[Mesa-users] Models with mesa_204.net do not attain the lgTmax value

Mon Dec 9 17:53:32 UTC 2024

Hello Amar,

I hope this response is not too late to be helpful.

Very nice to see you pushing mesa with big core-collapse networks! 'rel_run_E_error' represents the total relative energy error across the entire evolution of the model. The default cap in the test_suite you are running is is set to 0.01 or (1%), although you can relax this with controls:

For the whole run:
max_abs_rel_run_E_err = -1 ! default is unset, but it seems set to 1d-2 in your inlist.

For a given timestep:
limit_for_rel_error_in_energy_conservation = 1d-4 ! default
hard_limit_for_rel_error_in_energy_conservation = 1d-3  ! default

Considering your model has 580 retries, it would be good to make sure your timestep is being controlled by something physical throughout most of the evolution. I like the delta_lgRho_cntr_limit" or using Tmax or Tcenter or dx_nuc_drop_limit controls. Hitting the "max_abs_rel_run_E_err" stopping condition might indicate some lack of convergence in your model unless there is something atypical going on, although you are free to relax it if it doesn't affect your science.

> Please provide your valuable suggestions. In addition to the above-described model, four other models (with masses 12, 13, 13, and 14 ) were also simultaneously running in the same Mac Studio in different terminal tabs. The time taken by the models to reach the above stage is close to 75 days!!! Is it normal, or can I do some tweaks to improve the computational speed? I left "OMP_NUM_THREADS" unset while running MESA. Your help with this will be highly appreciated.

A few comments here.
-I would definitely set OMP_NUM_THREADS or else it's probably defaulting to something lower then you'd like?
- I also have an otherwise identical mac-studio to yours (except it is an m1). I would personally not recommend running more than 2 models at one time, and would recommend giving each model at least 10 cores, "export OMP_NUM_THREADS=10" for each. In my experience, my m1 mac studio is relatively slow single-core performance compared to most modern x86 processors (in MESA) and hence it's not necessarily going to be quick in this regard.
- When running a cc model with say mesa_204.net or mesa_206.net, I typically run these on my institution's supercomputer. Generally I run these models with 16 cores (on amd epyc nodes), and they take ~2 weeks to run to core-collapse at a moderate resolution ( ~ 20 k timesteps, 3-4k zones at cc). My mac studio is half as fast at best (so 1 month if i dedicate the whole computer to one model). In this regard, if you can get access to some other computing resources, it could help accelerate your project timeline.
- Also, it's good to check if your models can reach cc with something like approx21 first, before committing resources toward a large network run.
- Lastly, if you are evolving big networks, perhaps ensure you are adopting a lower "op_split_burn_minT" than in the test_suite (There should be a comment in the inlist saying this). The bigger the network, the more solver variables and equations need to be solved, and the stiffer the Jacobian matrix becomes to solve.  I adopt op_split_burn_minT = 2.8d9 in my large network runs, but it can be more conservative to adopt a lower number (2d9 or 1d9). If you don't do this, your model can stall at these high temperatures and produce large energy errors.

I hope this was helpful!

-EbF

________________________________
From: Mesa-users <mesa-users-bounces at lists.mesastar.org> on behalf of Amar Aryan via Mesa-users <mesa-users at lists.mesastar.org>
Sent: Tuesday, October 29, 2024 4:07 AM
To: mesa-users <mesa-users at lists.mesastar.org>
Subject: [Mesa-users] Models with mesa_204.net do not attain the lgTmax value

Dear MESA users,

I am trying to utilize one of the most extensive nuclear reaction networks (mesa_204.net<http://mesa_204.net/>) while evolving a solar metallicity 14 solar mass star from ZAMS to core-collapse. I am using mesa-r24.03.1 on a Mac Studio (M2 Ulta, Sonoma) and 128GB of RAM with mesa-r24.03.1. I utilize the 20M_pre_ms_to_core_collapse test_suit directory. Mentioned below are some more set-up details:

inlist_common:
time_delta_coeff = 0.8
mesh_delta_coeff = 0.8
mesh_delta_coeff_for_highT = 0.9

inlist_mass_Z_wind_rotation:
new_omega_div_omega_crit = 5d-2
Dutch_scaling_factor = 0.1

The model successfully passes all the inlists files up to "inlist_to_core_c_burn", but it will terminate while running "inlist_to_lgTmax" by displaying:

stop because abs rel_run_E_err exceeds limit (max_abs_rel_run_E_err)

    0.1000038039E-01    0.1000000000E-01

__________________________________________________________________________________________________________________________________________________

       step    lg_Tmax     Teff     lg_LH      lg_Lnuc     Mass       H_rich     H_cntr     N_cntr     Y_surf   eta_cntr   zones  retry

    dt_days    lg_Tcntr    lg_R     lg_L3a     lg_Lneu     lg_Mdot    He_core    He_cntr    O_cntr     Z_surf   gam_cntr   iters

   age_days    lg_Dcntr    lg_L     lg_LZ      lg_Lphoto   lg_Dsurf   CO_core    C_cntr     Ne_cntr    Si_cntr  v_div_cs       dt_limit

__________________________________________________________________________________________________________________________________________________

       6700   9.529648   3105.067  12.469148  11.238136  13.573740   7.965064   0.000000   0.000000   0.321152   2.990948   2800    580

 9.8410E-05   9.529648   3.103883   6.013303  11.441592 -99.000000   5.608676   0.000000   0.000026   0.014165   2.680830     19

 6.2126E+09   7.697737   5.130743 -99.000000  18.473546  -8.748448   3.715466   0.000001   0.000000   0.367928 -0.503E-02  max increase

                               rel_E_err    5.0880938722980180D-07

                       log_rel_run_E_err       -1.9999834801025218

terminated evolution: abs_rel_run_E_err

termination code: exceeded max abs rel_run_E_err

Please provide your valuable suggestions. In addition to the above-described model, four other models (with masses 12, 13, 13, and 14 ) were also simultaneously running in the same Mac Studio in different terminal tabs. The time taken by the models to reach the above stage is close to 75 days!!! Is it normal, or can I do some tweaks to improve the computational speed? I left "OMP_NUM_THREADS" unset while running MESA. Your help with this will be highly appreciated.

With best regards,
Amar
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.mesastar.org/pipermail/mesa-users/attachments/20241209/92799e6c/attachment.htm>