GIZMO/STARFORGE Version History

This page summarizes physics-relevant changes between tagged GIZMO versions for STARFORGE simulations run with SINGLE_STAR_STARFORGE_DEFAULTS plus full feedback and cooling. Changes are grouped by topic; commit hashes are cited for traceability.

The three canonical tags are:

  • v1.1 — commit 8d56d9c3 (2021-12-16)

  • v1.2 — commit 30459741 (2023-07-22)

  • v1.3 — commit 168cdb3d

Pre-v1.1

Ionizing-photon transport bugfix (3ae730f1, 2021-12-14)

A one-line fix to the M1 face-flux gate in radiation/rt_diffusion_explicit.h:

- if((scalar_i>0)&&(scalar_j>0)&&...)
+ if((scalar_i+scalar_j>0)&&...)

The old code required both cells across a face to have non-zero photon energy, so the I-front could never propagate into previously-empty neutral gas. HII-region expansion in STARFORGE runs predating 2021-12-14 was effectively frozen. A companion commit (4dc827f5) floors Rad_E_gamma at MIN_REAL_NUMBER to prevent any code path from re-encountering the same pathology.

Self-consistent RT boundary condition (a40f791e, 2021-06-17)

Introduced rt_apply_boundary_conditions: cells within 10 % of the box edge have Rad_E_gamma[k] clamped to band-resolved ISRF values each step. Already present in v1.1; earlier runs saw M1 photons hitting the box edge as vacuum.

v1.1

GALSF_USE_SNE_ONELOOP_SCHEME forced on (8d56d9c3)

The v1.1 head commit itself forces the FIRE-2 one-loop mechanical-feedback coupling for any run with SINGLE_STAR_FB_WINDS or SINGLE_STAR_FB_SNE. Unchanged through v1.3 — all three versions use 1-loop coupling.

EOS_SUBSTELLAR_ISM on but buggy

The flag was a default in v1.1 but implemented the integrated γ using the differential γ formula from Vaidya 2015, producing unphysical behaviour in the molecular regime. Disabled in v1.2; correctly reimplemented in v1.3.

Reduced speed of light default: 1.0e-4

v1.1 → v1.2

Radiation / RT

Dust temperature and IR opacity refactor

Emission opacity (depends only on Tdust) separated from absorption opacity (depends on Trad and Tdust). New routines rt_eqm_dust_temp() and gas_dust_heating_coeff(). Improved low-T (<10 K) opacity extrapolation with a β=2 law (relevant for opacity-limit collapse).

Reduced speed of light raised from 1e-4 to 3e-4 (d2b68c2f)

3× increase. Affects I-front propagation speed, M1 timestep cost, and radiation-pressure-driven dynamics. Effective only if not overridden in Config.sh.

RT photon re-injection after accretion (2b9dbc9e, then made default in 56944088)

When a sink swallows a cell, the accreted Rad_E_gamma is stored and re-emitted on the next RT source step in the IR bin. Corrects a systematic photon loss in optically-thick accretion envelopes at low RSOL. Made a default for SINGLE_STAR_STARFORGE_DEFAULTS with RT_INFRARED in 56944088. Not present at all in v1.1.

Strömgren-radius criterion bugfix (24109dbc)

rt_sourceinjection_evaluate was comparing neighbor density instead of the star’s local density to decide whether to downgrade photons to IR. Shells surrounding HII regions were never ionized under the old code.

Cosmic-ray background and attenuation

Three staged commits build the current prescription:

  1. 1396c4bc — CR energy density scaled linearly with InterstellarRadiationFieldStrength (note: a typo RT_ISRF_BRACKGROUND silently broke the H₂ branch).

  2. 877d35dd — Switched to √ISRF scaling (ζCR ∝ ΣSFR0.5, Crocker+20/Krumholz+23); CR coupling consolidated into a single function.

  3. 87599421 — Column-density attenuation added: flat for NH < 10²¹ cm⁻², then NH-1 power law with exponential cutoff above ~100 g cm⁻².

Net effect: dense cloud cores get lower ζCR → higher molecular fraction and less CR heating compared with v1.1.

Sink particles and feedback

SINGLE_STAR_DIRECT_GRAVITY_RADIUS = 1000 AU as default (56944088)

All star–star gravity within 1000 AU is computed exactly. Changes binary/multiple dynamics.

Sink-formation criteria at high density (d1ce905e + 93ed5116)

  • Prevents spurious twin-sink formation within 0.1 AU of an existing sink.

  • Fixes an overflow in the optically-thick column-density expression above 10¹⁴ cm⁻³ that pushed cooling into wrong logic.

  • Fixed flag logic so the optically-thick virial limiter (v_fast → 0.2 km/s above n > 10¹³ cm⁻³) activates when RT_INFRARED is on. In v1.1 this branch was silently disabled for RT runs, making sink formation very slow in dense cores.

Spawning logic bug (4892f8ad)

Max_Unspawned_MassUnits_fromSink was compared in cell-count units against a mass in mass units, giving a wrong trigger threshold for jet/wind spawning.

Equal-mass sink merging fix (51c94489)

Strict inequality prevented equal-mass sinks from merging; edge case for symmetric ICs or twin binaries.

Massive-star timestep before SN (517c40fa)

Ensures massive stars do not skip gas-interaction steps immediately before going SN, preventing unphysical single-step dump of accumulated ejecta mass.

BH_Mass bookkeeping in SINGLE_STAR_FB_WINDS (d8d122ad)

In the local area-weighted coupling path, BH_Mass was not decremented alongside Mass when wind mass was lost, causing luminosity/wind-rate calculations to drift.

Spawned-wind direction seeding bug (e573db86)

The angular-grid spawn mode was keying off ID_child_number vs. ID_generation with a swapped convention from an earlier commit, misaligning the basis-rotation cadence for jet spawning.

EOS_SUBSTELLAR_ISM disabled (b53e7e1c, 2022-10-27)

Deliberately removed from STARFORGE defaults because the v1.1 implementation used the wrong γ formula. v1.2 falls back to ideal-gas γ = 5/3 in the molecular regime.

v1.2 → v1.3

Thermodynamics

EOS_SUBSTELLAR_ISM correctly reinstated

Now uses the self-consistent molecular+atomic EOS with full T-dependent γ(T,ρ). H₂ physics moved to hydrogen_molecule.c.

EOS_PRECOMPUTE infrastructure

New set_eos routine caches T, γ, P; get_pressure is now a getter. convert_u_to_temp rewritten with a Newton iteration using the analytic heat capacity, plus secant fallback.

Bracketed root-finder (bracketed_rootfind.h)

Applied to the fH₂ solver, the dust-temperature solver, and the top-level cooling iteration. Faster convergence and better diagnostics on failure.

Dust solver improvements

Gas–dust–radiation coupling improved for the optically-thick core regime. Complete dust sublimation now assigns MAX_DUST_TEMP (10⁴ K) and zeroes dust abundances/processes. CO cooling now uses a Whitworth-2018 LVG optically-thick limiter.

Stellar feedback

Wind mass-loss rates use main-sequence luminosity only (b07a4b5c + related)

Wind Ṁ is now driven by the main-sequence luminosity, not total (MS + accretion) luminosity. The model fits do not include accretion luminosity; actively accreting massive protostars have weaker winds in v1.3 than in v1.2.

Cosmic rays

Radioactive-decay floor on ζCR (8e36c561)

Adds += 1e-21 · (Z/Z☉) to Get_CosmicRayIonizationRate_cgs, providing a metallicity-scaled minimum ζCR in deeply shielded gas where column attenuation otherwise drives it to zero. Applies to all METALS runs (broader than the commit message implies).

CR energy density returned 0 for particle index 0 (7fee2822)

Off-by-one: if(i<=0) → if(i<0). The first particle on each MPI rank always got uCR = 0 in v1.2, causing stochastic single-cell anomalies in H₂ chemistry and CR heating after every domain decomposition.

Other

z_background / Redshift_RT_Background parameterfile entry

Runtime control of the redshift used to compute the background UV/IR in RT_ISRF_BACKGROUND.