Assumption Generation for Asynchronous Systems by Abstraction Refinement

Compositional verification provides a way for deducing properties of a complete program from properties of its constituents. In particular, the assume-guarantee style of reasoning splits a specification into assumptions and guarantees according to a given inference rule and the generation of assumptions through machine learning makes the automatic reasoning possible. However, existing works are purely focused on the synchronous parallel composition of Labeled Transition Systems (LTSs) or Kripke Structures, while it is more natural to model real software programs in the asynchronous framework. In this paper, shared variable structures are used as system models and asynchronous parallel composition of shared variable structures is defined. Based on a new simulation relation introduced in this paper, we prove that an inference rule, which has been widely used in the literature, holds for asynchronous systems as long as the components’ alphabets satisfy certain conditions. Then, an automating assumption generation approach is proposed based on counterexample-guided abstraction refinement, rather than using learning algorithms. Experimental results are provided to demonstrate the effectiveness of the proposed approach.