SAGE26 Parameter Reference

This document is the canonical reference for every parameter accepted by SAGE26 parameter files (input/*.par). Parameter files are parsed by src/core_read_parameter_file.c.

Syntax: ParameterName  value  % optional comment

Lines beginning with % are comments. Required parameters must be present; optional parameters take the listed default if omitted.

---

Output

Parameter	Type	Required	Default	Description
FileNameGalaxies	string	yes	—	Base name for output files (e.g. model → model_0.hdf5).
OutputDir	string	yes	—	Directory for galaxy output. Created if absent.
OutputFormat	string	no	sage_hdf5	sage_hdf5, sage_binary, or lhalo_binary_output. The last converts any supported input tree format to lhalo-binary and writes no galaxy catalogue.
NumOutputs	int	no	-1	Number of snapshot outputs; -1 = all snapshots.
SaveFullSFH	0/1	no	0	Store per-snapshot SFR history arrays (SFHMassDisk, SFHMassBulge).

---

Simulation

Parameter	Type	Required	Default	Description
TreeType	string	yes	—	Merger tree format: lhalo_binary, lhalo_hdf5, consistent_trees_ascii, consistent_trees_hdf5, genesis_hdf5, gadget4_hdf5.
TreeName	string	yes	—	Tree file basename (files are named TreeName.N).
SimulationDir	string	yes	—	Directory containing tree files.
FileWithSnapList	string	yes	—	File listing snapshot scale factors, one per line.
FirstFile	int	yes	—	First tree file index to process.
LastFile	int	yes	—	Last tree file index to process (inclusive).
NumSimulationTreeFiles	int	yes	—	Total number of tree files (may differ from FirstFile–LastFile range).
LastSnapshotNr	int	yes	—	Index of the final snapshot in the tree files.
Omega	double	yes	—	Matter density parameter Ω_m.
OmegaLambda	double	yes	—	Dark energy density parameter Ω_Λ.
BaryonFrac	double	yes	—	Universal baryon fraction f_b = Ω_b / Ω_m.
Hubble_h	double	yes	—	Dimensionless Hubble parameter h (H₀ = 100 h km/s/Mpc).
PartMass	double	yes	—	N-body particle mass in 10¹⁰ M_sun/h.
BoxSize	double	yes	—	Simulation box side length in Mpc/h.

---

Units

Parameter	Type	Required	Default	Description
UnitLength_in_cm	double	yes	—	1 internal length unit in cm. Typically 3.08568e+24 (= Mpc/h).
UnitMass_in_g	double	yes	—	1 internal mass unit in g. Typically 1.989e+43 (= 10¹⁰ M_sun).
UnitVelocity_in_cm_per_s	double	yes	—	1 internal velocity unit in cm/s. Typically 100000 (= km/s).

---

Physics switches

Parameter	Type	No	Default	Values and meaning
SFprescription	int	no	1	Star formation prescription: 0=Croton+06; 1=Blitz & Rosolowsky 06 H₂; 2=Somerville+25 SFR; 3=Somerville+25 SFR+H₂; 4=Krumholz & Dekel 12; 5=KMT 09; 6=Krumholz 13; 7=Gnedin & Draine 14.
AGNrecipeOn	int	no	2	AGN feedback: 0=off; 1=empirical; 2=Bondi-Hoyle; 3=cold cloud accretion.
SupernovaRecipeOn	0/1	no	1	SN feedback: 0=off; 1=Croton+16 reheating/ejection.
ReionizationOn	0/1	no	1	Reionization suppression of infall: 0=off; 1=Kravtsov+04 analytic fit.
DiskInstabilityOn	0/1	no	1	Disk instability: 0=off; 1=Toomre criterion drives bulge and BH growth.
CGMrecipeOn	0/1	no	1	Two-regime CGM model: 0=off (classical C16 cooling only, including its rapid cold-accretion branch at r_cool > R_vir); 1=on.
KarpovModeOn	0/1	no	0	Metallicity of SN-reheated and ejected gas: 0=the full Karpov+23 recipe; 1=a low-metallicity floor at Z/Z_sun = 0.01. Only acts when the reheating/ejection path computes a metallicity for the outflow.
FIREmodeOn	0/1	no	1	FIRE stellar feedback: 0=off; 1=on.
EnhancedStarFormationOn	int	no	1	Feedback-free burst (FFB) galaxies: 0=off; 1=Li+24 mass threshold with their eq. 3 sigmoid, so a halo is FFB with probability f_ffb(Mvir, z); 2=Boylan-Kolchin (2025) acceleration threshold g_max > g_crit, with log-normal scatter applied to the halo concentration. Both apply to centrals regardless of the halo’s CGM/hot regime, and both take a fresh draw each snapshot from a hash of the halo’s MostBoundID and the snapshot, so classifications are reproducible run to run. See Star formation and feedback.
ConcentrationOn	int	no	3	Halo concentration method: 0=off; 1=Ishiyama+21 table; 2=V_max/V_vir; 3=V_max/V_vir with infall freeze for satellites.
BulgeSizeOn	int	no	3	Bulge radius model: 0=off; 1=Shen+2003 eq.33; 2=Shen+2003 eq.32; 3=Tonini+2016 (separate merger and instability channels, mass-weighted average).
StarburstColdGasOn	0/1	no	1	Include cold gas contribution during merger starbursts.
FixingGas	0/1	no	1	Cold-gas fixes (see investigation/gas_metals/README.md). 1: every H2 prescription (SFprescription 1, 3-7) forms stars at SFR = M_mol / t_dep (M_mol = H2gas / 0.74, helium included) instead of its efficiency x H2 / t_dyn law; nothing else changes. t_dep is set by H2DepletionTimeModel.
H2DepletionTimeModel	0-3	no	2	FixingGas 1 only. 0: t_dep = 1 Gyr (Lagos+18 Table 3 default nu_SF = 1 Gyr^-1). 1: Krumholz, McKee & Tumlinson (2009) eq. 10, t_dep = 2.6 Gyr (Sigma_gas/85 Msun pc^-2)^(+-0.33). 2: Krumholz (2013) eqs. 15-16, t_dep = t_ff/eps_ff, t_ff = pi^(1/4)/sqrt(8) sigma_g/(G (Sigma_GMC^3 Sigma_gas)^(1/4)), sigma_g = 8 km/s, Sigma_GMC = 85 Msun pc^-2, eps_ff = 0.01. 3: Krumholz, Dekel & McKee (2012) eqs. 2, 4, 8, 9, t_dep = min(t_ff,GMC, t_ff,T)/eps_ff with t_ff,GMC as in 2 and t_ff,T = sqrt(3 pi^2 Q^2 / (32 (beta+1) phi_P)) / Omega, Q = 1, beta = 0, phi_P = 3, Omega = Vmax/r. Models 1-3 are evaluated per annulus and need H2RadialIntegrationOn = 1.

---

FFB parameters

Parameter	Type	Required	Default	Description
EnhancedSFEfficiency	double	no	0.2	Maximum star formation efficiency during FFB bursts. 0.2 matches observations; 1.0 is the theoretical maximum.

---

H₂ star formation parameters

Parameter	Type	Required	Default	Description
H2DiskAreaOption	int	no	1	Disk area for H₂ surface density: 0=π r_s²; 1=π (3 r_s)²; 2=2π r_s² (central Σ₀).
H2RadialIntegrationOn	0/1	no	1	Use radial ring integration for H₂ fraction (more accurate, slower).
H2RadialNBins	int	no	25	Number of radial bins for the ring integration.
H2RadialRMaxFactor	double	no	5.0	Outer integration radius as a multiple of the disk scale radius.

---

Model parameters

Star formation

Parameter	Units	Default	Description
SfrEfficiency	dimensionless	0.05	Cold/H2 gas consumption efficiency per dynamical time. Used by SFprescription 0, 1, 4, 5, 7 unconditionally, and by 6 (K13) only in the single-slab path (H2RadialIntegrationOn=0). Unused by 2 and 3 (Somerville+25 use their own density-modulated epsilon_cl) and by 6 in the radial path (uses the K13 local depletion time natively).
RecycleFraction	dimensionless	0.43	Fraction of stellar mass instantaneously recycled to cold gas.
Yield	dimensionless	0.025	Fraction of stellar mass returned as metals.
FracZleaveDisk	dimensionless	0.0	Fraction of newly produced metals transferred directly to hot gas.

Supernova feedback

Parameter	Units	Default	Description
FeedbackReheatingEpsilon	dimensionless	2.9	Mass of cold gas reheated per unit of stellar mass formed (Martin 1999).
FeedbackEjectionEfficiency	dimensionless	0.3	Fraction of SN energy deposited into hot gas for ejection.
EnergySN	erg	1.0e51	Energy per supernova event.
EtaSN	M_sun⁻¹	5.0e-3	Number of supernovae per solar mass of stars formed.

AGN feedback

Parameter	Units	Default	Description
RadioModeEfficiency	dimensionless	0.08	AGN radio-mode heating efficiency (AGNrecipeOn=2).
QuasarModeEfficiency	dimensionless	0.005	AGN quasar-mode wind heating efficiency (AGNrecipeOn > 0).
BlackHoleGrowthRate	dimensionless	0.015	Fraction of cold gas accreted onto the BH during mergers (AGNrecipeOn > 0).

Mergers

Parameter	Units	Default	Description
ThreshMajorMerger	dimensionless	0.3	Mass ratio above which a merger is classified as major.
ThresholdSatDisruption	dimensionless	1.0	M_vir-to-baryonic mass ratio below which a satellite is disrupted rather than merged.
MergerTimeFactor	dimensionless	0.75 (published value 2.0)	Scales the Binney & Tremaine (1987) dynamical-friction merger timescale. It also sets how accreted stellar mass splits between the intracluster component and the BCG: when a satellite’s subhalo is lost from the tree, its stars go to the ICS if the clock is still running (MergTime > 0) and onto the central if it has expired.

Gas cycling

Parameter	Units	Default	Description
ReIncorporationFactor	dimensionless	0.19	Scales the reincorporation velocity threshold: Vcrit = 354.26 km/s * ReIncorporationFactor, so it is not a mass fraction despite the name. Reincorporation runs only where Vvir > Vcrit, at a rate (Vvir/Vcrit - 1) * EjectedMass / t_dyn. The 354.26 = V_SN/sqrt(2) follows from V_SN = 501 km/s (Paper I eq. 11, from eta_SN = 5e-3 Msun^-1 and E_SN = 1e51 erg). An earlier version used V_SN = 630 km/s, giving 445.48, and 0.15 was fitted against that; 0.19 restores the same effective threshold (445.48*0.15 = 66.8 vs 354.26*0.19 = 67.3 km/s) with the correct V_SN.

Cooling and cold streams

Parameter	Units	Default	Description
MShockMsun	Msun	6.0e11	Dekel & Birnboim (2006) virial-shock stability mass. Used twice: it sets the CGM/hot regime classification in determine_and_store_regime(), and it sets the cold-stream criterion in cooling_recipe_hot(). Both must see the same value, so change it here rather than in either site.

Reionization

Parameter	—	Default	Description
Reionization_z0	—	8.0	Characteristic redshift for reionization suppression (Kravtsov+04).
Reionization_zr	—	7.0	Width parameter for reionization suppression.

See the FFB parameters section above for EnhancedSFEfficiency.

---

Numerical time resolution

The per-snapshot physics loop (cooling, star formation, feedback) is integrated with an adaptive number of sub-timesteps: the count scales with deltaT / t_dyn, so a snapshot interval spanning several halo dynamical times is resolved with more substeps (bounded by a STEPS floor and a MAX_STEPS cap). Tying the effective resolution to t_dyn rather than the raw snapshot cadence is what lets one calibration transfer across simulations with different output spacing.

Parameter	Type	Default	Description
SubstepResolution	double	1.0	Runtime multiplier on the adaptive-substep floor and cap. Calibration-locked numerical knob, not a physics choice — the model is calibrated at 1.0; do not change it for science runs without recalibrating. Coarse steps over-cool (cooling outruns the AGN r_heat response before it can react), so raising the resolution lowers the massive-end SMF and total stellar mass. The shift from 1.0 to fully converged is only ~0.1 dex at the massive end (within typical observational SMF scatter), but runtime grows ~linearly with the substep count. Use higher values only for deliberate convergence / resolution studies. Note that the reported SfrDisk/SfrBulge are normalised by the substep count actually integrated (SubstepsUsed), not by STEPS; before September 2026 they were divided by STEPS, which scaled the reported SFR by effective_steps/STEPS and made this knob useless for convergence testing. The CGM/hot and FFB draws are keyed on the halo and snapshot, so changing the substep count does not change which haloes are classified as hot or FFB.

Convergence note. The substep dependence is a long-standing property of SAGE’s cooling/feedback operator-splitting (it is present, and slightly stronger, with the classic CGMrecipeOn=0 cooling), not something introduced by the two-regime CGM physics. Because the effect is only ~0.1 dex on the SMF, the calibrated 1.0 model sits within observational constraints; a converged model would need at most a light retune and would show more simulation-consistent behaviour, at higher compute cost.

---

MPI forest distribution

Parameter	Type	Required	Default	Description
ForestDistributionScheme	string	no	generic_power_in_nhalos	How forests are distributed over MPI tasks: uniform_in_forests, linear_in_nhalos, quadratic_in_nhalos, exponent_in_nhalos, generic_power_in_nhalos.
ExponentForestDistributionScheme	double	no	0.7	Exponent for exponent_in_nhalos or generic_power_in_nhalos schemes.

---

Removed parameters

SAGE26 was developed with a wide set of switches so alternative prescriptions could be compared. For release, the prescriptions the papers use are the only ones the code carries, and their parameters are gone.

An unrecognised tag is a startup error, not a warning. A parameter file written against an earlier version will abort with Tag 'X' not allowed or multiply defined until the lines below are deleted.

Values that are still meaningful physical constants now live as named constants in the source, cited at their definition. They are listed here so a reader who goes looking for a knob can find where it went.

Removed parameter	Fate
ColdStreamCeilingOn	Dekel & Birnboim (2006) eq. 39 (the former mode 1) is the only cold-stream prescription.
StreamMassFactor	DB06_STREAM_MASS_FACTOR = 3.0 in src/model_cooling_heating.c.
StreamThresholdWidthDex	STREAM_THRESHOLD_WIDTH_DEX = 0.15 in src/model_cooling_heating.c.
StreamZCritWidth	Removed with the smoothed-gate modes.
TrackICSAssembly	Always on: ICS_disrupt, ICS_accrete and ICS_sum_mt are always accumulated.
GasDiskRadiusFactor	GAS_DISK_RADIUS_FACTOR = 1.0 in src/model_starformation_and_feedback.c.
FeedbackFreeModeOn	Renamed EnhancedStarFormationOn, with modes reduced to 0/1/2. The old mode 4 (BK25 with log-normal concentration scatter) is now mode 2; old modes 2, 3, 5, 6, 7 and 8-11 are gone.
FFBMaxEfficiency	Renamed EnhancedSFEfficiency. Still tunable.
FFBConcSigma	FFB_CONC_SIGMA = 0.2 in src/model_regimes.c.
FFBThresholdSlope	FFB_THRESHOLD_SLOPE = -6.2 in src/model_regimes.c.
FFBIgnoreRegime	Always on: the FFB criteria apply whatever the halo’s CGM/hot regime.
FFBRandomMode	Always a fresh draw each snapshot.
RegimeRandomMode	Always a fresh draw each snapshot.
FFBFeedbackDelayMyr, FFBCloudClumping, FFBCloudClumpingDisk, FFBStreamRadiusFraction, FFBShellSoundSpeedKms, FFBToomreQ, FFBSigmaCritMsunPc2	Removed with the Dekel+23 shell and disc criteria (old FFB modes 8-11).
SNEnergyConservationOn	Always on: both the reheating and the ejection term are bounded by the supernova energy available.
MaxSNEnergyCoupling	MAX_SN_ENERGY_COUPLING = 2.0 in src/model_misc.h.
RedshiftPowerLawExponent	FIRE_REDSHIFT_EXPONENT = 1.25, defined in both src/model_starformation_and_feedback.c and src/model_mergers.c.

The same names are also gone from the Header/Runtime attributes of the HDF5 output, which records tunable parameters. Analysis scripts that read them should use the constants above instead.