Static analyses are mostly designed to show the absence of bugs: if the analysis reports no alarms then the program won’t exhibit any unwanted behaviours. To this aim they manipulate over-approximations of program semantics and, inevitably, they often report some false alarms. Recently, O’Hearn proposed Incorrectness Logic, that is based on under-approximations, as a formal method to find bugs that only reports true alarms. In this paper we aim to answer one important question raised by O’Hearn, namely which role can Abstract Interpretation play for the development of under-approximate tools for bug catching. In principle, Abstract Interpretation based static analyses can be defined for computing over-approximations as well as under-approximations, but in practice, most techniques exploited the former while few attempts developed the latter. To show why it is difficult to design effective under-approximation abstract domains, we first propose the new definitions of non emptying functions and highly surjective function family and then we formally prove the limits of under-approximation analysis by showing the non existence of abstract domains able to approximate such functions in a non trivial way. Our results outline the limits of under-approximation Abstract Interpretation and clarify, for the first time, why over- and under- approximation analyzers exhibited such a different development.

interpretation is a framework to design sound static analyses by over-approximating the set of program behaviours. While over-approximations can prove correctness, they cannot witness incorrectness because false alarms may arise. An ideal, but uncommon, situation is completeness of the abstraction that can ensure no false alarm is introduced by the abstract interpreter. Local Completeness Logic is a proof system that can decide both correctness and incorrectness of a program: any provable triple $$\vdash _{A}[P]~\textsf{c}~[Q]$$ ⊢ A [ P ] c [ Q ] in the logic implies completeness of an intensional abstraction of program $$\textsf{c}$$ c on input P and is such that Q can be used to decide (in)correctness. However, completeness itself is an extensional property of the function computed by the program, while the above intensional analysis depends on the way the program is written and therefore not all valid triples can be derived in the proof system. Our main contribution is the study of new inference rules which allow one to perform part of the intensional analysis in a more precise abstract domain, and then to transfer the result back to the coarser domain. With these new rules, all (extensionally) valid triples can be derived in the proof system, thus untying the set of provable properties from the way the program is written.

We formulate, in lattice-theoretic terms, two novel algorithms inspired by Bradley’s property directed reachability algorithm. For finding safe invariants or counterexamples, the first algorithm exploits over-approximations of both forward and backward transition relations, expressed abstractly by the notion of adjoints. In the absence of adjoints, one can use the second algorithm, which exploits lower sets and their principals. As a notable example of application, we consider quantitative reachability problems for Markov Decision Processes.