Particle sent to self error about sparta

[stanmoore1]

In a different code, we also had issues with axisymmetric. It seems to be more prone to round error issues. @sjplimp and @mgallis, what do you think?

[stanmoore1]

I talked to @sjplimp about this and he would like to understand why this is happening. @aborner1 do you have a reproducer input script?

[aborner1]

I only have large cases that reproduce it, I would need to see if I can generate something small that ideally can be run on a small number of procs, otherwise it's going to be a nightmare to debug.

[stanmoore1]

@aborner1 I have assigned this issue to you. Once we have a reproducer I will reassign it to @sjplimp. Thanks.

[stanmoore1]

This issue is possibly fixed by #194, @aborner1 can you confirm?

[aborner1]

@stanmoore1 Will test the new code on large axisymmetric simulations. It wasn't something I could reproduce, so I will never be able to fully confirm, but unless the error reappears, I agree that this sounds like a logical fix.

[stanmoore1]

Sounds good. We used fix check grid/check 1 error to debug #194, that is a good option to turn on if you are seeing issues.

[stanmoore1]

And global comm/sort yes so it is reproducible.

[stanmoore1]

@aborner1 I'm going to close this issue, but please reopen a new issue if you see the "Particle sent to self error" again.

[aborner1]

@stanmoore1 @sjplimp I just encountered this error on a 3D run. So it is definitely still an issue in some cases. Hard to reproduce, as, once again, it happens on a large simulation after a few hours. Can a similar bug fix be implemented for 3D?

[stanmoore1]

Can a similar bug fix be implemented for 3D?

There is no analogue for that axisymmetric issue in 3D, i.e. we tried in straight 2D and it works fine when a particle lands on the boundary.

Here is how I would recommend debugging this issue:

• Turn on global comm/sort yes so the runs are reproducible
• Turn on fix check grid/check 1 error to better diagnose grid cell problems
• Write out a restart file periodically, say every 10,000 steps
• When the simulation crashes after a few hours, increase the output frequency of restart files, i.e. write out a restart file every 1000 steps, and restart from the last checkpoint file to "zoom in" on the bug
• Keep zooming in until you have a reliable reproducer that runs in a hundred timesteps or less
• Send all the files (input and restart) to us for further debugging

[stanmoore1]

I just got another "particle outside of cell" error in a large 2D axisymmetric simulation, even with the fix in #194. I'll try to debug.

[stanmoore1]

#221 should fix any issues for axisymmetric by just deleting the particles.

[stanmoore1]

@aborner1 can you try out #221 on your axisymmetric case? It should be a fix-all for these types of issues. Not sure why you would see issues in 3D though.

[stanmoore1]

@aborner1 #221 should fix all axisymmetric issues so I'm going to close this issue. If you see more issues please let us know.