Performing simulations recently on Marconi, I noticed an unusual numerical instability. With commit 8eba66af5ea631d6ee9be4ec13d067afdf04ce47 (close enough to the master of stella to be of interest, I hope), I was testing the performance of stella with multiscale kx ky grids. I noticed that after several restarts the simulations grind to a halt with ever reducing time steps and increasing amplitudes. The potential spectrum looks like the attached figure:
comparephikytlow-power-es0-tprim-1.75-0.pdf
the input file for this case is here with other output files (but not the netcdf files):
multiscale-case.zip
To determine if this fault results from the multiscale nature of the simulation, or for some other reason, I drastically reduced the kx ky resolution with the following results. We still find a numerical instability:
comparephikytlow-power-es-tprim-1.75-0-full-range.pdf
This output and input files for this case are here:
Performing simulations recently on Marconi, I noticed an unusual numerical instability. With commit 8eba66af5ea631d6ee9be4ec13d067afdf04ce47 (close enough to the master of stella to be of interest, I hope), I was testing the performance of stella with multiscale kx ky grids. I noticed that after several restarts the simulations grind to a halt with ever reducing time steps and increasing amplitudes. The potential spectrum looks like the attached figure: comparephikytlow-power-es0-tprim-1.75-0.pdf the input file for this case is here with other output files (but not the netcdf files): multiscale-case.zip
To determine if this fault results from the multiscale nature of the simulation, or for some other reason, I drastically reduced the kx ky resolution with the following results. We still find a numerical instability: comparephikytlow-power-es-tprim-1.75-0-full-range.pdf
This output and input files for this case are here:
singlescale-case.zip
Comments and further investigations of this issue are welcomed.