Large-scale nonlinear programming

Researchers: Anders Forsgren and Göran Sporre, in cooperation with Philip E. Gill (UCSD).

Sponsor: The Swedish Natural Science Research Council (NFR).

The goal of this project is the development of computationally efficient methods for solving large sparse nonlinear optimization problems. We focus on methods that utilize second-derivatives, since we expect such methods to prove more robust and efficient than methods that only use first-derivative information.

Recent algorithmic work has been directed towards penalty-barrier methods for general nonlinear programming methods. A method, solving a primal-dual system of equations at each iteration, utilizing an augmented penalty-barrier merit function, has been developed. The current focus of Sporre's research is on primal-dual interior methods, convergence properties in particular. A report relating this research to interior methods for quadratic programming has been written [R8]. A paper on optimality conditions for semidefinite programming has been published [A12].


Research 1999/2000
KTH optimization laboratory
Models and methods for structural optimization
Last update: February 21, 2001 by Anders Forsgren, anders.forsgren@math.kth.se.