An Empirical Study of Probabilistic Arc Consistency

We report on an empirical evaluation of a new probabilistic heuristic for constructive search in constraint satisfaction problems. The heuristic is based on the estimation of solution probability. We show empirically that this heuristic is more accurate than related heuristics, and reduces the number of consistency checks and backtracks in constructive search by up to several orders of magnitude. Our results also show that the time required to estimate solution probabilities is less than the time required for search using other well-known heuristics as the problem size increases.