Picongpu: Weird assymetrical electron density in the LWFA example ( with dev branch ).

[pordyna]

I noticed that the LWFA example (default setting, 4.cfg configuration file) produces weird outputs compared to what it was doing before (current master branch 0.4.3).

Run on hemera ( tried on both k20 and gpu partitions ).
For people with access, here you can find the full simulation outputs (2020_03_26_dev and 2020_03_26_0-4-3):
/bigdata/hplsim/scratch/ordyna35/runs/LWFA

Bellow, for comparison, two exemplary pngs generated with the png plugin (XY plane).
Dev (900th simulation step, ):

0.4.3 (900th simulation step):

This is not only a problem with the png plugin, electron density dumped by the hdf5 or adios looks the same.

@PrometheusPi, @sbastrakov, @psychocoderHPC

[pordyna]

To me it looks like some of the electrons would stop moving :sweat_smile:

[sbastrakov]

Thanks for reporting @pordyna !

[sbastrakov]

I am looking into this, as this might be a bug to fix before the release.

[sbastrakov]

Intermediate result: this is definitely a bug that needs to be fixed urgently. Introduced to dev some time between 5 February and 3 March. I will narrow it down and fix, think today.

[PrometheusPi]

@pordyna Thank you for reporting this issue. It is awesome you found this bug.
I just checked my LWFA simulations I ran tonight and the same error occurs (k20/hemera - 2D only, 16GPUs):

Seems to affect a fixed number of cells at the absorbers.

[sbastrakov]

@PrometheusPi I think I figured out the bug. Since Feb. 20 the exponential absorber strength was very wrong. Making a PR to fix in a moment.

[PrometheusPi]

@sbastrakov Excellent! I am looking forward to your pull request.
(I did not see any recent changes in picongpu/share/picongpu/examples/LaserWakefield/include/picongpu/param/grid.param)

[sbastrakov]

I confirm that #3161 caused the bug and #3218 fixes it.
The error was introduced to the code side, not the setup.

[sbastrakov]

The fixing PR is merged, closing.