VibeMathedMath problems solved with AI

Navier–Stokes Millennium Prize problem: finite-time breakdown with smooth forcing

For the three-dimensional incompressible Navier–Stokes equations with positive viscosity, do there exist smooth divergence-free initial data and smooth external forcing for which a global smooth physically reasonable solution does not exist? The Clay Millennium Prize formulation allows this to be established either on R^3 (statement C) or on the periodic three-torus (statement D).

Result
Proved(see note)
Status
Candidate (review pending)
AI contribution
AI-discovered
Method
Construction
Field
PDEs; fluid dynamics; singularity formation
Posed by
Charles L. Fefferman, Clay Mathematics Institute
Year posed
2000
Years open
26y
Solved
2026-09-05
Model
Unnamed internal OpenAI model
Vendor
OpenAI
Collaborators
Verification
Lean-verified
Publication
Preprint
Significance
87 / 100
Disclosed cost
Wikipedia
No dedicated article

What was actually shown

OpenAI claims to construct a smooth finite-energy solution of the three-dimensional incompressible Navier–Stokes equations, starting from a fluid at rest and driven by smooth forcing, that develops a finite-time singularity. The construction is claimed both on R^3 and in the periodic setting, thereby establishing statements C and D of Fefferman's official Clay formulation. If the correspondence between the released proof/formalization and the Clay statements survives independent review, this resolves the Navier–Stokes Millennium Prize problem. It does not establish finite-time blow-up for the unforced Navier–Stokes equations; the Clay resolution comes through the smooth-forcing alternatives C/D.

What the AI did

OpenAI launched a multi-agent research effort using an unnamed internal model significantly more capable than GPT-6 Astra. Roughly 10,000 concurrent agents explored the Navier–Stokes Millennium Prize problem, with separate groups assigned the different Clay formulations and related subproblems. After the system independently found an unforced 3D Euler blow-up construction, that result was supplied to the Navier–Stokes agents. Codex was used to consolidate useful intermediate insights across groups. The system produced the analytical finite-time singularity construction after approximately 88 hours. GPT-6 Astra was subsequently used to formalize and verify the result in Lean.

Verification

Audited here on 9 September 2026. Statement anchor: OpenAI's ComparatorChallenges/NavierStokes.lean was diffed against google-deepmind/formal-conjectures at commit 8bf45ed, the commit its header names. The two compared theorems, navier_stokes_breakdown_R3 and navier_stokes_breakdown_periodic, are byte-identical to DeepMind's, as are the decay, periodicity and energy conditions they depend on; the only differences in the file are nine deleted lines carrying Formal Conjectures' alternatives (A) and (B), which OpenAI does not claim. The statement is therefore anchored to an independent third party's formalisation of Fefferman's conditions, not to one written by the prover. Counts taken independently: 2,486 Lean files, 616,276 lines, zero sorry outside the deliberate challenge placeholders, zero axiom declarations, zero native_decide, zero unsafe. comparator.json permits only propext, Quot.sound and Classical.choice and enables the independent nanoda kernel; the solution module states both theorems verbatim and discharges them via ComparatorBridge. In the paper, Theorem 1.1's force lies in C^inf_c(R^3 x (0,inf)), compactly supported in space and time, satisfying Fefferman's decay condition (5) outright. Not rebuilt here. No mathematician has read the argument: released 8 September 2026, no referee, and Clay's own process requires publication in a refereed journal plus two years of general acceptance.

Sources

Submitted by LucidHawk551 on

Changelog3 changes

Discussion1

AmberJackal71409 Sep 2026, 13:22 UTC

Should the unforced Euler equation blowup be moved into a separate result? It is an important result in its own right.

0