search

ECP-WarpX / WarpX

WarpX is an advanced electromagnetic & electrostatic Particle-In-Cell code.
https://ecp-warpx.github.io
Other
304 stars 194 forks source link

Possible bug related to domain decomposition #4970

Open Yin-YinjianZhao opened 5 months ago

Yin-YinjianZhao commented 5 months ago

Recently, we are using WarpX to do some 2D simulations. We found that when numprocs = [4,16] is changed, the results look very different. For example, the three plots below used numprocs = [1,64], numprocs = [4,16], numprocs = [16,4], respectively. The mobility diagnostics are: numprocs = [1,64]: mu_ave_pict_rMPI=1 numprocs = [4,16]: mu_ave_pict_rMPI=4 numprocs = [16,4]: mu_ave_pict_rMPI=16

I wonder if similar bugs have been detected before @RemiLehe @ax3l @dpgrote ? Note the cell sizes are:

nx = 96
nz = 256

We will try to use only multiples of 2 and test again different domain decompositions.

ax3l commented 5 months ago

Hi @Yin-YinjianZhao,

Thank you for reporting this! We are not aware of an existing bug of this kind, so you probably found something new. Can you please share a small reproducer input + analysis with us?