MESA version 23051
Comparing this inlist (attached as .txt) for a science case against inlist_1.3M_ms_high_Z from the test_suite case of the same name, I have systematically removed each difference in physics (replace with basic.net, use standard opacities, etc) to try to isolate the reason the science inlist fails. I have not found any change that enables successful passage through the helium flash
inlist_issue.txt
However, introduction of the following control parameters does enable passage through the he flash:
convergence_ignore_equL_residuals = .true. relax_use_gold_tolerances = .true.
Despite confirmation from other devs that this control is needed to get models in this regime through the he flash (as of some point after release 10398 and the solver controls overhaul),
there is no documentation to accompany convergence_ignore_equL_residuals = .true.
I think a few things need to happen:
-
the effect of this and similar controls needs to be documented. As far as I can tell, the meaning of this control is to shut off consideration of discrepancies in the luminosity equation entirely (?).
a. Corollary: is that safe??
-
It needs to be made loud and clear that this setting is necessary for (most) AGB models. I advocate for introducing it in inlist_1.3M_ms_high_Z even though it is not required in that case, just so users are aware of it. They will almost certainly consult this test_suite case when trying to figure out their own convergence issues for AGBs. It has been pointed out to me that this setting is in the 1M_pre_ms_to_wd test_suite case, but that is not the first place I personally would look for usable AGB settings.
-
We need a test case for AGBs (@giulia-cinquegrana and I are doing this, stand by)
-
Ideally, we need a better solution than (even a well-documented) magic flag for dealing with this. If the era of tighter tolerances made getting through the he flash impossible without telling the solver to ignore one of the stellar structure equations(?)...that seems like something we should fix.
end rant!
MESA version 23051
Comparing this inlist (attached as .txt) for a science case against
inlist_1.3M_ms_high_Zfrom thetest_suitecase of the same name, I have systematically removed each difference in physics (replace withbasic.net, use standard opacities, etc) to try to isolate the reason the science inlist fails. I have not found any change that enables successful passage through the helium flashinlist_issue.txt
However, introduction of the following control parameters does enable passage through the he flash:
convergence_ignore_equL_residuals = .true. relax_use_gold_tolerances = .true.Despite confirmation from other devs that this control is needed to get models in this regime through the he flash (as of some point after release 10398 and the solver controls overhaul),
there is no documentation to accompany
convergence_ignore_equL_residuals = .true.I think a few things need to happen:
the effect of this and similar controls needs to be documented. As far as I can tell, the meaning of this control is to shut off consideration of discrepancies in the luminosity equation entirely (?).
a. Corollary: is that safe??
It needs to be made loud and clear that this setting is necessary for (most) AGB models. I advocate for introducing it in
inlist_1.3M_ms_high_Zeven though it is not required in that case, just so users are aware of it. They will almost certainly consult thistest_suitecase when trying to figure out their own convergence issues for AGBs. It has been pointed out to me that this setting is in the1M_pre_ms_to_wdtest_suitecase, but that is not the first place I personally would look for usable AGB settings.We need a test case for AGBs (@giulia-cinquegrana and I are doing this, stand by)
Ideally, we need a better solution than (even a well-documented) magic flag for dealing with this. If the era of tighter tolerances made getting through the he flash impossible without telling the solver to ignore one of the stellar structure equations(?)...that seems like something we should fix.
end rant!