Megathread: Ghost cell free wall boundary issues
Opened this issue · 2 comments
Import of issue #9 from the old repository, originally reported by me.
Nick Gibbons created an issue 2021-12-03
Commit b802b66 swapped to the experimental “type 1” versions of the wall boundary conditions: WallBC_WithSlip, WallBC_NoSlipFixedT, and WallBC_NoSlip_Adiabatic, which do not use ghost cells to implement wall behaviour. This thread is for collecting specific simulations where the ghost-cell free treatment seems to cause problems.
Fabian Zander reports timestep degradation on the flying cube simulation collected in the examples:
The '1' BC actually causes the simulation to essentially stop. It doesn't actually fail, but the time step starts approaching 0 after quite some time in the simulation at which point I've been manually halting the simulation. This doesn't happen at all with the '0' BC simulation.
Step= 150000 t= 1.590e-02 dt= 1.930e-22 cfl=0.40 WC=51835.3 WCtFT=61042856941512818688.0 WCtMS=10315224.9
This issue affects my 11 species 2T blunt body calculations. Failure can be reproduced using the 1cm blunt cone “uniform/one” simulation, with conditions v=6km/s and h=35km. Failure occurs at a CFL of 50, timestep 820, due to invalidate density errors thrown during the preconditioner evaluation in the NK solver.
Tested in commit 948bac2 (Note that commit 85efde7 fixes the problem)