Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsAccording to a recent arXiv critique, parts of OpenAI’s Lean formalization do not match the corresponding claims in its written Navier–Stokes proof. That is a documented concern about translation—not a verdict that the entire proof is wrong, that Lean failed to check its encoded theorem, or that the claimed result settles the intended problem.
What does “mistranslated” mean in this case?
OpenAI says it produced a written proof of finite-time singularity for the Navier–Stokes equations and a formalization in Lean, a proof assistant. Lean checks whether a theorem follows from the definitions and proof steps represented in its formal environment. But checking the formal proof does not, by itself, establish that the formal statement faithfully captures the theorem described in natural-language mathematics. That correspondence is a separate question at the center of the critique. OpenAI’s announcement describes its claimed result and formalization; the arXiv paper examines whether the code matches the prose.
As an Amazon Associate I earn from qualifying purchases.
What differences did the paper identify?
A derivative estimate in Lemma 8.6
The paper’s authors compare a written estimate in Lemma 8.6 with the corresponding Lean declarations. They argue that the written estimate claims control with one fewer input derivative than the formalized estimate appears to require: in their account, the difference is between an m+4 and an m+5 derivative requirement. This is the authors’ technical comparison, not an independent audit of every part of the formalization. Read the paper’s analysis.
A pressure-flux bound
The same authors report a mismatch between the written pressure-flux bound and its Lean counterpart. They say the formal estimate depends on an additional quantity that does not appear in the written bound, and that the argument in the two versions differs. The paper presents these as examples supporting its concern about correspondence; they should not be inflated into a claim that every Lean declaration is defective. The paper details the comparison.
#1 Best Overall
Which questions are still separate?
The episode involves several questions that are easy to collapse into one another. Evidence about one does not automatically answer the others:
| Question | What it asks | What the cited material establishes |
|---|---|---|
| Does the Lean development prove its encoded theorem? | Do the formal proof steps follow from the definitions and assumptions in the Lean environment? | OpenAI says it supplied a Lean formalization. The critique discussed here focuses on correspondence with the prose; it does not establish that Lean’s checking of the encoded theorem failed. OpenAI’s account; the critique. |
| Does the encoded theorem match the written theorem? | Does the formal statement faithfully express what the natural-language proof claims? | The arXiv authors identify differences in the derivative estimate and pressure-flux bound described above. Their paper. |
| Is the written proof valid? | Do the mathematical arguments in the prose establish the claimed result? | The cited critique raises concerns about particular claims and arguments. The reviewed reporting does not supply a completed independent verdict on the entire written proof. Science News’ reporting. |
| Does the result address the intended problem? | Is the mathematical setting the one experts consider relevant to the Navier–Stokes question? | This is a separate dispute about scope and significance, not a consequence of a prose-to-code mismatch. Scientific American’s reporting. |
Has a human verified the proof?
At the time of Science News’ 2026 report, mathematical physicist Gregory Eyink of Johns Hopkins University said: “I don’t think anyone has completely verified the proof yet, certainly not on the human side.” That quotation describes the state of review when the report was published; it is not a guarantee about the status of review at a later date. The arXiv critique is a substantive technical examination, but its publication should not be mistaken for a completed independent verification of every part of the proof. Science News’ report.
Is this also a dispute over which Navier–Stokes problem was solved?
Yes, but it is a different dispute. Scientific American reported criticism that the result may concern a variant some experts consider disconnected from physical reality or less interesting. That criticism concerns whether the chosen setting answers the intended mathematical or physically motivated question; it does not establish whether the Lean code faithfully represents the written proof. Scientific American covers the scope debate.
What does OpenAI say about the project’s origins?
OpenAI says the work began after its team heard a rumor it later connected to Tristan Buckmaster and Levent Alpöge, whose result OpenAI characterizes as concerning forced Euler. The company says it offered them access to its prompts and later proof, and recognizes their priority on forced Euler. These are OpenAI’s descriptions; the cited sources do not independently resolve every question about priority or access to data. OpenAI also describes the group working on its Navier–Stokes result as involving on the order of 10,000 concurrent agents and says it does not intend to claim the Millennium Prize for the result. Those details describe the company’s account, not evidence that the proof is correct. OpenAI’s announcement.
Quick Recap
Rank #4
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
