forest-channel-d3q27¶
英語 | 日本語
Wall-modeled turbulent half-channel / flat-plate BL on the D3Q27 cumulant path -- Stage 1 (dynamic) of docs/design/forest-d3q27-wall-model-design.md: does the ported near-wall tau-bump recover the LOG LAW u+ = (1/kappa) ln(y+) + B (kappa=0.41, B=5.2) in a real turbulent flow, with the first cell in the log layer (y+ ~ 30- 100, near-wall UNRESOLVED)? No new stream/forcing: it reuses the committed domain-BC step (no-slip WALL-MODELED floor, free-slip top + spanwise, uniform-or- digital-filter turbulent inflow, anti-bounce-back pressure outlet) exactly as forest_cube_canopy but with a FLAT floor (no cubes). The spanwise+streamwise+time mean U(z) is fit to the log law and compared to a wmodel=0 negative control.
概要: 壁モデル付きの半チャネル。第1セルの y+ を 30–100 に置き、壁関数が対数則を再現するかを確認する。
Geometry¶

Animation¶
Fields¶
この計算が最後の出力ステップで書き出した場をそのまま描いたもの。3D ケースは中央断面。


Build & run¶
cmake -S native -B native/build && cmake --build native/build -j
./native/build/orheo-forest-channel-d3q27
このページの図を作った計算のコスト(Miyabi-G Hopper GPU (GH200, -stdpar=gpu)): 45.4 s wall。
Validation¶
| Tier | T2 — 単点比較 |
| QoI | u+ 対数則 |
| Reference | Kim-Moin-Moser (1987) / Moser (1999) |
| Gate | — ゲートなし |
Source & run details¶
| Item | Value |
|---|---|
| Target | orheo-forest-channel-d3q27 |
| Source | native/benchmarks/forest_channel_d3q27.cpp |
| Lattice | D3Q27 |
| GPU twin | yes (orheo-forest-channel-d3q27-gpu) |
| Modules | forest_field.hpp, forest_geometry.hpp, forest_inflow.hpp, forest_step_lat.hpp |
| Gallery run | 2026-08-30 @ 38ad1eaa |
| Args | 24 4000 |
| Wall time | 45.39 s |
| CPU time | 45.26 s |
| Device | Miyabi-G Hopper GPU (GH200, -stdpar=gpu) |
Computation log¶
この計算の実際の出力(先頭と末尾)。
=== wall-modeled channel D3Q27 (wmodel=2, digital-filter turbulence) ===
H=24 nu=6.000e-05 tau=0.50018 Re_H=2e+04 u_tau(law)=0.0023 y+(z=1)=38.2 domain 240x48x32 (blocks 30x6x4)
step 0/4000 umax=0.0515 0s
step 1000/4000 umax=0.0625 9s
step 2000/4000 umax=0.0642 18s
step 3000/4000 umax=0.0688 31s
step 4000/4000 umax=0.0652 45s
=== done: u_tau=0.0010 (y+ first=17.4); log-law |u+ - law| mean over 6 log-window pts = 17.779 (gate < 0.5) [off] ===