TL;DR
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
An arXiv preprint submitted Oct. 6, 2026, argues that a Lean proof associated with OpenAI’s announced Navier–Stokes blow-up result does not faithfully represent the natural-language argument. The authors also contend that resolving ambiguity during mathematical autoformalisation can be computationally intractable; the paper is a preprint, and its claims have not been independently established by the source material.
A preprint submitted to arXiv on Oct. 6, 2026, argues that a Lean formalisation associated with OpenAI’s announced proof of blow-up for the Navier–Stokes equations does not match the natural-language argument it is presented as verifying. The authors say the case illustrates a broader limitation: mechanically checking a formal proof does not, on its own, establish that the proof faithfully captures the original mathematical text.
The paper, titled Navier–Stokes Lost in Translation, examines autoformalisation, in which an AI system translates mathematics written in ordinary language into a formal language such as Lean. A proof assistant can check whether the resulting formal argument follows the rules of its formal system. The authors argue that this check cannot validate the translation’s meaning unless the translation itself is faithful.
In its abstract, the paper says it provides examples of AI mistranslations of mathematical statements and proofs into Lean, producing differences between the original text and what the formal system verifies. It identifies the OpenAI Navier–Stokes announcement as one example and states that the formalised Lean proof does not correspond to the natural-language proof of blow-up. This is the authors’ finding; the supplied source does not include an independent assessment or a response from OpenAI.
The authors also make a theoretical argument about ambiguity in mathematical prose. They say resolving ambiguities needed for a semantically faithful translation can sit arbitrarily high in the Solvability Complexity Index hierarchy, which they describe as SCI = ∞. In informal terms, they argue that the general task can exceed the difficulty of any computational problem, including the Halting problem. The paper is a 25-page arXiv submission, not a peer-reviewed publication according to the supplied listing.
Why Formal Proof Checks Can Mislead
The dispute concerns what a successful proof-assistant check does and does not establish. A Lean verifier can confirm that a formal proof follows from the definitions and assumptions encoded in Lean. But readers also need confidence that those definitions and assumptions express the intended mathematical claim. If the translation changes the meaning, a valid formal proof may certify a different proposition from the one stated in ordinary language.
That distinction matters as researchers and technology companies use AI to translate mathematical work into machine-checkable form. Formal verification can help identify errors in a formal argument, but this paper warns that it cannot automatically settle whether the formal argument preserves the source text. For the high-profile Navier–Stokes example, the authors’ claim, if upheld, would raise questions about the relationship between OpenAI’s announcement and the Lean artefact cited in connection with it. The preprint alone does not resolve the mathematical status of the claimed blow-up result.
mathematical proof assistant software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Mathematical Prose to Lean
Autoformalisation converts statements and reasoning written in natural language into a formal system, where terms and proof steps must follow precise rules. Lean is a proof assistant used to construct and check such formal arguments. The appeal is that a machine can verify the formal proof against the system’s rules, reducing reliance on informal inspection of each step.
The paper focuses on the stage before that check: interpreting the original text. Mathematical writing can leave assumptions, definitions, or scope implicit, and a translation must decide how to represent them. The authors’ central point is that verification of the translated object and verification of the original argument are separate tasks. Their abstract says they illustrate this with several practical examples, including the OpenAI-associated Navier–Stokes proof, but the supplied material does not give the examples’ full technical details.
The Navier–Stokes equations describe fluid motion and are central to mathematical analysis and physics. The source identifies OpenAI’s result as an announced proof of blow-up, meaning the development of a singularity in a solution. The preprint addresses whether a formalised proof corresponds to its natural-language counterpart; it does not, in the provided abstract, establish a general resolution of the underlying Navier–Stokes problem.
“The formalised Lean proof does not correspond to the NL proof of blow-up of solutions to the Navier-Stokes equations.”
— The authors of the arXiv preprint, in its abstract
Questions About the Translation Claim
The supplied source is the preprint’s abstract and submission record, so it does not provide enough detail to independently evaluate the authors’ technical analysis or reproduce their comparison of the natural-language proof and Lean formalisation. The listing describes a v1 submission and does not indicate peer review. No response from OpenAI or an independent mathematician is included in the source material.
It is also unclear from the abstract what precise Lean statement was checked, which parts of the translation are said to diverge, or whether the authors’ conclusion applies to the full announced argument or to a particular formalised version. The preprint’s broad complexity claim is the authors’ characterization; its implications for practical autoformalisation systems remain to be assessed. The source does not establish whether OpenAI’s underlying mathematical claim is correct or incorrect.
What Review Could Establish
The next step is scrutiny of the preprint’s detailed examples and formal analysis by other researchers. Readers will need to compare the relevant natural-language statements, definitions, assumptions and Lean code to determine whether the alleged mismatch is accurately described. Any response or clarification from OpenAI could also explain how the formalisation relates to its announced proof.
The arXiv record gives the paper’s submission date as Oct. 6, 2026, and identifies it as a 25-page preprint. The supplied material names no scheduled review, conference presentation or follow-up publication. Until further analysis is available, the reported mismatch should be attributed to the paper’s authors, while the status of the announced Navier–Stokes proof remains unresolved by this source.
Key Questions
The authors say the Lean proof does not correspond to the natural-language proof of blow-up associated with OpenAI’s announcement. That is the preprint’s claim, not an independent finding in the supplied source.
Does Lean verify the original natural-language proof?
Lean checks a proof written in its formal language against formal rules and definitions. The paper argues that this does not by itself show that the formal version faithfully represents the original mathematical text.
Has the paper been peer reviewed?
The supplied arXiv record identifies it as a 25-page preprint submitted on Oct. 6, 2026. The source does not report peer review.
No. The preprint challenges the correspondence between a natural-language argument and a Lean formalisation. The supplied material does not settle the mathematical correctness of the announced result.
Source: hn
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
