Nonlinear programming via König's maximum theorem (Q328441)

The authors establish a result which is an equivalent reformulation of the finite-dimensional Hahn-Banach theorem. This result generalizes the König Maximum theorem. The latter result can be stated as follows: Assume there is a finite family \(S_1,\dots, S_n\) of sublinear functions defined in a vector space \(E\), a nonempty and convex subset \(X\) of \(E\), and a linear function \(L\) such that \[ L\leq \max_{i=1,\dots,n} S_i \text{ on } X. \] Then there is a convex combination of the \(S_i\)'s that majorizes \(L\). The authors show how (a generalization of) this theorem can be used to establish some central results in nonlinear programming that are improvements upon their classical versions. The analysis uses infsup-convexity, which is a natural notion of convexity arising in the context of minimax inequalities. The authors prove that their Theorem 2.3 is equivalent to the finite dimensional Separation theorem. As a consequence, the authors establish sharp duality between optimal solutions of constrained variational problems and Lagrange multipliers, as well as some results of Karush/Kuhn-Tucker and Fritz-John type for nonlinear programming problems.