Lurch: a Lightweight Alternative to Model Checking

Formal methods, including model checking, is powerful but can be costly, in terms of memory, time, and modeling effort. Difficult problems, similar to the verification problem addressed by model checking, have been shown to exhibit a phase transition, suggesting that an easy range of problem instances might be solved much faster and with much less memory using a new type of model checker based on partial, random search. Here we compare the performance of Lurch, our prototype random search model checker, to the popular tools SMV and SPIN. The tools’ performance is compared for a range of randomly generated models based on a simple tic-tac-toe game. Our results suggest that Lurch might be used in place of existing tools for systems too large or too difficult to model small enough for conventional model checking.