Closed zingale closed 5 months ago
reproducer:
in DustCollapse:
make DIM=3 -j 20
then use the attached inputs
inputs_3d_monopole_regtest.txt
and run as:
mpiexec -n 20 ./Castro3d.gnu.MPI.ex inputs_3d_monopole_regtest max_step=20
note that the reproducer sets a fixed_dt
to force problems and a retry.
this definitely seems to be related to doing amr.subcycling_mode=None
here's the issue. When we reject a step and retry w/o subcycling, it seems that we are not resetting LevelData[level]->get_state_data(State_Type).curTime();
on the finer levels.
Then when we enter into get_new_grav_vector()
, we have a mismatch between the time
that is passed in on level 1 and the time that is inferred from State_Type
.
So I think in the reset for retry, we need to update all of the levels above us as well if we are not subcycling.
When running with AMR with no subcycling and monopole gravity, if we trigger a retry, then we get errors in the
make_radial_gravity
of the form:I've seen this with the massive star problem but I can also reproduce with DustCollapse (see below)