On the Integration of Heuristics with Column-Oriented Models for Discrete Optimization

Abstract: Column-oriented models are today common in the eld of discrete optimization, and there is an increasing interest in using such models as a basis for heuristic solution methods. The common theme of this work is to explore some possibilities to integrate heuristic principles and column-oriented models for discrete optimization problems.In the rst paper, we consider a resource allocation problem for cellular systems. We propose a strong column-oriented formulation and a corresponding column generation method, as well as an enhanced column generation scheme for this problem. The enhanced scheme is composed of a stabilization technique, an approximate column generation principle, and, for nding integer solutions, a heuristic that is embedded in the column generation scheme.The second paper provides a new and strong convexied formulation of the xed charge transportation problem. This formulation is obtained by integrating the concepts of Lagrangian decomposition and column generation. It is shown both theoretically and practically that this integration yields a formulation which is stronger than three other convexied formulations of the problem.