A New Relaxation for Tree-Based Problems and Minimum Power-Cost Spanning Trees

Abstract

In this paper, we investigate lower bounds of tree-based optimization problems in order to obtain effective exact algorithms, such as branch-and-bound algorithms. Our new approach inherits the development of dynamic programming algorithms for constrained shortest path problems, as they occur as subproblems in Lagrangian relaxation algorithms and column generation-based algorithms for variants of the Vehicle Routing Problem. In the q-route relaxation, paths must satisfy a capacity constraint while the elementarity constraint is relaxed, that is, paths may contain cycles. An analogue of q-routes for tree optimization problems are q-arbs, a structure that relaxes elementarity for arborescences. We introduce a generalized formulation of q-arbs for a broad class of tree-based problems, and apply the neighborhood restrictions of so-called ng-routes to them to obtain tighter bounds. We apply the new dynamic programming approach to the Minimum Power-Cost Spanning Tree Problem and show empirically that the resulting bounds are often better than traditional LP-based lower bounds of (mixed) integer programming models.