Handling of Single Particles / Perfectly Flat / Line Beams

[ax3l]

While running in debug (#43), I realized the ResizeMesh function does assert the simulation:
https://github.com/ECP-WarpX/impactx/pull/19/files#r778820245

0::Assertion `hi > lo' failed, file "/home/axel/src/impactx/build/_deps/fetchedamrex-src/Src/Base/AMReX_Algorithm.H", line 107, Msg: "Error - calling bisect but lo and hi don't describe a reasonable interval." !!!

I temporarily disabled calling it and enabled debug builds in CI (#43). The function probably needs to be fixed.

[RemiLehe]

Thanks for reporting this.
Could you post the input script that created this error?
Is this a case with no macroparticles? (Also: when there are no macroparticles: should we just skip the space charge calculation altogether?)

[ax3l]

Thanks for looking into this! It's the example we have checked in for CI: examples/input_fodo.in That one should have 1 particle: we hard-code init of one particle atm.

[ax3l]

Discussed on Slack:

since we cannot calculate min/max/std for <2 particles in a sim, we should dynamically disable:

• grid scaling from ResizeMesh
• deposition
• Poisson solve
• space charge push

as soon as our particle container drops <2 particles.

[ax3l]

Actually, the problem is more complicated:

If we initialize particles in a line or plane, e.g., with no delta in x or y or z in a "toy" distribution - then we will also get problems sizing a mesh accordingly along those "flat" dimensions.

The same is true if someone accidentally initializes two particles in a sim - on the same position (and using same momentum + charge/mass ratio).

I think the general solution would be:

• calculate the beam extent
• check if it's "zero" for a dimension and do not resize the mesh, deposit or Poisson solve in those dimensions.
  • alternatively/maybe simpler: assume homogeneity/periodicity in those dimensions.

[ax3l]

@RemiLehe on Slack:

One thing to be careful about: if we do not resize the mesh, we should also make sure to never call Redistribute (in any part of the PIC loop), otherwise particles will be removed.

👍