Bertrand Mazure, Lakhdar SaÏs, Éric Grégoire
In this paper, tabu search for SAT is investigated from an experimental point of view. To this end, TSAT, a basic tabu search algorithm for SAT, is introduced and compared with Selman et al. Random Walk Strategy GSAT procedure, in short RWS-GSAT. TSAT does not involve the additional stochastic process of RWS-GSAT. This should facilitate the understanding of why simple local search methods for SAT work. It is shown that the length of the tabu list plays a critical role in the performance of the algorithm. Moreover, surprising properties about the (experimental) optimal length of the tabu list are exhibited, raising interesting issues about the nature of hard random SAT problems.