Decomposition branching for mixed integer programming

Abstract

We introduce a novel and powerful approach for solving certain classes of mixed integer programs (MIPs): decomposition branching. Two seminal and widely used techniques for solving MIPs, branch-and-bound and decomposition, form its foundation. Computational experiments with instances of a weighted set covering problem and a regionalized p-median facility location problem with assignment range constraints demonstrate its efficacy: it explores far fewer nodes and can be orders of magnitude faster than a commercial solver and an automatic Dantzig-Wolfe approach.