Decomposition for compositional verification

Compositional verification is based on the idea that the correctness check of a complex system can be divided into smaller verification tasks for its components. In this paper, we show how to decompose a specification into components when either no such decomposition is given, or when the given composition does not lend itself to an efficient compositional verification. Our decomposition is the starting point for an application of the L learning algorithm, generating assumptions for an assume-guarantee reasoning. We prove correctness of the decomposition as well as present experimental results using the model checker FDR2 as the teacher during learning.