On Lagrange multipliers of trust-region subproblems (Q1014888): Difference between revisions
From MaRDI portal
Changed an Item |
Changed an Item |
||
Property / describes a project that uses | |||
Property / describes a project that uses: GQTPAR / rank | |||
Normal rank |
Revision as of 07:43, 28 February 2024
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | On Lagrange multipliers of trust-region subproblems |
scientific article |
Statements
On Lagrange multipliers of trust-region subproblems (English)
0 references
29 April 2009
0 references
Due to their good convergence properties, trust-region methods are widely used. When computing the next iterate, one solves an unconstrained trust-region subproblem that involves also Lagrange multipliers. Several solution approaches, as for example the Steihaug-Toint CG method [\textit{T. Steihaug}, SIAM J. Numer. Anal. 20, 626--637 (1983; Zbl 0518.65042); \textit{Ph. L. Toint}, in: Proc. IMANA Conf., Reading 1980, 57--88 (1981; Zbl 0463.65045)] only require approximations of the Lagrange multipliers. This paper proves a property of the Lagrange multipliers generated when using a Krylov subspace method for the solution of the subproblems. That is, it is shown that the Lagrange multipliers grow monotonically. The paper also suggests improved solution techniques that take this additional property of the Lagrange multipliers into account.
0 references
trust region subproblem
0 references
Lagrange multipliers
0 references
Krylov spaces
0 references