Optimization and Systems Theory Seminar
Friday, December 15, 2000, 11.00-12.00, Room 3721, Lindstedtsv. 25


Andreas Wächter
Department of Chemical Engineering
Carnegie Mellon University
Pittsburgh, Pennsylvania
USA
E-mail: andreasw@andrew.cmu.edu

An interior point algorithm for large-scale nonlinear programming

A primal-dual interior point algorithm for nonlinear optimization problems (NLPs) will be presented. The method follows a line search approach, where the search directions are computed using a non-orthogonal decomposition scheme. This allows an efficient exploitation of problem structures as they are encountered in many chemical engineering applications. We will present numerical results for dynamic optimization problems, where the NLPs obtained by discretization using collocation on finite elements have up to 800,000 variables.

In this talk we will discuss the issue of global convergence in more detail. Many of the current interior point NLP methods can be shown to fail to converge to a feasible point on a simple, well-posed problem. We will present a line search procedure, where the traditional merit function is replaced by the notion of a filter which has recently been proposed by Fletcher and Leyffer. This method guarantees convergence to a feasible point, and in numerical tests seems generally more efficient and robust than our previous merit function approach.


Calendar of seminars
Last update: December 6, 2000 by Anders Forsgren, anders.forsgren@math.kth.se.