Subgoal Alternation in Model Elimination

The order in which subgoals are selected and solved has a strong influence on the search space of model elimination procedures. A general principle is to prefer subgoals with few solutions over subgoals with many solutions. In this paper we show that the standard selection methods are not flexible enough to satisfy this principle. As a generalization of the standard paradigm the new method of subgoal alternation is presented and integrated into the theorem prover SETHEO. Among other advantages, subgoal alternation also provides more look-ahead information about the needed proof resources than the standard method; this information can be used for search pruning. The evaluation of the new technique on a large number of formulae shows a significant improvement in performance.