Mathlib-integrated formalization of the modern FLT proof with reusable Galois-representation infrastructure
Large but known engineering, now helped by agentic tools: 38, above low-precision training (13) and below the research-grade formal-verification problems around 48, because integration with Mathlib's design and review is harder than producing a certificate. Application value 29, just below RH (30), reflects reusable Langlands-programme infrastructure for formalization. Leverage 36: the library would gate further formalized arithmetic geometry but nothing mathematical depends on it.
The problem
Teach a computer to check the whole modern proof of Fermat's Last Theorem, as reusable pieces of a shared library.
- 1Wiles proved FLT using deep number theory
- 2Computer checking needs all that theory in Lean
- 3Build modular forms and Galois tools in Mathlib
- 4Result: trusted proof and reusable library
Goal. Formalize the modern proof of Fermat's Last Theorem (FLT) as a Lean 4 library merged into Mathlib. The proof strategy is: a hypothetical solution to gives an elliptic curve (the ) with strange properties; the theorems of Wiles and Taylor–Wiles show such a curve would be modular; and Ribet's level-lowering theorem shows no such modular curve exists. The formalization project (led by Kevin Buzzard at Imperial College London, with an EPSRC grant and a public ) is aimed at the heart of this: the theorem, which identifies a deformation ring of Galois representations with a Hecke algebra acting on modular forms.
Path to understanding
foundations → problemThe research entries this problem rests on, in the order to read them. Each links to its explainer.
- Start with
- Step 2
- Step 3
- Step 4
- Elliptic curvesTheory · host
- Then
- Mathlib-integrated formalization of the modern FLT proof with reusable Galois-representation infrastructureThe problem
State of the art
In progress. The Imperial project started in October 2024 with an EPSRC grant and a public blueprint. Buzzard's December 2024 report, two months in, describes formalizing a modern R = T (modularity lifting) argument building on Diamond, Fujiwara, Kisin, Taylor and Scholze, with ingredients such as crystalline cohomology and divided powers still missing from Mathlib, and stresses that the aim is the general theorems rather than FLT alone. On 4 September 2026 Anthropic's agentic formalization produced a 13-million-line certificate of FLT following the Darmon–Diamond–Taylor route; it compiles about twenty times slower than Mathlib and is not integrated with it. Buzzard checked its final statement with Comparator and judged that it tells us essentially nothing new mathematically. The integrated formalization of the modern proof, including reusable automorphic infrastructure, remains to be done and is not covered by the certificate.
Realistic applications
Reusable formal infrastructure for the Langlands program (Galois representations, Hecke algebras, deformation rings) so that further arithmetic theorems can be formalized without rebuilding it. Also a test bed for scaling Mathlib development.
What counts as solved
The Imperial blueprint is fully formalized with no sorry, merged into Mathlib or an agreed library compatible with it, the theorem FermatLastTheorem derived from the integrated library, and the library compiles within a stated budget (for example at most 3 times Mathlib's build time).
How it gets verified
Lean proofA formal proof checked by Lean 4 / Mathlib.
Goal: the theorem FermatLastTheorem (Mathlib defines FermatLastTheoremFor and has the cases n = 3 and n = 4) derived from the Imperial College FLT project at github.com/ImperialCollegeLondon/FLT, merged into Mathlib or a compatible library. Done means no sorry, the library builds within the stated budget (for example 3 times Mathlib's build time), and CI is green on a pinned Lean and Mathlib version.
Claims under verification
+ add a claimWhere the work is
Almost all effort is formalization engineering: API design, refactoring, upstreaming and review.
Bottlenecks
ranked by votes- #1EngineeringMathlib integration and review capacity
Maintainers must review and merge definitions that touch many Mathlib areas; a 13-million-line certificate cannot be merged as is.
Community grades
Sign in to add your own grade.
Connected problems
No connections yet.
Progress
- Sep 4, 202613M-line certificate appearssource
Does not replace the integrated library; gives a reference proof to port from.
- Dec 11, 2024Buzzard progress reportsource
Describes the reduction to automorphic input and what remains.
- Oct 1, 2024Project startssource
EPSRC-funded Imperial FLT project begins (month approximate).