# Is Navier-Stokes solved: a blowup proof is claimed but not yet verified

> Alpöge and Buckmaster say smooth flows blow up with no outside force, but the certificate behind the proof is unreleased and Clay's process takes years.

URL: https://claudegpt.si/news/is-navier-stokes-solved  
Published: 2026-11-21T06:57:40.366Z  
Author: ClaudeGPT Si Newsroom  
Section: Research  
Topics: Navier-Stokes, Mathematics, Anthropic, Claude, OpenAI, Lean  
Publisher: ClaudeGPT Si

## Key takeaways

- On November 20, 2026, Levent Alpöge and Tristan Buckmaster claimed smooth 3D Navier-Stokes flows can blow up with no outside force, which would disprove Fefferman's statements (A) and (B).
- The certificate the proof rests on, and the interval-arithmetic check of it, have not been released, so the claim is not yet verified.
- Clay's rules require publication in a refereed outlet, two years of scrutiny and general acceptance before a $1 million prize can be considered.

Navier-Stokes is not solved yet, though a proof is now claimed. On November 20, 2026, Levent Alpöge of Anthropic and Tristan Buckmaster of New York University posted a [paper](https://github.com/tristanbuckmaster/navier-stokes) claiming to settle the Millennium Prize problem, built with Anthropic's Claude and partly checked in the Lean proof assistant. As of November 21, the numerical certificate the proof rests on is unreleased, no independent expert has confirmed the proof, and the [Clay Mathematics Institute](https://www.claymath.org/millennium/navier-stokes-equation/) lists the problem as active.

The claim is serious enough that Terence Tao wrote about it the same day. It is not verified.

## What the Navier-Stokes problem asks

The Navier-Stokes equations describe how a fluid like water or air moves. They are Newton's law, force equals mass times acceleration, applied to every small piece of a fluid, plus a rule that the fluid can't be squeezed.

The open question is whether a smooth flow in three dimensions stays smooth forever, or whether its speed can become infinite after a finite time. Mathematicians call that second outcome "blowup."

The [official problem description](https://www.claymath.org/wp-content/uploads/2022/06/navierstokes.pdf), written for Clay by Charles Fefferman, asks for a proof of any one of four statements:

| Statement | Says | Where | Outside force |
| --- | --- | --- | --- |
| (A) | Every smooth starting flow stays smooth forever, with bounded energy | All of 3D space | None |
| (B) | Every smooth starting flow stays smooth forever | A periodic box (the torus) | None |
| (C) | Some smooth starting flow and smooth force lead to breakdown | All of 3D space | Allowed |
| (D) | Some smooth starting flow and smooth force lead to breakdown | A periodic box | Allowed |

Note the asymmetry. The breakdown statements let the solver add an outside force, as long as it is smooth and fades away as Fefferman specifies. That detail shaped everything that happened in 2026.

The prize is $1 million. Clay's [rules](https://www.claymath.org/wp-content/uploads/2022/03/millennium_prize_rules_0.pdf) set aside $7 million for the seven Millennium problems, $1 million each, and judge a Navier-Stokes answer in either direction by the same procedure.

## Why the problem is so hard

Fefferman's description sums up what was known. In two dimensions, flows stay smooth. In three, they stay smooth if the starting flow is small enough, or for a short time. In 1934 Jean Leray proved that "weak" solutions, a looser kind that allows rough behavior, always exist, but nobody could show they are unique or smooth. Fefferman judged that standard methods were not enough and that deep new ideas would probably be needed.

In a paper first posted in [February 2014](https://arxiv.org/abs/1402.0290), Tao built an averaged version of the equations that keeps the same energy law, and showed it can blow up. His conclusion: any proof that the real equations stay smooth must use finer structure than the standard tools see.

Proving blowup is hard in a different way. Computers can suggest a singularity, but numbers alone prove nothing. Jiajie Chen and Thomas Hou showed one route: a [computer-assisted proof](https://arxiv.org/abs/2210.07191) of blowup from smooth data for the 3D axisymmetric Euler equations, the frictionless cousin of Navier-Stokes, in a domain with a boundary. Euler isn't on Clay's list, and a boundary isn't Clay's setting, but the template of computing an approximate solution and then proving a true one sits nearby is the one the 2026 work follows.

## The forced claim: OpenAI, September 8, 2026

On September 8, 2026, OpenAI said an internal model had constructed Navier-Stokes blowup, according to [Scientific American](https://www.scientificamerican.com/article/the-inside-story-behind-openais-million-dollar-math-proof/). The magazine reported the work came from a swarm of about 10,000 agents that reportedly used $15 million in computing.

The catch was the force. The proof uses a carefully tuned external force that stirs the fluid until it blows up, building on a program by Diego Córdoba and Luis Martínez-Zoroa. Statements (C) and (D) allow that, and Scientific American wrote that the result technically met Clay's conditions. Many mathematicians still saw the tuned force as a loophole in the problem's wording.

It also turned bitter. Buckmaster said OpenAI's agents followed a path he and Alpöge had built, and suspected a leak. OpenAI denied it.

## The unforced claim: Alpöge and Buckmaster, November 20, 2026

The new paper closes the loophole if it holds. It says there are smooth starting flows in 3D space, and on the torus, that blow up in finite time with no force at all. In Fefferman's terms, that disproves (A) and (B), and since zero is an allowed force, proves (C) and (D). [Our news report](https://claudegpt.si/news/navier-stokes-solved-unforced-blowup-claim) covers the announcement.

How it works:

- The authors look for a flow that repeats itself at smaller and smaller scales as it nears the blowup time, a "discretely self-similar" solution.
- They compute an approximate version and pack it, with explicit inequalities, into what they call a certificate.
- They prove a true solution sits close to it, and that suitable smooth starting flows converge to it and blow up with it, at the rate (T − t) to the power −1/4 in an energy-type norm, where T is the blowup time.

The paper's AI statement says: "We used Anthropic models (Claude) throughout this work, and no others." The agents got a curated dossier of leads, then guidance, and a draft generated from the Lean proof was rewritten with Claude under the authors' direction. Scientific American [reported](https://www.scientificamerican.com/article/ai-just-solved-the-real-navier-stokes-problem/) that the write-up is unpolished because, after the OpenAI episode, the pair felt they could not wait.

## What has been checked, and what hasn't

According to the paper and the repository, Theorems 1.1 and 7.1 are formalized in Lean 4, a program that checks every logical step of a proof by machine. The repository's library is about 2,270 files and 600,000 lines. Four gaps remain:

1. **The certificate is not public.** The paper gives only its hash, a fingerprint for confirming the file once released, and says it is coming soon.
2. **The key computation sits outside Lean.** Checking the certificate's inequalities uses interval arithmetic, which bounds rounding errors rigorously. It isn't formalized or in the repository.
3. **The Lean statements trail the paper.** They match an earlier version of it, and the formal theorem has no blowup rate. Patches made on November 20 were not rebuilt.
4. **Human review has barely started.** Tao wrote on [his blog](https://terrytao.wordpress.com/2026/11/20/unforced-euler-singularity-unforced-navier-stokes-singularity-and-the-generalized-riemann-hypothesis/) that the strategy is broadly clear, but "more work needs to be done to create a proper, professionally written proof." He rated a companion unforced Euler result, by Alpöge, Buckmaster and Matei Coiculescu, as further along.

## What has to happen before it counts as solved

First, the mathematics. The certificate has to come out, others have to rerun the interval arithmetic, and the patched Lean package has to build. Then fluid experts have to read the argument and find a gap or not.

Then Clay, which is slow by design. A proposed solution must appear in a qualifying refereed journal, survive at least two years after publication, and win general acceptance among mathematicians. Only then does Clay decide whether to convene a special advisory committee of at least three members. It does not take direct submissions.

Clay also says it weighs whether a solution depends crucially on earlier published insights, and can credit that work in an award. With OpenAI's forced result and the Córdoba and Martínez-Zoroa program in the background, credit may be contested.

Even in the best case, no prize can come before late 2028. The first real test comes sooner: the day the certificate is released.

## Frequently asked questions

### Has Navier-Stokes been solved?

Not in a verified way. Alpöge and Buckmaster claimed unforced blowup on November 20, 2026, but their certificate is unreleased and experts are still reviewing the paper. Clay lists the problem as active.

### What is the difference between forced and unforced Navier-Stokes blowup?

Forced blowup uses an outside force that stirs the fluid, which Clay's breakdown statements (C) and (D) allow. Unforced blowup happens with no force at all, which would disprove statements (A) and (B) and is the outcome most mathematicians saw as the real question.

### Did OpenAI solve Navier-Stokes?

OpenAI said on September 8, 2026, that its model constructed Navier-Stokes blowup, but with a finely tuned external force. Scientific American reported that it technically met Clay's conditions while many experts saw the force as a loophole.

### How much is the Navier-Stokes Millennium Prize?

$1 million. Clay's rules set aside $7 million for the seven Millennium problems, $1 million each, and a Navier-Stokes resolution in either direction is judged by the same procedure.

## Sources

- [Finite-time blow-up for the Navier–Stokes equations (paper and Lean formalization)](https://github.com/tristanbuckmaster/navier-stokes), L. Alpöge and T. Buckmaster
- [Existence and smoothness of the Navier–Stokes equation (official problem description)](https://www.claymath.org/wp-content/uploads/2022/06/navierstokes.pdf), Clay Mathematics Institute
- [Millennium Prize description and rules](https://www.claymath.org/wp-content/uploads/2022/03/millennium_prize_rules_0.pdf), Clay Mathematics Institute
- [Navier-Stokes equation](https://www.claymath.org/millennium/navier-stokes-equation/), Clay Mathematics Institute
- [Unforced Euler singularity, unforced Navier-Stokes singularity, and the Generalized Riemann Hypothesis](https://terrytao.wordpress.com/2026/11/20/unforced-euler-singularity-unforced-navier-stokes-singularity-and-the-generalized-riemann-hypothesis/), Terence Tao
- [Finite time blowup for an averaged three-dimensional Navier-Stokes equation](https://arxiv.org/abs/1402.0290), arXiv
- [Stable nearly self-similar blowup of the 2D Boussinesq and 3D Euler equations with smooth data I: Analysis](https://arxiv.org/abs/2210.07191), arXiv
- [AI just solved the 'real' Navier-Stokes problem](https://www.scientificamerican.com/article/ai-just-solved-the-real-navier-stokes-problem/), Scientific American
- [The inside story behind OpenAI's million-dollar math proof](https://www.scientificamerican.com/article/the-inside-story-behind-openais-million-dollar-math-proof/), Scientific American
