Large-Scale Nonlinear Optimization with Inexact Step Computations

Andreas Waechter
IBM Thomas J. Watson Research Center

We present an interior-point method in which the search direction can be computed by an iterative linear solver. We discuss termination criteria for the inexact solution of the augmented system that are tailored to the context of nonlinear optimization, particularly the need to handle non-convexity and loss of regularity. The practical performance of the method is demonstrated on several large-scale examples, derived from the discretization of PDE-constrained optimization problems.

This is joint work with Frank Curtis, Olaf Schenk and Johannes Huber.

Presentation (PDF File)

Back to Workshop II: Numerical Methods for Continuous Optimization