Ant Colony Optimization for Tree Decompositions

Instances of constraint satisfaction problems can be solved efficiently if they are representable as a tree decomposition of small width. Unfortunately, the task of finding a decomposition of minimum width is NP-complete itself. Therefore, several heuristic and metaheuristic methods have been developed for this problem. In this paper we investigate the application of different variants of Ant Colony Optimization algorithms for the generation of tree decompositions. Furthermore, we extend these implementations with two local search methods and we compare two heuristics that guide the ACO algorithms. Our computational results for selected instances of the DIMACS graph coloring library show that the ACO metaheuristic gives results comparable to those of other decomposition methods such as branch and bound and tabu search for many problem instances. One of the proposed algorithms was even able to improve the best known upper bound for one problem instance. Nonetheless, for some larger problems the best existing methods outperform our algorithms.