Sets of best \(L_ 1\) approximants (Q914092): Difference between revisions

For \(X=L_ 1(\Omega,{\mathcal A},\mu)\), C a proximinal convex subset of X and \(f\in X\), denote by P(f) the set of all best \(L_ 1\)-approximants to f in C. The paper is concerned with the following problem: If \(f_ 1,f_ 2\in X\) are close, how close are the sets \(P(f_ 1)\) and \(P(f_ 2)\) in the Hausdorff metric \(d_ h?\) By a result of \textit{T. Shintani} and \textit{T. Ando} [Z. Wahrsch. Verw. Gebiete 33,33-39 (1975; Zbl 0298.41016)] if \({\mathcal B}\) is a sub-\(\sigma\)-algebra of \({\mathcal A}\) then \(C=L_ 1(\Omega,{\mathcal B},\mu)\) is proximinal in X. The authors show by an example that in this case, \(f_ n\to f\) in \(L_ 1\) doesn't imply \(d_ h(P(f_ 1),P(f_ n))\to 0\) but, for every \(\epsilon >0\) there exists \(N>0\) such that \(d_ h(g,P(f))<\epsilon\) for all \(g\in P(f_ n)\) and all \(n\geq N\). Another proximinal set studied in this paper is that of all nondecreasing functions in \(L_ 1[0,1]\). In this case \(d_ h(P(f)\), \(P(g))<\epsilon\), in the uniform metric, for all \(f,g\in L_ 1[0,1]\) such that \(| f(x)-g(x)| <\epsilon /8,\) for all \(x\in [0,1]\).