Generalized Pattern Search methods for a class of nonsmooth optimization problems with structure

https://doi.org/10.1016/j.cam.2008.10.047Get rights and content
Under an Elsevier user license
open archive

Abstract

We propose a Generalized Pattern Search (GPS) method to solve a class of nonsmooth minimization problems, where the set of nondifferentiability is included in the union of known hyperplanes and, therefore, is highly structured. Both unconstrained and linearly constrained problems are considered. At each iteration the set of poll directions is enforced to conform to the geometry of both the nondifferentiability set and the boundary of the feasible region, near the current iterate. This is the key issue to guarantee the convergence of certain subsequences of iterates to points which satisfy first-order optimality conditions. Numerical experiments on some classical problems validate the method.

Keywords

Generalized Pattern Search
Nonsmooth optimization
Linear constraints
Clarke’s differential calculus

Cited by (0)

This work was supported by Dip. Energetica, Univ. di Firenze, and MIUR, Rome, Italy.