The Enormous Theorem Checked by Computer in 24 Days

The Enormous Theorem, Checked by Computer in 24 Days

Some proofs in mathematics are so long that no single person has read them all. The Classification of Finite Simple Groups is one of them. It lists the basic building blocks of symmetry — the simple groups from which all finite groups can be built — and its proof runs to roughly 15,000 pages, in more than 500 articles, by about 100 mathematicians, built up over decades. Mathematicians call it the Enormous Theorem.

On 26 October 2026, Axiom Math, a company based in Palo Alto, announced that it had produced what it calls the first complete computer-checked version of that proof, written in the Lean proof assistant — a language in which a computer verifies every logical step. Axiom says swarms of AI agents, directed by its engineers and mathematicians, produced about 43.1 million lines of Lean code in 24 days.

What does “computer-checked” mean?

A normal mathematical proof is written for humans: words, diagrams and equations, with some steps left for the reader to fill in. A formal proof in Lean is different. Every definition and every deduction is written in a precise formal language, and the computer checks that each step really follows from the ones before it. Nothing is left as “obviously true”.

That does not mean a computer discovered the theorem, or that the theorem was in doubt. The classification itself was already accepted by mathematicians. What Axiom says it has done is translate the existing, human-written argument into a form a machine can check from start to finish — and, in doing so, found places where the published argument needed repair.

A gap found — and a new argument to close it

According to Axiom’s own account, its agents found counterexamples to a claim used in one part of the proof, the Unbalanced Group Theorem, as presented in John Walter’s 1986 memoir. The counterexamples did not overturn the theorem itself, Axiom says, but they left a key inference unjustified, so a different argument was needed for that step.

Axiom says group theorists who had helped develop the classification, including Richard Lyons and Ronald Solomon, advised on the repair, and that its system produced and formalised a new preservation argument. Axiom’s paper presents both that longer argument and a shorter deduction found afterwards. The formalisation builds on Mathlib, the shared Lean mathematics library, and incorporates about 1.2 million lines from an earlier project, Qiuzhen-CFSG.

Reports of the announcement also say the process flagged more than 2,000 errors in the existing literature, most of them minor, with 34 classified as notable deviations. Those figures come from Axiom’s account of its own work.

The important caveat

A machine check proves that the encoded deductions follow from the encoded assumptions. It does not, by itself, prove that the encoded statement is exactly the theorem mathematicians intended — that still needs human understanding. And as of 27 October 2026, Axiom’s chief executive had said no group outside the company had yet independently rebuilt and checked the 43-million-line proof; Axiom was publishing its build files so that others could try. Independent reproduction is the next step that will show how robust this claim is.

Even with that caveat, the story is a good one for students: checking is part of mathematics, not an afterthought. A proof is not finished when it looks convincing — it is finished when every step survives being checked, however long that takes. Sometimes checking finds a gap, and fixing the gap makes the mathematics stronger.

If your child enjoys puzzles, patterns and “prove it” thinking, maths tutoring can build that checking habit step by step — from clear working and careful reasoning to tackling harder problems with confidence. Book your first lesson and compare verified maths tutors, online, in person or in a group class.

Sources: