Pull request to put the Isca Mars configurations into Isca's master

[sit23]

The Isca Mars configurations haven't ever been merged into the master, and I'd like to make sure this is done ASAP.

In this P/R I have put in my grey-radiation Mars and Socrates Mars configurations without dust as test cases that can be run with the trip tests. These are the ones that are described in this paper:

https://www.mdpi.com/2073-4433/10/12/803

I'd also like to get in Emily Ball's Isca-Mars configurations that built on my Mars stuff, but I wanted to do this first. This P/R is ready for merging (pending passing the trip tests).

[sit23]

Having put all of the Mars-related modifications into this P/R, nearly all of the trip tests. However, I worked with Claude to find the causes of these trip test failures, and have identified and fixed all of them. The summary of this investigation (generated by Claude) is given below:

Summary

Three genuine numerical regressions between mars_dev_gfort_local_heating_2026 and master were tracked down and fixed. All three shared the same shape: a Titan-motivated code change was added as the only code path instead of being gated behind its own (already-existing, default-off) namelist option, so it silently ran for every experiment — including ones that never asked for it — and broke bit-reproducibility against master.

• frierson and realistic_continents_variable_qflux diverged from master (tiny for frierson, substantial for realistic_continents_variable_qflux — ps off by up to 1.7).
• bucket_model diverged substantially (ps off by 4.5, temp by 0.5).
• variable_co2_grey diverged moderately (ps off by 0.023).

All three are now bit-identical to master. Socrates-based trip tests (socrates_aquaplanet, socrates_aquaplanet_cloud, ape_aquaplanet), previously untestable locally due to an unrelated environment issue, now run and pass as well.

Fix 1 — spectral_dynamics.F90: grid-tracer accumulation order (commit 3cdb2e93)

update_tracers's grid-tracer branch was restructured to support a new Titan option (do_spec_tracer_filter, spectral damping of the grid tracer) by accumulating the horizontal+vertical advection tendencies into dt_tr and recomputing tr_future from it once at the end, instead of incrementing tr_future directly after each advection step. Mathematically equivalent, but a different floating-point summation order — and that changed bit-level results for every moist run, not just Titan ones, since do_spec_tracer_filter defaults to .false..

Fix: branch on do_spec_tracer_filter. The new accumulate-then-filter path runs unchanged when the option is actually requested; everything else falls back to the original direct-increment formulation, byte-for-byte.

Fix 2 — idealized_moist_phys.F90: unconditional bucket-water spectral diffusion (commit 3cdb2e93)

Same underlying pattern, bigger effect. The bucket-water tendency block gained an unconditional call to a new diffuse_surf_water() subroutine, regardless of damping_coeff_bucket (which defaults to 0., i.e. "no damping requested"). The subroutine round-trips the tendency field through a spherical harmonic transform and back — and even at zero damping, that round-trip isn't a no-op: it truncates the field to the spectral resolution, silently low-pass-filtering real grid-scale structure (e.g. land/ocean boundaries in bucket_depth). That's a genuine physical change, not roundoff, which is why bucket_model's divergence was far larger than the others.

Fix: only call diffuse_surf_water() when damping_coeff_bucket > 0.

Fix 3 — two_stream_gray_rad.F90: seasonal frac-of-day precision loss (commit ea8e155b)

Different in kind from the first two: not an accidental regression. Commit 7f4201dd ("Important fix for two-stream") correctly fixed frac_of_day to account for total elapsed seconds rather than just the current day's remainder — needed for solar day lengths > 86400s (Mars, Titan), where the model's internal "days" bookkeeping (fixed 86400s calendar units) and the actual solar day length diverge. Master never needed this and still uses the original formulation. For the standard Earth day length, the new formulation still runs and still gets the right answer, but only after mod()-ing away an accumulated whole-days component — introducing a small but real precision loss relative to just using the bounded seconds-of-day value directly. That's what broke variable_co2_grey (which sets do_seasonal: True) against master.

Fix: branch on day_in_s == 86400.. Earth-standard day length gets the exact pre-7f4201dd formulation back, bit-for-bit; non-Earth day lengths keep 7f4201dd's correction unchanged, preserving correct Mars/Titan seasonal timing.

Socrates trip tests

Previously blocked locally by an unrelated environment issue: GFDL_SOC points to a nonexistent path, and scratch codebase clones (which don't inherit the untracked trunk symlink our real checkout already has set up correctly) picked up a dangling symlink to that broken location. No code change — just needed trunk repointed at the real local Socrates source per clone, same method already in use in this checkout. All three Socrates test cases (socrates_aquaplanet, socrates_aquaplanet_cloud, ape_aquaplanet) pass against master.

Known remaining issue (not fixed in this PR)

top_down_test shows a small, stable divergence from master (ucomp ~1.2e-7, vor ~1.1e-13) that is not caused by any of the above — confirmed by reverting all candidate files from the relevant merge and finding no change. Bisecting it further is blocked by an unrelated double free or corruption crash in old commits (likely in mixed_layer.F90's heat-capacity spinup code), which would need real memory debugging rather than a mechanical fix to work around. Given this test case sees little use, it's left as the one known open item rather than pursued further here.

Test plan

• frierson vs master: bit-identical
• bucket_model vs master: bit-identical
• realistic_continents_variable_qflux vs master: bit-identical
• variable_co2_grey vs master: bit-identical
• socrates_aquaplanet, socrates_aquaplanet_cloud, ape_aquaplanet vs master: bit-identical
• held_suarez, MiMA, realistic_continents_fixed_sst, variable_co2_rrtm, column_test vs master: unaffected, still bit-identical
• top_down_test vs master: still diverges (pre-existing, unrelated, documented above — not addressed in this PR)

[sit23]

The plan for this branch now is to attempt to track down the cause of the top-down test divergence bug. If this can be found, a fix will be applied, and then this branch can be merged.