Article ID: | iaor2001995 |
Country: | Germany |
Volume: | 87 |
Issue: | 2 |
Start Page Number: | 317 |
End Page Number: | 350 |
Publication Date: | Jan 2000 |
Journal: | Mathematical Programming |
Authors: | Todd M.J., Wagner M. |
This paper investigates quasi-Newton updates for equality-constrained optimization. Using a least-change argument we derive a class of rank-3 updates to approximations of the one-sided projection of the Hessian of the Lagrangian which keeps the appropriate part symmetric (and possibly positive definite). By imposing the usual assumptions we are able to prove 1-step superlinear convergence for one of these updates. Encouraging numerical results and comparisons with other previously analyzed updates are presented.