A remark on almost everywhere convergence of convolution powers (Q1305180)

For an \(L^p\) function \(f\) on a probability space \((X,B,m)\), an invertible \(m\)-preserving transformation \(\tau\), and a fixed probability measure \(\mu\) on the integers, define the convolution \(\mu f\) by \((\mu f)x=\sum_{k=-\infty}^{\infty} \mu(\{k\})f(\tau^kx)\). The question of when the convolution powers \(\mu^nf\) converge (and in what sense) has been widely studied. \textit{A. Bellow, R. Jones} and \textit{J. Rosenblatt} [Ergodic Theory Dyn. Syst. 14, No. 3, 415-432 (1994; Zbl 0818.28005)] gave sufficient conditions for pointwise a.e. convergence in terms of the Fourier transform of \(\mu\) and an angular ratio condition; the latter condition is shown to hold for example if \(\mu\) has zero expectation and finite second moment. In contract, they also showed that if \(\mu\) has non-zero expectation and finite second moment then strong sweeping out occurs, a dramatic failure to converge. Here the author discusses the relationship between three different technical conditions on \(\mu\) that relate to convergence questions. In addition, a specific question from the work of Bellow, Jones and Rosenblatt is answered, via the construction of a measure \(\mu\) with \(\sum k^s\mu(\{k\})<\infty\) for all \(s<2\), zero expectation, and with strong sweeping out.