duality without constraint qualification in nonsmooth optimization在非光滑优化二元性约束资格.pdfVIP

DUALITY WITHOUT CONSTRAINT QUALIFICATION IN NONSMOOTH OPTIMIZATION S. NOBAKHTIAN Received 7 May 2005; Revised 26 July 2005; Accepted 1 January 2006 We are concerned with a nonsmooth multiobjective optimization problem with inequal- ity constraints. In order to obtain our main results, we give the definitions of the gener- alized convex functions based on the generalized directional derivative. Under the above generalized convexity assumptions, sufficient and necessary conditions for optimality are given without the need of a constraint qualification. Then we formulate the dual problem corresponding to the primal problem, and some duality results are obtained without a constraint qualification. Copyright © 2006 Hindawi Publishing Corporation. All rights reserved. 1. Introduction A multiobjective problem is a problem where two or more objective functions are to be minimized on an implicitly constrained feasible set. In such a problem for optimality conditions and duality results, we often deal with constraint qualifications. A constraint qualification assumes some regularity of the constraint functions near the optimal so- lution, in particular to exclude a cusp on the boundary of the feasible region. In some approaches to multiobjective optimization problems, the necessary conditions for effi- ciency are derived under the same constraint qualifications as in nonlinear programming with a scalar-valued objective function. Constraint qualifications are often assumed but do not always hold. Some of the well-known qualifications are the Kuhn-Tucker qualifi- cation, Slater qualifications [1], and Mangasarian-Fromovitz qualification [7]. Weir and Mond [8] defined dual problems for the scalar-valued programming prob- lems where the usual convexity requirement for duality was relaxed and a constraint qual- ification was not needed. They established their results using differentiability. Then in [9] Mond and Weir extended their results for multiobjective programs.