I make no claims to mathematical mastery but after hearing about the Navier-Stokes vortex proposition, I thought it might be worth applying the technique elsewhere.
The Spaghetti Test: What Happens to the Navier–Stokes Singularity in a Magnetic Field?
AI slop, labeled as such. Drafted with a language model, [edit: definitely not] edited by a human who [absolutely does not] own[s] the errors. Posted because the question seems worth asking in public and I haven't seen anyone ask it yet. If I've missed prior work, tell me and I'll update.
The occasion
On September 8 OpenAI posted a writeup and a Lean formalization claiming finite-time blowup for 3D incompressible Navier–Stokes: smooth fluid at rest, smooth forcing, finite energy throughout. If it holds up, that settles Clay statements C and D — the forced-breakdown statements. Unforced regularity (A and B) is untouched, and Palasek has already pointed to a viscous-dissipation obstacle on the unforced side. The singular object, in the popular description, is a vortex filament that spirals inward and elongates without bound. Spaghetti.
The byline says "OpenAI." The discovery has more names than that, and I'm treating it as joint:
- Diego Córdoba and Luis Martínez-Zoroa built the program — forced blowup by making widely separated spatial scales interact, first for the incompressible porous medium equation with rough forcing (2024–25). Everything downstream is this method.
- Tristan Buckmaster (NYU) and Levent Alpöge (Anthropic) pushed the forcing from rough to smooth and carried it to Boussinesq and 3D Euler, Lean-verified August 22, posted September 7, with Matei Coiculescu as coauthor on the IPM paper. They report hypo-dissipative Navier–Stokes done and awaiting Lean. Their proofs were LLM-generated and human-rewritten, and they have been blunt about the quality of the raw model output.
- OpenAI's math team under Sébastien Bubeck, plus an unnamed internal model running on the order of 10,000 agents, took the forced Euler result and got the viscous term through. Lean via GPT-6 Astra. OpenAI's own writeup concedes priority on forced Euler to Alpöge–Buckmaster and says it won't claim the prize.
- Upstream: Hou–Luo (2014), Elgindi (2021), Chen–Hou (2022), Tao's averaged model (2016) for the vortex-stretching lineage. Buckmaster's public position is that Martínez-Zoroa deserves a Fields Medal for the forcing idea.
There is a priority dispute running in parallel. I'm not adjudicating it here. The mathematical lineage above is not disputed by anyone.
One thing the press coverage flattens: "spaghetti" is the picture, not the engine. Per Tao's read of the Alpöge–Buckmaster Boussinesq paper, the construction is a stack of high-frequency layers whose amplitude-frequency dynamics reduce to a simple ODE, with a lot of cutoff machinery around it. Any physical argument about this singularity has to survive contact with that ODE, not with a cartoon filament.
Why MHD
Nobody should be surprised the singularity is a vortex. Beale–Kato–Majda (1984): Navier–Stokes blows up if and only if ∫‖ω‖∞ dt diverges. Vorticity is the only quantity the equations permit to concentrate.
Magnetohydrodynamics is the one place where "electromagnetism also has vortices" is exact rather than a metaphor. The induction equation
∂B/∂t = ∇×(u×B) + η∇²B
is the vorticity equation with B written where ω goes:
| Fluid | MHD |
|---|---|
| Kelvin/Helmholtz: vortex lines frozen in | Alfvén: field lines frozen in |
| vortex stretching, ω ∝ length | flux conservation, B ∝ length |
| kinetic helicity ∫u·ω (Moffatt) | magnetic helicity ∫A·B (Woltjer) |
| Beltrami flow, ω ∥ u | force-free field, J ∥ B |
| viscosity ν | resistivity η |
| Beale–Kato–Majda: ∫‖ω‖∞ | Caflisch–Klapper–Steele: ∫(‖ω‖∞ + ‖J‖∞) |
The last row is the tell. MHD's blowup criterion has two terms. A singularity can live in the vorticity, as in the fluid, or in the current — and current concentrates in sheets, not strings.
The dictionary also breaks in one important place. Navier–Stokes spaghetti works because ω stretches itself: u is the Biot–Savart inverse of ω. In MHD that closure is gone. In Elsässer variables z± = u ± B, ideal MHD is ∂z±/∂t + (z∓·∇)z± = −∇p — no self-advection anywhere, only cross-stretching, and cross-stretching has an off switch: when u ∥ B locally the nonlinearity vanishes identically (Walén's exact Alfvén-wave solutions). MHD turbulence is known to seek that state (Boldyrev 2006). Navier–Stokes stretches harder as it collapses; MHD tends to switch its own nonlinearity off.
Question 1: Does a threaded field regularize the singularity? (This is Goldreich–Sridhar, applied.)
I want to be upfront that the mechanism here is not mine. It is critical balance from Goldreich & Sridhar 1995.
The Navier–Stokes singularity runs on axial strain — ∂_z u_z, variation of the flow along the filament. Thread the filament with an axial field B∥ and that becomes variation along B. Critical balance says a magnetized fluid permits exactly as much variation along B as makes the Alfvén crossing time equal the eddy turnover time, and no more: k∥ v_A ~ k⊥ δv. Applied to the collapsing core, with k⊥ ~ 1/r and δv ~ v_θ, the axial strain rate is capped at
γ ~ (v_θ / r) · (v_θ / v_A)
instead of the hydrodynamic γ ~ v_θ / r. The field throttles the engine by one factor of the Alfvén Mach number M = v_θ / v_A.
The direction of M under stretching is fixed by kinematics regardless of the detailed profile: for any incompressible elongation of the core, circulation conservation gives v_θ ∝ 1/r while flux conservation gives B∥ ∝ 1/r², so v_A grows faster than v_θ and M falls. Whatever the initial M₀, the core goes sub-Alfvénic at some finite elongation, past which further stretching costs magnetic tension energy (B²/μ₀, growing as B∥²) faster than the swirl can supply it. The forcing that drives the Navier–Stokes spiral is now working against a rubber band whose stiffness grows with extension.
The escape hatch is resistive: flux leaks out of the core on a timescale r²/η, and r is shrinking. That race — throttled stretching rate against resistive diffusion across a collapsing core — is decided by the self-similar exponents of the actual construction, which I don't have. It's also where the magnetic Prandtl number Pm = ν/η enters.
What I'm actually claiming. Critical balance, applied to the Córdoba–Martínez-Zoroa forced singularity, predicts regularization for any smooth B with nonzero flux through the collapsing core, in ideal MHD and in resistive MHD with Pm not too small. The swirl doesn't vanish; it converts to torsional Alfvén waves along the core.
What I'm not claiming. A number. An earlier draft of this post had ω_max ∝ B₀⁻², derived by treating the core as a material tube of fixed volume. That assumption gives core kinetic energy growing without bound, which contradicts the finite-energy property of the proof — the core is not a material tube, fluid spirals in and out of it. The sign of the argument survives (v_A outruns v_θ under any incompressible stretching); the exponent does not until it's redone against the real profile. There should be a power law in ω_max versus B₀. I don't yet know what it is.
Why GS95 doesn't already answer this. Critical balance is a statement about a statistical steady state in strong turbulence with δB ~ B₀ at the outer scale. It says nothing about whether a specific smooth solution reaches infinity in finite time, and it doesn't cover the weak-field regime where the core starts super-Alfvénic and crosses over mid-collapse. Whether critical balance can serve as a regularity mechanism for a constructed singularity — and whether it does so for arbitrarily small B₀ — is a question GS95 doesn't pose. Critical balance is itself a conjecture (Boldyrev's dynamic alignment already amends it), so leaning on it here is a claim, not a citation.
Question 2: If MHD blows up, is it a sheet rather than a string?
This one I think is open and unasked.
The Alpöge–Buckmaster Boussinesq result is the nearest neighbor. 2D Boussinesq and 2D MHD are structural siblings — vorticity plus a transported field that sources it — and 2D Boussinesq has long been the model problem for axisymmetric Euler with swirl. So the mathematician's natural next question in the Córdoba–Martínez-Zoroa program is: does the multiscale construction produce finite-time blowup for 2D or 3D MHD with smooth forcing? I don't see that posted anywhere.
My guess at the answer, framed as a conjecture: any finite-time singularity of ideal MHD produced this way lives in the current, not the vorticity. ∫‖J‖∞ dt diverges while the vortex channel is shut down by alignment in the core. Physically: the spiral wraps any perpendicular field into nested sheets, and sheets are what MHD actually makes. The Orszag–Tang vortex — the canonical MHD test problem, literally a vortex — is the initial-condition family to probe this with, because it is known to form sheets.
What's already on record (verify before citing; I'm working from memory on the older items)
- Goldreich & Sridhar, ApJ 1995 — critical balance. Boldyrev, PRL 2006 — dynamic alignment.
- Caflisch, Klapper, Steele, Comm. Math. Phys. 1997 — the two-term MHD blowup criterion.
- Kerr & Brandenburg, PRL 1999 — numerical evidence for an ideal-MHD singularity, sheet-type. Grauer & Marliani, PRL 2000 — higher resolution, same class of data, growth looks exponential rather than finite-time. The vortex-vs-sheet ambiguity has been live for 25 years.
- Frisch, Pouquet, Sulem, Meneguzzi 1983 — original ideal-MHD singularity search; Orszag & Tang, JFM 1979 for the initial condition.
- Serrin-type blowup criteria for compressible MHD exist that are independent of B. Those are criteria, not constructions: they say B can't be the only thing that blows up, not that B can't prevent blowup. Not to be confused with Question 1.
- Global regularity of 2D ideal MHD is still open. 2D Euler is fine; adding B makes even two dimensions hard, because u can stretch B in the plane. MHD is not "Navier–Stokes plus a term." It is a different animal with the same skeleton.
The concrete proposal
On paper. Take the amplitude-frequency ODE from the Alpöge–Buckmaster Boussinesq paper and add the Alfvén term — the restoring force a threaded B∥ exerts against the axial strain of each layer. Does the ODE still run off in finite time? That is a two-line calculation for someone who has read the paper, and it either kills Question 1 or promotes it.
On a GPU. Periodic box (statement D is on the torus, so that's the right setting). Pseudo-spectral; Dedalus is enough. Implement the forcing from the writeup. Add a uniform B₀ẑ along the spiral axis, Pm = 1, scan B₀ over three decades. Track ‖ω‖∞, ‖J‖∞, cross-helicity, and the local u–B angle in the core. Then repeat with B₀ perpendicular.
Predictions: axial B₀ — ‖ω‖∞ peaks and turns over, peak value a decreasing power of B₀ (exponent to be determined), energy appearing in torsional Alfvén modes. Perpendicular B₀ — no vorticity peak, ‖J‖∞ growth in a sheet, growth exponential rather than algebraic-to-a-pole.
The magnetized run is cheaper than the unmagnetized one. You only resolve the core down to the Alfvén crossing, not to zero. Arrested singularities are the easy ones.
Why anyone should care
If Question 1 resolves yes, critical balance is not just a turbulence phenomenology but a regularity mechanism, and a 1995 astrophysics paper has something to say about a 2026 Millennium result. If Question 2 resolves as conjectured, the MHD singularity problem was never a vortex problem — it's a reconnection problem, and the 25-year numerical stalemate is because everyone was looking for spaghetti in a system that only makes lasagna. Either way, the next question is not "does MHD blow up" but "which of its two channels does the geometry permit," and the fluid result just handed us the initial condition to ask it with.
I'll run the axial case locally once the forcing profile is extractable from the proof. If someone beats me to it — or has already done the ODE — post it and I'll update this with your name on it.
