Searcharxiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.08144

Navier-Stokes lost in translation: Why Lean verification of AI autoformalisation does not guarantee correct natural language proofs

Alexander Bastounis·Fabian Circelli·Anders C. Hansen
View original source ↗Read source document ↗Export metadata ↓

Abstract

Autoformalisation is increasingly used to verify mathematical texts, including those generated by AI, as in OpenAI's announced proof of blow-up of solutions to the Navier-Stokes equations. In this process, an AI system translates the text from a natural language (NL) into a formal language such as Lean. Once this translation is done, the argument expressed in the formal language can easily be mechanically verified. The purpose of this article is to demonstrate why this process may offer no confidence in the original NL argument, owing to the various difficulties in performing the translation semantically faithfully. In particular, we highlight that the problem of resolving ambiguities in mathematical NL text, which is necessary in order to provide semantically faithful translation, is arbitrarily high up in the Solvability Complexity Index (SCI) hierarchy/arithmetical hierarchy (the SCI $= \infty$). Hence, informally, providing semantically faithful AI autoformalisation is harder than any computational problem including the Halting problem (which has SCI $= 1$). To demonstrate the effect of this result we provide several examples of AI mistranslations of NL statements and proofs into Lean in practice, resulting in mismatches between NL proofs and their Lean `verifications'. These include OpenAI's announced Navier-Stokes proof. In particular, we show that the formalised Lean proof does not correspond to the NL proof of blow-up of solutions to the Navier-Stokes equations.

Explore related subjects

math.APcs.AImath.LO

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Alexander Bastounis, Fabian Circelli, Anders C. Hansen. 2026-10-06. Navier-Stokes lost in translation: Why Lean verification of AI autoformalisation does not guarantee correct natural language proofs. https://arxiv.org/abs/2610.08144

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Global-in-time Well-posedness of Classical Solutions to the Vacuum Free Boundary Problem for the 1-D Viscous Saint-Venant System with Large Data

We establishes the global existence and uniqueness of classical solutions to the vacuum free boundary problem for the 1-D viscous Saint-Venant system with a general class of large initial data. Since the fluid depth $ρ$ vanishes on the moving boundary, the momentum equations degenerate in both time evolution and spatial dissipation, potentially causing singularities in velocity derivatives and making classical solutions difficult to construct. By exploiting the intrinsic degenerate-singular structures, we identify admissible initial depth profiles for which $ρ_0^α$ belongs to $H^3$ and vanishes as the distance to the moving boundary, with $\frac{1}{3}<α\le 1$. In particular, for $α=1$, it satisfies the physical vacuum boundary condition but violates the BD entropy condition. First, we introduce new weighted nonlinear energy functionals involving lower- and higher-order derivatives, based on the balance between pressure and spatial dissipation. Second, we derive new global weighted $L^p$ estimates, $2\le p\le\infty$, for the effective velocity $v=u+(\logρ)_y$, where $y$ is the Eulerian coordinate, using the transport properties of its evolution equation. These estimates differ from those of Bresch-Desjardins (Comm. Math. Phys. 238 (2003), 211-223) and Kanel (Differ. Uravn. 4 (1968), 721-734) and are crucial for $α=1$, where the BD entropy condition fails. Finally, we establish weighted energy estimates for first-order velocity derivatives with weights involving powers of $ρ_0$ and $η_x$, where $x$ is the Lagrangian coordinate and $η$ the flow map. These estimates give an upper bound for $η_x$ on every finite time interval without any smallness assumption on the initial data. Further singular or degenerate weighted energy estimates yield the desired global regularity.

math.AP↗
arXiv2026

Desingularization of vortex sheets for the 2D Euler equations

We show how to regularize vortex sheets by means of smooth, compactly supported vorticities that asymptotically evolve according to the Birkhoff-Rott vortex sheet dynamics. More precisely, consider a vortex sheet initial datum $ω^0_{\mathrm{sing}}$, which is a signed Radon measure supported on a closed curve. We construct a family of initial vorticities $ω^0_\varepsilon \in C^\infty_c(\mathbb{R}^2)$ converging to $ω^0_{\mathrm{sing}}$ distributionally as $\varepsilon \to 0^+$, and show that the corresponding solutions $ω_\varepsilon(x,t)$ to the 2D incompressible Euler equations converge to the measure defined by the Birkhoff-Rott system with initial datum $ω^0_{\mathrm{sing}}$. The regularization relies on a layer construction designed to exploit the key observation that the Kelvin-Helmholtz instability has a strongly anisotropic effect: while vorticities must be analytic in the "tangential" direction, the way layers can be arranged in the "normal" direction is essentially arbitrary.

math.AP↗
arXiv2026

Global-in-time convergence from bipolar Euler-Poisson equations to unipolar ones

In this paper, the Cauchy problem for the multi-dimensional (M-D) bipolar Euler-Poisson equations with far field vacuum is considered. Based on physical observations and some elaborate analysis of this system's intrinsic structures, for a class of smooth initial data that are of small scaled density but possibly large mean velocity, we give one rigorous global-in-time convergence proof for regular solutions from M-D bipolar Euler-Poisson equations to M-D unipolar Euler-Poisson equations through the vanishing electron-to-ion mass-ratio limit. Here the initial scaled density is required to vanish in the far field, and the spectrum of the Jacobian matrix of the initial mean velocity stays uniformly away from the nonpositive real axis. In order to deal with singular limits of this kind, the global-in-time uniform tame estimates of regular solutions to M-D bipolar Euler-Poisson equations with respect to the mass ratio are established, based on which the corresponding error estimates in smooth function spaces between the two systems considered are given. To this end, we establish global a priori estimates for solutions with compactly supported initial densities, uniformly in the mass ratio and the initial support radii. Compactness then yields global regular solutions for general initial densities.

math.AP↗