code/DATASET.md

# Parametric rotating-cube CFD dataset

Generated with OpenFOAM ESI v2512 (`pimpleFoam`, AMI dynamic mesh, laminar) per the
LOCKED design in `CLAUDE.md`. Reproduces/extends the phi-GNN 3D rotating-cube case
(Gao, Cheng, Jaiman) as a 2D parameter sweep for a downstream neural operator.

## Grid (25 cases, complete)
- **Re** ∈ {50, 100, 150, 200, 300} — varied by viscosity only (`nu = 1/Re`).
- **psi** ∈ {0, 15, 30, 45, 60}° — inflow elevation out of x-y plane,
  `U_inf = (cos psi, 0, sin psi)`; azimuth `beta = 0` fixed.
- Constant across all cases: `U_inf|=1`, `c=1`, physical `omega=0.25` rad/s
  (`omega*=0.25`), `dt*=5e-3`, sampling every 8 steps (`dt*_sample=0.04`), endTime `t*=200`.

## Storage layout (per case `runs/Re{Re}_psi{psi}/`)
- Decomposed across 12 processors (scotch). **U and p only** are written per snapshot
  (`writeObjects` FO); the rigid-rotation mesh is stored **once** in
  `processor*/constant/polyMesh` (+ `rotating` cellZone). Reconstruct node positions
  downstream from `theta = omega * t` on the `rotating` zone; **U is absolute-frame**.
- **4991 snapshots/case** over `t* = 0..200` (binary). ~58 GB/case, **~1.4 TB total**.

## Force coefficients — stationary average over t* ∈ [150, 200] (~1250 samples)
Aref = c² = 1, magUInf = 1, dragDir = (1,0,0), liftDir = (0,1,0); Cs = axial (z) force.
`*_std` = RMS fluctuation about the mean (vortex-shedding amplitude).

| Re \ psi |   0°  |  15°  |  30°  |  45°  |  60°  |
|---------:|------:|------:|------:|------:|------:|
| **Cd** | | | | | |
| 50  | 2.334 | 2.270 | 2.094 | 1.790 | 1.321 |
| 100 | 1.718 | 1.690 | 1.620 | 1.428 | 1.038 |
| 150 | 1.513 | 1.497 | 1.407 | 1.289 | 0.916 |
| 200 | 1.423 | 1.413 | 1.307 | 1.075 | 0.843 |
| 300 | 1.358 | 1.354 | 1.297 | 1.096 | 0.641 |
| **Cl (Magnus, from +z spin)** | | | | | |
| 50  | -0.287 | -0.288 | -0.282 | -0.240 | -0.160 |
| 100 | -0.337 | -0.345 | -0.358 | -0.292 | -0.190 |
| 150 | -0.381 | -0.369 | -0.392 | -0.328 | -0.198 |
| 200 | -0.404 | -0.350 | -0.345 | -0.315 | -0.196 |
| 300 | -0.407 | -0.323 | -0.302 | -0.272 | -0.242 |
| **Cs (axial force)** | | | | | |
| 50  | 0.000 | -0.653 | -1.163 | -1.535 | -1.811 |
| 100 | 0.000 | -0.503 | -0.844 | -1.109 | -1.308 |
| 150 | 0.000 | -0.463 | -0.824 | -0.991 | -1.155 |
| 200 | 0.000 | -0.450 | -0.827 | -1.087 | -1.095 |
| 300 | 0.000 | -0.430 | -0.816 | -1.032 | -1.149 |
| **Cd_std (shedding amplitude)** | | | | | |
| 50  | 0.129 | 0.115 | 0.085 | 0.049 | 0.020 |
| 100 | 0.153 | 0.139 | 0.116 | 0.073 | 0.032 |
| 150 | 0.163 | 0.150 | 0.133 | 0.086 | 0.040 |
| 200 | 0.168 | 0.157 | 0.136 | 0.089 | 0.043 |
| 300 | 0.165 | 0.185 | 0.136 | 0.092 | 0.037 |

### Physics check (both axes carry distinct physics — as designed)
- **Cd ↓ with Re** (drag crisis trend) and **Cd ↓ with psi** (cross-flow component
  `cos psi` shrinks as inflow tilts toward the spin axis).
- **Cs grows with psi** from 0 → ~-1.8 — the genuine axial through-flow force that has
  no in-plane analogue; this is the new physics psi was chosen to span.
- **Shedding amplitude (Cd_std) shrinks with psi** — oblique/axial inflow suppresses the
  2D vortex street — and grows with Re. psi=0 columns are the classic 2D shedding limit.
- **Cl < 0** everywhere: Magnus-type lift from the fixed +z rotation sense.

## Laminar-validity note (for dataset docs)
The no-turbulence-model (laminar) assumption is most strained at the **high-Re / low-psi
corner** (Re=300, psi=0–15), where shedding fluctuation peaks (Cd_std ≈ 0.16–0.19) and
oblique rotation is least stabilising. These cases ran stably to t*=200 with bounded
Courant, but treat Re=300 (esp. psi ≤ 15) as the resolution/validity ceiling of the
current mesh; increasing psi monotonically stabilises the wake (Cd_std → 0.04 at psi=60).

## Validation provenance
- Phase 1 (static cube, low-blockage): Cd = 0.978 vs paper 0.935, bracket [0.854, 1.122] — PASS.
- Phase 2 (rotating AMI mechanics): AMI weights 1.0→1.0002, continuity ~1e-13, ~0.6 s/step/12-core.