# Navier-Stokes blowup hunt, round 2

Date: 2026-11-20. Code: `compute/navier_stokes_hunt_r2.py` (reuses the
solver in `compute/navier_stokes.py` and the field builders in
`compute/navier_stokes_hunt.py`). Results:
`data/navier_stokes_hunt_r2.json`. Full run log:
`data/navier_stokes_hunt_r2.log`. Protocol for both cases: nu = 0.001,
dt = 0.01, final time 10.0, 1000 steps, pseudo-spectral with 2/3
dealiasing and RK4 on a periodic cube [0, 2*pi]^3.

## What ran

Case 1, Kida-Pelz at 64^3. The round 1 strongest case rerun at higher
resolution, same field: u = sin x cos y cos z, v = cos x sin y cos z,
w = -2 cos x cos y sin z. A timing probe before the full run measured
0.83 s per step at 64^3, projecting about 831 s for 1000 steps.

Case 2, high-amplitude anti-parallel vortex tubes at 48^3. The round 1
tube case (gamma = 2.0, core radius 0.5, displacement 0.1) decayed
monotonically with growth factor 1.0, but vortex stretching is quadratic
in vorticity while viscous diffusion is linear, so that null may have
been an amplitude threshold rather than intrinsic tameness. Here the
circulation is tripled (gamma = 6.0), the core narrowed to 0.4, and the
Crow-instability seed displacement doubled to 0.2 sin(y) in opposite
directions, which raises the nonlinear stretching roughly ninefold
relative to diffusion and sharpens the collision gradients. Velocity is
recovered from the vorticity by the same Fourier Biot-Savart law as
round 1. This stresses enstrophy growth through a localised tube
collision and reconnection, a different axis from the global single-mode
coupling of Kida-Pelz.

## Results

| Case | Enstrophy init / final / peak | Enstrophy growth | Max vorticity init / final / peak | Vorticity growth | Energy init / final | Smooth | Runtime |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Kida-Pelz 64^3 | 1.125 / 13.968 / 21.216 | 18.86 | 3.000 / 19.900 / 34.519 | 11.51 | 0.375 / 0.149 | yes | 1516.71 s |
| Tubes, gamma 6.0, 48^3 | 0.227 / 0.221 / 0.232 | 1.02 | 5.987 / 5.728 / 6.251 | 1.04 | 0.0437 / 0.0392 | yes | 310.66 s |

## What the numbers say

Kida-Pelz enstrophy growth across resolutions: 14.23 at 32^3, 18.29 at
48^3, 18.86 at 64^3. The step from 48^3 to 64^3 adds only about 3
percent, so the enstrophy growth is saturating with resolution rather
than running away. Peak vorticity growth is more resolution sensitive:
8.06 at 32^3, 7.97 at 48^3, 11.51 at 64^3, with the peak reaching 34.52
against an initial 3.0. Final enstrophy at 64^3 (13.97) is slightly
below the 48^3 final value (14.35) while the peak is slightly above
(21.22 vs 20.57). Both cases stayed smooth to t = 10 with all
diagnostics finite. No blowup was observed in either case.

The high-amplitude tubes stayed essentially flat: enstrophy growth
1.02 and vorticity growth 1.04, with enstrophy ending slightly below
its initial value. Tripling the circulation, narrowing the core, and
doubling the Crow seed did not activate the collision. The round 1
tube null was therefore not an amplitude threshold in this range, at
least at nu = 0.001 and t = 10; the configuration decays slowly rather
than stretching.

Runtime note: the 64^3 wall time of 1516.71 s is well above the 831 s
probe projection because the Yang-Mills round 2 run and a Riemann PARI
scan were running concurrently on the same 1.5 shared CPUs during most
of case 1 (load average near 3.0, this process near 30 percent CPU).
Case 2 ran after those finished and took 310.66 s, close to the round 1
48^3 time of 369.92 s.

## What the numbers do not say

These are small-grid, unforced, fixed-time runs. They can never prove
or disprove a continuum statement, and they say nothing about Clay
statements (A) or (B). A finite enstrophy growth of about 19x that
saturates with resolution is a hint that this initial condition at this
viscosity is not blowing up by t = 10, not a proof of global regularity,
and the vorticity peak rising at 64^3 shows the small scales are not
fully resolved. Longer times, lower viscosity, and higher resolution
would be needed before any stronger statement could even be attempted,
and a genuine blowup claim would need a continuum-limit argument, not a
single-grid diagnostic.
