Unexplained end of run

[PLoren-Aguilar]

We are tying to run the ccsn_IIp example, and the run systematically ends at some point, with one single message saying "Killed ./star". No extra logs or messages. We are not sure what the cause could be, but we know it systematically occurs at the same step. Any help (even if it just any debug suggestion) would be amazing. We don't know where to start.

Many thanks,

Pablo Loren-Aguilar

[warrickball]

I won't have an opportunity to post much detail any time soon but know that this is almost certainly because you are running out of RAM. If you're running on a virtual machine or other environment that controls memory (like an HPC scheduler job), try increasing the allocation. Otherwise, you could try setting num_steps_for_garbage_collection in &star_job to something like 100 or 1000 to see if that helps, but it might just be that your stellar model is too large. This can happen if you're tracking a large number of chemical species at a large number of meshpoints.

[PLoren-Aguilar]

Many thanks, this certainly seems to be the case after monitoring the RAM consumption. Unfortunately, garbage collector didn't help. Is there any documentation that could help us learning how to lower the number of meshpoints and/or chemical species? Thanks!

[fxt44]

for pre-2023 versions of mesa the size of the matrix is set by the 4-byte integer variable lwork in star/private/star_newton.f90 in get_newton_work_sizes. the upshot, assuming i did the substitutions correctly, is

memory = (n_isotopes + 5) * n_cells * (3*(n_isotopes + 5) + 9) < 2^31

where n_isotopes is the number of isotopes and n_cells is the number of cells.
given niso, one can solve for the maximum number of cells nz before this memory limit is hit.

in 2023 rob farmer did some reaggangment of the work arrays

30d37de

that should, in principle, allow a larger problem size of

max_isotopes = sqrt(2^31 / n_cells)

[fxt44]

a system kill due to insufficient ram comes up occasionally. see for example https://lists.mesastar.org/pipermail/mesa-users/2023-January/014259.html .