Tearing based automatic abstraction for CTL model checking

Abstract

In this paper we present the tearing paradigm as a way to automatically abstract behavior to obtain upper and lower bound approximations of a reactive system. We present algorithms that exploit the bounds to perform conservative ECTL and ACTL model checking. We also give an algorithm for false negative (or false positive) resolution for verification based on a theory of a lattice of approximations. We show that there exists a bipartition of the lattice set based on positive versus negative verification results. Our resolution methods are based on determining a pseudo-optimal shortest path from a given, possibly coarse but tractable approximation, to a nearest point on the contour separating one set of the bipartition from the other.