New Branch-and-Bound Rules for Linear Bilevel Programming

Abstract

A new branch-and-bound algorithm for linear bilevel programming is proposed. Necessary optimality conditions expressed in terms of tightness of the follower’s constraints are used to fathom or simplify subproblems, branch and obtain penalties similar to those used in mixed-integer programming. Computational results are reported and compare favorably to those of previous methods. Problems with up to 150 constraints, 250 variables controlled by the leader, and 150 variables controlled by the follower have been solved.