A forward-backward abstraction refinement algorithm

refinement-based model checking has become a standard approach for efficiently verifying safety properties of hardware/software systems. Abstraction refinement algorithms can be guided by counterexamples generated from abstract transition systems or by fixpoints computed in abstract domains. Cousot, Ganty and Raskin recently put forward a new fixpoint-guided abstraction refinement algorithm that is based on standard abstract interpretation and improves the state-of-the-art, also for counterexample-driven methods. This work presents a new fixpoint-guided abstraction refinement algorithm that enhances the Cousot-Ganty-Raskin's procedure. Our algorithm is based on three main ideas: (1) within each abstraction refinement step, we perform multiple forward-backward abstract state space traversals; (2) our abstraction is a disjunctive abstract domain that is used both as an overapproximation and an underapproximation; (3) we maintain and iteratively refine an overapproximation M of the set of states that belong to some minimal (i.e. shortest) counterexample to the given safety property so that each abstract state space traversal is limited to the states in M.