The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
Notable: “read from a cached version of … The Internet”
Does this solution confirm that, provided all the conditions are met, we will be able to create a spacecraft that will fly at a speed many times greater than the speed of light, using vortex displacement, which can be achieved — I’m exaggerating, of course, but thanks to the force of two rotating magnets? And one more thing: according to Einstein’s theories, this could be considered a time machine!
Welcome to the community!
I’ve got a physics background (from ages ago) but I am by no means an expert. But it doesn’t relate to any FTL travel, and Navier-Stokes is non-relativistic so it wouldnt hold for matter moving close to the speed of light. It’s more about the general characteristics of turbulances and singularities, in a mathematical sense. Like not a singularity of the universe but more something like, when the rate at which velocity changes from point A to point B becomes infinite.
This is seriously impressive. The interesting part isn’t just that AI generated a proof, but that multiple agents were able to work together and produce a formal Lean version that can actually be checked.
If this holds up under independent mathematical review, it could be a pretty major milestone for AI-assisted mathematics. I’m especially curious to see how human mathematicians validate the key steps and whether the approach can be applied to other long-standing problems.
The combination of agents + formal verification feels like where things are getting really interesting.
There is a striking similarity between the vortical structure illustrated in this announcement and the “vortex string” geometry reported by R. C. Y. Mui, D. G. Dommermuth, and E. A. Novikov in 1996.
I have in mind Figure 1 of their paper, “Conditionally averaged vorticity field and turbulence modeling,” Physical Review E 53, 2355–2359 (1996). Figure 1 shows a thin, strongly twisted/spiraling vortical structure with an elongated axial component. The paper explicitly discusses this structure as a “vortex string” and relates its twisting component to vortex stretching.
The resemblance to the structure described here — a vortex that spirals inward and becomes increasingly elongated, while the energy remains finite — seems unusually close. The similarity is not simply that both involve vortex filaments; the overall twisting, elongation, and axial stretching are visually and geometrically quite striking.
The 1996 work was a conditional/statistical description of intense vortical structures in turbulence, based on DNS, and was not a construction of a finite-time Navier–Stokes singularity. I am therefore not suggesting that it proves the present result.
Rather, I wonder whether it would be useful to compare the actual local vorticity and velocity fields, as well as their scaling laws, in the 1996 vortex-string structure and in the present construction. There may be an interesting connection here that is worth examining.
I’m way out of my depth here but I think the quantitative comparison is absolutely worth doing!
One angle that’s gone mostly unmentioned in the coverage: the construction is entirely Eulerian. It shows a similarity region collapsing to a point and a norm diverging. Whether any actual fluid parcel rides that region in is a separate question, and the paper doesn’t address it.
It turns out to be answerable exactly. Taking the material derivative of their similarity coordinate on the axis, there’s a distinguished root where advection exactly cancels the coordinate drift — so a genuine Lagrangian lineage sits there for all preterminal time. And because the transverse velocity-gradient block in their Eq. (4.5) is scaling-plus-rotation and nothing else, those blocks commute at different times, the time-ordering collapses, and you get the deformation spectrum in closed form rather than as estimates.
The material neighborhood contracts like ϑ² while the Eulerian core radius goes like ϑ^(1/2), so the parcel is being crushed considerably faster than the structure containing it.
Worth stressing this is downstream of their theorem — if the construction has a problem, none of it survives. I wrote it up with the derivations and a reproducibility script: 10.5281/zenodo.22685519
There seems to be much confusion about what OpenAI actually did here, so allow me to provide some context.
The Millennium Prize question is not “Can someone solve the Navier–Stokes equation?” The question was much more specific:
If you start with a perfectly well-behaved fluid, is it guaranteed to remain well-behaved forever?
According to OpenAI’s claims, the answer appears to be no, as they’ve found a solution that develops a singularity in finite time, aka it breaks while still fulfilling the required starting conditions. This is broadly consistent with what most physicists would expect intuitively. The most interesting thing here appears to be that AI has solved a math problem that humans had failed to resolve for decades.