ingham’s theorem
in the following form.
Theorem 4.2. Let Θ be a nonnegative even function on R which decreases to 0 at infinity.
There exists a compactly supported function f on Hn whose Fourier transform satisfies the
decay condition
Z
(k + n − 1)! −√(2k+n)|λ| Θ(√(2k+n)|λ|)
|λ| n
|fb(a, z)|2 dz ≤ c e
Cn k!(n − 1)!
R∞
if and only if 1 Θ(t)t−1 dt < ∞.
The proof of Ingham’s theorem given in [5] can be easily modified to prove this version.
We can also prove other refined versions stated and proved elsewhere. For more on Ingham’s
theorem and its close relatives we refer to [9] and the references therein.

Acknowledgments
References
[1] F. Astengo, Di Blasio and F. Ricci, Gelfand transforms of polyradial Schwartz functions on the Heisen-
berg group, J. Funct. Anal. 251 (2007), no. 2, 772-791.
[2] F. Astengo, B. Di Blasio and F. Ricci, Gelfand pairs on the Heisenberg group and Schwartz functions,
J. Funct. Anal. 256 (2009), no.5, 1565-1587.
[3] F. Astengo, B. Di Blasio and F. Ricci, Fourier transform of Schwartz functions on the Heisenberg group,
Studia Math., 214 (2013), no.3, 201-222.
[4] F. Astengo, B. Di Blasio and F. Ricci, Paley-Wiener theorems for the U (n)-spherical transform on the
Heisenberg group, Ann. Mat. Pura Appl. (4), 194 (2015), no. 6, 1751-1774.
[5] S. Bagchi, P. Ganguly, J. Sarkar and S. Thangavelu, An analogue of Ingham’s theorem on the Heisenberg
group, Preprint (2022).
[6] H. Bahouri, J-Y. Chemin and R. Danchin, A frequency space for the Heisenberg group Ann. Inst.
Fourier (Grenoble), 69 no. 1 (2019), 365-407.
[7] C. Benson, J. Jenkins and G. Ratcliff, The spherical transform of a Schwartz function on the Heisenberg
group, J. Funct. Anal. 154(1998), no. 2, 379-423.
 FOURIER TRANSFORM ON THE HEISENBERG GROUP 21

[8] G. B. Folland, Harmonic Analysis in Phase Space, Ann. of Math. Stud., vol. 122, Princeton University
Press, Princeton, N.J., 1989.
[9] P. Ganguly and S. Thangavelu, Analogues of theorems of Chernoff and Ingham on the Heisenberg group,
J. d’Analyse Math. (to appear)
[10] D. Geller, Fourier analysis on the Heisenberg group, I. Schwartz space, J. Func. Anal. 36 (1980), no.
2, 205-254.
[11] S. Helgason, Groups and Geometric Analysis, Academic Press, (1984).
[12] S. Helgason, Geometric analysis on symmetric spaces, Mathematical surveys and monographs, AMS,
Vol 39, (1994).
[13] R. Rawat, P. K. Ratnakumar and S. Thangavelu, A restriction theorem for the Heisenberg motion
group, Studia Math. 126(1) (1997), 1-12.
[14] F. Ricci, Schwartz functions on the Heisenberg group, spectral multipliers and Gelfand pairs, Rev. Un.
Mat. Argentina 50 (2009), no. 2, 175-186.
[15] E. M. Stein and G. Weiss, Interpolation of operators with change of measures, Trans. Amer. Math. Soc.
87 (1958), 159-172.
[16] E. M. Stein and G. Weiss, Introduction to Fourier Analysis in Euclidean spaces, Princeton Univ. Press,
Princeton, N.J., 1971. x+297 pp.
[17] R. S. Strichartz, Harmonic analysis as spectral theory of Laplacians, J. Funct. Anal. 87 (1989), no. 1,
51-148.
[18] R. S. Strichartz, Lp harmonic analysis and Radon transforms on the Heisenberg group, J. Funct. Anal.
96 (1991), no. 2, 350-406.
[19] G. Szego, Orthogonal polynomials, Am. Math. Soc. Colloq. Pub. 23, Providence, RI (1967).
[20] M. Taylor, Noncommutative harmonic analysis. Mathematical Surveys and Monographs, 22 American
Mathematical Society, Providence, RI, (1986).
[21] S. Thangavelu, Lectures on Hermite and Laguerre expansions, Mathematical Notes 42. Princeton Uni-
versity Press, Princeton, NJ, 1993.
[22] S. Thangavelu, On Paley-Wiener theorems for the Heisenberg group, J. Funct. Anal. 115 (1993), no. 1,
24-44.
[23] S. Thangavelu, Harmonic analysis on the Heisenberg group, Progr. Math.Vol. 159 Birkhäuser, Boston,
MA, 1998
[24] S. Thangavelu, An introduction to the uncertainty principle. Hardy’s theorem on Lie groups. With a
foreword by Gerald B. Folland, Progress in Mathematics 217. Birkhäuser, Boston, MA, 2004

(S. Thangavelu) Department of Mathematics, Indian Institute of Science, 560 012 Banga-
lore, India
Email address: [email protected]