Investigation Of Single-Update Memoryless Cg-Methods

Abstract

Two new CG-methods are proposed in this work for nonlinear unconstrained optimization problems which are considered as memoryless Variable Metric (VM) methods. They are derived for the inexact line searches and evaluated numerically against Shanno's two memoryless VM-methods. The results indicate that, the respective new methods have practical improvements on the Shanno's methods. Moreover, they are clearly more efficient than the Perry's extended CG-method.