Fixed Euclidean space, Newtonian time, and omnium flow

OpenAI’s Navier–Stokes construction drives an incompressible fluid on fixed Euclidean space with Newtonian time to a finite-time blowup of its velocity, from rest, with a smooth force of compact support, and presents it as Fefferman’s alternative (C). FDT reads the fluid’s pressure gradient as the medium’s clock factor, its enthalpy per count, and the source’s equation as the slow limit of the medium’s free fall. On the fixed flat domain a maintained velocity bound and positive viscosity give ‖∇u(t)‖₂² ≤ ‖∇u₀‖₂² exp(C_E²t/ν) and global strong continuation, and the helical average of the omnium supplies the bound c wherever the fluid velocity is the resolved transport in one flat frame. The construction’s blowup continues the slow limit past c while its rulers, clocks and differential operator stay fixed.

Download the paper (PDF, 35 pages, 0.6 MB) Go to the figure

The problem as posed

The construction and its background

The Clay problem asks whether smooth, divergence-free initial data for the incompressible Navier–Stokes equations, on ℝ³ or on a three-torus, always extend to a smooth global solution of finite energy, or whether they can break down. The OpenAI manuscript states in its Theorem 1.1 that for every viscosity ν > 0 there are a smooth force of compact support and smooth velocity and pressure fields on ℝ³ × [0, 1), starting from rest and supported in a fixed compact set, whose kinetic energy stays bounded while the supremum of the velocity diverges as t → 1. It presents this as Fefferman’s alternative (C), breakdown on ℝ³, and the periodic alternative (D) through its Corollary 10.6. The force is designed backwards from the constructed velocity and pressure, extended smoothly past the blowup time, and held fixed for every comparison solution, and the source’s energy identity, with its Lemma 10.4, keeps the work and the dissipation finite: the blowup comes with finite forcing work.

The construction’s gradient, Laplacian, volume measure and norms are Euclidean and its time coordinate is Newtonian. In the source’s fixed background the problem is

∂_tu + (u·∇)u = −∇π + ν∆u + f∇·u = 0, ∂_tδ_ij = 0

There is no dynamical lapse, spatial metric or source equation in this system. Increasing velocity, rearranging pressure and performing work leave its rulers and its time parameter fixed by definition. Along the path the source follows into the blowup the leading swirl grows as τ^−1/2−h, with τ = 1 − t and the source’s exponent 0 < h < 1/100, on a radius that shrinks as τ^1/2, while the axis stays smooth at every earlier time.

What the construction fixes and what it states, as the paper reads it, with the class of each entry. Exponents and normalizations are the source’s.
Item	Value	Class
Domain and time	ℝ³ × [0, 1), Euclidean space, Newtonian time	construction
Viscosity	constant, ν > 0	construction
Initial velocity	zero	construction
Force	smooth, compact support, designed from the solution	construction
Work and dissipation	finite, by the energy identity	stated
Swirl on the blowup path	τ^−1/2−h, 0 < `h` < 1/100	stated
Radius of the path	shrinks as τ^1/2	stated
Clay alternative	(C) on ℝ³; (D) on the three-torus	stated

The FDT reading

0.1c The medium’s clock factor, 0.1d the strand rate, I-⁠3 the evolving metric

The pressure gradient is the medium’s clock factor

The source writes the fluid as a medium pushed by pressure forces across a fixed background. In FDT a medium’s patterns move free of every force in the rescaled metric 𝔪²g_μν, whose clock factor is the enthalpy per count, 𝔪 = 𝔥_N/𝔥_N,0 with 𝔥_N = (E + PV)/N, and the pressure force is the bookkeeper’s reading of their common rescaling (0.1c). For a stress tensor of perfect form at fixed entropy per count, with 𝔓^μν = c²g^μν + 𝖴^μ𝖴^ν,

(E/V + P) ∇_𝖴𝖴^μ = −𝔓^μν∂_νP, dP = n d𝔥_N⇒ ∇_𝖴𝖴^μ = −𝔓^μν∂_ν ln 𝔥_N 0.1c

For a liquid of constant density, 𝔥_N = m₀c²(1 + π/c²), and the slow limit is ∂_tu + (u·∇)u = −∇π, up to corrections of relative order |u|²/c² and π/c². The source’s equation is that slow limit, its viscous term the first-order stress of a causal relativistic fluid and its force the drive’s exchange. The blowup continues the limit past |u| ∼ c and π ∼ c², where the medium’s patterns keep their resolved transport below c.

Continuation on the fixed flat domain

Take the fixed-domain equation on ℝ³ or a flat three-torus with viscosity ν > 0, divergence-free data in H^s, s ≥ 3, and its maximal strong solution on [0, T_*). If the Euclidean velocity stays bounded by a constant C_E on the whole maximal interval, pairing the equation with −∆u and applying Young’s inequality give the derivative estimate (Theorem 3.1):

Y′ + νZ ≤ (C_E²/ν) Y, Y = ‖∇u‖₂², Z = ‖∆u‖₂²‖∇u(t)‖₂² ≤ ‖∇u₀‖₂² exp(C_E²t/ν) Theorem 3.1

The energy equality bounds ‖u‖₂, so the H¹ norm stays bounded up to any finite candidate endpoint, the local strong theory restarts the solution past it, and T_* = ∞. The solution is unique and smooth for positive times. This is the classical continuation criterion, proved in full: Leray’s lower bounds near a putative blowup, the endpoint ∫₀^T‖u‖_∞² dt ≤ C_E²T < ∞ of the Prodi–Serrin class, and Fefferman’s own record that the velocity becomes unbounded near any finite blowup time. The source’s solution has an unbounded velocity, so it lies outside the theorem’s hypothesis: the two theorems are consistent. A forcing in L²(0, T; L²) on every finite interval keeps the argument, with Y′ + νZ ≤ (2C_E²/ν)Y + (2/ν)‖f‖₂². Positive viscosity is essential: at ν = 0 the dissipative term the estimate spends is gone.

The bound the omnium supplies

At a resolved point the strands of the omnium’s closed patterns run at c: a closed pattern’s strand runs on a helix, translating at v and circulating at u with v² + u² = c², and keeps time by circulating (0.1d). Resolved transport is the normalised average of the strand velocities over their local directions τ̂, and the triangle inequality bounds it (Proposition 5.3); the spread of the directions about the average obeys an exact identity (Proposition 5.4):

u_loc = c ∫ τ̂ dμ, |u_loc| ≤ c∫ |cτ̂ − u_loc|² dμ = c² − |u_loc|²

Symmetric closed circulation gives zero net transport while every segment of the strand moves at c; as |u_loc| approaches c the spread vanishes and the directions align, the null limit, which timelike transport never reaches. Where the fluid’s Euclidean velocity is this resolved transport in one maintained flat orthonormal frame, C_E = c is admissible and the estimate holds with c²t/ν in the exponent: finite-time breakdown is excluded in that class (Corollary 3.2).

The helical average. Strand directions drawn from the centre of a sphere of radius c, every strand running at c; the teal arrow is their average, the resolved transport u_loc, which never leaves the sphere. On the right the transport and the square root of the spread are the two legs of a right triangle in a circle of radius c: |u_loc|² + spread = c², exactly. Ninety-six directions with equal weights; the closed-circulation set is antipodally symmetric, so its average is zero. The alignment carries the directions toward one axis by a conformal map of the sphere; it is a drawing parameter and carries no physics of its own. Drag to turn the sphere.

0.1d, I-⁠5

Finite energy bounds no velocity

The construction keeps its integrated kinetic energy finite while its velocity supremum diverges. The two are separate statements about a field. For a smooth, compactly supported, divergence-free profile w, reference scales V and L, and concentration λ ≥ 1, the family

u_λ(x) = Vλ^3/2 w(λx/L)‖u_λ‖₂² = V²L³‖w‖₂², ‖u_λ‖_∞ = Vλ^3/2‖w‖_∞

keeps its L² norm while its supremum grows without bound, independently of any evolution equation. The same split appears in a stationary matter field: the singular branch of an attractive inverse-square Gross–Pitaevskii potential has finite enclosed norm, 2π𝔘ε near the origin, while its local density diverges, and its gradient and quartic energies leave the finite-energy domain. A finite integrated norm bounds no local density.

Concentration at fixed energy. The speed of an axisymmetric swirl, w = (r/L) e^{−(r² + z²)/L²} in the azimuthal direction, against distance from its axis on its mid-plane, with V = 1 m s⁻¹ and L = 1 m; the dashed grey curve is λ = 1. Move the concentration: the profile narrows and rises at the same kinetic energy, 492 J in water, and its peak travels along the dotted line ‖u‖_∞ ∝ λ^3/2 until it crosses c, the bound of the helical average, at λ = 7.88 × 10⁵. The Gaussian tail stands in for compact support at the drawn precision.

0.1d, I-⁠7

Rulers and clocks in the physical continuation

In FDT clocks and rulers are closed omnium patterns whose adjustment is the geometry (0.1a), and a medium is not a fluid pushed by pressure forces across a fixed background (0.1c). With lapse 𝖭, spatial metric 𝗁_ij and shift 𝖻ⁱ, a fluid element’s velocity relative to the observers normal to the slices, 𝗏ⁱ = (dxⁱ/dt + 𝖻ⁱ)/𝖭, sets the clock it keeps:

dτ = 𝖭 dt √(1 − v_phys²/c²)|u_coord|_E ≤ 𝖭_maxc/√h_min + 𝖡_max ≡ C_E

Converting the local bound c to a Euclidean coordinate bound uses all three frame quantities, with 𝖭_max bounding the lapse, 𝖡_max the shift and h_min the lowest eigenvalue of the spatial metric; the calibration 𝗁 = δ, 𝖭 = 1, 𝖻 = 0 gives C_E = c. The fluid, the apparatus that drives it and the four-force density between them enter one conserved total source, every channel once (I-⁠7), and that source drives the metric:

∇_μT_fluid^μν = 𝒬_drv^ν, ∇_μT_drv^μν = −𝒬_drv^νG_μν[g] = κT_tot,μν I-⁠3

On an evolving spatial metric the gradient norm of a velocity field picks up terms the fixed-domain estimate never sees: the derivative of the metric in the norm, the derivative of the connection, and the derivative of the volume measure. The coupled continuation theorem, with a causal matter and apparatus model, its data and the estimates for the coupled metric and fluid evolution, is to be derived (D-⁠9).

The two phases and the seam

The density coordinate is the share of the three-sphere, |z₁|² = α_σ (0.3a). The Clifford torus α_σ = ½ is the seam; it splits the support into the matter phase, α_σ > ½, compressed omnium and closed braids, and the spacetime phase, α_σ < ½, decompressed substrate. Crossing the seam exchanges the regimes: the seam 𝔍 exchanges the two moduli at fixed phases and fixes the Clifford torus pointwise, as a jacket turned inside out, while the coordinate swap (z₁, z₂) ↦ (z₂, z₁) fixes there only one great circle. On the seam, and only there, the two readings of an interface agree; where the exchange is a symmetry of a declared interface model, its energy 𝒲 is stationary at the seam, 𝒲′(½) = 0, and the opposing tractions balance. A mode crosses an interface without reflection exactly when the impedances on its two sides match, r = (Z_out − Z_in)/(Z_out + Z_in). A change of chart, however smooth, preserves every invariant norm, an unbounded one included: no relabelling removes a divergence.

What FDT states

Commitments the record can end

The paper proves continuation on the fixed flat domain and states the physical continuation’s domain. Its commitments, each with the result that ends it, and its open items:

0.1c
The pressure force is the clock factor’s gradient. For a medium of perfect form at fixed entropy per count the pressure force is the bookkeeper’s reading of the gradient of ln 𝔥_N, and the Euler equation is its slow limit. A medium whose patterns fall otherwise than in 𝔪²g_μν, with every channel counted, ends this reading.
0.1d
No resolved transport above c. Every resolved transport of the omnium is an average of strand velocities of speed c, so in its local orthonormal frame its speed stays at or below c, strictly below for timelike transport. A resolved transport measured above c in its own frame ends this reading.
0.8
No physical singularity. No variable of the chain vanishes, force is capped by F_max, and the density coordinate lives on the open interval. A physical fluid whose velocity, vorticity or density diverges at a point in finite time ends the reading.
I-⁠3
The fixed Euclidean equation, recovered. At bulk speed small against c, with weak metric perturbations and a curvature length large against the sampled region, the incompressible equation and its strong solutions are recovered. A laboratory flow in that regime departing from the incompressible equation by more than its material and thermodynamic reduction allows, with every channel counted, ends the reading.
D-⁠9
The coupled continuation theorem. A causal matter and apparatus model with its data, and the estimates for the coupled metric and fluid evolution, are to be derived (D-⁠9).
D-⁠11
Persistence of the flat-frame bound. The persistence of the identification of the helical average with the Euclidean velocity, for arbitrary classical data, is to be derived (D-⁠11).
D-⁠5, D-⁠7
Couplings and rates. The density of the matter action, the seam transmission couplings and pairing strengths (D-⁠5), and the phase-exchange rates with the quantitative jet account (D-⁠7), are to be derived.

Readings declined

Use 6 Data yes, interpretation not necessarily

Standard reading

A fluid is a medium pushed by pressure forces across a fixed Euclidean space with Newtonian time.

FDT reading

The medium’s patterns free-fall on clocks and rulers rescaled together by its enthalpy per count, and the pressure force is the bookkeeper’s reading of that rescaling (0.1c, I-⁠7); rulers, clocks and geometry evolve with the fluid and its drive (0.1a).

c is the speed of light in vacuum, a constant the fluid never meets.

c is the strand rate of the omnium. Every strand of a closed pattern runs at c, so the resolved transport of a fluid is an average that stays below it (0.1d).

Convergence

Use 2 Same source, same data

For the same complete source, initial and boundary data and gauge, the coupled equations return the same curvature as general relativity (I-⁠3). The limiting map by which FDT recovers the measured record of laboratory flows is the medium window, completed by the first-order viscous stress Π^μν of a causal relativistic fluid and the drive’s exchange 𝒬^ν:

∇_𝖴𝖴^μ = −𝔓^μν∂_ν ln 𝔥_N + 𝔓^μ_ν(𝒬^ν − ∇_λΠ^νλ)/(n𝔥_N)→ ∂_tu + (u·∇)u = −∇π + ν∆u + f |u|²/c², π/c², α_ext → 0⁺

with Π^ij → −ρ₀ν(∂_iu_j + ∂_ju_i), 𝒬ⁱ → ρ₀fⁱ and n𝔥_N/c² → ρ₀ on the flat laboratory chart. The limits are those of the measured record: bulk speeds small against c, pressures small against ρ₀c², a gravitational window small for the laboratory pattern, small metric and frame changes over the measurement interval, and a controlled incompressible, constant-viscosity reduction of the material. With the same reduced equation, viscosity, forcing, data and pressure convention, the same strong solution is recovered on its existence interval. Every laboratory flow, turbulence measurement and vortex experiment stands as measured.

What FDT declines is the fixed background, not the fluid record. FDT is a true extension of general relativity and quantum mechanics: where the three converge, FDT reproduces their results; where it declines their assumptions, here a Euclidean space and a Newtonian time that do not respond to the fluid, it is under no obligation to reproduce them.