bench_velocity_mg_256 Namespace Reference

Functions
	cavity (N, nz, Re, dt, mode, U=1.0, nsteps=200, vcyc=4, vel_iter=60, outer=2, pit=40)
	main ()

Detailed Description

Velocity-MG benchmark on the DOMAIN-BC paths (lid cavity 256^3, backward-facing step high-res 2D).

Unlike the IBM packed case (which has a dt ceiling from the row-scaled cut/solid coarse-coupling), the
cavity/BFS have NO immersed solid, so the velocity-MG coarse operator is geometry-exact -> it should be
both (a) FASTER than RB-GS at high resolution (MG kills the stiff-diffusion error in O(1) V-cycles vs RB-GS
O(N) sweeps), and (b) free of any dt restriction (a standard Helmholtz MG is unconditionally stable).

Reports, per case:
  - SPEEDUP: wall-clock + step count to reach steady, RB-GS vs velocity-MG (diffusion-only) at the same dt,
    and the velocity-MG run with a LARGE dt (pseudo-transient acceleration -> far fewer steps).
  - NO-DT-RESTRICTION: implicit-FOU + upwind velocity-MG at a sweep of growing dt; must stay finite and keep
    converging (CFL = U*dt up to O(1000)), where explicit advection and the IBM packed case would diverge.

Function Documentation

cavity()

bench_velocity_mg_256.cavity	(		N,
			nz,
			Re,
			dt,
			mode,
			U = 1.0,
			nsteps = 200,
			vcyc = 4,
			vel_iter = 60,
			outer = 2,
			pit = 40
	)

Lid-driven cavity (N x N x nz), run a FIXED nsteps; return wall-clock + the converged-so-far centreline
umin. mode: 'rbgs' | 'vmg' (diffusion-only) | 'vmg_fou' (implicit-FOU + upwind).

Definition at line 26 of file bench_velocity_mg_256.py.

Referenced by main().

main()

bench_velocity_mg_256.main

(

)

Definition at line 55 of file bench_velocity_mg_256.py.

References cavity(), and main().

Referenced by main().