Experiments with Multiple Abstraction Heuristics in Symbolic Verification

Abstract

AbstractIn this work we investigate a symbolic heuristic search algorithm in a model checker. The symbolic search algorithm is built on a system that manipulates binary decision diagrams (BDDs). We study the performance of the search algorithm in terms of the number of BDD operations, size of the BDDs, number of nodes they contain and run-time. We study the heuristic distribution of the state space, we measure effort by computing the mean heuristic value, and we compare single and multiple heuristics. In the case of multiple heuristics, we consider admissible and non-admissible merge strategies. We experiment on problems from a variety of domains. We find that multiple heuristics can perform significantly worse than single heuristics in symbolic search in at least one domain. In general, the effect of the heuristics on the symbolic search in the different domains varies markedly, and we conjecture that the different behaviour is caused by intrinsic differences in the characteristics of the state space.KeywordsModel CheckHeuristic SearchGoal StateSolution PathBinary Decision DiagramThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.