Optimality conditions for non-lipschitz generalized convex programming via clarke-rockafellar gradients
Thurai. Kugendran, L. McLinden · Optimization · 1998
For nonlinear programs with non-Lipschitz. generalized con\ex data functions. we develop various explicit first-order sufficient and /or necessary optimality conditions. These involve a natural generalization of the well known Karush-Kuhn-Tucker conditions, but with the familiar gradient condition modified so as to involve asymptotic (i.e. singular), as well as ordinary, Clarke-Rockafellar generalized gradients. In this way we cover situations in which the sets of ordinary generalized gradients are empty or unbounded, which can occur even at points where the functions are finite everywhere nearby. Along wit the use of asymptotic gradients, the novelty here lies in the identification of weak hypotheses on the data functions suitable for deriving such optimality results. In particular. the notions of protoconvexity is found to play a central role. along with the more familiar notions of quasiconvexity and’ pseudoconvexity