Skip to content
State of Proof

MathematicsTheoretical result

Faltings’ Theorem Formalized

Axiom; the public repository states that roughly 430 modules were written by AI agents and have not yet been reviewed by humans.

Source

Tracked

Updated

Source record; see version

A translucent deep-blue surface with three openings carries thin amber paths and teal points over a subtle grid.
Fig. 01Conceptual illustration · not a proof certificate, a mathematical construction, or a measurement
About this illustration

When rational paths cannot continue forever

The openings introduce genus, while the amber paths and teal points suggest rational points on a curve. This is invented geometry, not a specific curve, proof diagram, calculation or measurement.

Source · Axiom announcement, theorem statement; Mordell README, `Faltings.finite_setOf_comp_eq_id_of_two_le_genus` and LeanEval endpoint

The idea, in plain language

A landmark finiteness theorem enters a proof checker

What it is
Axiom reports a Lean formalization of Faltings’ theorem: curves of genus at least two over a number field have finitely many rational points.
Who did it
Axiom; its source says about 430 proof modules were written by AI agents and are not yet human-reviewed.
What it could mean
A formal version could make a landmark result easier to inspect mechanically. It does not replace expert review of the theorem statement, dependencies, or argument.
See the check plan

What the paper says.

Axiom reports a Lean 4.35 formalization of the Bombieri–Vojta route to Faltings’ theorem, built on its Stacks Project library. The public source gives both a scheme-theoretic theorem and a Mathlib-only LeanEval endpoint, while explicitly saying the code has not yet been reviewed by humans.

Why we tracked this work

source-reviewed intake only; independent examination has not started.

Foundational Arithmetic geometry · Formalized mathematics · AI-assisted discovery

What could change?

Some algebraic curves can have many rational solutions, but not infinitely many once their shape has enough holes. Axiom reports encoding that landmark result in Lean. The formal source is public; independent review remains open.

If it holds up

Specialists could inspect a machine-checkable route through a central theorem in arithmetic geometry and reuse its formal infrastructure.

If it does not

A mismatch in definitions, dependencies, or the proof chain would narrow the formal claim without overturning Faltings’ established theorem.

What is on the record?

Discovery recorded. State of Proof has not yet examined this claim.

Source
Faltings’ Theorem Formalized Mathematical research record · Source record; see version
Version
2026-11-12 announcement and repository snapshot at Git commit 82a30cb7857f6e3cfdb1be91f5533c0ccb57d493.
Evidence reported
Manuscript; independent examination not started
Available artifacts
Pinned AxiomMath/Mordell commit 82a30cb7857f6e3cfdb1be91f5533c0ccb57d493; Lean 4.35; Mordell/MordellConjecture/Curve.lean, Mordell/MordellConjecture/FunctionField.lean, and Mordell/LeanEval/Faltings.lean. The README reports no sorry, nativedecide, or added axiom, and lists only propext, Classical.choice, and Quot.sound for the cited endpoint. No State of Proof build, Lean replay, or manuscript code execution occurred.
Highest-risk dependency
Analyst judgment: the statement-to-classical-theorem mapping, the extensive AI-produced dependency closure, and the unreviewed Bombieri–Vojta chain are load-bearing. Publisher-reported zero-sorry and axiom outputs are not a State of Proof replay, specialist proof review, peer review, or a verdict on Faltings’ theorem.

The next independent check.

Compare the ordinary theorem statement and the LeanEval endpoint clause by clause with Faltings’ theorem, especially the definitions of curve, genus and rational point. Independently replay the pinned dependency closure in an isolated environment and inspect theorem axioms and imports; distinguish a successful kernel check from human review of statement correspondence.

No proof docket yet

A docket is the public record of checks and open questions. This paper does not have one yet; the check plan above describes work still to do.

Explore existing proof dockets

Current boundary

Intake record only; examination not started. State of Proof has only recorded this intake. The paper's full argument remains unexamined; the next bounded check is: Compare the ordinary theorem statement and the LeanEval endpoint clause by clause with Faltings’ theorem, especially the definitions of curve, genus and rational point. Independently replay the pinned dependency closure in an isolated environment and inspect theorem axioms and imports; distinguish a successful kernel check from human review of statement correspondence.

How validation works