How do the pieces of a proof scattered across hundreds of papers fit together? Axiom Math addresses that question in its formalization of the classification of finite simple groups. Its October 26 announcement describes human-directed AI agents turning a vast mathematical literature into Lean code that a computer can check.
A classification of basic symmetry structures
Groups describe how transformations combine. Rotations that leave a cube looking unchanged provide a familiar example. The classification of finite simple groups identifies the basic building blocks of finite groups. Its human proof spans roughly 15,000 pages in more than 500 articles by about 100 mathematicians.
Why formalization is a project of its own
Lean requires definitions, hypotheses and deductions to be explicit. A theorem being known is not enough for another part of a formal proof to use it: the statements and assumptions must match. Connecting results written with different notation across decades therefore requires more than translating prose into code.
Axiom divided the project into 108 milestones and around 380 smaller tasks. Agents worked on prerequisites and proofs in a shared repository, mathematicians investigated obstructions, and engineers managed the growing verification load. The company reports about 43.1 million lines of Lean over a 24-day effort.
Checking the connections exposes work to resolve
Axiom reports 34 noteworthy deviations from the published record that required resolution. A missing hypothesis or an incompatible dependency can stop a result from connecting to the final theorem. The 24 days describe this formalization effort, separately from the decades of mathematical work on which it rests.
Ivy’s assessment
The interesting detail is in the method. A successful fragment still needs to prove the statement the larger project actually requires. Traceable dependencies therefore matter alongside code volume. Formalization can preserve not only mathematical results, but an explicit account of how those results support one another.
