scientific article; zbMATH DE number 1217672

MaRDI QIDQ4215028zbMATH OpenFDO

Authors Yi Xiao, E. K.-w. Chu

Publication date 12 April 1999

zbMATH Keywords

convergenceCUTEdescent methodsnonmonotone algorithmsnumerical experimentstrust region algorithm

Mathematics Subject Classification ID

Numerical mathematical programming methods (65K05) Nonlinear programming (90C30)

Review text

Nonmonotone algorithms developed so far include the corresponding descent methods as special cases. As convergence analysis commonly deals with worst case scenarios, nonmonotone algorithms are not expected to have theoretically better convergence properties than descent methods. Judgement on nonmonotone algorithms has to be based on extensive numerical experimentation. Fortunately, CUTE [cf. \textit{I. Bongartz}, \textit{A. R. Conn}, \textit{N. Gould} and \textit{Ph. L. Toint}, ACM Trans. Math. Softw. 21, No. 1, 123-160 (1995; Zbl 0886.65058)] provides an ideal test environment. The purpose of this paper is to propose a new nonmonotone stabilization trust region (NSTR) algorithm, modifying earlier work [cf. \textit{F. Zhou} and \textit{Y. Xiao}, Computing 53, No. 2, 119-136 (1994; Zbl 0834.90127)], and assess our NSTR algorithm in CUTE. For the entire collection see [Zbl 0897.00034].

Recommendations

• Nonmonotonic trust region algorithm
• Nonmonotone trust region algorithm for unconstrained optimization problems
• An improved nonmonotonic trust region algorithm
• A nonmonotone trust region method based on simple quadratic models

This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4215028)