Continuous metric selection and multivariate approximation (Q909178)

Let G be a finite dimensional subspace of a normed space X and let \(P_ G\) denote the metric projection of X onto G; that is, \(P_ G(x)\equiv \{g\in G:\) \(d(x,G)=\| x-g\| \}\). The author shows that if S is an uncountable compact metric space, then C(S,G) is proximinal in C(S,X) if and only if \(P_ G\) has a continuous selection. In addition, he shows that if T is a compact metric space with only a finite number of cluster points, then C(T,G) is proximinal in C(T,X) if and only if \(P_ G\) satisfies the following semicontinuity property: for any \(x\in X\), there is \(g\in P_ G(x)\) such that if \(x_ n\to x\), then \(d(g,P_ G(x_ n))\to 0\). Using a result proved independently by \textit{A. Brown} [J. Approx. Theory 57, 48-68 (1989; Zbl 0675.41037)] and \textit{N. V. Zhivkov} [J. Approximation Theory 56, No.1, 59-71 (1989; Zbl 0675.41045)], the author shows that there exists a normed space X and a finite dimensional subspace G of X such that C(T,G) is proximinal in C(T,X) but such that \(P_ G\) does not have a continuous selection. A main tool in the proofs is a characterization by Brown of continuous selections.