Improving Performance of Automated Forward Deduction System EnCal on Shared-Memory Parallel Computers

The performance of an automated forward deduction system is crucial to its applicability. Since a forward deduction system working for discovery has no explicitly specified proposition or theorem given previously as goal, it often deduces many redundant intermediates, i.e., instances of those that have previously deduced. Therefore, how to reduce redundant intermediates is a general and difficult issue for any forward deduction system. In particular, since a parallel forward deduction system has multiple threads, the issue is more crucial to its performance. In this paper, we present a new algorithm to detect redundant intermediates in order to improve performance of EnCal, an automated forward deduction system for general-purpose entailment calculus. We have implemented EnCal based on the new algorithm on a shared-memory parallel computer and our experiment showed this algorithm is effective.