Cut3d failed error about sparta

[stanmoore1]

@aborner1 thanks for the bug reports. We'll look into these.

[stanmoore1]

@aborner1 I can reproduce this. I get a hang and this message: Cut3d failed in cell ID: 542217403. Using a printf statement I see that the error flag is:

    if (errflag == 2)
      error->one(FLERR,"EF: Singlet BPG edge not on cell face");

It is supposed to error out but it never makes it there because it hangs or segfaults in the failed_cell() function.

[aborner1]

@stanmoore1 I just reproduced this bug with the latest master and it still hangs and returns the Cut3d message.

[stanmoore1]

@sjplimp can you take a look?

[stanmoore1]

@aborner1 did #97 fix this issue?

[stanmoore1]

This still fails with latest master:

Step CPU Np 
       0            0        0 
     100 6.8187714e-05        0 
     200 0.00012922287        0 
     300 0.00018811226        0 
     400 0.00024700165        0 
     500 0.00030422211        0 
     600 0.0003631115        0 
     700 0.00042104721        0 
     800 0.00047922134        0 
     900 0.00053811073        0 
Cut3d failed on proc 0 in cell ID: 1073
  lo corner 0.084 0.04 0.02
  hi corner 0.0855 0.0425 0.0225
  # of surfs = 7 out of 606
  surfs: 533 0.0812961 534 0.0812961 537 0.0808713 574 0.0854356 577 0.0851887 578 0.0851887 605 0.0851887
ERROR on proc 0: EF: Singlet BPG edge not on cell face (../cut3d.cpp:611)

[aborner1]

Can confirm this as well as "No positive volumes in cell" error (which I believe is related) still happen with the latest master. This is the easiest reproducer case I have ever created for this problem.

[stanmoore1]

@aborner1 this appears to be a numerical precision issue. You are getting a triangle that is too small to be numerically meaningful. This seems to be from the sphere just barely touching the box bounds in the +/- y and z directions, which makes four minuscule triangles:

  1.96262e-17 min triangle edge length
  3.43546e-20 min triangle area

If I push out the box bounds by epsilon then it works fine:

create_box          0.06 0.09 0.01999999999999999 0.07000000000000001 0.01999999999999999 0.07000000000000001

  0.00251578 min triangle edge length
  2.74751e-06 min triangle area

[sjplimp]

@aborner1 @stanmoore1 Alternatively does the problem go away if the sphere is translated by an epsilon different amount so that the clipping does not produce any tiny triangles? I wonder if there is a better way to clip to avoid epsilon-size triangles?

[stanmoore1]

@sjplimp yes translating the sphere by epsilon also works:

read_surf data.sphere scale 0.025 0.025 0.025 trans 0.09 0.04500000000000001 0.04500000000000001 clip

  0.00251578 min triangle edge length
  2.74751e-06 min triangle area

[stanmoore1]

Translating the other way still makes tiny triangles but they are larger than epsilon, so it succeeds because it has enough numerical precision to do the calcs:

read_surf data.sphere scale 0.025 0.025 0.025 trans 0.09 0.04499999999999999 0.04499999999999999 clip

  7.37615e-17 min triangle edge length
  1.37243e-19 min triangle area

[stanmoore1]

Fixed by #244. Closing #7.