kbc-cube-d3q27¶
English | Japanese
« Home · « Turbulence (checked against DNS/LES)
Wall-mounted cube in an open channel on the D3Q27 forest AMR-LBM -- the bluff-body / urban test of the KBC entropic operator (docs/kbc-entropic- design.md §11: "the natural next step is a D3Q27 Ahmed/urban run comparing pure KBC vs MRT+Smagorinsky at resolution, where §S2 predicts KBC earns its keep"). S2 showed pure KBC is implicit-LES adequate on decaying Taylor-Green; this asks the same question on a real, under-resolved SEPARATED wake -- does PURE KBC (parameter-free) match the production LES baseline (cumulant + Smagorinsky) without a tuned Smagorinsky constant, at the SAME resolution?
Geometry¶

Animation¶
Fields¶
The fields this computation wrote at its last output step, plotted as they are. 3D cases show the mid-plane.


Build & run¶
cmake -S native -B native/build && cmake --build native/build -j
./native/build/orheo-kbc-cube-d3q27
Cost of the computation that made the figures on this page (Miyabi-G Hopper GPU (GH200, -stdpar=gpu)): 67.4 s wall.
Validation¶
| Tier | T2 — single-point comparison |
| QoI | Cd, recirculation length L_r/h |
| Reference | Wall-mounted cube (qualitative comparison with literature values) |
| Gate | — no gate |
Source & run details¶
| Item | Value |
|---|---|
| Target | orheo-kbc-cube-d3q27 |
| Source | native/benchmarks/kbc_cube_d3q27.cpp |
| Lattice | D3Q27 |
| GPU twin | yes (orheo-kbc-cube-d3q27-gpu) |
| Modules | forest_field.hpp, forest_geometry.hpp, forest_step_lat.hpp |
| Gallery run | 2026-08-30 @ 38ad1eaa |
| Args | `` |
| Wall time | 67.38 s |
| CPU time | 67.25 s |
| Device | Miyabi-G Hopper GPU (GH200, -stdpar=gpu) |
Computation log¶
The actual output of this computation (head and tail).
=== wall-mounted cube, D3Q27 forest (cumulant + Smagorinsky) ===
h=16 fine cells (8 coarse), Re_h=4000 ulat=0.060 tau0=0.50036 cs=0.100 collide_sel=1 max_level=1
domain 144x56x40 level-0 cells (18x7x5 blocks, kNb=8), open top; cube x[32,40] y[24,32] z[0,8]
forest: 1953 blocks, max_level=1; interior L0=225792 L1=774144
solid raster: 7600 cube cells (halfway bounce-back)
=== time-stepping (16000 coarse steps; averaging from 8000) ===
step 0 t*=0.00 umax=0.0725 fx=14.676 cd=31.8498 0.0 MLUPS
step 200 t*=1.50 umax=0.0881 fx=0.43762 cd=0.9497 375.5 MLUPS
step 400 t*=3.00 umax=0.0864 fx=0.51514 cd=1.1179 403.7 MLUPS
step 600 t*=4.50 umax=0.0880 fx=0.44292 cd=0.9612 414.0 MLUPS
step 800 t*=6.00 umax=0.0854 fx=0.56608 cd=1.2285 417.9 MLUPS
step 1000 t*=7.50 umax=0.0838 fx=0.50143 cd=1.0882 421.4 MLUPS
step 1200 t*=9.00 umax=0.0824 fx=0.52874 cd=1.1474 423.8 MLUPS
step 1400 t*=10.50 umax=0.0847 fx=0.50107 cd=1.0874 425.5 MLUPS
...
step 14800 t*=111.00 umax=0.0863 fx=0.48377 cd=1.0498 425.7 MLUPS
step 15000 t*=112.50 umax=0.0837 fx=0.48629 cd=1.0553 425.6 MLUPS
step 15200 t*=114.00 umax=0.0848 fx=0.4934 cd=1.0707 425.4 MLUPS
step 15400 t*=115.50 umax=0.0845 fx=0.49586 cd=1.0761 425.3 MLUPS
step 15600 t*=117.00 umax=0.0841 fx=0.49501 cd=1.0742 425.2 MLUPS
step 15800 t*=118.50 umax=0.0832 fx=0.49625 cd=1.0769 425.1 MLUPS
step 16000 t*=120.00 umax=0.0919 fx=0.49604 cd=1.0765 425.0 MLUPS
=== done: operator=cumulant cs=0.100 | Cd_mean=1.0541 Lr/h=3.233 umax_final=0.0919 [stable] ===
wrote kbc_cube_out/{kbc_cube.csv,slice_ymid.csv,wake_line.csv,summary.csv}