Code Improvements for Model Elimination Based Reasoning Systems

Abstract

We have been investigating ways in which the performance of model elimination based systems can be improved and in this paper we present some of our results. Firstly, we have investigated code improvements based on local and global analysis of the internal knowledge base used by the theorem prover. Secondly, we have looked into the use of a n lists to represent ancestor goal information to see if this gives a performance boost over the traditional two list approach. This n list representation might be thought of as a simple hash table. Thirdly, we conducted initial investigations into the effect of rule body literal ordering on performance. The results for the code improvements show them to be worthwhile, producing gains in some example problems. Using the n list representation gave mixed results: for some examples it improved execution speed, in others it degraded it. A rule body literal ordering that placed instantiated goals (including hypotheses) early in the bodies of rules showed an improvement in execution time.