As an example, in .../benchmarks/03-fluidbed/Size0016 changing ‘0.0032’ to ‘0.0031’ or ‘0.0033’ throughout yields an assertion error
Assertion 'amrex::almostEqual(dx,dy) && amrex::almostEqual(dy,dz)' failed, file ".../src/setup/mfix_calc_bc_areas.cpp", line 178. (This running with GPU support enabled, but I doubt this matters.)
Is this an intended and/or well-documented constraint, or something that should be remedied?
Currently*, MFIX-Exa requires a cubic fluid mesh. So changing the x and z domain lengths from 3.2mm to 3.1mm would make dy > dx = dz. If you really needed that mesh size, you can change the y domain extent from 0.01 to 50*0.0031/16 = 0.0096875 throughout, and then it will run.
*NOTE: Jordan did mention at the last user’s meeting that a recent change should make it so that we could use non-cubic grids. If there is a strong need for that, please make that case which might encourage future development in that area.
I don’t expect that we’ll support non-isotropic meshes anytime soon because the AMReX solvers we rely on currently do not support it.
Furthermore, the online documentation does state:
MFIX-Exa simulations with a non-uniform mesh will not run.
Thanks—I completely forgot they had to be square.