Approximate duality (Q2483045): Difference between revisions

The authors consider a convex programming problem involving inequality, equality and set constraints with functions defined on a Hausdorff locally convex topological vector space together with its Lagrangian dual. They prove the basic connections between the set of \(\epsilon\)-minimizers, the set of \(\epsilon\)-Kuhn-Tucker vectors, the set of points satisfying \(\epsilon\)-Kuhn-Tucker conditions and the set of \(\epsilon\)-saddle points, and show how the optimal value of the primal problem relates to the value of the Lagrangian at approximate minimizers and approximate Kuhn-Tucker vectors. These results are generalizations of some results of Chap. 28 in the book by \textit{R. T. Rockafellar} [Convex analysis. (Princeton), N. J.: Princeton University Press, (1970; Zbl 0193.18401)]. They also show how this framework can be used for quadratic convex programming problems and apply the results to support vector machines from learning theory.