Workshop Semidefinite Programming: Applications and Algorithms
ZIB-Berlin, November 15-17, 1998


Florian Jarre

"A QQP-Minimization Method for Semidefinite and Smooth Nonconvex Programs"

In many applications, semidefinite programs with nonlinear equality constraints arise. These problems are nonconvex, and the common interior-point results no longer apply. We give two such examples to emphasize the importance of this class of problems. To solve such problems we propose an interior-point method in which the usual linear systems of the Newton step and of the predictor step are replaced by simple quadratically constrained quadratic programs. These QQP's are of a special structure and can be solved directly. In addition, the QQP's allow for a special line search which effects that the algorithm is applicable to nonconvex programs. Some convergence results are given, and some preliminary numerical examples show the behaivour of the QQP-steps for nonconvex minimization.


helmberg@zib.de
Last Update: September 16, 1998