expansion-loss2d¶
English | Japanese
« Home · « Multi-fidelity (1D ⇄ FVM ⇄ LBM)
Backward-facing step / sudden expansion: reattachment length and the measured loss coefficient of the expansion itself.
Build & run¶
cmake -S native -B native/build && cmake --build native/build -j
./native/build/orheo-expansion-loss2d
expansion_loss2d [CONFIG] (KEY = VALUE lines; see named_args.hpp)
Cost of the computation that made the figures on this page (Miyabi-G Grace CPU (72 cores, -stdpar=multicore)): 26.0 s wall.
Validation¶
| Tier | T1 — automatic gate |
| QoI | reattachment_length |
| Reference | Armaly et al. (1983) J.Fluid Mech 127:473 -- reattachment x_r/S (laminar) |
| Gate | ✅ direct gate |
Source & run details¶
| Item | Value |
|---|---|
| Target | orheo-expansion-loss2d |
| Source | native/benchmarks/expansion_loss2d.cpp |
| Lattice | D2Q9 |
| GPU twin | no |
| Modules | channel2d.hpp, stations.hpp |
| Gallery run | 2026-07-28 @ c9b0572c |
| Args | `` |
| Wall time | 26.02 s |
| CPU time | 1867.19 s |
| Device | Miyabi-G Grace CPU (72 cores, -stdpar=multicore) |
Computation log¶
The actual output of this computation (head and tail).
=== backward-facing step: reattachment + measured K (D2Q9) ===
h1=40 h2=78 S=38 (ER=1.940) Nx=1840 Ny=78
Re=200 (on h1: V1 h1/nu) Re_Dh=400 (on Dh = 2 h1, the
convention Armaly et al. use -- twice this one)
ulat=0.0400 nu=0.008000 tau=0.5240
Borda-Carnot K=(1-A1/A2)^2=0.2348 -- the INVISCID idealisation,
printed for the signed gap, not used as a gate.
step 20000 x_r/S= 8.49 V1=0.04030 V2=0.02127 (V1/V2=1.895, geom 1.940) K_raw=0.5060 K=-0.2675 mass 2.814% dn grad/exact=2.014
step 40000 x_r/S= 8.45 V1=0.03999 V2=0.01990 (V1/V2=2.010, geom 1.940) K_raw=0.4769 K=-0.1161 mass 2.979% dn grad/exact=1.244
step 60000 x_r/S= 8.10 V1=0.04010 V2=0.02080 (V1/V2=1.928, geom 1.940) K_raw=0.4856 K=0.0374 mass 1.162% dn grad/exact=0.705
step 80000 x_r/S= 8.11 V1=0.04003 V2=0.02051 (V1/V2=1.952, geom 1.940) K_raw=0.5976 K=-0.0234 mass 0.143% dn grad/exact=1.347
step 100000 x_r/S= 8.18 V1=0.04002 V2=0.02043 (V1/V2=1.959, geom 1.940) K_raw=0.5096 K=0.0775 mass 0.429% dn grad/exact=0.679
step 120000 x_r/S= 8.08 V1=0.04006 V2=0.02065 (V1/V2=1.940, geom 1.940) K_raw=0.5424 K=0.0384 mass 0.522% dn grad/exact=0.918
step 140000 x_r/S= 8.14 V1=0.04003 V2=0.02047 (V1/V2=1.956, geom 1.940) K_raw=0.5412 K=0.0333 mass 0.301% dn grad/exact=0.929
...
-> K is NOT trustworthy: a fit window is not in developed
flow, so extrapolating it onto the step plane is
extrapolating the wrong line. Lengthen Ldown (the
redevelopment length grows with Re) and re-run.
RESULT ER=1.940 Re=200 h1=40 S=38 xr_S=8.1203 K_measured=0.048961 K_energy=0.384539 K_raw=0.535958 K_borda=0.234775 rel_borda=-0.791458 alpha_up=1.5347 alpha_dn=1.7028 mass_imbal=0.000885
-> couple it in: orheo-multifidelity-duct with k_lbm=0.0490
(or build a sweep into ktable=Q1:K1,...)