Section: New Results

Formalisation and verification

Participants : Pierre-Louis Curien, Lucien David, Emilio Jesús Gallego Arias, Kailiang Ji, Pierre Letouzey, Jean-Jacques Lévy, Cyprien Mangin, Daniel de Rauglaudre, Yann Régis-Gianas, Alexis Saurin, Matthieu Sozeau.

Proofs and surfaces

The joint work of Pierre-Louis Curien with Jovana Obradović (former PhD student of the team and now postdoc in Prague), Zoran Petrić and other Serbian colleagues on formalising proofs of incidence theorems (arising by repeated use of Menelaus theorem) by means of a cyclic sequent calculus, has been submitted to a journal, and has been presented at the conference Topology, Algebra, and Categories in Logic (TACL) 2019, Nice, in June 2019 [53].

A Coq formalisation of the first-order predicate calculus

In relation with a logic course for master students, Pierre Letouzey made a Coq formalisation of the first-order predicate calculus. The logical rules are expressed in a natural deduction style (with explicit contexts). Pierre Letouzey proposed two low-level representations of formulas : one based on quantifiers with names, the other using “locally nameless” techniques. The equivalence between the two settings has been proved correct. Using this deep embedding, Pierre Letouzey formalised in Coq the whole course notes (prepared some years ago by Alexandre Miquel), including the completeness theorem for this logic. This development is available at https://gitlab.math.univ-paris-diderot.fr/letouzey/natded.

A Coq formalisation of circular proofs and their validity condition

During the summer 2019, Alexis Saurin supervised Lucien David's M1 internship on formalizing in Coq circular proofs and their meta-theory. This work built on Xavier Onfroy's previous work as well as on Pierre Letouzey's formalisation of the predicate calculus in natural deduction mentioned above. While the previous work by Xavier Onfroy was both contributing to the proof theory part and the ω-automata part (which is need for the decidability theorem), Lucien David completely focused on the the proof theory side. In particular, he was able to improve significantly on Xavier Onfroy's formalisation by using ideas from Letouzey's formalisation of natural deduction and by interacting with Pierre Letouzey and Alexis Saurin. This development is available at https://github.com/LuluDavid/CircularProofsValidity.

Lexing and regular expressions in Coq

Pierre Letouzey and Yann Régis-Gianas revisited in Coq classical techniques about lexing and regular expressions. In particular, regular expressions (with complement and conjunction) have been formalised, as well as their Brzozowski derivatives, and the finiteness theorem due to Brzozowski : a given regular expression admits only a finite number of derivatives (up to some equivalence). Both the general equivalence (based on language identity) and practical approximations (similarities) has been considered (and proved decidable). From that, the algorithms building recognizing automata (with derivatives as states) have been formalised and proved, leading to the minimal automata when using the general equivalence (but at a high cost), or to practical approximations of the minimal automata when using various similarities. This work is still ongoing. For instance, the correctness proof of a particular similarity used in an existing implementation (ml-ulex) is quite elusive for the moment. They also plan to extend this development up to a full-scale tool a la ocamllex in Coq.

Real Numbers as sequences of digits in Coq

Daniel de Rauglaudre has been continuing the formalisation of real numbers defined as sequences of digits in any radix with the LPO axiom/oracle (Limited Principle of Omniscience). Although the operations (additions and multiplications) work with this method, the proof of associativity of addition needs more work to be achieved. This development is available at https://github.com/roglo/coq_real/.

Category theory in Coq

Daniel de Rauglaudre started an implementation in Coq of Category theory in Coq, using in particular theorems coming from HOTT (HOmotopy Type theory) that he implemeted some years ago. Several notions around Categories have been defined. For example, Yoneda Lemma, among others. This development is available at https://github.com/roglo/mycoqhott/.

Number theory in Coq

Daniel de Rauglaudre started and almost completed the formalisation in Coq of the proof of Euler's Product Formula, stating that the Riemann zeta function, which is a sum on all the natural numbers, is also a product on all the prime numbers. He also added several theorems about the prime numbers. This development is available at https://github.com/roglo/coq_euler_prod_form.

Proofs of algorithms on graphs

Jean-Jacques Lévy and Chen Ran (a PhD student at the Institute of Software, Beijing) pursued their work about formal proofs of graph algorithms. Their goal is to provide proofs of algorithms checked by computer and human readable. In 2019, they presented at ITP 2019 a joint paper with Cyril Cohen, Stephan Merz and Laurent Théry on this work [37]. This article compared formal proofs in three different systems (Why3, Coq, Isabelle/HOL) of Tarjan (1972) linear-time algorithm computing the strongly connected components in directed graphs.

The current work is to have a proof of the implementation of this algorithm with imperative programming and memory pointers. They also planed to produce formal proofs of other abstract algorithms such as the Hopcroft-Tarjan (1972) linear-time algorithm for planarity testing in undirected graphs.

Certified compilation and meta-programming

Matthieu Sozeau participates to the CertiCoq project led by Andrew Appel at Princeton (https://www.cs.princeton.edu/~appel/certicoq) whose aim is to verify a compiler from Coq's Gallina language down to CompCert C-light which provides itself a certified compilation path to assembly language. Together with Yannick Forster at the University of Saarbrucken and the MetaCoq team, Matthieu Sozeau focused the verification of type-checking and erasure which were previously trusted parts of the system. The new verified erasure function fills a gap in the proof of correctness of compilation from Gallina terms down to C-light. The whole compiler can be run on realistic examples (the erasure phase does take most of the compilation time and should be optimised further).

In collaboration with Xavier Denis (Paris Diderot), Yann Régis-Gianas formalised and built a compiler for Mtac2. A paper is in preparation.