Key takeaways

  • Levent Alpöge and Tristan Buckmaster posted a paper on November 20, 2026, claiming smooth 3D Navier-Stokes flows blow up in finite time with no outside force.
  • The authors say they used Anthropic's Claude throughout, and parts of the proof are formalized in Lean 4.
  • The certificate the proof depends on, and the interval-arithmetic check of it, have not been released, so the claim is unverified.

Levent Alpöge of Anthropic and Tristan Buckmaster of New York University say they have proved that the Navier-Stokes equations can blow up without any outside force. They posted a 79-page paper, dated November 20, 2026, with a Lean formalization in a public GitHub repository. If the claim holds, the Navier-Stokes problem is solved in the form most mathematicians care about. It is not yet checked.

The Navier-Stokes equations describe how water and air move. The open question was whether a smooth flow can develop a point of infinite speed in finite time. In September, OpenAI said its AI agents had shown blowup, but only with a carefully tuned external force stirring the fluid. Many mathematicians called that a loophole. This paper says it closes it.

What the paper claims

The paper says there are smooth, compactly supported starting flows in ordinary 3D space, and smooth flows on a periodic box (the torus), whose solutions blow up in finite time with no force applied. In the official problem description by Charles Fefferman, that disproves statements (A) and (B) and proves (C) and (D), the two "breakdown" outcomes.

The method builds on a solution that repeats itself at shrinking scales, a "discretely self-similar" solution. The authors find an approximate version by computer, which they call a certificate, then prove that a true solution sits close to it. The blowup rate they give is (T − t) to the power −1/4 in the energy-like Ḣ¹ norm, where T is the blowup time.

Claude did much of the work

The paper has a short AI statement: "We used Anthropic models (Claude) throughout this work, and no others." The authors say the agents got a curated list of leads, then guidance. A first draft was generated from the Lean proof and then rewritten with Claude "under our direction." The acknowledgments thank Anthropic for supporting the collaboration.

That makes this the second Navier-Stokes blowup result this fall credited to AI, after OpenAI's forced version in September. Buckmaster has said OpenAI's work leaned on his and Alpöge's approach. OpenAI denies that, Scientific American reported.

What is checked and what isn't

The formal parts are real but incomplete:

  • Theorems 1.1 and 7.1 are formalized in Lean 4 with Mathlib, a library of about 600,000 lines, according to the repository. Lean is a proof assistant that checks every logical step by machine.
  • The certificate, the numerical object the whole proof rests on, is not released. The paper gives only its hash and says it "will be released soon."
  • The check that the certificate meets the required inequalities uses interval arithmetic and sits outside the Lean proof. It is not in the repository either.
  • The repository notes the Lean statements were patched on November 20 and the patched package was not rebuilt.

Terence Tao wrote on his blog on November 20 that the paper is "technically human-readable" and its strategy "broadly clear," but that "more work needs to be done to create a proper, professionally written proof." He ranked it behind a companion result on the Euler equations by Alpöge, Buckmaster and Matei Coiculescu, which is further along.

The same day, Alpöge also claimed a proof of the Riemann hypothesis with Claude. Two Millennium Prize claims in one day from one person deserve more scrutiny, not less.

What happens next

Two things decide whether this stands. The certificate has to come out so others can rerun the interval check, and experts have to read the argument. The Clay Mathematics Institute still lists the problem as active, and its page does not mention either claim.

Sources

  1. Finite-time blow-up for the Navier–Stokes equations (repository and paper)L. Alpöge and T. Buckmaster
  2. Unforced Euler singularity, unforced Navier-Stokes singularity, and the Generalized Riemann HypothesisTerence Tao
  3. AI just solved the 'real' Navier-Stokes problemScientific American
  4. The inside story behind OpenAI's million-dollar math proofScientific American
  5. Navier-Stokes equationClay Mathematics Institute