Regularity conditions in differentiable vector optimization revisited (Q1035923): Difference between revisions

The positive linear dependence is used in vector optimization both to restart already known regularity conditions and to get new results along this point of view, without convexity assumption. The new conditions are easily computable and useful in the development of algorithms, particularly in the definition of termination criteria. The positive linear dependence and independence allow a direct analysis of the interrelations among the gradients of the objective functions and the constraints. The distinct notions of regularity that may occur in the vector optimization field emerge naturally in connection with the Fritz-John necessary condition for optimality. Approaches involving separation sets, contingent cones, linearized approximations to the feasible set, etc., are related to the Fritz-John condition indirectly by means of alternative results like Motzkin's theorem. For the general vector optimization, a sufficient condition for weak regularity is given. Also, a necessary and sufficient condition for total regularity is proposed for the vector optimization problems. A very useful diagram of the interrelations of regularity conditions for the vector optimization problem concludes the paper.