Navier-Stokes can blow up with no outside force, Buckmaster and Alpöge say

Two months after OpenAI’s forced version, the pair posted their Navier-Stokes blow-up proof on GitHub, with a 55-page Euler paper alongside and some files still missing.


Navier-Stokes can blow up with no outside force, Buckmaster and Alpöge say

Anthropic logo, Guilherand-Granges, France, March 2025

Image Credits Credit: PhotoGranary02 / Shutterstock

The Navier-Stokes equations can blow up in finite time with no outside force acting on the fluid, say Tristan Buckmaster of New York University and Levent Alpöge of Anthropic. They posted the proof, heavily assisted by AI, on GitHub on Friday.

The equations describe how fluids such as water and air move. Blow-up means a smooth flow develops a point where the solution breaks down in finite time. Whether that can happen is one of the Clay Mathematics Institute’s $1m Millennium Prize Problems.

In September, OpenAI published its own proof of blow-up. Its construction applies a smooth force to the fluid. The new claim removes the force.

What was released

The repository holds a 79-page paper, dated 20 November, and a formalisation in Lean, a program that checks every logical step. The Lean library runs to about 2,270 files and 600,000 lines.

The repository also lists what is missing. It does not yet include a certificate of numerical data whose hash appears in the paper. The paper says the authors will release it. The computation that checks the certificate sits outside the formalisation. The authors also patched the Lean files on 20 November but did not rebuild the patched package.

Buckmaster explained the rush in a Mastodon post.

“After last time, we couldn’t risk waiting, so the Navier-Stokes write-up is unpolished and heavily LLM-assisted. Fortunately, once the numerical profile is in hand, the proof follows a well-trodden computer-assisted path,” Buckmaster wrote.

Tao’s first read

Terence Tao, the UCLA mathematician, wrote on his blog that the paper is technically human-readable. Its strategy is broadly clear, he said: build an approximate solution numerically, then use rigorous perturbation theory to turn it into an exact one. A properly written proof still needs more work, he added.

An Euler paper too

The same day, Alpöge and Buckmaster released a 55-page paper with Matei Coiculescu, a junior fellow at NYU. It proves blow-up for the Euler equations, which describe fluids without viscosity, also with no outside force. Tao called it a human-readable proof, and it comes with a Lean formalisation.

OpenAI got there first on Euler. In September it said nearly 100 AI agents had resolved the unforced Euler question in about 50 hours. Before OpenAI’s paper came out, the new paper says, Alpöge and Buckmaster had finished an unreleased second proof with forcing that used the same mechanism.

The authors say they used Claude and Codex throughout. Anthropic models wrote the first proof without force, and the team then rewrote it with the models over several revisions.

After last time

In September, Buckmaster and Alpöge posted their forced results one day before OpenAI announced its proof. The timing turned into a dispute. OpenAI later said its investigation found that Buckmaster’s Codex prompts could not have influenced its system.

Alpöge’s Friday did not stop there. He also claimed a proof of the Riemann hypothesis, found with Claude.

Get the TNW newsletter

Get the most important tech news in your inbox each week.

Thanks! Check your inbox to confirm your subscription.

By subscribing you agree to our Privacy Statement.

Published
Back to top