On theorems of Beurling and Hardy for the Euclidean motion groups (Q812888): Difference between revisions

A generalization of Beurling's uncertainty principle has been proved by \textit{A. Bonami, B. Demange} and \textit{Ph. Jaming} [Rev. Mat. Iberoam. 19, 23--55 (2003; Zbl 1037.42010)]. Theorem A. Let \(f \in L^2({\mathbb R}^n)\) and \(\hat{f}\) the Fourier transform of \(f\) . Suppose that for some \(N \geq 0\) \[ \int_{{\mathbb R}^n}\int_{{\mathbb R}^n}\displaystyle{\frac{|f(x)|}{(1+|x|)^{N}} \frac{|\hat{f}(y)|}{(1+|y|)^N}e^{|(x,y)|}dxdy} < \infty . \] Then \(f=0\) whenever \(N \leq n\). If \(N > n\), then the above holds if and only if \(f\) can be written as \( f(x)=P(x)e^{-(Ax,x)/2} \), where \(A\) is a real positive definite matrix and \(P\) is a polynomial of degree \(\leq (N-n)/2\). Hardy's uncertainty principle says that if \(f \in L^1({\mathbb R}^n)\) satisfies \(|f(x)| \leq c(1+|x|)^mp^n_{s}(x)\), \(|\hat{f}(y)| \leq c(1+|y|)^me^{-t|y|^2}\) , where \(p^n_{t}(x)=(4\pi t)^{-n/2}e^{-|x|^2/4t}\) is the heat kernel of the Laplacian \(\Delta_{n}\) on \({\mathbb R}^n\), then \(f=0\) whenever \(s <t\). When \(s=t\), \(f(x)=Q(x)p^n_{t}(x)\) where \(Q\) is a polynomial of degree \( \leq m\). It is clear that this result follows immediately from Theorem A. In this paper the authors establish an analogue of Beurling's theorem for the Fourier transform on the Euclidean motion group \(M(n)\), and they also prove the most general version of Hardy's theorem. Let \(M(n)\) be the semidirect product of \({\mathbb R}^n\) with \(K=SO(n)\). The group law is given by \( (x,y)(y,k')=(x+k\cdot y, kk'),\) \(x, y \in {\mathbb R}^n, \;\;k, k' \in K\). Consider \(M=SO(n-1)\) as a subgroup of \(K\) leaving the point \(e=(1,0, \cdots, 0)\) fixed. Then all the irreducible unitary representations of \(M(n)\) relevant for the Plancherel theorem are parameterized (up to unitary equivalence) by pairs \((\lambda, \sigma)\) where \(\lambda >0\) and \(\sigma \in \hat{M}\), the unitary dual of \(M\) ( when \(n=2\) there is only one parameter, namely \(\lambda >0\)). The group Fourier transform of \(f \in L^1(M(n))\) is then defined by \[ \hat{f}(\lambda, \sigma)= \int_{M(n)}f(x,k)\pi_{\lambda, \sigma}(x,k) dx dk, \] where \(\pi_{\lambda, \sigma}(x,k)\) is the irreducible unitary representation of \(M(n)\) associated to the parameter \((\lambda, \sigma\)). Let \(||\hat{f}(\lambda,\sigma)||_{HS}\) stand for the Hilbert-Schmidt operator norm of \(\hat{f}(\lambda, \sigma)\). An analogue of Beurling's theorem is the following: Theorem 1. (a) Let \(f \in L^1 \cap L^2(M(n))\) and assume that \[ \int_{M(n)}\int^{\infty}_{0}\displaystyle{ \frac{||\hat{f}(\lambda, \sigma)||_{HS}}{(1+\lambda)^n} \frac{|f(x,k)|}{(1+|x|)^n}}e^{\lambda|x|}\lambda^{n-1}d\lambda dx dk <\infty \] for every \(\sigma \in \hat{M}\). Then \(f=0\). (b) Let \(f \in L^1 \cap L^2(M(2))\) and assume that \[ \int_{(z, e^{i\phi}) \in M(2)}\int_{0}^{\infty}\displaystyle{ \frac{||\hat{f}(\lambda)||_{HS}}{(1+\lambda)^3}\frac{|f(z,e^{i\phi})|} {(1+|z|)^3} }e^{\lambda|z|}\lambda d\lambda dz d\phi < \infty. \] Then \(f\) can be represented as \( f(z,e^{i\phi})= g(e^{i\phi})e^{-a|z|^2} \) for some \(g \in L^2(T)\) and \(a> 0\), where \(T\) stands for the circle group which is identified with \(SO(2)\). A general version of Hardy's theorem is the following: Theorem 2. (a) Let \(f \in L^1(M(n))\) satisfy the following two conditions \( |f(x,k)| \leq c(1+|x|)^{N}p^{n}_{s}(x),\;\;(x,k) \in M(n) \), \[ ||\hat{f}(\lambda, \sigma)||_{HS} \leq c(1+\lambda)^{N}e^{-t\lambda^2}, \;\;(\lambda, \sigma) \in {\mathbb R}_{+}\times \hat{M}. \] Then \(f=0\) whenever \(s <t\). When \(s=t\), \(f\) can be expressed as a finite linear combination of functions of the form \( P_{m,j}(x)(-\Delta_{n+2m})^{(j-m)/2}p_{t}^{n+2m}(x)g_{mj}(k)\), where \(P_{m,j}\) are solid harmonics of degree \(m\) and \(g_{mj}\) are certain bounded functions in \(L^2(K)\). The fractional powers of the Laplacian are defined via the Fourier transform: \[ (-\Delta_{n})^{j/2}f(x) = (2\pi)^{-n/2}\int_{{\mathbb R}^n} |\xi|^{j}\hat{f}(\xi)e^{ix\cdot \xi} d\xi. \] The ranges of \(m\) and \(j\) are restricted by the condition \[ |P_{m,j}(x)||(-\Delta_{n+2m})^{(j-m)/2}p_{t}^{n+2m}(x)| \leq c(1+|x|)^Np_{t}^n(x). \] (b) Let \(f \in L^1(M(2))\) satisfy the estimate \( |f(z,e^{i\phi})| \leq cp_{t}^2(z), \;\;z \in {\mathbb R}^2. \) Further assume that for every \(\lambda > 0\), and \(k,j \in {\mathbb Z}\) \[ |(\hat{f}(\lambda)e_{k}, e_{j})| \leq c_{kj}\lambda^{|k-j|}e^{-t\lambda^2}. \] Then \(f\) can be represented as \[ f(z, e^{i\phi})=p^2_{t}(z)(\int_{{\mathbb R}^2}f(z, e^{i\phi})dz). \]