Relaxation methods for pessimistic bilevel optimization

TL;DR

This paper studies relaxation methods for solving smooth pessimistic bilevel optimization problems with convex lower levels, providing convergence analysis and numerical comparisons for various relaxation techniques.

Contribution

It introduces and compares relaxation methods for the KKT reformulation of pessimistic bilevel problems, a novel application beyond existing MPCC studies.

Findings

Convergence results for global and local solutions.

Numerical illustrations of relaxation algorithms.

Preliminary performance comparisons.

Abstract

We consider a smooth pessimistic bilevel optimization problem, where the lower-level problem is convex and satisfies the Slater constraint qualification. These assumptions ensure that the Karush-Kuhn-Tucker (KKT) reformulation of our problem is well-defined. We then introduce and study the (i) Scholtes, (ii) Lin and Fukushima, (iii) Kadrani, Dussault and Benchakroun, (iv) Steffensen and Ulbrich, and (v) Kanzow and Schwartz relaxation methods for the KKT reformulation of our pessimistic bilevel program. These relaxations have been extensively studied and compared for mathematical programs with complementatrity constraints (MPCCs). To the best of our knowledge, such a study has not been conducted for the pessimistic bilevel optimization problem, which is completely different from an MPCC, as the complemetatrity conditions are part of the objective function, and not in the feasible set of the problem. After introducing these relaxations, we provide convergence results for global and local optimal solutions, as well as suitable versions of the C- and M-stationarity points of our pessimistic bilevel optimization problem. Numerical results are also provided to illustrate the practical implementation of these relaxation algorithms, as well as some preliminary comparisons.