[timlichtenberg]

Description

Option 2 of #963: Zalmoxis Anderson acceleration (interior_struct.zalmoxis.use_anderson) is now on by default, and the structure call without temperature data (the setup solve and the equilibration) keeps it on. That call still turns use_jax off, because the internal temperature mode converges to a different structure on the JAX path; Anderson needs no such exclusion on either path. The four input files that pinned use_anderson = false as the old default now carry true, and the schema and the config docs follow.

This PR also raises the fwl-aragog floor to 26.10.08, the release that caches the jitted CVODE RHS and Jacobian and the parsed EOS tables (FormingWorlds/aragog#73).

Together these are options 1, 2 and 8 of #963; option 3 (vectorised energy integrals) follows separately. Outputs are not bit-identical to main: the changes move the coupled trajectory at the level of a 1 % change of the Aragog rtol, which is the acceptance level set in #963.

Validation of changes

Paired coupled Earth runs on the same machine, started together, this branch with the paired aragog branch against main and against main with rtol x 1.01.

Anderson default alone (option 2), dynamic structure:

• Phi 0.05 crossing time +0.26 % against main; the median trajectory differences stay below 1.13 times the rtol spread, the maxima below the rtol spread.
• Structure parity on a 20-step dynamic run with 3 re-solves, same step sequence on both arms: R_int agrees to 2e-7 and M_int to 1e-6 (JAX path), R_int to 6e-8 and M_int to 3e-6 (numpy path). All structure solves converge on both arms.
• Wall time of that 20-step run: 24.3 to 22.2 min on the JAX path, 36.7 to 12.9 min on the numpy path (use_jax = false), where every re-solve runs with Anderson and the density seed.

All three options together: the static run goes from 26 to 11 min. In the dynamic runs, every median difference on a uniform time grid stays within twice the spread of two rtol perturbations, and the Phi 0.05 crossing (+0.15 to +0.17 %) lies inside their range (+0.004 % and +1.64 %). Single-column maxima exceed that spread at local features: an F_cmb sawtooth at 1.8 to 3.8 kyr in one run that does not repeat in a second run, and the onset time of the T_magma drop near 50 kyr.

Tests: tests/interior_struct/test_zalmoxis.py (the guard keeps Anderson and turns off JAX for a call without temperature data, over use_anderson x outer_solver), tests/interior_struct/test_zalmoxis_outer_solver_config.py (use_anderson reaches the solver configuration), tests/config/test_struct.py (default). The nightly science validation on this branch with the aragog 26.10.08 release from PyPI is green (25 of 25 jobs): the unit tier reports 5081 passed and the integration tier 206 passed on both runners.

Tested on macOS 26 (Apple silicon), Python 3.12.

Checklist

• I have followed the contributing guidelines
• My code follows the style guidelines of this project
• I have performed a self-review of my code
• My changes generate no new warnings or errors
• I have checked that the tests still pass on my computer
• I have updated the docs, as appropriate
• I have added tests for these changes, as appropriate
• I have checked that all dependencies have been updated, as required

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.66%. Comparing base (4fc66f8) to head (bef40f8).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #971      +/-   ##
==========================================
- Coverage   93.97%   93.66%   -0.32%     
==========================================
  Files         128      128              
  Lines       20891    20890       -1     
  Branches     3515     3515              
==========================================
- Hits        19633    19567      -66     
- Misses        908      914       +6     
- Partials      350      409      +59     

Flag	Coverage Δ
nightly	93.63% <100.00%> (+0.08%)	⬆️
unit-tests	92.28% <100.00%> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.