Question About Subcycles

[joehellmers]

On a few of the science codes I've seen failure due to "Error: too many subcycles. !!!" For a couple of them I've simply added castro.max_subcycles=20 for example and that fixed the problem, but now I'm getting it on subchandra. I suppose this is a message related to convergence. I'm concerned that perhaps I'm doing something wrong.

My execution in my Slurm job looks like this

mpirun -n 64 Castro2d.gnu.MPI.SMPLSDC.ex inputs_2d.NSE
castro.max_subcycles=30
amr.plot_file=<project_folder>/subchandra/plt
amr.check_file=<project_folder>/subchandra/chk

Also I changed the tolerance for the tol_hse to 1.e-8 in the _prob_params file.

I guess I can increase the maximum subcycles and/or modify more tolerances, but before I waste a bunch of time I'm wondering if there is something else I should do.

I'm using Castro v26.07.

Thanks.

[zingale]

What problem are you running?

A retry can happen for a number of reasons, including a burn failure, violating the CFL at the end of the timestep, species not summing to 1, or more. The code should tell you the reason for the retry. When a retry happens, the new state is thrown out and it retries the timestep with a smaller dt.

There are a number of options that can control this:
https://amrex-astro.github.io/Castro/docs/timestepping.html#retry-mechanism

Assuming that the code is failing when integrating the reactions, there are a few things you can try:

• turn on the burning retry -- this will retry the burn by swapping the Jacobian type before kicking the failure back to Castro. You can enable this with integrator.use_burn_retry=1.
• loosen the integration tolerance
• try a different integrator -- we've had success with the Rosenbrock integrator, which was recently added. To use this, compile with INTEGRATOR_DIR=Rosenbrock

[joehellmers]

This is for subchandra. The message is burn unsuccessful.

I'll try your recommendations.