Structure-aware computation of predicate abstraction
Alessandro Cimatti, Jori Dubrovin, Tommi A. Junttila, Marco Roveri · 2009
The precise computation of abstractions is a bottleneck in many approaches to CEGAR-based verification. In this paper, we propose a novel approach, based on the use of structural information. Rather than computing the abstraction as a single, monolithic quantification, we provide a structure-aware abstraction algorithm, based on two complementary steps. The first, highlevel step exploits the structure of the system, and partitions the abstraction problem into the combination of several smaller abstraction problems. This is represented as a formula with quantifiers. The second, low-level step exploits the structure of the formula, in particular the occurrence of variables within the quantifiers, and applies a set of low-level rewriting rules aiming at further reducing the scope of quantifiers. We experimentally evaluate the approach on a substantial set of benchmarks, and show significant speed ups compared to monolithic abstraction algorithms.