コンテンツにスキップ

kbc-tgv3d

英語 | 日本語

« Home · « 乱流(DNS/LES 照合)

D3Q27 Taylor-Green vortex at Re=1600 -- the canonical 3D turbulence-decay benchmark, used here as the referee for the KBC-vs-Smagorinsky LES question (docs/design/kbc-entropic-design.md §6/§7-②). Fully periodic box [0,2pi]^3, and -- crucially -- NO explicit Smagorinsky (cs_smag = 0) for ANY operator: this isolates each operator's intrinsic dissipation. Four D3Q27 collision operators are compared on the volume-mean dissipation rate eps(t) = -dE_k/dt:

概要: 同じ TGV を 4 種の衝突演算子で解き、各演算子が持つ「本質的な散逸」を比較する。SGS モデルを入れずに演算子だけの性質を見るのが狙い。

Geometry

kbc-tgv3d 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。

kbc-tgv3d umag

kbc-tgv3d ux

Build & run

cmake -S native -B native/build && cmake --build native/build -j
./native/build/orheo-kbc-tgv3d

kbc_tgv3d N Re U0 t_end_star [csv_prefix]

このページの図を作った計算のコスト(Miyabi-G Hopper GPU (GH200, -stdpar=gpu)): 29.1 s wall。

Validation

Tier T2 — 単点比較
QoI 散逸率 eps(t)
Reference Brachet (1983) / van Rees (2011) DNS
Gate — ゲートなし

Source & run details

View source

Item Value
Target orheo-kbc-tgv3d
Source native/benchmarks/kbc_tgv3d.cpp
Lattice —
GPU twin yes (orheo-kbc-tgv3d-gpu)
Modules lbm_d3q27.hpp
Gallery run 2026-08-30 @ 38ad1eaa
Args ``
Wall time 29.14 s
CPU time 28.81 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)。

kbc_tgv3d N=128 Re=1600 U0=0.050 L=20.372 nu0=6.366e-04 tau0=0.50191  8148 steps (t*<=20)  cs_smag=0 (pure operators)
  [BGK] peak eps=0.01355 at t*=8.84
  [MRT] peak eps=0.01237 at t*=8.34
  [CUM] peak eps=0.01204 at t*=8.84
  [KBC] peak eps=0.01272 at t*=8.84

=== D3Q27 TGV3D Re=1600  peak dissipation (ref ~0.0125 at t*~9) ===
  BGK   eps_max = 0.01355  at t* = 8.84
  MRT   eps_max = 0.01237  at t* = 8.34
  CUM   eps_max = 0.01204  at t* = 8.84
  KBC   eps_max = 0.01272  at t* = 8.84