Regarding Failed to build required coarse EB level X !!!

[milin333martin]

I have been trying to build an STL in an EB_CNS based code. The issue is while building the STL file. Wherein, I get the error :
amrex::Abort::17::Failed to build required coarse EB level 1 !!! SIGABRT
I tried moving around the EB domain to see if it fixes the issue, but I am unsuccessful. I tried reducing the max level of refinement to 0 and 1, wherein the error doesnt occur but the built geometry is rather garbage. Any higher the value of max levels and I end up with this problem.

I'd greatly appreciate any help on how to get rid of this problem.

For your reference, these are some of the contents of my input file:
`amrex.fpe_trap_invalid=0

max_step = 1
stop_time = 1.e-6

geometry.is_periodic = 0 0 0
geometry.coord_sys = 0 # 0 => cart, 1 => RZ 2=>spherical
geometry.prob_lo = 0.0 0.0 0.0
geometry.prob_hi = 0.02 0.02 0.02
amr.n_cell = 64 64 64

cns.lo_bc = 4 5 5
cns.hi_bc = 2 5 5

cns.cfl = 0.2 # cfl number for hyperbolic system

REFINEMENT / REGRIDDING

cns.tag_probspecific = 1
amr.max_level = 5 # maximum level number allowed
cns.refine_max_dengrad_lev = 2
amr.ref_ratio = 2 2 2 2 2 2 2 2 # refinement ratio
amr.regrid_int = 2 2 2 2 2 2 2 2 # how often to regrid
amr.blocking_factor_x = 16
amr.blocking_factor_y = 16
amr.blocking_factor_z = 16

amr.max_grid_size_x = 64
amr.max_grid_size_y = 64
amr.max_grid_size_z = 64

amr.n_error_buf_x = 2 # number of buffer cells in error est
amr.n_error_buf_y = 2
amr.n_error_buf_z = 2
amr.grid_eff = 0.98 # what constitutes an efficient grid
amr.refine_grid_layout = true
amr.regrid_on_restart = 1
amr.compute_new_dt_on_regrid = 1

eb2.geom_type = stl
eb2.stl_file = B6.stl
eb2.stl_scale = 1
eb2.stl_center = 0.005 0.0098 0.00098
eb2.reverse_normal = 0
eb2.stl_use_bvh = 1

Important EB settings

eb2.max_grid_size = 32
eb2.cover_multiple_cuts=1
eb2.extend_domain_face = 1
cns.refine_cutcells = 1

`
And the STL file is:
B6.zip

Thank you in advance for your help

[WeiqunZhang]

Could you clarify a few things?

eb2.reverse_normal = 0 is probably a typo. It should read eb2.stl_reverse_normal = 0. But the real question is whether you want to reverse the normal or not. Since the default is not, the typo is harmless if eb2.stl_reverse_normal = 0 is what you want. However, most of your domain is completely covered. If I make a 2d slice, the shape is basically a triangle inside a square. Both are very small. Everything outside the square and inside the triangle are covered. The open space between the two is tiny. Is that really what you want to simulate? If so, why not make the domain much smaller since most of the domain is covered.

[WeiqunZhang]

This is what I see. All those blue regions are covered.

[milin333martin]

Thank you again for looking into the issue. The reverse normal was indeed a typo. But I think my definition for the flow region is reversed. The objective of the simulation is to study detonation moving from a wide domain to a much smaller narrow passage. So in our case only the region between the square and triangle must be covered, while outside the square and inside the triangular region, we are supposed to have fluid. The reason why we can’t make the domain smaller is to have at least one detonation cell within the domain (y -dirn especially).

[WeiqunZhang]

Okay. Now the issue is the narrow region is too narrow. Even with 256^3 cells, the triangle region has exactly one open cell surrounded by a number of cut cells. So I don't think using 64^3 cells at level 0 could work. Even we are able to build the EB, I don't think the algorithm could work.

That being said, you could try to build EB with EB2::Build(...., bool build_coarse_level_by_coarsening = false, ...). For hyprebolic problems with EB surface fully refined, we don't need to build coarse levels by coarsening.