Non-linear optimization and large-scale problems

Jean Charles Gilbert (1)

1 INRIA Rocquencourt

This paper recalls some basic concepts of unconstrained optimization techniques and reviews some useful methods for solving large-scale problems: conjugate gradient methods without critical line-searches, limited memory quasi-Newton methods and the partitioned quasi-Newton method. We shall discuss their efficiency, their convergence properties and their respective advantages and disadvantages.

Mots clés: conjugate gradient; large-scale optimization; limited memory quasi-Newton method; partitioned quasi-Newton method; unconstrained optimization

Domaines: Mathématiques [math]

Jean Charles Gilbert. Non-linear optimization and large-scale problems. Engineering Optimization, 1991, 18 (1-3), pp.5-21. ⟨10.1080/03052159108941009⟩. ⟨hal-04135649⟩