Dissipation of convolution powers in a metric group (Q2385606): Difference between revisions

Let \(\mu\) be a regular probability measure on a locally compact group \(G\), and \(\mu^n\) denote the \(n\)-th convolution power. \textit{Y. Derriennic} [Ann. Inst. Henri Poincaré, Nouv. Sér., Sect. B 12, 111--129 (1976; Zbl 0353.60075)] and \textit{A. Mukherjea} [Z. Wahrscheinlichkeitstheor. Verw. Geb. 33, 273--284 (1976; Zbl 0304.60004)] proved that if \(\mu\) is not supported on a compact subgroup, then \(\mu^n(C)\to 0\), as \(n\to \infty\), for every compact set \(C\subset G\). The author gives an example of a metric group, for which the above property is not valid, and finds a necessary and sufficient condition for its validity. Another example shows that a noncompact metric group \(G\) can also support a probability measure \(\mu\), such that \(\text{supp} \mu =G\), and the concentration functions of \(\mu\) do not converge to zero, again in contrast to the case of locally compact groups [\textit{W. Jaworski, J. Rosenblatt} and \textit{G. Willis}, Math. Ann. 305, No. 4, 673--691 (1996; Zbl 0854.43001)]. Sufficient conditions for the convergence of the concentration functions to zero are found.