Extensions of Radstrom's lemma with application to stability theory of mathematical programming (Q1078809)

Let E be the Euclidean space, A, B convex sets in E, U the closed unit ball in E and h the Hausdorff distance on \(2^ E\), the power set of E. The author considers relationships between \(h(A+X,B+X)\) and h(A,B) where X is a subset of E. If \(A+\lambda U\) and \(B+\lambda U\) are closed for all \(\lambda >0\) then Radstrom's lemma states that \(h(A+X,B+X)=h(A,B)\) whenever X is bounded. The author shows that \(h(A+X,B+X)\leq h(A,B)\) under the Radstrom's assumptions (or if A, B are both closed) for any set \(X\subseteq E\). However, \(h(A+X,B+X)=h(A,B)\) whenever A,B\(\subseteq L\) and X is an arbitrary set in \(L^{\perp}\) (L is a linear subspace). Moreover the following theorem is proved. Let A and B be nonempty line- free closed convex sets with identical recession cones and \(A^ 1=\overline{co}\) (ext A), \(B^ 1=\overline{co}\) (ext B) both bounded. Then \(h(A,B)\leq h(A^ 1,B^ 1)\), where the right-hand-side distance is attained as \(d(x^*,y^*)\) where one of the following conditions must hold: (1) \(x^*\in ext A^ 1\) and \(y^*\in\) an exposed face of \(B^ 1\), (2) \(y^*\in ext B^ 1\) and \(x^*\in\) an exposed face of \(A^ 1\). In the first instance \(y^*\) is unique for a given \(x^*\in ext A^ 1\), while in the second instance \(x^*\) is unique for a given \(y^*\in ext B^ 1.\) Finally, the results are applied to study a stable behaviour of a point- to-set map defining a constraint set in a perturbed mathematical programming problem.