A selection principle with applications in convex analysis and optimization (Q1118149)

A convex function \(\phi\) defined on a locally convex vector space E with values in an ordered locally convex vector space F is known to be subdifferentiable at each of its points of continuity provided that (i) F is order-complete or (ii) F has the monotonic sequence property, the positive cone \(F_+\) is based, and E is an Asplund space. In the present paper we establish the subdifferentiability of \(\phi\) in the case where neither any order-completeness assumption is imposed on F nor E has to be a differentiability space, but where every positive linear functional \(\tilde z\in F^{'*}\) is assumed continuous for \(\sigma (F',F)\). This includes the case where F is a reflexive (F)-space as well as the case where \(F_+\) is weakly compactly well-based. Our method of proof makes use of a selection principle for linear set-valued mappings. Part two of our paper presents a second application of the selection method. Here we obtain an infinite dimensional version of the Kuhn-Tucker theorem for convex programming.