k-Step Relative Inductive Generalization

We introduce a new form of SAT-based symbolic model checking. One common idea in SAT-based symbolic model checking is to generate new clauses from states that can lead to property violations. Our previous work suggests applying induction to generalize from such states. While effective on some benchmarks, the main problem with inductive generalization is that not all such states can be inductively generalized at a given time in the analysis, resulting in long searches for generalizable states on some benchmarks. This paper introduces the idea of inductively generalizing states relative to k-step over-approximations: a given state is inductively generalized relative to the latest k-step overapproximation relative to which the negation of the state is itself inductive. This idea motivates an algorithm that inductively generalizes a given state at the highest level k so far examined, possibly by generating more than one mutually k-step relative inductive clause. We present experimental evidence that the algorithm is effective in practice.