A method to eliminate redundant case-splittings in MGTP and its evaluation

We present a new method to eliminate redundant inferences caused by case-splitting in MGTP. There are two types of redundancies eliminated by the method: One is that the same sub-proof tree may be generated at several descendant nodes after a case-splitting occurs. Another is caused by useless model candidate extensions with irrelevant non-Horn clauses. Both types of redundancies can be eliminated by combining the folding-up function that conveys unit lemmata obtained from solved descendant nodes to a proper antecedent node and the NHM function that suppresses useless model extensions. The method has been implemented in MGTP. We evaluate its effects on a UNIX workstation for some typical problems taken from the TPTP problem library.