Formula Normalizations in Verification

Abstract We apply and evaluate polynomial-time algorithms to compute two different normal forms of propositional formulas arising in verification. One of the normal form algorithms is presented for the first time. The algorithms compute normal forms and solve the word problem for two different subtheories of Boolean algebra: orthocomplemented bisemilattice (OCBSL) and ortholattice (OL). Equality of normal forms decides the word problem and is a sufficient (but not necessary) check for equivalence of propositional formulas. Our first contribution is a quadratic-time OL normal form algorithm, which induces a coarser equivalence than the OCBSL normal form and is thus a more precise approximation of propositional equivalence. The algorithm is efficient even when the input formula is represented as a directed acyclic graph. Our second contribution is the evaluation of OCBSL and OL normal forms as part of a verification condition cache of the Stainless verifier for Scala. The results show that both normalization algorithms substantially increase the cache hit ratio and improve the ability to prove verification conditions by simplification alone. To gain further insights, we also compare the algorithms on hardware circuit benchmarks, showing that normalization reduces circuit size and works well in the presence of sharing.