Math. Z. 242,761–779 (2002) Digital Object Identifier (DOI) 10.1007/s002090100379
Revisiting Hardy’s theorem for the Heisenberg group
S. Thangavelu Stat-Math Division,Indian Statistical Institute,8th Mile Mysore Road, Bangalore,560 059,India (e-mail: [email protected])
Received: 9 January 2001 / in final form: 17 April 2001 Published online: 1 February 2002 – c Springer-Verlag 2002
Dedicated to Eli Stein on his 70th birthday
Abstract. We establish several versions of Hardy’s theorem for the Fourier transform on the Heisenberg group. Let fˆ(λ) be the Fourier transform of a n ˆ ∗ ˆ function f on H and assume f(λ) f(λ) ≤ c pˆ2b(λ) where ps is the heat kernel associated to the sublaplacian. We show that if |f(z,t)|≤cpa(z,t) then f =0whenever a
1 Introduction and the main results
The aim of this paper is to establish an analogue of Hardy’s theorem for the Fourier transform on the Heisenberg group. This is a continuation of the author’s earlier paper [22] in which a version of Hardy’s theorem was proved. That theorem in turn has its roots in the ‘Hardy’s theorem for the Weyl transform’ which appeared in the author’s monograph [21]. If this is so one wonders,as the author often does himself,why he is visiting Hardy’s theorem again and again,each time reformulating and refining his earlier results. We have no explanation to offer except possibly the compulsion of the demanding critic within himself who never leaves the author at peace until his theorems,if not their proofs,are straight from the ‘Book’. So,like a painter,with an extra stroke or two we try to change the shade a bit here and there in an attempt to draw closer to the truth. 762 S. Thangavelu
We can think of Hardy’s theorem as an uncertainty principle which roughly says that a function and its Fourier transform both cannot be de- caying fast at infinity unless,of course,the function is identically zero. Our first result,Theorem 1.1 below gives the optimal version of this uncertainty principle on the Heisenberg group. As the Fourier transform on the Heisen- berg group is operator valued we have to measure the decay of the Fourier transform in terms of the Hermite semigroup. We can also think of Hardy’s theorem as one characterising the heat kernel associated to the sublaplacian. Unlike the Euclidean case the heat kernel on the Heisenberg group is not explicitly known. What is known is its partial Fourier transform in the cen- tral variable and our result Theorem 1.2 characterises exactly this partial Fourier transform of the heat kernel. Consider the Fourier transform fˆ(λ),λ ∈ R,λ =0of a function f n n on the Heisenberg group H . Let pt(z,s), (z,s) ∈ H ,t > 0 be the heat n kernel associated to the sublaplacian L on H . The Fourier transform of pt −tH(λ) 2 2 is given by pˆt(λ)=e where H(λ)=−∆ + λ |x| is the (scaled) Hermite operator on Rn. As an analogue of Hardy’s theorem for the group Fourier transform on Hn we offer: Theorem 1.1. Let f be a measurable function on Hn which satisfies the 2 m estimate |f(z,s)|≤c (1 + |z| ) pa(z,s) for some a>0,m≥ 0. Further ˆ ∗ ˆ m assume that f(λ) f(λ) ≤ cH(λ) pˆ2b(λ), for some b>0 and for all λ =0. Then f =0whenever a 2 2 |f(x)|≤c(1 + |x|2)me−a|x| , |fˆ(ξ)|≤c(1 + |ξ|2)me−b|ξ| (1.1) where fˆ(ξ) is the Fourier transform of f given by − n −ix·ξ fˆ(ξ)=(2π) 2 f(x)e dx Rn 1 −a|x|2 then (i) f =0whenever ab > 4 ; (ii) f(x)=cp(x)e where p(x) is a 1 polynomial of degree ≤ 2m when ab = 4 and (iii) there are infinitely many 1 linearly independent functions satisfying the conditions (1.1) when ab < 4 . To see the analogy of this result with Theorem 1.1 we only have to rewrite conditions (1.1) in terms of the heat kernel − n − 1 |x|2 pt(x)=(4πt) 2 e 4t associated to the standard Laplacian ∆ on Rn. The above result for the Euclidean Fourier transform was first established by Hardy [6] for n =1and m =0. Now several versions of this interesting Revisiting Hardy’s theorem for the Heisenberg group 763 result are known,see [13] and [23]. In [22] we have established Theorem b 1.1 when m =0and a< 2 . The optimal result a p+q j for every k ∈ N and P (z)=|z| Ypq(z ). Then (i) when a = b, f = ∞ ϕ ∗3 pa with ϕˆ ∈ L (R); (ii) when a>bthere are infinitely many linearly independent functions satisfying the conditions of the theorem. Refinements of the classical Hardy theorem have been established in [23] using spherical harmonics. These results in [23] are similar in spirit to the Helgason’s treatment of the Paley-Wiener theorem [7]. The above two theorems are analogues of these refinements. Since the perfect symmetry between f and fˆ is lost when we move away from the Euclidean Fourier transform we have two versions as above instead of one in the Euclidean case. The problem of establishing an analogue of Hardy’s theorem for Fourier transforms on Lie groups started with the work of Sitaram and Sundari [18]. For other versions of Hardy’s theorem for semi-simple Lie groups see Cowling et al [3] and Sengupta [16]. Analogues of Hardy’s theorem for the Heisenberg group have been studied in Sitaram et al [17] and Thangavelu [21] and [22]. For step two nilpotent Lie groups see the works of Bagchi and Ray [2] and Astengo et al [1]. General nilpotent Lie groups were considered in Kaniuth and Kumar [9]. Symmetric spaces were treated in Narayanan and Ray [12],solvable extensions of H-type groups by [1]. See also the recent works [14] and [15] of Sarkar for semi-simple Lie groups. Analogues of Theorems 1.2 and 1.3 have been established for noncom- pact rank one symmetric spaces in [23]. There we have also proved a version of Hardy’s theorem for the spectral projections associated to the Laplace- Beltrami operator. We will state and prove a similar theorem for the spectral projections associated to the sublaplacian in Sect. 4. The plan of the paper is as follows. In Sect. 2 we collect relevant material on the Heisenberg group. Most results we need are contained in the paper [5] of Geller. We also refer to the monograph of Folland [4] and that of the author [21]. Theorems 1.1, 1.2 and 1.3 will be proved in Sect. 3. We are extremely thankful to Ms. Ashalata for typing the manuscript. We also wish to thank the referee for his careful reading of the manuscript and valuable suggestions. 2 Preliminaries on the Heisenberg group The Heisenberg group Hn is just Cn × R equipped with the group law 1 (z,t)(w, s)= z + w, t + s + Im(z · w) . 2 For each λ ∈ R,λ=0there is an irreducible unitary representation πλ(z,t) of Hn realised on L2(Rn). These representations are explicitly given by iλt iλ(x·ξ+ 1 x·y) πλ(z,t)ϕ(ξ)=e e 2 ϕ(ξ + y), Revisiting Hardy’s theorem for the Heisenberg group 765 where ϕ ∈ L2(Rn) and z = x + iy. Each infinite dimensional irreducible unitary representation is equivalent to one of these. The group Fourier trans- form of a function f ∈ L1(Hn) is defined to be the operator valued function ˆ f(λ)= f(z,t)πλ(z,t)dz dt. Hn iλt The representation πλ satisfies πλ(z,t)=e πλ(z,0) and therefore ˆ λ f(λ)= f (z)πλ(z)dz, (2.1) C n where we have written πλ(z)=πλ(z,0) and ∞ f λ(z)= f(z,t)eiλtdt −∞ is the inverse Fourier transform of f in the t-variable. The formula (2.1) suggests that we consider Weyl tranforms of functions g on Cn. These are defined by Gλ(g)= g(z)πλ(z)dz. (2.2) C n If f ∗ g is the convolution of two functions on Hn defined by f ∗ g(z,t)= f((z,t)(w, s)−1)g(w, s)dw ds Hn then it is easily checked that λ λ λ (f ∗ g) (z)=f ∗λ g (z) where the λ-twisted convolution of f λ and gλ is given by λ λ λ λ i λ Im(z·w) f ∗λ g (z)= f (z − w)g (w)e 2 dw. C n λ λ λ λ It then follows that Gλ(f ∗λ g )=Gλ(f )Gλ(g ). Let Fλf be the λ- symplectic Fourier transform of a function f on Cn given by i λ Im(z·w) Fλf(z)= f(z − w)e 2 dw. (2.3) C n 766 S. Thangavelu n We define the Weyl correspondence of a function f on C by Wλ(f)= −1 Gλ(Fλ f). The most important result we need is the Hecke-Bochner type identity for the Weyl transform Gλ. In order to state this important formula we need to recall some definitions. For each pair of non-negative integers (p, q) let Hpq be the space of all harmonic polynomials of the form α β P (z)= cαβz z (2.4) |α|=p |β|=q n n where z ∈ C ,α,β ∈ N . Elements of Hpq are called bigraded solid harmonics of degree (p, q). Let Spq be the space of all restrictions of elements 2n−1 of Hpq to the unit sphere S . The elements of Spq are called bigraded spherical harmonics. Then L2(S2n−1) is the orthogonal direct sum of the j spaces Spq,p,q ≥ 0. Let {Ypq :1≤ j ≤ d(p, q)} be an orthonormal basis j for Spq. The corresponding elements of Hpq are denoted by Ppq. n n For each multiindex α ∈ N ,x∈ R ,let Φα(x) be the normalised Her- mite function which is an eigenfunction of the Hermite operator H with λ eigenvalue (2|α| + n) where |α| = α1 + ... + αn. Define Φα(x)= n 1 λ λ |λ| 4 Φα(|λ| 2 x) for λ =0so that H(λ)Φα =(2|α| + n)|λ|Φα where H(λ)=−∆ + λ2|x|2. We say that an operator T acting on L2(Rn) is λ λ λ λ δ radial if it is diagonalised by Φα and TΦα = c|α|Φα. Let Lk,δ >−1 be the Laguerre polynomials of type δ. We refer to Szego [20] for various properties δ of Lk. Define the Laguerre functions by n−1 n−1 1 2 − 1 |z|2 ϕ (z)=L |z| e 4 k k 2 (2.5) for z ∈ Cn. With these notations we are now in a position to state Geller’s result. Theorem 2.1. Suppose gP ∈ L1(Cn) or L2(Cn) where g is a radial func- q tion and P ∈Hpq. Then for λ>0,Gλ(gP)=(−1) Wλ(P )S where S is a λ λ λ λ radial operator whose action on Φα is given as follows : SΦα = c|α|(g)Φα λ where ck(g)=0for k