A multilevel algorithm for solving the trust-region subproblem

Résultats de recherche: Contribution à un journal/une revueArticle

114 Downloads (Pure)

Résumé

We present a multilevel numerical algorithm for the exact solution of the Euclidean trust-region subproblem. This particular subproblem typically arises when optimizing a nonlinear (possibly non-convex) objective function whose variables are discretized continuous functions, in which case the different levels of discretization provide a natural multilevel context. The trust-region problem is considered at the highest level (corresponding to the finest discretization), but information on the problem curvature at lower levels is exploited for improved efficiency. The algorithm is inspired by the method described in [J.J. More and D.C. Sorensen, On the use of directions of negative curvature in a modified Newton method, Math. Program. 16(1) (1979), pp. 1-20], for which two different multilevel variants will be analysed. Some preliminary numerical comparisons are also presented. © 2009 Taylor & Francis.
langue originaleAnglais
Pages (de - à)299-311
Nombre de pages13
journalOptimization Methods and Software
Volume24
Numéro de publication2
Les DOIs
étatPublié - 1 avr. 2009

Empreinte digitale

Trust Region Subproblem
Newton-Raphson method
Discretization
Modified Newton Method
Trust Region
Negative Curvature
Numerical Comparisons
Numerical Algorithms
Euclidean
Continuous Function
Objective function
Exact Solution
Curvature

Citer ceci

@article{3a0e5f5af1214b0aa96791b36696d9db,
title = "A multilevel algorithm for solving the trust-region subproblem",
abstract = "We present a multilevel numerical algorithm for the exact solution of the Euclidean trust-region subproblem. This particular subproblem typically arises when optimizing a nonlinear (possibly non-convex) objective function whose variables are discretized continuous functions, in which case the different levels of discretization provide a natural multilevel context. The trust-region problem is considered at the highest level (corresponding to the finest discretization), but information on the problem curvature at lower levels is exploited for improved efficiency. The algorithm is inspired by the method described in [J.J. More and D.C. Sorensen, On the use of directions of negative curvature in a modified Newton method, Math. Program. 16(1) (1979), pp. 1-20], for which two different multilevel variants will be analysed. Some preliminary numerical comparisons are also presented. {\circledC} 2009 Taylor & Francis.",
author = "Philippe Toint and D. Tomanos and {Weber Mendonca}, Melissa",
note = "Publication code : FP SB010/2007/04 ; QA 0002.2/001/07/04",
year = "2009",
month = "4",
day = "1",
doi = "10.1080/10556780802571467",
language = "English",
volume = "24",
pages = "299--311",
journal = "Optimization Methods and Software",
issn = "1055-6788",
publisher = "Taylor & Francis",
number = "2",

}

A multilevel algorithm for solving the trust-region subproblem. / Toint, Philippe; Tomanos, D.; Weber Mendonca, Melissa.

Dans: Optimization Methods and Software, Vol 24, Numéro 2, 01.04.2009, p. 299-311.

Résultats de recherche: Contribution à un journal/une revueArticle

TY - JOUR

T1 - A multilevel algorithm for solving the trust-region subproblem

AU - Toint, Philippe

AU - Tomanos, D.

AU - Weber Mendonca, Melissa

N1 - Publication code : FP SB010/2007/04 ; QA 0002.2/001/07/04

PY - 2009/4/1

Y1 - 2009/4/1

N2 - We present a multilevel numerical algorithm for the exact solution of the Euclidean trust-region subproblem. This particular subproblem typically arises when optimizing a nonlinear (possibly non-convex) objective function whose variables are discretized continuous functions, in which case the different levels of discretization provide a natural multilevel context. The trust-region problem is considered at the highest level (corresponding to the finest discretization), but information on the problem curvature at lower levels is exploited for improved efficiency. The algorithm is inspired by the method described in [J.J. More and D.C. Sorensen, On the use of directions of negative curvature in a modified Newton method, Math. Program. 16(1) (1979), pp. 1-20], for which two different multilevel variants will be analysed. Some preliminary numerical comparisons are also presented. © 2009 Taylor & Francis.

AB - We present a multilevel numerical algorithm for the exact solution of the Euclidean trust-region subproblem. This particular subproblem typically arises when optimizing a nonlinear (possibly non-convex) objective function whose variables are discretized continuous functions, in which case the different levels of discretization provide a natural multilevel context. The trust-region problem is considered at the highest level (corresponding to the finest discretization), but information on the problem curvature at lower levels is exploited for improved efficiency. The algorithm is inspired by the method described in [J.J. More and D.C. Sorensen, On the use of directions of negative curvature in a modified Newton method, Math. Program. 16(1) (1979), pp. 1-20], for which two different multilevel variants will be analysed. Some preliminary numerical comparisons are also presented. © 2009 Taylor & Francis.

UR - http://www.scopus.com/inward/record.url?scp=61349197442&partnerID=8YFLogxK

U2 - 10.1080/10556780802571467

DO - 10.1080/10556780802571467

M3 - Article

VL - 24

SP - 299

EP - 311

JO - Optimization Methods and Software

JF - Optimization Methods and Software

SN - 1055-6788

IS - 2

ER -