Steering Augmented Lagrangian Methods
I propose an enhanced augmented Lagrangian algorithm for solving large-scale inequality constrained optimization problems. The novel feature of the algorithm is an adaptive update for the penalty parameter motivated by recently proposed techniques for exact penalty methods. This adaptive updating scheme improves the overall performance of the algorithm without sacrificing the strengths of the core augmented Lagrangian framework, such as its ability to be implemented matrix-free.
This latter strength is particularly important due to renewed interest in employing augmented Lagrangian algorithms for solving large-scale problems. I will provide preliminary computational results that illustrate how our method outperforms traditional augmented Lagrangian methods in terms of critical performance measures. This is joint work with Professor Frank E. Curtis from Lehigh University and Hao Jiang from Johns Hopkins University.