Compact convergence of \(\sigma\)-fields and relaxed conditional expectation (Q5945634)

The paper addresses fine issues of so-called Young measures, and translates the results into the probabilistic language in terms of (relaxed) conditional expectations. The Young measures, i.e., parameterized weakly* measurable families of probability measures, are identified as elements of a completion of the space of \(\sigma\)-fields. The linear structure of the underlying Lebesgue space of \(p\)-integrable functions is partly extended onto Young measures by continuous extension by using the narrow convergence (i.e., the usual weak* convergence of Young measures or, in the probabilistic language, the convergence induced by Prohorov's metric). Compactness and Kuratowski's convergence of subsets in this topology is addressed in detail. Considering \(L^p\)-Young measures, a so-called strong-narrow topology is introduced by refining the narrow topology by adding \(|\cdot|^p\) into the usual space of test functions, \(p\geq 1\). A combination of Mosco's type of these two topologies (together with an explicit boundedness requirement) is used for convergence of subsets of Young measures. Minimizer of \(\nu\mapsto\int_0^1\int_R|f(t)-x|^2\nu_t(dx) dt\) on a given subset \(Y\) of \(L^2\)-Young measures for a given \(f\in L^2(0,1)\), called relaxed conditional expectations \(E(f|Y)\), is shown to exist, be unique and stable. Several examples illustrate the general results.