A self correcting conjugate gradient algorithm

Abstract

In this paper we develop a new procedure for constructing a conjugate gradient direction equation. The new equation is a linear combination of two orthogonal vectors one of which is the negative gradient. This procedure is reduced to the method of Polak-Ribiere whenever line search is perfectly accurate. Otherwise, as reflected by our computational results, the method is more effective than any other conjugate gradient algorithm we have tested.

Keywords:

• Conjugate gradient
• line search
• unconstrained optimization
• descent direction

C.R. Categories:

• 3.15
• 5.12
• 5.15
• 5.41