Convergence in approximation and nonsmooth analysis (Q1089554)

The authors examine some stability questions associated with problems in approximation theory and nonsmooth analysis. They study what happens to various sets that often appear in those fields, when one perturbs in some sense the data determining them. Such sensitivity analysis is important for developing efficient numerical algorithms. In the second section they investigate the continuity of the map \(A\to P_ A(x)\), where \(P_ a(x)\) is the set of all best approximations to x from A. This problem was first considered by \textit{B. Brosowski, F. Deutsch} and \textit{G. Nürnberger} [ibid. 29, 261-277 (1980; Zbl 0483.41033)], who considered a family \(\{A_ t\}_{t\in T}\) of subsets of a normed linear space X parametrized by a topological space T and studied the continuity of \(t\to P_{At}(x)\). Recently \textit{M. Tsukada} [(*) ibid. 40, 301-309 (1984; Zbl 0545.41042)] addressed the same problem but with a nonparametrized method. Namely, he allowed the sets \(\{A_ n\}_{n\geq 1}\) to converge in some sense to A, and he examined what happens to the sequence \(\{P_{An}(x)\}_{n\geq 1}\). However, he limited himself to reflexive, strictly convex, smooth Banach spaces in which case the set \(P_ a(x)\) for A, a nonempty, closed, convex subset of X is a singleton (i.e., A is a Chebyshev set). On the one hand, the authors generalize the work of Tsukada (*) and, on the other hand, provide several new results on this issue. In Section 3, they investigate analogous questions in the context of nonsmooth analysis. Finally, in Section 4, they pass to prediction sequences in the Lebesgue-Bochner space \(L^ 1_ x(\Omega,\Sigma)\). Here they examine the convergence of those sequences as we vary the sub- \(\sigma\)-field of \(\Sigma\), not necessarily in a monotone way, and the function f(\(\cdot)\) that has to be approximated.