Skip to content
← Back to Mathematics

Mathematicians claim, with heavy AI help, a proof that 3D fluids left alone can blow up

Their 79-page paper comes with Lean code, but a key computation is not public yet and no referee has read it.

2fluid equations with no outside force now claimed to blow up

On 20 November 2026 Terence Tao reported two new results on fluids with no outside force. Levent Alpöge, Tristan Buckmaster and Matei Coiculescu, “with heavy AI assistance”, wrote a human-readable proof that the 3D incompressible Euler equations can blow up in finite time, meaning a smooth flow develops a singularity. Alpöge and Buckmaster, “again with heavy AI usage”, did the same for the 3D Navier–Stokes equations, the one behind a $1 million Clay Millennium Prize problem.

The Navier–Stokes paper states its setting. It builds smooth, compactly supported, divergence-free starting data on all of 3D space, and smooth data on the periodic cube (the torus), for which the force-free equations blow up. The authors say this disproves statements (A) and (B) of Charles Fefferman’s official problem statement, and so proves (C) and (D). The Euler paper builds smooth periodic starting data on the torus whose velocity stays bounded while the vorticity (local spin) at one point grows without limit.

Both come with Lean formalizations, in which a computer checks each step. The Navier–Stokes proof follows an expected route: numerically build an approximate solution that repeats itself under rescaling, then use rigorous perturbation theory to turn it into an exact one. The paper says the approximation is a “certificate” that will be released soon. Tao calls the paper “technically human-readable” and says more work is needed for a professionally written proof.

What it means

If the proofs hold, a fluid left alone can break down, the question most mathematicians wanted answered after the September results that needed an outside force. Referees and the Clay Institute have not weighed in, and no prize has been awarded.

Caveats

  • Neither paper has been refereed. Tao says the Navier–Stokes proof still needs a professionally written version.
  • The Navier–Stokes repository says the certificate and the interval-arithmetic computation that checks it are not included, so the Lean code does not cover that step. Its latest patch, from 20 November, had not been rebuilt.
  • The Euler Lean code follows an 18 November draft, not the arXiv version.
  • Tao’s post names no AI tool, and neither repository does. Alpöge works at Anthropic.
  • A comment on Tao’s post says OpenAI proved unforced Euler blow-up first. We found no OpenAI page saying so.
Suggest a correction