A Simulated Annealing Based Approach for the Roman Domination Problem

The Roman Domination Problem is an NP-hard combinatorial optimization problem on an undirected simple graph. It represents scenarios where a resource shall be economically distributed over its vertices while guaranteeing that each vertex has either a resource itself or at least one neighbor with a sharable surplus resource. We propose several (meta-)heuristic approaches for solving this problem. First, a greedy construction heuristic for quickly generating feasible solutions is introduced. A special feature of this heuristic is an optional advanced tiebreaker. This construction heuristic is then randomized and combined with a local search procedure to obtain a greedy randomized adaptive search procedure (GRASP). As an alternative, we further propose a simulated annealing (SA) algorithm to improve the solutions returned by the construction heuristic. As we observe different pros and cons for the GRASP and the SA, we finally combine them into a simulated annealing hybrid, which interleaves phases of greedy randomized construction and phases of simulated annealing. All algorithms are empirically evaluated on a large set of benchmark instances from the literature. We compare to an exact mixed integer linear programming model that is solved by Gurobi as well as to a variable neighborhood search from the literature. In particular the simulated annealing hybrid turns out to yield on average the best results, making it a new state-of-the-art method for the Roman domination problem.