From Global Depletion to Locked Enstrophy

Abstract

From Global Depletion to Locked Enstrophy A retraction, a change of variable, and two confounds in the vorticity-alignment reading of three-dimensional Navier–Stokes Driven by Dean Kulik 10 September 2026 Abstract On 8 September 2026 OpenAI announced a proposed resolution of statements (C) and (D) of the Clay formulation of the Navier–Stokes problem: a construction, with a Lean formalization, in which the three-dimensional incompressible equations with a smooth external force develop a finite-time singularity from smooth initial data at rest with bounded energy. The claim has not been through refereed publication or the subsequent period of general acceptance that the Clay prize rules require, and the Clay Mathematics Institute continues to list the problem as open. Nothing in the present paper independently verifies or disputes that construction. Whatever its final status, the construction is a direct counterexample to a claim made in this programme in February 2026, namely that positive viscosity guarantees global regularity. That claim is retracted here without qualification. This paper reports what survives, what replaces it, and one methodological result that was not anticipated. A direct numerical simulation engine, written independently, reproduces the July 2026 Taylor–Green measurements of this programme to three or four significant figures: the strain ledger closes to 10⁻¹⁵, the enstrophy-weighted vorticity alignment flips from the most-stretching strain eigenvector to the intermediate one at the onset of nonlinearity, and the production-to-dissipation ratio crosses below unity and stays there. Those measurements stand. The reading built on them does not. The globally averaged depletion ratio D = ⟨ω·Sω⟩/⟨λ₁|ω|²⟩ is shown to be nearly blind to the configuration it was introduced to exclude. At the enstrophy peak, where D = 0.197, as much as 63 per cent of the total enstrophy could be perfectly aligned with the most-stretching eigenvector without moving D at all, because compressed regions contribute down to a pointwise floor of λ₃/λ₁ ≈ −1.17 and subsidize the average. The measured locked fraction is 4–10 per cent. The constraint on alignment locking therefore comes from the locked-enstrophy fraction, a local quantity, and not from D. Substituting that variable and sweeping Reynolds number gives a decline over the range where the numerics are trustworthy — but two confounds bound what can be read from it, and both are reported here rather than left for a referee. The first is spatial resolution. At Re = 400, degrading the resolution parameter k_max·η from 1.11 to 0.72 changes the locked fraction by a factor of 2.4, while the physical change per doubling of Reynolds number is a factor of about 2.3. Resolution error and signal are the same size, so the apparent plateau at Re = 800–3200 on a fixed 64³ grid is set by the mesh rather than by the fluid, and the sweep cannot be completed at fixed grid at all. The second is the definition of the diagnostic itself. The locked fraction depends on an alignment cutoff, and the fitted exponent over Re = 100–400 runs from −0.46 at a cutoff of 0.70 to −1.95 at 0.98 — a range wider than any value within it. No single exponent characterises the decline, and the figure of −1.24 obtained at the inherited cutoff of 0.90 is a property of that choice. What is robust is the sign: the locked fraction falls with Reynolds number at every cutoff tested. What is also robust, and more informative than the exponent it replaces, is the systematic steepening — across Re = 100 to 400 the moderately aligned population thins by a factor of 1.9 while the near-perfectly aligned tail thins by 14.9. The alignment distribution is changing shape, not merely shifting, with the extreme tail depleting roughly eight times faster than the bulk. Finally, a structural observation that clarifies why the two halves of the Clay formulation fell so differently. Every term driving an incompressible fluid is a gradient: pressure enters as −∇p, body forces as −∇Φ, viscosity as diffusion. The advective term transports but does not force. The only push in the system is the external force f, and f is precisely what statements C and D permit and statements A and B forbid. The half of the problem containing a push was resolved in eighty-eight hours; the half without one remains open. 1. Context The Clay Millennium formulation of the Navier–Stokes problem, due to Fefferman, presents four statements. Statements (A) and (B) assert existence and smoothness of solutions on ℝ³ and on the torus ℝ³/ℤ³ respectively, with the external force identically zero. Statements (C) and (D) assert breakdown on those same domains, and they permit an external force subject to smoothness and decay conditions. Resolving any one of the four resolves the problem as posed. On 8 September 2026 OpenAI announced that a coordinated multi-agent system running an internal model had produced an analytical argument, together with a Lean formalization, proposed as establishing statements (C) and (D). The status of that claim is not settled: it has not completed refereed publication, the Clay prize rules require publication followed by a period of general acceptance in the mathematical community before an award is considered, and the Clay Mathematics Institute continues to list the problem as open. The discussion below takes the construction at face value in order to test this programme’s own earlier claims against it, and none of the results reported here depends on its ultimate validity except the retraction of Section 3, which would stand or fall with it. The construction exhibits a vortex whose leading profile is self-similar, with radial width contracting faster than axial length, spiralling inward while stretching along its axis, and whose velocity diverges in finite time while total kinetic energy remains bounded. The technique follows a lineage developed analytically and without computers by Martínez-Zoroa and Córdoba, who built singularity formation from successive amplification of oscillatory layers while keeping every derivative of the source uniformly bounded. The essential move in such a construction is stated plainly in the paper itself. For any incompressible velocity field and pressure, one may simply define the external force to be the residual of the momentum equation, at which point Navier–Stokes holds by construction. The entire difficulty is to choose a field that becomes singular while that residual, and every one of its derivatives, extends smoothly through the singular time. The individual terms are permitted to diverge; they must cancel. This bears directly on the present programme, which had two positions on record. The earlier one, from February 2026, asserted global regularity for all positive viscosity. The later one, from July 2026, made no such assertion and instead measured a geometric quantity in unforced decaying turbulence, marking the decisive bound explicitly open. The first is now falsified. The second requires a change of variable, which is the substance of this paper. 2. The vorticity ledger and the alignment reading Taking the curl of the momentum equation and contracting with the vorticity gives an exact budget for enstrophy, with no modelling and no closure assumption: ½ d/dt ⟨|ω|²⟩ = ⟨ω·Sω⟩ − ν⟨|∇ω|²⟩ The first term on the right is production by vortex stretching; the second is viscous dissipation. Write P and E for the two. In two dimensions the vorticity is a scalar and the stretching term vanishes identically, which is the entire structural difference between the solved planar problem and the open three-dimensional one. Incompressibility forces the trace of the strain rate tensor to vanish pointwise, so its three eigenvalues sum to zero at every point: λ₁ ≥ λ₂ ≥ λ₃ with λ₁ > 0 and λ₃ < 0. Production is maximised when the vorticity aligns with e₁, the most-stretching eigenvector. The July 2026 paper of this programme defined the depletion ratio D = ⟨ω·Sω⟩ / ⟨λ₁|ω|²⟩ as the fraction of the available stretching that the flow actually takes up, and argued that sustained alignment with e₁ constitutes a fixed point of the direction dynamics which the flow must refuse. It reported D ≈ 0.20 at the enstrophy peak and predicted that D would remain bounded away from unity as Reynolds number increased. The idea that vorticity-direction geometry, rather than vorticity magnitude alone, controls regularity is not new here. Constantin and Fefferman established in 1993 that a pointwise Lipschitz condition on the direction field of the vorticity, imposed where the vorticity is large, suffices to rule out singularity formation; the criterion is pointwise precisely because an averaged condition cannot bound a concentrating core. Ashurst and coworkers reported in 1987 that vorticity in turbulent flow aligns preferentially with the intermediate strain eigenvector rather than the most-stretching one, an observation reproduced in essentially every direct simulation since. The contribution claimed here is neither of those. It is the identification of which measured quantity carries the constraint, and its scaling. 3. Retraction The February 2026 document of this programme states that smooth solutions exist for all time, that regularity holds for ν > 0 because viscosity implements a release pathway at all scales, and that singularity formation requires suppression of that pathway, which physical fluids do not permit. It further presents a theorem asserting that solutions remain smooth for all time under the stated dynamics. That claim is retracted. The construction announced in September 2026 has viscosity active throughout, satisfies ν > 0, and becomes singular in finite time. No forcing qualifier appears anywhere in the February text, and the claim as written is therefore a universal statement with a counterexample. It is marked dead, not softened. The accompanyi