EB generation fails for STL files with certain EB options

[ankithadas]

EB Generation results in out of bounds values for EB boundary centroid (outside -0.5 to 0.5 range) for certain geometry types. To reproduce this, add

diff --git a/Src/EB/AMReX_EB2_3D_C.cpp b/Src/EB/AMReX_EB2_3D_C.cpp
index 863bfb501..6c64987ec 100644
--- a/Src/EB/AMReX_EB2_3D_C.cpp
+++ b/Src/EB/AMReX_EB2_3D_C.cpp
@@ -128,7 +128,19 @@ void set_eb_data (const int i, const int j, const int k,
     bcent(i,j,k,0) = (Bx + nx*vfrac(i,j,k)) * apnorminv * dx[1] * dx[2];
     bcent(i,j,k,1) = (By + ny*vfrac(i,j,k)) * apnorminv * dx[0] * dx[2];
     bcent(i,j,k,2) = (Bz + nz*vfrac(i,j,k)) * apnorminv * dx[0] * dx[1];
-
+    // check. The value put here is rather arbitrary...
+    bool a1 = bcent(i,j,k,0) > 2;
+    bool a2 = bcent(i,j,k,1) > 2;
+    bool a3 = bcent(i,j,k,2) > 2;
+    
+    if (a1 || a2 || a3) {
+        amrex::Print(amrex::Print::AllProcs) << "Bcent values (" << bcent(i,j,k,0) << ","
+                                           << bcent(i,j,k,1) << ","
+                                           << bcent(i,j,k,2) << ") at index "
+                            << "(" << i << "," << j << "," << k << ")\n";
+        amrex::Abort("Bcent values outside range!");
+    }
+    
     Real dx1 = dx[0];
     Real dx2 = dx1*dx1;
     Real dx3 = dx2*dx1;

and run amrex-tutorials/ExampleCodes/EB/STLtest with inputs

eb2.geom_type             = stl
eb2.stl_file              = B1-5-ascii.txt

eb2.stl_scale             = 1e-3                   # default is 1
eb2.stl_center            = 0      0      0     # default is (0,0,0)
eb2.stl_reverse_normal    = 0                   # default is 0.
eb2.cover_multiple_cuts   = 1
#eb2.small_volfrac=0.05

prob_lo        =   -0.6  -0.6  0.00     # Lo corner coordinates
prob_hi        =   +0.6   0.6  0.60    # Hi corner coordinates

ncells                = 128    128     64

max_grid_size         = 32

with STL file attached.
B1-5-ascii.txt

Adding small vfrac fixes this specific case. This problem seems to persist for other cases as well.

[WeiqunZhang]

What do you mean by "Adding small vfrac"? eb2.cover_multiple_cuts=1? If it's a multi-cut cell and the flag is false, it's a run time error, because the results will be garbage.

[ankithadas]

Sorry I meant adding eb2.small_volfrac=0.05 seems to fix that case. Actually, my previous test case was incorrect. I should not put assert statements inside the EB generation functions. I have made a proper reproducer

Input file

eb2.geom_type             = stl
eb2.stl_file              = B1-1-ascii.txt

eb2.stl_scale             = 1e-3                   # default is 1
eb2.stl_center            = 0      0      0     # default is (0,0,0)
eb2.stl_reverse_normal    = 0                   # default is 0.
eb2.cover_multiple_cuts   = 1
#eb2.small_volfrac=0.05

#prob_lo        =   -1.2  -1.2  0.00     # Lo corner coordinates
#prob_hi        =   +1.2   1.2  1.20    # Hi corner coordinates
prob_lo        =   -3.6  -1.2  0.00     # Lo corner coordinates
prob_hi        =   +3.6   1.2  1.20    # Hi corner coordinates


#ncells                = 128    128     64
ncells                = 384    128     64

max_grid_size         = 32

Code
main.cpp
To compile, replace amrex-tutorials/ExampleCodes/EB/STLtest/main.cpp with the above

STL file
B1-1-ascii.txt

The generated BCent value is > 1e12. It looks like a division by zero error. I believe it is due to some of the constants (sml and tiny) defined in AMReX_EB2_3D_C.cpp file. Changing sml to 1e-13 and tiny to 1e-14 seems to fix it. However, I am unsure what exactly is happening in this function to suggest this change.