c-white commented 5 years ago

I was running 1D shock tubes with refinement, and I noticed problems with one test where a magnetized rarefaction passed from fine to coarse cells. I first noticed this in SR, but the problem is entirely reproducible in Newtonian MHD. The problem seems to very much show up in By first (there is no Bx or Bz in this problem), eventually affecting all the variables.

To reproduce, first make athena/temp/ and place shock.txt as shock.athinput and plot_shock.txt as plot_shock.py there.

Next, compile and run two executables with different numbers of ghost zones:

./configure.py -b --prob shock_tube --coord cartesian --flux hlld --nghost 2 -hdf5
make clean
make -j
mv bin/athena bin/athena_2
bin/athena_2 -d temp -i temp/shock.athinput job/problem_id=shock_2

./configure.py -b --prob shock_tube --coord cartesian --flux hlld --nghost 4 -hdf5
make clean
make -j
mv bin/athena bin/athena_4
bin/athena_4 -d temp -i temp/shock.athinput job/problem_id=shock_4

Finally, overplot the results:

temp/plot_shock.py shock 8 Bcc2
open temp/shock.Bcc2.00008.png

shock bcc2 00008

The refined region runs from 0.3 to 0.7, and as the rarefaction passes 0.3 the transverse field glitches. Note we don't need all 4 ghost cells, since this is using PLM reconstruction. The problem persists when using PPM. It also exists when using PLM and 3 ghost zones (or even 10 ghost zones).