Eeect of Autarky Pruning on Random and Circuit Formulas: an Experimental Study

abstract Modoc was proposed by Van Gelder as an improvement to model elimination. The main contribution of Modoc is in its new pruning technique based on the concept of autarky, rst introduced by Monien and Speckenmeyer. Compared to programs based on model search, Modoc has since been observed to excel in speed on circuit formulas yet to fall behind on random formulas. This paper reports on a study conducted to explain this behavior in Modoc. The study is based on experiment in which the eeectiveness of the two pruning techniques used in Modoc|autarky pruning and lemma pruning|are examined on random and circuit formulas. We observe that the eeectiveness of autarky pruning diiers tremendously between the two classes of formulas. We also observe that for circuit formulas, autarkies, which are believed to be very few of, are more likely to be found by simplifying the formula for some partial truth assignment. This may lead to possible new ways to use autarkies to solve the satissability problem.