An exact algorithm for the bilevel mixed integer linear programming problem under three simplifying assumptions

Abstract

We present an exact algorithm for the bilevel mixed integer linear programming (BMILP) problem under three simplifying assumptions. Although BMILP has been studied for decades and widely applied to various real world problems, there are only a few BMILP algorithms. Compared to these existing ones, our new algorithm relies on fewer and weaker assumptions, explicitly considers finite optimal, infeasible, and unbounded cases, and is proved to terminate finitely and correctly. We report results of our computational experiments on a small library of BMILP test instances, which we created and made publicly available online.