The gap between probability and prevalence: Loneliness in vector spaces (Q4517482)

Let \(V\) be an infinite-dimensional, complete, metric vector space. A set \(S\) is called \textit{shy} if it is a subset of a universally measurable set \(S'\), for which \(\mu(S'+v)= 0\) for every \(v\in V\) and for some finite measure \(\mu\) on \(V\). This is a slight extension of related concepts of ``small'' set due to \textit{J. P. R. Christensen} [Topology and Borel structure. Amsterdam etc.: North-Holland (1974; Zbl 0273.28001)] and due to \textit{B. R. Hunt}, \textit{T. Sauer} and \textit{J. A. Yorke} [Bull. Am. Math. Soc., New Ser. 27, No. 2, 217--238 (1992; Zbl 0763.28009)]. The complement of a \textit{shy} set is a \textit{prevalent} set. It is shown that there is a gap between prevalence and likelihood. First, for any probability \(\mu\) on \(V\), there is a shy set \(C\) with \(\mu(C)= 1\). Second, when \(V\) is locally convex, any i.i.d. sequence with law \(\mu\) repeatedly visits neighborhoods of only a shy set of points if the neighborhoods shrink to \(0\) at any rate. It is commented why the results look natural in an infinite-dimensional space. The other comment is the connections to naive stochastic search in a Banach space, which is discussed in [\textit{P. W. Millar}, Prog. Probab. 30, 497--510 (1992; Zbl 0790.62049)].