Using Cost-Based Solution Densities from TSP Relaxations to Solve Routing Problems.

The Traveling Salesman Problem, at the heart of many routing applications, has a few well-known relaxations that have been very effective to compute lower bounds on the objective function or even to perform cost-based domain filtering in constraint programming models. We investigate other ways of using such relaxations based on computing the frequency of edges in near-optimal solutions to a relaxation. We report early empirical results on symmetric instances from TSPLIB.