Researchers claim AI-assisted Navier-Stokes proof without external force
The GitHub submission includes a 79-page paper and Lean files, but a numerical certificate is missing and patched files have not been rebuilt.

Two researchers say they have proved that the Navier-Stokes equations can develop a breakdown in finite time without an external force. Tristan Buckmaster, of New York University, and Levent Alpöge, of Anthropic, posted their heavily AI-assisted work on GitHub, with a paper dated November 20.
The equations model the movement of fluids, including air and water. The claimed result concerns whether an initially smooth flow can reach a point where its mathematical solution breaks down. That question is among the Clay Mathematics Institute’s Millennium Prize Problems. OpenAI published a proof in September involving an applied smooth force; the researchers say their construction requires none.
Alongside the 79-page paper, the repository contains a formalization using Lean, software for checking logical steps, spanning about 2,270 files and 600,000 lines. But a numerical-data certificate referenced by its hash in the paper has not yet been included. The authors say they will provide it. The certificate-checking computation is outside the formalization, and changes made to the Lean files on November 20 were not followed by a rebuild of the package.
UCLA mathematician Terence Tao wrote that he could broadly follow the approach: construct a numerical approximation, then establish an exact solution through rigorous perturbation theory. He nevertheless said further work was needed to produce a properly written proof. Buckmaster acknowledged on Mastodon that the write-up was unpolished and extensively assisted by large language models.




