Convergence of iterates of averages of certain operator representations and of convolution powers (Q923738)

Let G be a locally compact \(\sigma\)-compact group and let \(\mu\) be a probability measure on G that is adapted and strictly aperiodic. Moreover let T be a strongly continuous representation of G by isometries on a uniformly convex Banach space B. Then the strong integral \(P_{\mu}:=\int_{G}T(t)\mu (dt)\) yields a contraction on B. By uniform convexity the mean ergodic theorem applies to \(P_{\mu}\). The paper is concerned with the strong convergence of the sequence \((P^ n_{\mu})_{n\geq 1}\) (to the projection onto the subspace of \(P_{\mu}\)-invariant vectors). Whereas weak convergence of \((P^ n_{\mu})_{n\geq 1}\) always holds (Theorem 2.6) strong convergence is proved only under additional conditions. If G is compact then \((P^ n_{\mu})_{n\geq 1}\) converges strongly to \(P_ m\) in view of the Ito-Kawada theorem (where m denotes Haar measure on G). If \(\mu\) is normal then \((P^ n_{\mu})_{n\geq 1}\) strongly converges (Theorem 2.4). The main result of this paper (Theorem 2.8) is the strong convergence of \((P^ n_{\mu})_{n\geq 1}\) if \(\mu\) is spread-out (i.e. some convolution power \(\mu^ k\) is not singular). This property is well known from the theory of random walks on groups. These results are applied to the concentration function \(C(\mu,K):=\sup \{\mu (sK):\) \(s\in G\}\) of \(\mu\) where K runs through the compact subsets of G: If G is not compact and if \(\mu\) is normal or spread-out then \(\lim_{n}C(\mu^ n,K)=0\) (Theorems 3.3 and 3.5). There already exist profound investigations on the convergence of \((C(\mu^ n,K))_{n\geq 1}\) to 0 and on the rate of this convergence. Let us only mention the paper by \textit{K. Hofmann} and \textit{A. Mukherjea} [Math. Ann. 256, 535-548 (1981; Zbl 0471.60015)] where it is proved that \(\lim_ nC(\mu^ n,K)=0\) for all irreducible probability measures \(\mu\) on G provided G belongs to a large class of locally compact non-compact groups including the discrete, the Abelian and the Lie groups. Thus this result can in general not be compared with the results in the paper under review.