Repository navigation
Conversation
This now uses 128-bit precision for the integration, with 400 point quadrature per interval using the method of Gong et al. 2001. It also expands the density range down to 1.e-15 (from 1.e-12).
|
I ran a 15 I used the latest git with PRs #1103, #1112, and #1115. Here's the comparison of the central evolution:
The inlists are essentially those from: |
2014 CODATA |
|
Perhaps it's a good time to update the fundamental constants in MESA with recent CODATA used to generate the helm tables? in $MESA_DIR/const/public/const_def.f90? |
|
2022 CODATA is the current, but "The CODATA Task Group on Fundamental Constants (TGFC) is currently collecting data for the 2026 adjustment (closing December 31, 2026), with updated values expected in early 2027." so maybe waiting a few more months is worth it? |
|
one issue with constants is that there are a lot of places where duplicate (but slightly different) values of fundamental constants are hardcoded instead of pulled in from const/. It might make sense to centralize those first. a quick codex audit shows:
|
|
great point, I'll add that to the issue. |
|
@zingale , perhaps you can quickly use the eos plotter or something to plot the difference between the old and new helm tables to show the differences for a few different values of X,Y,Z. You'll want to regenerate the eos table images used in the docs using the plotter as well. |
|
sure, I'll try to do some plots tomorrow or Wed. |
|
there are microphysics where there is limited or no knowledge of the fundamental constants that were used. for example, several of the eos that mesa currently uses in its patchwork and the opacity tables. also challenging to catch are hardcoded compound constants comprised of some combination of the fundamental constants. |
|
The table size seems to have bloated quite a bit. It looks like it has been saved with excessive precision (i.e. beyond ~16 digits). This is not needed as MESA will load the table as doubles anyway, so this just increases the download size. |
|
I'll regenerate without the extra precision, using `{:.17g}' since you only worry about double. We need 17 digits for subnormal numbers. |

This regenerates helm_table.dat with 128-bit floating point / 400 quadrature points per integration interval and a wider density range (1.e-15 to 1.e-15) instead of (1.e-12 to 1.e15).
The code to generate this is here:
https://github.com/zingale/electron_positron_cxx/tree/main/generate_table
There will be floating point differences because of the change in precision / quadrature, and also (likely) because of the fundamental constants. I don't know what value of constants was used in the original table construction. The values I used come from SciPy (which in turn uses CODATA), and are here:
https://github.com/zingale/electron_positron_cxx/blob/main/src/fundamental_constants.H
It's easy to regenerate this if we want to change things.