Continuous selections as parametrically defined integrals (Q1002814)

The author proves the following selection theorem for lower-semicontinuous complete- and convex-valued mappings into metrizable (not necessarily locally convex) vector spaces: Let \((M, \rho)\) be a metric subset of a topological vector space \(Y\), let \(F: X \to M\) be a lower semicontinuous multivalued mapping of a paracompact space \(X\), let the values of \(F\) be \(\rho\)-complete, and let every point \(x_0 \in X\) have a neighborhood \(W(x_0)\) such that the family \(\{ F(x) \}_{x \in W(x_0)}\) is uniformly locally convex. Then there exists a continuous single-valued selection \(f : X \to M\) of the mapping \(F\), that is, \(f(x) \in F(x)\) for each \(x \in X\). Here a family \(\{ C_\alpha \}_{\alpha \in A}\) of convex subsets of a topological vector space \(Y\) is called uniformly locally convex if for every neighborhood \(U\) of zero in \(Y\), there exists a neighborhood \(V\) of zero in \(Y\) such that the convex hull of the set \((B_\alpha - B_\alpha) \cap V\) is contained in \(U\) for each \(\alpha \in A\) and each precompact subset \(B_\alpha\) of \(C_\alpha\). A continuous single-valued selection for this theorem is obtained by the integral of a certain single-valued map with respect to a probability measure associated with a Milyutin map. This theorem extends not only the classical Michael selection theorem [\textit{E. Michael}, Ann. Math. (2) 63, 361--382 (1956; Zbl 0071.15902)], but also Dobrowolski and van Mill's selection theorem [\textit{T. Dobrowolski} and \textit{J. v. Mill}, Fundam. Math. 192, 215--232 (2006; Zbl 1120.54016)] for lower-semicontinuous mappings with closed, convex, finite-dimensional values into metrizable (not necessarily locally convex) vector spaces.