AN IMPROVED SUBGRADIENT METHOD FOR CONSTRAINED NONDIFFERENTIABLE OPTIMIZATION

Abstract

Polyak's subgradient method for constrained nondifferentiable optimization problems is modified in one respect to improve the computational efficiency. The two consecutive projection operations in Polyak method are combined into a single projection operation. The resulting algorithm has a convergence property which is strictly stronger than that of the original Polyak method. A computational test shows significant improvement both in the number of iterations and in the CPU time.