Publications
Papers on combinatorics formalised in Lean 4, and my Master’s thesis on automated theorem proving. Lean code and preprints are linked where they are public.
Papers
-
[1]
LPBench: A Lean Benchmark for Research-Level Mathematical Theorem Proving
Preprint, Zenodo, November 2026, doi:10.5281/zenodo.24195212.
A benchmark of 117 research-level problems in combinatorics and graph theory, formalised in Lean 4.
-
[2]
10 Scalable Challenges in Combinatorics: A Public Benchmark Formalised in Lean
Extended abstract, AI4Math workshop at CICM 2026, Ljubljana; informal proceedings.
-
[3]
A Lean 4 Formalisation of OpenAI’s Proof of the Cycle Double Cover Conjecture
Lean code: GitHub, doi:10.5281/zenodo.21410822.
See also J. Carmesin, 10 Scalable Challenges in Combinatorics: A Public Benchmark Formalised in Lean.
theorem CDC_multi {W : Type*} {E : Type*} [DecidableEq W] [Finite W] [Fintype E] {G : Multigraph W E} (hb : G.Bridgeless) : ∃ (n : ℕ) (C : Fin n → E → ZMod 2), G.IsCycleDoubleCover C := by classical letI : Fintype W := Fintype.ofFinite W refine G.exists_isCycleDoubleCover_of_forall_cubic ?_ hb intro W' _ _ G' _ hcubic hbG obtain ⟨n, C, hC⟩ := G'.exists_cycleDoubleCover hcubic hbG exact ⟨n, C, G'.isCycleDoubleCover_toMultigraph hC⟩The final theorem of the formalisation, for every finite bridgeless multigraph (CDC/CDC_multigraph/Main_multi.lean). It depends only on Lean’s standard axioms.
Master’s thesis
-
[T]
Automated Theorem Proving for INMO: A Hybrid Framework for Combining Neural and Symbolic Reasoning
Master’s thesis, M.Tech. in Artificial Intelligence & Machine Learning, BITS Pilani · August 2025.