Analysis on an improved global convergence for a spectral conjugate gradient method

Release time:2016-04-22

Affiliation of Author(s):Central South University

Journal:International Journal of Applied Mathematics and Statistics

Key Words:Bench-mark problems - Conjugate gradient - Conjugate gradient algorithms - Convergence results

Abstract:In this paper, we establish the theory of global convergence for a spectral conjugate gradient algorithm recently developed by Z. Wan etc. An assumption, that the inequalities 0 < gTk gk-1 &le 2&par;gk&par;2 are satisfied for any k, is first investigated by numerical experiments. It is shown that such assumption holds only for k large enough in solving some benchmark problems, not for all ones. Another contribution of this paper is to obtain the same convergence result under some weaker assumptions.

First Author:Songhai Deng

Indexed by:Applied Research

Correspondence Author:Zhong Wan

Document Type:J

Volume:31

Issue:1

Page Number:20-26

ISSN No.:09731377

Translation or Not:no

Date of Publication:2013-04-01