Analytic factorization of Lie group representations (Q656214): Difference between revisions

Let \(G\) be a (finite-dimensional, real, connected) Lie group. The article under review is devoted to representations \(\pi\) of \(G\) on a Fréchet space~\(E\), i.e., continuous actions \(\pi\colon G\times E\to E\) by linear isomorphisms \(\pi(g,.)\) of~\(E\). Results are obtained for F-representations, i.e., representations such that \(\pi \colon G\times (E,p_n)\to (E,p_n)\) is continuous for some sequence of continuous seminorms~\(p_n\) on~\(E\) defining its topology. For such \(\pi\), the authors describe a certain convolution algebra \({\mathcal A}(G)\) of analytic functions \(\phi\) on \(G\) which acts on \(E\) via \(\Pi(\phi,v):=\int_G \phi(g)\pi(g,v)\, dg\). The latter action takes \(E\) into the space \(E^\omega\) of analytic vectors, and the authors obtain the following factorization result: Every \(v\in E^\omega\) can be written as a sum \[ v=\Pi(\phi_1,w_1)+\cdots+\Pi(\phi_k,w_k) \] with \(k\in {\mathbb N}\), \(\phi_1,\ldots,\phi_k\in {\mathcal A}(G)\), \(w_1,\ldots, w_k\in E^\omega\). Likewise, every element of \({\mathcal A}(G)\) is a sum \(\phi_1*\psi_1+\cdots+\phi_k*\psi_k\) with \(\phi_1,\psi_1,\ldots, \phi_k,\psi_k\in{\mathcal A}(G)\). (Theorem~1.1). If \(G={\mathbb R}\), then one can always take \(k=1\) (Lemma~6.5); the authors conjecture that the same holds in general. To define \({\mathcal A}(G)\), pick a left invariant Riemannian metric on~\(G\) and let \(d\) be the associated distance function on~\(G\). Let \({\mathcal R}(G)\) be the space of all continuous functions \(\phi\in C(G)\) which are superexponentially decaying in the sense that \(\sup\{|\phi(g)|e^{nd(g,1)}\colon g\in G\}<\infty\) for all \(n\in {\mathbb N}\). Then \({\mathcal R}(G)\) is an algebra under convolution, and the natural left and right actions of \(G\) on \({\mathcal R}(G)\) give rise to an F-representation \(L\otimes R\) of \(G\times G\) on \({\mathcal R}(G)\), whose space of analytic vectors is denoted \({\mathcal A}(G)\). Two further results deserve mention: By Theorem~6.1, every test function \(\theta\in C^\infty_c(G)\) can be written as \(\theta=\phi*\psi\) for certain \(\phi,\psi\in {\mathcal R}(G)\). In the case \(G={\mathbb R}^n\), this answers an open problem by \textit{L. A. Rubel, W. A. Squires} and \textit{B. A. Taylor} (see p.\ 560 in [Ann. Math. (2) 108, 553--567 (1978; Zbl 0402.32002)]). Finally, \(E^\omega\) is shown to coincide with the space of analytic vectors for the Laplace-Beltrami operator on~\(G\). Reviewer's comment: In the introduction, the authors claim that the actions of \(C^\infty_c(G)\) on \(E\) and \(E^\infty\) are continuous, for every representation \(\pi\) of a Lie group \(G\) on a Fréchet space~\(E\). However, the convolution action of \(C^\infty_c({\mathbb R})\) on \(C^\infty({\mathbb R})\) is discontinuous (see Proposition~7.1 in a preprint by L. Birth and the reviewer, \url{arXiv:1112.4729}), although it is associated with the smooth translation action \({\mathbb R}\times C^\infty({\mathbb R})\to C^\infty({\mathbb R})\). If~\(\pi\) is an F-representation, the pathology cannot occur.