A Trust Region Algorithm for Unconstrained Nonsmooth Optimization

欧宜贵,候定丕

DOI: https://doi.org/10.3969/j.issn.0253-2778.2001.06.003

2001-01-01

Abstract:The classical trust region algorithm for smooth nonlinear programming is extended to the nonsmooth case where the objective function is only lower semicontinuous in R n. At each iteration, the objective function is approximated by the first order approximate function that satisfies some assumptions, and the search direction is obtained by solving an easy problem. It is proven that each accumulation point of the iteration point sequence generated by this algorithm is a critical point of the primitive problem.