Complementation in Abstract Interpretation

The reduced product of abstract domains is a rather well known operation in abstract interpretation. In this paper we study the inverse operation, which we call complementation. Such an operation allows to systematically decompose domains; it provides a systematic way to design new abstract domains; it allows to simplify domain verification problems, like correctness proofs; and it yields space saving representations for domains. We show that the complement exists in most cases, and we apply complementation to two well known abstract domains, notably to the Cousot and Cousot's comportment domain for analysis of functional languages and to the complex domain Sharing for aliasing analysis of logic languages.