Hunting finite-time blowup at 64^3 and 96^3: enstrophy grows to 145.79x and decays. No blowup observed. Nothing solved.
The Navier-Stokes equations describe how a viscous fluid moves: velocity, pressure, and viscosity, with the incompressibility constraint that what flows in must flow out. The Clay problem asks whether smooth initial data always produce smooth solutions for all time, or whether a solution can blow up in finite time, with velocity or vorticity running to infinity. Nobody knows. This lab hunts for blowup numerically. The signal it watches is enstrophy, the mean square vorticity: if a solution were heading for blowup, enstrophy would run away. If enstrophy grows to a peak and then decays, the run stayed smooth. Every run here stayed smooth. That is evidence about the search, not a proof about the equations.
Solver: pseudo-spectral with real FFT, 2/3 dealiasing, RK4 time stepping, periodic cube [0, 2pi]3, dt 0.01, 1000 steps to T = 10 unless stated. The round 4 real-FFT solver was verified against the reference solver to near machine precision. Initial fields: Kida-Pelz high-symmetry flow, Taylor-Green vortex, perturbed Taylor-Green, and anti-parallel vortex tubes.
| Round | Kida-Pelz growth | Taylor-Green growth | Notes |
|---|---|---|---|
| 1 (Nov 20) | 18.29 at 483 (14.23 at 323) | perturbed TG 12.00 at 323 | First hunt, nu 0.001, tubes 1.00 |
| 2 (Nov 20) | 18.86, peak 21.22, final 13.97 | no TG run | 643, nu 0.001 |
| 3 (Nov 20) | 18.86, peak 21.22, final 1.65 at T = 20 | no TG run | 643 to T = 20, nu 0.001; 963 rejected on cost |
| 4 (Nov 20) | 40.33, peak 45.37, final 32.68 | 36.42 | 643, nu 0.0005 |
| 5 (Nov 20) | 64.46, peak 72.52 at t = 8.9, final 62.75 | 62.69, peak 23.51 at t = 9.9 | 643, nu 0.00025 |
| 6 (Nov 21) | 88.24, peak 99.27 at t = 9.4, final 97.28 | 87.56, peak 32.84 at t = 10.0 | 643, nu 0.000125; perturbed TG 78.67 vs 58.29; tubes 1.02 and 1.24 |
| 7 (Nov 21) | 145.79 at 963, peak 164.02 at t = 8.2, final 132.99 | 138.27 at 963, peak 51.85 at t = 9.8, final 51.48 | 963, nu 0.000125; TG 643 to T = 20: 88.56 |
navier_stokes_hunt_r2.json,
navier_stokes_hunt_r4.json, navier_stokes_hunt_r5.json,
navier_stokes_hunt_r6.json (viscosity_ladder_kida_pelz_64).Growth rises with resolution at fixed viscosity and does not saturate. Kida-Pelz goes from 88.24 at 643 to 145.79 at 963, ratio 1.65. Taylor-Green goes from 87.56 to 138.27, ratio 1.58. Perturbed Taylor-Green shows the same pattern at lower resolution: 58.29 at 483 to 78.67 at 643, ratio 1.35. The vortex tubes are dead as an attack: anti-parallel tubes grow 1.02, tight anti-parallel tubes 1.24, at nu 0.000125. They do not stretch at any tested viscosity.
navier_stokes_hunt_r6.json and
navier_stokes_hunt_r7.json.The round 3 cost wall is gone. Round 3 rejected a 963 run on a 10-step probe of the old reference solver: 17.323 s/step, peak RSS 660.7 MB, extrapolating to 4.81 h for 1000 steps, and about 8.8 h once the measured 1.83x probe underestimate is applied (receipt rcpt_0045). Round 6 re-measured with the real-FFT solver: 1.716 s/step, 10 steps in 17.16 s, peak RSS 379.4 MB, extrapolated 1715.5 s, or 0.48 h, for 1000 steps. Round 7 ran the full 963: Kida-Pelz in 1357.6 s, Taylor-Green in 1351.5 s, and a dual run of Kida-Pelz 963 produced an identical field hash. The wall was the old solver, not the physics.
At T = 10 the growth looks like it might keep climbing. At T = 20 it does not. Kida-Pelz at 643, nu 0.000125: peak enstrophy 99.27 (growth 88.24x) at t = 9.4, final enstrophy 42.43 (37.7x) at T = 20, decaying. Taylor-Green at the same viscosity and resolution: peak enstrophy 33.21 (growth 88.56x) at t = 11.0, final 14.94 at T = 20, decaying. Taylor-Green peaks later than Kida-Pelz, which is why the T = 10 snapshot understates it: at T = 10 Taylor-Green had reached only 32.84. Both runs were repeated with identical field hashes: the Kida-Pelz T = 20 dual run matches hash 35e01ce6..., and the Kida-Pelz 963 dual run matches hash 6445c2c1....
navier_stokes_hunt_r6.json (Kida-Pelz T = 20) and
navier_stokes_hunt_r7.json (Taylor-Green T = 20).Receipt chain: 66 receipts, chain valid, head
bdebd1e7d93dd108bffe0ba01d499e603db81a8c17027bb0aad36a70f73dacdd
(verification receipt rcpt_0066). Navier-Stokes receipts:
| Receipt | What it covers |
|---|---|
| rcpt_0045 | Round 3: 963 rejected on measured timing (17.323 s/step, peak RSS 660.7 MB, extrapolated 4.81 h, about 8.8 h with the 1.83x underestimate); fallback Kida-Pelz 643 to T = 20 |
| rcpt_0060 | Round 6: nu 0.000125 deep dive, 643, T = 10 and T = 20, viscosity ladder, tubes, 963 cost probe |
| rcpt_0062 | Round 6 chain verification |
| rcpt_0063 | Round 7: full 963 at nu 0.000125, Taylor-Green 643 to T = 20 |
| rcpt_0066 | Round 7 chain verification; chain head bdebd1e7 |
Data files:
data/navier_stokes_hunt_r6.json (SHA-256
b6a88cb2dbcd02e89300814dbf3f7d8f1d3b83d8b05faaabf00e7adb0e420254),
navier_stokes_hunt_r7.json (SHA-256
5ac93078cac19b6f0986b7148bb845fee3d83ef3bc174b40211a15ec984e6490).
Earlier rounds: navier_stokes_hunt.json,
navier_stokes_hunt_r2.json,
navier_stokes_hunt_r3.json,
navier_stokes_hunt_r4.json,
navier_stokes_hunt_r5.json. Dual-run field hashes:
Kida-Pelz T = 20, 35e01ce6...; Kida-Pelz 963,
6445c2c1....
Taylor-Green at 963 to T = 20, to confirm transience at the higher resolution, or lower viscosity (nu 0.0000625) at 643, to see whether the ladder keeps climbing. A third move is a new initial field, since the tubes are dead and Kida-Pelz and Taylor-Green both decay.
We simulated a fluid in a box, made it less and less viscous, and watched how violently it churns. Each time the churning grew stronger, but every time it peaked and calmed down again, even in the biggest simulation we could run. Nothing blew up. That does not prove fluids can never blow up. It proves this way of hunting for a blowup has not found one yet.
No blowup, no singularity, no proof that smooth solutions exist for all time, and no disproof either. We did not test any viscosity below 0.000125, any resolution above 963, or any initial field other than the four tried.