Automatic Synthesis of Assumptions for Compositional Model Checking

We present a new technique for automatically synthesizing the assumptions needed in compositional model checking. The compositional approach reduces the proof that a property is satisfied by the parallel composition of two processes to the simpler argument that the property is guaranteed by one process provided that the other process satisfies an assumption A. Finding A manually is a difficult task that requires detailed insight into how the processes cooperate to satisfy the property. Previous methods to construct A automatically were based on the learning algorithm L∗, which represents A as a deterministic automaton and therefore has exponential worst-case complexity. Our new technique instead represents A as an equivalence relation on the states, which allows for a quasi-linear construction. The model checker can therefore apply compositional reasoning without risking an exponential penalty for computing A.