Navier–Stokes: unforced blow-up claimed (Claude)
Awaiting review
Importance: historic (5 of 5)Status
- Claim
Awaiting review
- Our reporting
- Medium confidence
- Verification
- Lean-verified
- Importance
- Historic (5 of 5)
- Last verified
- 22 November 2026
Your AI and this story
- GPT-6.2 Astra173 days after its cutoff
- Claude Opus 5.5143 days after its cutoff
- Gemini 4 Hydrogen234 days after its cutoff
- Grok 4.7173 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 143 days before it.
Key facts
- Abstract (paper of 20 Nov, github.com/tristanbuckmaster/navier-stokes): ‘We prove that there exist smooth, compactly supported, divergence-free initial data on R³, and smooth divergence-free initial data on the torus T³, for which the solutions of the three-dimensional Navier–Stokes equations without force blow up in finite time. This disproves the statements (A) and (B) of the Millennium Prize problem as formulated by Fefferman’
- Mechanism: solutions converge in self-similar variables to a backward discretely self-similar solution and blow up at the rate (T − t)^(−1/4) in Ḣ¹; the profile comes from a ‘certificate’ (an approximate solution meeting explicit inequalities), checked by interval arithmetic (128-bit, Arb/FLINT) outside Lean
- Not yet checkable: the certificate ‘will be released soon’ (hash 3f7a335b…b22c5 given); the repository’s Lean package (≈2,270 files, 600,000 lines; axioms limited to propext, Quot.sound, Classical.choice in its Comparator configs) got a patch on 20 Nov that ‘was not rebuilt when this repository was assembled’
- AI statement (§1.6, verbatim): ‘We used Anthropic models (Claude) throughout this work, and no others.’ Agents were given ‘a curated dossier of potential leads’; a draft was generated from an initial Lean proof with Scott Armstrong’s ‘enslopification’ skills, then rewritten with Claude under the authors’ direction
- Companion paper (Alpöge, Buckmaster, Matei Coiculescu, cims.nyu.edu/~tristanb/cc.pdf, 54 pages): smooth periodic initial data for the 3D incompressible Euler equations whose solution blows up in finite time (bounded velocity, vorticity at the origin → ∞), built from localized Craik–Criminale waves; Lean formalization at github.com/mcoiculescu/EulerBlowupPublic; AI statement: ‘We used Claude and Codex throughout this work’
- Provenance note in the Euler paper: before OpenAI’s unforced Euler paper, the first two authors had an unreleased forced-Euler proof using ‘the same Craik–Criminale amplification mechanism as OpenAI’s paper’; Buckmaster’s account of his communications with OpenAI is cited
- Terence Tao (blog, 20 Nov): the Navier–Stokes paper is ‘technically human-readable’ and its strategy ‘broadly clear’ (numerically build an approximate discretely self-similar solution, then rigorous perturbation theory), but ‘more work needs to be done’ for a professionally written proof
- Scientific American (20 Nov): frames it as closing the loophole in OpenAI’s September forced blow-up, whose ‘finely tuned force’ many experts saw as a gap; no independent expert yet called the proof correct
Show 3 more
- Buckmaster on Mastodon (20 Nov, read in the original): ‘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.’
- Status on 22 Nov (~12:00 UTC): the repository still has only its two commits of 20 Nov; the certificate is not released and the patched Lean package has not been reported rebuilt; no specialist has publicly called the proof correct. Unlike the Riemann claim, no community verification effort has formed around it yet
- Reception: it drew far less attention than the same-day Riemann claim; Scott Armstrong (whose ‘enslopification’ skills drafted the paper) posted ‘In lesser news, Tristan and Levent claim to have proved blow-up of UNFORCED Navier-Stokes’ (212k views)
What happened
On 20 November 2026, the same day as his Riemann Hypothesis claim, Levent Alpöge (Anthropic) and Tristan Buckmaster (NYU) posted a paper saying that the three-dimensional Navier–Stokes equations without external force can blow up in finite time from smooth, compactly supported initial data. That is the negative answer to the Clay Millennium problem, in the form most experts said OpenAI’s forced result of September had not reached. A Lean package comes with it, but the numerical certificate the proof rests on is only promised, with its hash published. A companion paper with Matei Coiculescu claims finite-time blow-up for the unforced Euler equations on the torus.
Why it matters
If it holds up, a second Millennium problem falls in the same autumn, again by a human–AI team, and it settles the version of the Navier–Stokes question that the September dispute left open. Until the certificate is out and checked, and specialists have read the 79 pages, it remains a claim. Terence Tao calls the strategy clear but the write-up unfinished.
What is disputed or not yet verified
| Verification | Unrefereed; Lean packages public but the Navier–Stokes certificate is unreleased and its interval-arithmetic check lies outside Lean; patched Lean not rebuilt at release |
|---|
Sources
8 sources from 7 sites. Numbers match the chips in the text.
8 sources: 3 primary, 2 press, 3 reactions
Primary
- Alpöge and Buckmaster: Finite-time blow-up for the Navier–Stokes equations (paper + Lean)github.com, paper
- Alpöge, Buckmaster, Coiculescu: Finite-time singularity formation for the 3D Euler equations (PDF)cims.nyu.edu, paper
- Euler blow-up Lean formalizationgithub.com, code
Press
- Scientific American: AI just solved the ‘real’ Navier–Stokes problemscientificamerican.com, press
- TechCrunch: AI claimed a solution to the Riemann hypothesis, and a human did too (covers the Navier–Stokes release)techcrunch.com, press
Reactions
- Terence Tao: unforced Euler singularity, unforced Navier–Stokes singularity, and GRHterrytao.wordpress.com, discussion
- Tristan Buckmaster on Mastodon: the two papers (20 Nov)mastodon.social, discussion
- Scott Armstrong on X: ‘In lesser news…’ (212k views)x.com, discussion
Changes
- Buckmaster’s Mastodon post read in the original; status check (certificate still unreleased, repo unchanged); reception
- Filed from the paper, the repository README, Tao’s post and press coverage