A Theory of Hints in Model Checking

摘要

Model checking, in particular symbolic model checking, has proved to be extremely successful in establishing properties of finite state programs. In most cases, the proven properties are safety properties stating that the program never executes outside a specified set of states. But another important class of properties, progress (liveness) properties, which state that program execution eventually reaches some specified set of states, has been difficult to model-check as they, typically, involve doubly-nested fixpoint computations. In this paper, we propose that progress properties can be checked more efficiently if they are accompanied by hints as to why they hold. We develop a theory in which hints are given as regular expressions over the actions of the program. We derive a number of inference rules and algebraic properties of hints. Empirical evidence suggests that hints can significantly improve the efficiency of model checking.