Lemmaless Induction in Trace Logic

Abstract

We present a novel approach to automate the verification of first-order inductive program properties capturing the partial correctness of imperative program loops with branching, integers and arrays. We rely on trace logic, an instance of first-order logic with theories, to express first-order program semantics by quantifying over program execution timepoints. Program verification in trace logic is translated into a first-order theorem proving problem where, to date, effective reasoning has required the introduction of so-called trace lemmas to establish inductive properties. In this work, we extend trace logic with generic induction schemata over timepoints and loop counters, reducing reliance on trace lemmas. Inferring and proving loop invariants becomes an inductive inference step within superposition-based first-order theorem proving. We implemented our approach in the Rapid framework, using the first-order theorem prover Vampire. Our extensive experimental analysis shows that automating inductive verification in trace logic is an improvement compared to existing approaches.