Navier–Stokes

A guide to the forced blow-up of September 2026, and to whether a fluctuating fluid could reach it.

On 8 September 2026 OpenAI released a 166-page construction of a smooth, compactly supported force under which the three-dimensional Navier–Stokes equations, started from rest, develop unbounded velocity at time one while the kinetic energy stays bounded. That is alternative (C) of the Clay problem as Fefferman stated it in 2000. This site exists to understand the construction, one mechanism at a time, and then to ask a physical question about it.

The question is this. Real fluids are not the continuum of the equations. They carry thermal noise, they have a molecular scale, and any turbulent fluid has fast motions that a model of the slow flow treats as fluctuations. Would a fast-fluctuating version of Navier–Stokes, arguably the more accurate model, still blow up under this force? The fluctuations pages collect what is known and what is not.

The three regions of the construction at one instant before the singular time. Fluid spirals into a column that thins faster than it shortens; the column's own momentum balance closes, the exterior's closes, and the annulus between them does not. Pulses in the annulus close it.

The result

Write $\tau = 1 - t$ for the time remaining. The theorem is stated for every viscosity.

$$\partial_t u + (u\cdot\nabla)u - \nu\Delta u + \nabla p = f,\qquad \nabla\cdot u = 0,\qquad u(\cdot,0)=0,$$

with $f \in C_c^\infty(\mathbb{R}^3\times(0,\infty))$, $u$ and $p$ smooth on $[0,1)$ with support in a fixed compact set, $\sup_{t<1}\|u(t)\|_{L^2}<\infty$ and $\limsup_{t\uparrow 1}\|u(t)\|_{L^\infty}=\infty$.

In English: a fluid at rest, pushed by a smooth push that switches off before the singular time and vanishes outside a ball, can be made to spin up without bound at a single point while carrying only a bounded amount of energy. No smooth solution with the same force and bounded energy can therefore exist for all time. Compact support gives the periodic version as a corollary, which is alternative (D).

The construction has one idea and a great deal of bookkeeping. The idea is that the force is whatever the momentum equation leaves over once you have chosen the flow. The difficulty is to choose a flow that blows up while that leftover stays smooth. The residual trick page states this precisely and the scales, core, pulses and corrections pages follow the paper's own order.

Scope

The theorem settles the forced alternatives (C) and (D). The unforced alternatives (A) and (B), which is what most people mean by the regularity problem, are untouched. The force is smooth but it is not real analytic, and Constantin, Ignatova and Vicol have since shown that any construction with this one's anisotropic scaling and axisymmetric core needs a non-analytic force that stays active at the singular point. The Lean formalization builds from source with no sorry and only the three standard axioms, checked here on 24 September; the formal statement page compares it with Fefferman's conditions clause by clause. The Clay Mathematics Institute describes the problem as apparently settled and its evaluation as ongoing. The status page keeps these facts dated.

The fluctuation question

The construction seeds each pulse with an exponentially small force and lets the background shear amplify it. It then cancels the singular part of the residual to every order. Both steps are fine-tuned. A fluid with thermal noise seeds the same instabilities at random phase and with far larger amplitude, a fluid with a molecular scale stops resolving the core once its radius reaches a mean free path, and an equation with transport noise is now known to have global smooth solutions with high probability, force included. None of this finds an error in the proof. All of it bears on whether the proof describes a fluid. The fluctuations page sorts the candidate modifications by what is proved, the molecular limit page computes where a real fluid stops following the equations, and the seeds page shows the amplification arithmetic.

Where to start. Read the problem for the four alternatives, then the residual trick. The literature map places the construction among its ancestors, and the timeline records September 2026.

Cite

@misc{openai2026navierstokes,
  title  = {Finite Time Blowup for Navier--Stokes},
  author = {{OpenAI}},
  year   = {2026},
  note   = {Manuscript, 166 pp., released 8 September 2026},
  url    = {https://openai.com/index/navier-stokes-solution/}
}

Comments and corrections are welcome as issues on the repository.