The Pulse
New Paper Tests the Conditions Behind OpenAI’s Navier–Stokes Claim
A paper by Peter Constantin, Mihaela Ignatova and Vlad Vicol examines two structural features of OpenAI’s proposed finite-time Navier–Stokes singularity. The authors show that the same behavior cannot produce a singularity when the external
AI.info Team ·
Two specific features of OpenAI’s proposed Navier–Stokes singularity are enough to force regularity under a different assumption about the external force, according to a paper posted September 17 by Peter Constantin, Mihaela Ignatova and Vlad Vicol.
The authors do not claim to disprove OpenAI’s construction. Their result is narrower: if a solution has the same anisotropic bounds and exact axisymmetry in its collapsing core, and if the body force is real analytic while remaining bounded in the required way, then the putative singularity cannot occur. The paper therefore places a concrete restriction on the kind of forcing OpenAI’s construction must use.
Two features at the center of the dispute
The paper, Regularity of asymptotically axisymmetric solutions to the 3D Navier-Stokes equations with analytic forcing, starts from the construction described by OpenAI’s September 8 announcement. OpenAI says its internal system produced a proof that a three-dimensional incompressible fluid, initially at rest and subject to a smooth external force, can develop unbounded velocity in finite time while retaining finite energy.
Constantin, Ignatova and Vicol isolate two properties they identify in that construction. The angular mean of the solution follows anisotropic Type II bounds, and the flow becomes exactly axisymmetric in a collapsing core region. Their paper studies solutions that share those properties but operate under a real-analytic body force.
That distinction matters because “smooth” and “real analytic” describe different levels of regularity. A smooth function can have derivatives of every order without being determined locally by a convergent power series. Real analyticity imposes a stronger local structure, and the authors use that structure to rule out the singular behavior in the setting they analyze.
Zooming in on the proposed singularity
The proof proceeds by examining the flow at progressively smaller scales near the point where the singularity would form. The authors use the anisotropic length scales supplied by the assumed bounds, rather than treating the collapse as uniform in every direction.
That rescaling produces what the paper calls ancient limits: limiting solutions defined over an unbounded interval extending backward in time. The authors then use the equations governing those limits to obtain additional rigidity. Under the paper’s hypotheses, the limiting behavior is incompatible with a genuine singular point, so the original solution remains regular there.
The abstract does not present the result as a general theorem that all forced Navier–Stokes solutions remain regular. It applies to solutions with the two selected structural properties and to forcing with the stated analytic and boundedness conditions.
What the result says about OpenAI’s force
The paper’s main consequence is conditional but direct. In OpenAI’s construction, or in any related construction with the same two properties whose force remains bounded in C2 up to the singular time, the force cannot both be real analytic in the spatial variables and support the proposed singularity.
The authors also conclude that the force cannot vanish identically near the singular point under those conditions. Put differently, the external forcing must retain a more complicated local structure than an analytic force that stays controlled through the approach to blowup.
That conclusion does not address every possible way to construct a finite-time singularity. It restricts a specific mechanism: anisotropic Type II behavior, an exactly axisymmetric collapsing core and a force that remains bounded in the relevant norm.
OpenAI’s claim remains broader than this paper
OpenAI’s announcement describes a vortex that spirals inward and stretches along its axis as the central region contracts. The company says the velocity grows without bound while the kinetic energy stays finite, with acceleration, pressure, momentum transfer and viscosity becoming large but cancelling in a precise way.
OpenAI also says the result concerns a smooth external force rather than an unbounded force inserted by hand. The company presents the work as a resolution of the finite-time blowup alternative in the Clay Mathematics Institute’s formulation, but states that it does not intend to claim the Millennium Prize for the result.
The new paper does not independently verify or reject OpenAI’s full write-up or its Lean formalization. Its contribution is a mathematical obstruction to one class of forcing compatible with the geometry and scaling behavior OpenAI describes.
Why the forcing assumption carries so much weight
External forcing is not a minor detail in this episode. OpenAI’s result concerns the forced three-dimensional equations, while the Clay problem is commonly framed around whether smooth initial data can remain regular for the unforced system. The force supplies energy and structure to the flow, even though OpenAI says the construction keeps the total kinetic energy finite.
By showing that analyticity blocks the proposed singularity under the selected conditions, the authors narrow the available design space for such examples. A successful construction of that form must use forcing that is smooth but not real analytic near the singular point, or must depart from one of the geometric and scaling assumptions examined in the paper.
That is the concrete outcome of the September 17 preprint: not a verdict on OpenAI’s entire claim, but a proof that the claim’s described mechanism cannot survive an analytic forcing assumption with bounded C2 behavior.