Superlinear convergence of a reduced BFGS method with piecewise line-search and update criterion

Jean Charles Gilbert 1
1 PROMATH - Mathematical Programming
Inria Paris-Rocquencourt
Abstract : We show the q-superlinear convergence of a reduced BFGS method for equality constrained problems, using eventually only one constraint linearization per iteration. The local method is globalized either with a standard arc-search or when an update criterion is satisfied, with a piecewise line-search. The aim of the latter technique is to realize generalized Wolfe conditions, which allow the algorithm to maintain naturally the positive definiteness of the generated matrices. We show that if the sequence of iterates converges, the convergence is q-superlinear. No assumption is made on the speed of convergence of the sequence of iterates or on the boundedness of the sequence of generated matrices. The main difficulty is to show that the ideal step-size is accepted after finitely many steps.
Type de document :
Rapport
[Research Report] RR-2140, INRIA. 1993
Liste complète des métadonnées

https://hal.inria.fr/inria-00074532
Contributeur : Rapport de Recherche Inria <>
Soumis le : mercredi 24 mai 2006 - 15:41:36
Dernière modification le : vendredi 25 mai 2018 - 12:02:05
Document(s) archivé(s) le : mardi 12 avril 2011 - 17:25:01

Fichiers

Identifiants

  • HAL Id : inria-00074532, version 1

Collections

Citation

Jean Charles Gilbert. Superlinear convergence of a reduced BFGS method with piecewise line-search and update criterion. [Research Report] RR-2140, INRIA. 1993. 〈inria-00074532〉

Partager

Métriques

Consultations de la notice

233

Téléchargements de fichiers

57