Characterization of the weakly efficient solutions in nonsmooth quasiconvex multiobjective optimization (Q2182861): Difference between revisions

The simultaneous minimization of a finite set of functions \(f_i\) in some nonempty closed convex \(X\subset\mathbb{R}^n\) is studied subject to the nonpositivity of some functions \(g_j\). All \(f_i\) and \(g_j\) are assumed continuous quasiconvex from \(\mathbb{R}^n\) to \(\mathbb{R}\). They derive six KKT-like necessary and one sufficient condition for a point to be a weakly efficient solution in terms of various kinds of subdifferentials (Greenberg-Pierskalla, Penot, star, Plastria, Gutiérrez, Suzuki-Kuroiwa). For example, if \(x\) is weakly efficient and all \(f_i\) and all \(g_j\) with \(g_j(x)=0\) are Lipschitz on their strict lower level sets at \(x\), then \(0\) is a sum of some vector normal to \(X\) at \(x\), some Penot subgradients at \(x\) of these active \(g_j\)'s and a convex combination of Plastria subgradients of the \(f_i\) at \(x\). Other similar necessary weak optimality conditions hold under other assumptions on the \(f_i\) and/or \(g_j\) combined with appropriate constraint qualifications, e.g. Slater like. On the other hand, if at some feasible \(x\), the null vector may be expressed as a sum of star subgradients of all the \(f_i\), Penot subgradients of all the active \(g_j\)'s and a normal vector to \(X\), then \(x\) is weakly efficient.