Memory-Efficient Tactics for Randomized LTL Model Checking

We study model checking of LTL properties by means of random walks, improving on the efficiency of previous results. Using a randomized algorithm to detect accepting paths makes it feasible to check extremely large models, however a naive approach may encounter many non-accepting paths or require the storage of many explicit states, making it inefficient. We study here several alternative tactics that can often avoid these problems. Exploiting probability and randomness, we present tactics that typically use only a small fraction of the memory of previous approaches, storing only accepting states or an arbitrarily small number of “token” states visited during executions. Reducing the number of stored states generally increases the expected execution time until a counterexample is found, but we demonstrate that the trade-off is biased in favor of our tactics. By applying our memory-efficient tactics to scalable models from the literature, we show that the increase in time is typically less than proportional to the saving in memory and may be exponentially smaller.