コンテンツにスキップ

forest-cube-canopy

英語 | 日本語

« Home · « 外部流れ・風工学

Resolved cube-array urban canopy on the D3Q27-cumulant forest AMR-LBM -- Stage 4 of docs/design/forest-d3q27-design.md (the CityLBM-reproduction target).

概要: 整列立方体群(λ_p=0.25)による都市キャノピー。Coceal らの DNS と比較する。個々の建物ではなく群としての抵抗を見る。

Geometry

forest-cube-canopy geometry

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。

forest-cube-canopy p_pa

forest-cube-canopy umag_ms

Build & run

cmake -S native -B native/build && cmake --build native/build -j
./native/build/orheo-forest-cube-canopy

このページの図を作った計算のコスト(Miyabi-G Grace CPU (72 cores, -stdpar=multicore)): 89.8 s wall。

Validation

Tier T2 — 単点比較
QoI キャノピー内速度プロファイル
Reference Coceal et al. (2006) DNS; Cheng & Castro (2002)
Gate — ゲートなし

Source & run details

View source

Item Value
Target orheo-forest-cube-canopy
Source native/benchmarks/forest_cube_canopy.cpp
Lattice —
GPU twin yes (orheo-forest-cube-canopy-gpu)
Modules cube_canopy_geom.hpp, forest_geometry.hpp, forest_inflow.hpp, forest_step_lat.hpp
Gallery run 2026-07-27 @ 7628c0cf
Args 8 12000 0.05 5000 0.10 .
Wall time 89.79 s
CPU time 3545.74 s
Device Miyabi-G Grace CPU (72 cores, -stdpar=multicore)

Computation log

この計算の実際の出力(先頭と末尾)。

=== forest cube-array urban canopy (D3Q27 cumulant + Smagorinsky, Stage 4) ===
aligned array lambda_p=0.25 (Coceal et al. 2006 / Cheng & Castro 2002): 12x2 cubes, h=8 fine cells (4 coarse)
domain 128x16x16 level-0 cells (16x2x2 blocks, kNb=8), top=4h; fetch 2+2 tiles; Re_h=5e+03 ulat=0.050 tau0=0.50012 cs=0.10 turb=1
forest: 260 blocks, max_level=1; interior cells L0=18432 L1=114688
solid raster: 24000 cells (face-snapped halfway bounce-back)
inflow law: u*=0.0050 d=2.40 z0=0.240 (lattice; U(4h)=ulat)
inflow ON (digital-filter turbulence), 16x16 level-0 inlet cells, filter scales L=h
=== time-stepping (12000 coarse steps; averaging from 6000) ===
step      0/12000  t_conv=0.000  umax=0.0496  Fx_p=12.72  u*=0.0455 | 0.0s  0.0 MLUPS
step    500/12000  t_conv=0.195  umax=0.0556  Fx_p=0.4153  u*=0.0082 | 5.7s  21.8 MLUPS
step   1000/12000  t_conv=0.391  umax=0.0618  Fx_p=-0.2342  u*=0.0000 | 10.7s  23.1 MLUPS
step   1500/12000  t_conv=0.586  umax=0.0648  Fx_p=0.1075  u*=0.0042 | 15.2s  24.5 MLUPS
step   2000/12000  t_conv=0.781  umax=0.0674  Fx_p=0.4198  u*=0.0083 | 20.4s  24.3 MLUPS
step   2500/12000  t_conv=0.977  umax=0.0707  Fx_p=-0.09712  u*=0.0000 | 25.9s  24.0 MLUPS
...

=== QoI summary ===
u*: drag=0.0019 stress-peak=0.0027 | log-law fit: u*=0.0060 z0/h=0.067 d/h=1.00 (short-fetch fit sees the inflow outer layer; trust drag/stress)
canopy deficit U(0.5h)/U(3.5h)=0.169  inflection at z/h=1.12  Qout/Qin=0.9906
MEM(once-per-cycle)/pressure force ratio = 4.378  (published drag = pressure integral; see header)
wrote ./slice_ycube.csv
wrote ./slice_ygap.csv
wrote ./slice_ycube_mean.csv
wrote ./slice_ygap_mean.csv
=== done (33.2 MLUPS; outputs in .) ===