Tristan Buckmaster and Levent Alpöge say they proved finite-time blowup for three-dimensional Navier-Stokes with no external force, using heavy help from AI. They published a 79-page PDF and a Lean 4 package on November 20, 2026. Scientific American reports the result as the zero-force answer many experts wanted after OpenAI''s forced-flow proof in September. Nothing in this post verifies the mathematics. explainx.ai separates what the authors posted from what anyone has checked.
TL;DR: claimed versus verified
| Question | Answer |
|---|---|
| What is claimed? | Smooth initial data on R³ (and periodic analogues) can yield finite-time blowup with zero forcing |
| Primary artifacts? | PDF on GitHub, Lean tree, Buckmaster Mastodon note per Scientific American |
| AI role? | Authors describe the write-up as heavily LLM-assisted |
| Independent verification? | None cited in Scientific American beyond the authors |
| Numerical certificate? | Not released; repo README says it will come soon |
| Lean rebuild status? | Nov 20 patch applied; package not rebuilt when repo went public |
| Clay / Millennium status? | Still open on Clay''s site |
Why the forcing loophole matters
Navier-Stokes on R³ is a Millennium Prize problem. The official statements split existence and smoothness questions from breakdown questions that can involve external forces. OpenAI''s September construction proved blowup with a crafted smooth force that stirs the fluid. Many mathematicians treated that as valid relative to one formalization but unsatisfying physically, because real fluids are not pushed by a hidden forcing term tuned to create a singularity.
Scientific American frames Buckmaster and Alpöge''s new work as closing that gap: blowup without importing such a force. If correct, it would address the version experts describe as the stronger Navier-Stokes headline, not the forced variant OpenAI documented.
Our Cambridge PhD explainer and credit dispute timeline already tracked the September fight over scooping, co-authorship, and data access. Today''s release is a new claim, not a resolution of that dispute.
Rounded price tag on a wave line, illustrating fluid flow metaphors behind Navier-Stokes blowup claims
Image: Library reference — fluid-flow imagery for the Navier-Stokes Millennium thread.
What the GitHub repository contains
The tristanbuckmaster/navier-stokes README, dated November 20, 2026, lists:
| Path | Contents |
|---|---|
paper/finite-time-blow-up-navier-stokes.pdf | 79-page paper version dated Nov 20 |
lean/ | Lean v4.33.1 package with Mathlib pin in lakefile.toml |
lean/README.md | Maps formal statements to paper theorems |
lean/NOTICE | Plain-language statement guide |
The README also documents honest limits:
- No certificate data in the repo; the hash in the paper points to a file not yet published.
- Interval arithmetic verification (128-bit, Arb through FLINT) is outside Lean and not included.
- A November 20 patch changed cell-layer bounds; the patched package was not rebuilt when the repository was assembled.
That combination means builders should not treat a green local lake build as the full story until the authors ship the certificate and rebuild instructions.
What Scientific American reports
Peter Hall''s November 20 article in Scientific American is the mainstream secondary source Techmeme relayed. Key reported points explainx.ai can attribute without treating them as proved math:
- Buckmaster posted on Mastodon around 8:37 A.M. EST with the proof.
- The release was rushed because, after last time, the pair could not risk waiting.
- The write-up is unpolished and heavily AI-assisted.
- The numerical certificate would be released soon.
- The work aims at the unforced blowup question OpenAI''s result left open.
Scientific American does not, in the material we read, quote an external mathematician endorsing the proof. That absence is itself signal: unlike Alpöge''s same-day Riemann drop, where MIT''s Andrew Sutherland published a live Lean check page, this Navier-Stokes repo had no parallel public replay from a named independent verifier at launch.
Lean formalization versus the missing certificate
Computer-assisted blowup proofs in this family typically pair:
- A numerical profile (often approximately self-similar at shrinking scales).
- Rigorous perturbation showing a true solution nearby.
- Interval arithmetic certifying inequalities the formal prover cannot alone supply.
The repository README separates steps one and three from what is inside Lean. The formal theorems, per lean/NOTICE, assume a periodic similarity profile exists; the certificate should establish that assumption via external computation.
Terence Tao''s blog discussions of similar claims (cited in community write-ups, not as verification of this paper) describe the method class as numerical profile plus rigorous closure. Until the certificate file ships and third parties replay the interval code, readers should file this under claimed computer-assisted proof, not Clay closed.
Chain of nodes ending in a green check, echoing formal proof steps not yet fully verified
Image: Library reference — formal steps still need independent replay for Navier-Stokes.
Comparison to OpenAI''s September proof
| Dimension | OpenAI (Sept 2026) | Buckmaster / Alpöge (Nov 20, 2026) |
|---|---|---|
| Force term | Uses smooth external forcing in the construction | Claims none needed |
| Public Lean | Published openai/NavierStokesAndEuler with scrutiny | Large lean/ tree, not rebuilt after patch |
| External certificate | OpenAI documented its pipeline; separate critiques on scope | Hash only; file missing at publish |
| Independent replay | Months of community Lean/paper comparisons | Too early; no named endorser in Scientific American |
| Credit narrative | Dispute with Buckmaster | Same authors return with different claim |
Readers searching has Navier-Stokes been solved should expect three layers: headline tweets, formal artifact drops, and slow expert acceptance. Only the last layer moves Clay''s official problem page.
What developers and math watchers should do
If you run Lean 4:
- Clone tristanbuckmaster/navier-stokes and read
lean/README.mdbefore attempting a full build; expect large Mathlib downloads and long compile times. - Watch for a follow-up commit that rebuilds after the November 20 patch; the README warns the public tree may not match the patched statements yet.
If you follow AI math:
- Treat Alpöge''s Riemann thread and this Navier-Stokes thread as independent artifacts released hours apart; see our Masullo human GRH post for a third same-day storyline.
- Re-read Millennium fact-check patterns before updating internal slides.
If you write about it publicly:
- Attribute blowup claims to Buckmaster and Alpöge''s release plus Scientific American reporting.
- Do not state Clay closure until Clay or refereed consensus says so.
What people are asking
Is this the real Millennium problem? The authors'' framing targets the unforced blowup storyline experts wanted. Confirming that their theorems match Clay''s exact statements still requires expert reading, not README line counts alone.
Did AI write the proof? The authors characterize the text as LLM-assisted and rushed. That affects exposition quality, not whether the underlying inequalities hold. Verification is mathematical, not editorial.
Does this validate Anthropic over OpenAI? Both labs appear in the September credit fight; this release is authored by Buckmaster (NYU) and Alpöge (Anthropic employee) on their own timeline. It is not an Anthropic corporate announcement. OpenAI''s earlier result remains on its own terms.
How Clay wording maps to blowup claims
The Clay Navier-Stokes problem is not a single sentence. Prize materials describe existence and smoothness for solutions starting from smooth data with no external force, and separate breakdown formulations that can admit forcing terms. That split is why OpenAI could publish a forced blowup while experts still asked whether unforced fluids can diverge.
Buckmaster and Alpöge''s headline is the unforced branch: they argue you can pick smooth, compactly supported, divergence-free initial velocity on R³ so the Navier-Stokes evolution loses bounded energy control in finite time. The periodic torus analogues in the Lean NOTICE file are part of the same family of statements in the paper.
None of that substitutes for checking that each theorem in the PDF matches Fefferman''s official problem text line by line. Past AI math episodes, including scope critiques of OpenAI''s Lean export, show green formal checks can still miss headline mismatch. Readers should copy the Main.lean theorem statements into a notebook and compare them to Clay''s published options before repeating solved in a headline.
What to watch when the certificate lands
When the authors publish the promised certificate file, experts will likely:
- Verify the hash in the paper against the downloaded artifact.
- Replay the interval arithmetic pipeline the README describes (Arb via FLINT, 128-bit precision).
- Rebuild Lean after the November 20 patch and confirm the certificate still satisfies the patched cell bounds.
- Ask whether the numerical profile survives peer review separate from the formal bookkeeping.
Until those steps produce public logs, the honest status line is interesting claim with incomplete artifacts, not Millennium closed. explainx.ai will add an update section when a certificate URL or a named referee statement appears on an official page.
How this fits explainx.ai coverage
November 20 stacked multiple unsolved-math headlines into one news cycle. explainx.ai already hosts:
- Alpöge Riemann claimed vs verified with Sutherland checks.
- OpenAI Navier-Stokes swarm and dispute.
- Early Claude Navier-Stokes rumor fact-check.
This post adds the unforced blowup claim as its own event. If Buckmaster releases the certificate or a named mathematician publishes a refutation or endorsement, we will update here first.
What experts and builders flagged at launch
Community write-ups at launch emphasized the missing certificate and the README note that the patched Lean package was not rebuilt, the same gaps explainx.ai flagged when OpenAI''s forced proof shipped without settling the unforced case. Treat social posts as sentiment, not verification, until interval-arithmetic logs and referee reports appear.
Related reading
- OpenAI Navier-Stokes agent swarm and credit fight
- AI did not solve Navier-Stokes: explainer
- Alpöge Riemann Hypothesis claimed vs verified
- Claude Navier-Stokes rumor fact-check
- Millennium Prize AI claims fact-check hub patterns
- Masullo human GRH draft same day
Sources for this article reflect what tristanbuckmaster/navier-stokes, Scientific American, and claymath.org stated as of November 21, 2026. Mathematics status can change quickly when certificates or referee reports appear.
