On the splitting problem for selections (Q1018354): Difference between revisions

For a space \(Y\), the set of all non-empty subsets of \(Y\) is denoted by \(2^Y\). For a space \(X\) and a metric space \((Y, d)\), a mapping \(f : X \to Y\) is called a selection (resp.\ \(\varepsilon\)-selection for \(\varepsilon>0\)) of a mapping \(F : X \to 2^Y\) if \(f(x ) \in F(x)\) (resp.\ \(\inf \{ d(f(x) ,y ) : y \in F(x)\} <\varepsilon\)) for each \(x \in X\). The splitting problem for selections was posed by \textit{D. Repovš} and \textit{P. V. Semenov} [J.\ Math.\ Anal.\ Appl.\ 334, 646--655 (2007; Zbl 1128.54012)] and stated as follows: Let \(X\) be a space, \(Y\), \(Y_1\) and \(Y_2\) Banach spaces and \(F_1 :X\to 2^{Y_1}\) and \(F_2 :X\to 2^{Y_2}\) (lower semi-) continuous mappings with closed convex images. Let \(L:Y_1\oplus Y_2\to Y\) be a continuous linear surjection and \(f:X\to Y\) a continuous selection of the composite mapping \(L(F_1;F_2) : X \to 2^Y\) defined by \(L(F_1;F_2)(x)=\{ y\in Y : y=L(y_1;y_2),\; y_1\in F_1(x),\; y_2\in F_2(x)\} \) for each \(x \in X\). Then, do there exist continuous selections \(f_i : X \to 2^{Y_i}\) of \(F_i\), \(i=1,2\), such that \(f(x) =L(f_1(x), f_2(x))\) for each \(x \in X\)? In the paper under review, the authors prove that for finite-dimensional Banach spaces \(X\), \(Y\), \(Y_1\) and \(Y_2\), the answer to the splitting problem is affirmative if the graph of the mapping \((F_1; F_2) : X \to 2^{Y_1 \oplus Y_2}\) is a \(P\)-set in the sense of \textit{M. V. Balashov} [Math. Notes 71, No.~3, 295--304 (2002; Zbl 1052.52004)], or if the mappings \(F_i\), \(i=1,2\) are continuous in the Hausdorff metric, the images of \(F_i\), \(i=1,2\) are strictly convex or a single-point, and the kernel of \(L\) is not parallel to \(Y_i\), \(i=1,2\). On the other hand, they show there are compact convex subsets \(A\) and \(B\) of \(\mathbb{R}^3\) such that there is no pair of continuous mappings \(a: A+B \to A\) and \(b: A+B \to B\) on the Minkowski sum \(A+B\) of \(A\) and \(B\) satisfying \(a(c) + b(c) =c\) for each \(c \in A +B\), which answers a question of \textit{D. Repovš} and \textit{P. V. Semenov} [op. cit.] and shows that the splitting problem may have no solution even if \(X\) is compact and \(Y\), \(Y_1\) and \(Y_2\) are \(\mathbb{R}^3\). Furthermore, the authors give an affirmative solution for the approximate splitting problem for Lipschitz continuous selections in Hilbert spaces: Let \(X\) be a convex subset of a Hilbert space, \(Y\), \(Y_1\) and \(Y_2\) Hilbert spaces, and \(F_1 :X\to 2^{Y_1}\) and \(F_2 :X\to 2^{Y_2}\) uniformly continuous mappings (in the Hausdorff metric) with closed convex bounded images such that \(\sup_{x \in X} \text{diam} (F_1(x); F_2(x)) < \infty\). Let \(L:Y_1\oplus Y_2\to Y\) be a continuous linear surjection and \(f:X\to Y\) a Lipschitz continuous selection of \(L(F_1;F_2)\). Then, for each \(\varepsilon>0\), there exist Lipschitz continuous \(\varepsilon\)-selections \(f_i : X \to 2^{Y_i}\) of \(F_i\), \(i=1,2\), such that \(f(x) =L(f_1(x), f_2(x))\) for each \(x \in X\).