[Mesa-users] PPISN evolution with large nuclear network: timestep shrinks in late burning stages
Farag, Ebraheem
ebraheem.farag at yale.edu
Tue Jul 22 08:25:25 UTC 2025
Hello Shuo,
I'm sorry you're running into these convergence issues. It seems you're pushing the code toward its limits, and as such, I don't have any simple suggestions for improving the convergence of your model. In fact, there might not be a "simple" solution to your issue.
If your timesteps were already sufficiently small (your model was stalling), that could have triggered a net integration failure, however It's not clear.
If you plan to continue investigating this failure, perhaps plot the profile of the model at each timestep (perhaps use pgstar) and look for issues, plot the evolution of the timestep size and retry count to find the specific place where the model begins to fail instead of looking at the final error message (which might occur much later when the model has been failing for many steps). Perhaps turn on "report_solver_progress = .true." and find the specific zone which contains large residuals triggering your retries and timestep dives (maybe it's in the core, maybe it's not in the core?). I would suggest using a photo restart from an earlier point before the crash to quickly diagnose this and save yourself time, just ensure the model wasn't already crashing at the restart point. These are all methods I use to investigate my own stellar models.
I hope this was somewhat helpful, goodluck!
-EbF
________________________________
From: µÔË· <zhaishuo at bao.ac.cn>
Sent: Tuesday, July 15, 2025 10:25 PM
To: Farag, Ebraheem <ebraheem.farag at yale.edu>; mesa-users at lists.mesastar.org <mesa-users at lists.mesastar.org>
Subject: Re: Re: [Mesa-users] PPISN evolution with large nuclear network: timestep shrinks in late burning stages
Dear Ebraheem,
Thank you again for your helpful suggestions. I¡¯ve completed tests of both approaches you mentioned and would like to briefly report the outcomes.
1. Enabling gold_tolerances with maxT_for_gold_tolerances = 1.5d9
With this adjustment, the model produced fewer pulses during the late stages.
However, the timestep still collapsed significantly in the final burning phases.
Additionally, the core trajectory in the log ¦Ñc¨Clog Tc diagram seemed to deviate from the expected path, possibly indicating convergence difficulties or a numerically unstable evolution.
2. Enabling both gold_tolerances and op_split_burn, with op_split_burn_min_T = 1.5d9
Unfortunately, this combination caused the code to crash.
The run crashed with the following error: netint failed: abs(h)<stpmin. It seems the nuclear energy integration failed due to excessively small step sizes.
Many energy-related quantities also became NaN, suggesting a numerical instability after turning on op_split_burn.
The details of the error message is shown below:
[
netint failed: abs(h).lt.stpmin 8.3076475457893520E-170 9.9999999999999997E-029
s% error_in_energy_conservation 6048 NaN
s% total_energy 6048 -4.5618413321028718D+51
s% total_energy_mdot 6048 0.0000000000000000D+00
s% total_energy_start 6048 -4.5619930982387049D+51
s% total_energy_sources_and_sinks 6048 NaN
s% total_nuclear_heating 6048 NaN
s% total_non_nuc_neu_cooling 6048 1.8372787518776513D+47
s% total_irradiation_heating 6048 0.0000000000000000D+00
s% total_extra_heating 6048 0.0000000000000000D+00
s% dt*L_surf 6048 2.1836361470470933D+38
s% dt*L_center 6048 0.0000000000000000D+00
L_surf 6048 6.9194044851879179D+39
s% Fr(1) 6048 NaN
s% Lc(1) 6048 NaN
s% Lt(1) 6048 NaN
sum L 6048 NaN
]
If you have any thoughts on the possible cause of the failure with op_split_burn, or further ideas on stabilizing the timestep with large networks, I¡¯d be very grateful.
The log ¦Ñc¨Clog Tc diagrams for both cases are included in the attachment.
Thanks and regards,
Shuo
-----ÔʼÓʼþ-----
·¢¼þÈË:"Farag, Ebraheem" <ebraheem.farag at yale.edu>
·¢ËÍʱ¼ä:2025-06-24 14:00:43 (ÐÇÆÚ¶þ)
ÊÕ¼þÈË: µÔË· <zhaishuo at bao.ac.cn>, "mesa-users at lists.mesastar.org" <mesa-users at lists.mesastar.org>
³ËÍ:
Ö÷Ìâ: Re: [Mesa-users] PPISN evolution with large nuclear network: timestep shrinks in late burning stages
Hello,
Perhaps see section 10.2 in MESA VI, Jermyn et al. 2023<https://ui.adsabs.harvard.edu/abs/2023ApJS..265...15J/abstract>. If the only difference is the nuclear network, and no other issue in the model is being exhibited, the culprit is likely stiff nuclear burning from adopting a 280 isotope nuclear network. Perhaps try using op_split_burn. Controls in the latest version are found here<https://docs.mesastar.org/en/24.08.1/reference/controls.html#split-burn>, but refer to the defaults in r11701 for specifics. I'm not sure how will operator splitting will play out with the extreme hydrodynamics of a ppisn, but it will likely make the burning more manageable. I see you have gold_tolerances = .false. As well, so turning that back on and potentially playing with these might help:
! From your inlist
relax_use_gold_tolerances = .false.
use_gold_tolerances = .false.
maxT_for_gold_tolerances = 5d9
-EbF
________________________________
From: Mesa-users <mesa-users-bounces at lists.mesastar.org> on behalf of µÔË· <zhaishuo at bao.ac.cn>
Sent: Monday, June 23, 2025 3:57 AM
To: mesa-users at lists.mesastar.org <mesa-users at lists.mesastar.org>
Subject: [Mesa-users] PPISN evolution with large nuclear network: timestep shrinks in late burning stages
Dear all,
I am currently simulating the evolution of pulsational pair-instability supernova (PPISN) progenitors using MESA version r11701.
My setup is based on the same inlists and run_star_extras.f90 modifications as used in Farmer et al. (2020).
I employ a large nuclear reaction network (280 isotopes) to resolve detailed nucleosynthesis during advanced burning stages.
However, I¡¯ve encountered a critical issue: during late evolution (log Tc > 9.4, log ¦Ñc ~ 8), the timestep drops sharply (e.g., log dt ¡Ö -21),
causing the model to either freeze or progress extremely slowly.
This does not occur when using a smaller reaction network under otherwise identical conditions.
In addition to standard timestep and mesh controls, I have tried tuning the following parameters:
- Pextra_factor
- opacity_max
- mesh_delta_coeff
- max_dq
- min_D_mix
Despite these adjustments, the timestep collapse persists when using the large reaction network.
To facilitate feedback, I¡¯ve uploaded a compressed test folder to Google Drive. It includes:
- the custom nuclear reaction network,
- initial metallicity settings,
- relevant inlists,
- history.data and log.txt from the problematic run.
You can download the test setup here:
https://drive.google.com/file/d/15fuWWYKaEFllDBMEoFTPU1WhEM3IsNg2/view?usp=sharing
If further output (e.g., selected profiles) would be helpful, I¡¯m happy to share.
Any suggestions on how to stabilize timestep behavior in this context would be greatly appreciated.
Thanks and regards,
Shuo Zhai,
The National Astronomical Observatories of the Chinese Academy of Sciences
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.mesastar.org/pipermail/mesa-users/attachments/20250722/35263a64/attachment.htm>
More information about the Mesa-users
mailing list