A Farkas lemma for difference sublinear systems and quasidifferentiable programming (Q1315435)

It is well-known that the Farkas-Minkowski theorem can be formulated for sublinear (i.e., positively homogeneous) functions \(f: \mathbb{R}^ n\to \mathbb{R}\) and \(g= (g_ 1, \dots, g_ m): \mathbb{R}^ n\to \mathbb{R}^ m\) by means of the subdifferentials in the following manner: The system \(\{g(x) \leq 0\), \(f(x)<0\}\) is not solvable if and only if \(0\in \partial f(0)+ \overline {\text{cone}} \bigcup^ m_{i=1} \partial g(0)\). In the present paper some generalizations of this theorem are presented where 1. \(g\) (partly also \(f\)) is a vector function from a Banach space in an ordered Banach space, 2. \(f\) and \(g\) are difference sublinear (DSL), i.e., the difference of sublinear functions, 3. inhomogeneous systems are discussed. All the results are used for the construction of asymptotic first order necessary optimality conditions of Karush-Kuhn-Tucker type for quasidifferentiable programming problems. In short examples some special problems of difference convex programming, quasidifferentiable programming, convex minimization, and semi-infinite convex (concave) programming are discussed. At the end of the paper the authors give some remarks to generalized Robinson stability conditions.