We present an LTL model checker whose code has been completely verified using the Isabelle theorem prover. The checker consists of over 4000 lines of ML code. The code is produced using the Isabelle Refinement Framework, which allows us to split its correctness proof into (1) the proof of an abstract version of the checker, consisting of a few hundred lines of ``formalized pseudocode'', and (2) a verified refinement step in which mathematical sets and other abstract structures are replaced by implementations of efficient structures like red-black trees and functional arrays. This leads to a checker that, while still slower than unverified checkers, can already be used as a trusted reference implementation against which advanced implementations can be tested.
- Esparza, J., Lammich, P., Neumann, R., Nipkow, T., Schimpf, A., & Smaus, J.-G. (2013). A Fully Verified Executable LTL Model Checker. Lecture Notes in Computer Science, 463–478. https://doi.org/10.1007/978-3-642-39799-8_31
- Open access
- Projekt home page