Numerical Experience with a Reduced Hessian Method for Large Scale Optimization

L. Biegler, J. Nocedal, C. Schmid and D. Ternet

The reduced Hessian SQP algorithm analyzed by Biegler, Nocedal and Schmid is developed in this paper into a practical method for large-scale optimization. The novelty of the algorithm lies in the incorporation of a correction vector that approximates the cross term ZWYp_y. This improves the stability and robustness of the algorithm without increasing its computational cost. The paper studies how to implement the algorithm efficiently, and presents a set of tests illustrating its numerical performance. An analytic example, showing the benefits of the correction term, is also presented.

Report OTC 97/06 Optimization Technology Center, July 1997