SAGE26 — Semi-Analytic Galaxy Evolution

SAGE26 is a C99 semi-analytic code for modelling galaxy formation in a cosmological context. It is a major update to Croton et al. (2016), adding a two-regime CGM model, FIRE stellar feedback, feedback-free burst (FFB) galaxies, multiple H2-based star formation prescriptions, and extended bulge/ICS tracking.

SAGE reads N-body merger trees, evolves galaxies through cosmic time using semi-analytic prescriptions, and writes galaxy catalogues in HDF5 or binary format. It runs on any simulation whose trees are in a supported format and contain a minimum set of halo properties. Test trees for the Mini-Millennium Simulation are provided.

---

What is new in SAGE26

Feature	Parameter	Reference
Two-regime CGM model with self-regulating precipitation	CGMrecipeOn	Dekel & Birnboim (2006), Voit (2015)
FIRE stellar feedback	FIREmodeOn	Muratov et al. (2015)
Feedback-free burst galaxies	FeedbackFreeModeOn	Li et al. (2024), Boylan-Kolchin (2025)
7 H2 star formation prescriptions	SFprescription	BR06, KMT09, KD12, K13, GD14, S25
Separate merger/instability bulge tracking	BulgeSizeOn	Tonini et al. (2016)
ICS assembly tracking	TrackICSAssembly	—
Full star formation history arrays	SaveFullSFH	—
Concentration	ConcentrationOn	Ishiyama+21 lookup table, Vmax/Vvir, Vmax/Vvir with subhalo infall freeze
ConsistentTrees, Genesis, Gadget-4 tree readers	TreeType	—
HDF5 output + libsage.so for Python/PSO	OutputFormat	—

---

Install

Dependencies

Package	Required	Notes
C99 compiler (gcc or clang)	Yes
GSL	Yes	required for tests
HDF5	Optional	HDF5 tree reading and output
MPI	Optional	parallel execution

Build

git clone https://github.com/MBradley1985/SAGE26.git
cd SAGE26
make                   # serial build -- produces ./sage and libsage.so
make USE-MPI=yes       # MPI-parallel build (switches compiler to mpicc)
make USE-HDF5=yes      # enable HDF5 support
make MEM-CHECK=yes     # address/UB sanitizers for debugging (gcc only)
make clean             # remove all build artefacts

---

Quickstart

./first_run.sh                          # create output dirs, download Mini-Millennium trees
./sage input/millennium.par             # run the model (serial)
mpirun -np 4 ./sage input/millennium.par  # run in parallel
python plotting/allresults-local.py     # z=0 diagnostic plots
python plotting/allresults-history.py   # multi-redshift diagnostics

Full details on the parameter file format, output format, and all physics switches are in docs/parameters.md.

---

Physics options

Every switch below is parsed by src/core_read_parameter_file.c. Full descriptions, defaults, and the model-parameter (non-switch) list are in docs/parameters.md.

Star formation (SFprescription)

Value	Prescription
0	Croton et al. (2006) original
1	Blitz & Rosolowsky (2006) H2
2	Somerville et al. (2025) SFR
3	Somerville et al. (2025) SFR + H2
4	Krumholz & Dekel (2012)
5	Krumholz, McKee & Tumlinson (2009)
6	Krumholz (2013)
7	Gnedin & Draine (2014)

AGN feedback (AGNrecipeOn)

Value	Mode
0	Off
1	Empirical (Croton+2016 radio mode)
2	Bondi-Hoyle accretion
3	Cold-cloud accretion

Supernova feedback (SupernovaRecipeOn)

Value	Mode
0	Off
1	Croton+2016 reheating/ejection

Reionization (ReionizationOn)

Value	Mode
0	Off
1	Kravtsov+2004 analytic suppression of infall

Disk instability (DiskInstabilityOn)

Value	Mode
0	Off
1	Toomre criterion drives bulge growth, instability starbursts, BH growth

Two-regime CGM model (CGMrecipeOn)

Galaxies below the Dekel & Birnboim (2006) shock mass are placed in Regime 0 (CGM / precipitation regime); those above are in Regime 1 (hot-halo / classical cooling regime). Each regime uses a dedicated cooling recipe.

Parameter	Values	Effect
CGMrecipeOn	0/1	0=off (classical C16 cooling only); 1=on. Carr+2022 exact copy.

Adaptive time integration (SubstepResolution)

The snapshot interval is integrated with a substep count that scales with deltaT / t_dyn, so high-redshift snapshots spanning several dynamical times are resolved with more substeps (bounded by a STEPS floor and MAX_STEPS cap) rather than a fixed count.

Parameter	Values	Effect
SubstepResolution	double	Runtime multiplier on both the adaptive-substep floor and cap; default 1.0. Sweep for convergence / N-invariance testing without recompiling

FIRE stellar feedback (FIREmodeOn)

Value	Mode
0	Off
1	Muratov+2015 FIRE mass loading

Feedback-free burst galaxies (FeedbackFreeModeOn)

Value	Mode
0	Off
1	Li+2024 sigmoid
2	Boylan-Kolchin+2025 (Ishiyama+21 concentration)

Halo concentration (ConcentrationOn)

Value	Method
0	Off
1	Ishiyama+2021 lookup table
2	V_max / V_vir from the simulation
3	V_max / V_vir with subhalo infall freeze for satellites

H2 star formation (auxiliary switches)

Parameter	Values	Effect
H2DiskAreaOption	1–3	Disk area for H2 surface density: 1=π r_disk²; 2=π (3 r_disk)²; 3=2π r_disk²
H2RadialIntegrationOn	0/1	Radial ring integration for H2 fraction (slower, more accurate)
H2RadialNBins	int	Number of radial bins for ring integration

Mergers and ICS tracking

Parameter	Values	Effect
TrackICSAssembly	0/1	Record satellite disruption / accretion contributions to ICS

Output

Parameter	Values	Effect
OutputFormat	string	sage_hdf5 or sage_binary
SaveFullSFH	0/1	Track per-snapshot SFR history arrays
NumOutputs	int	Number of snapshot outputs; -1 = all snapshots

Supported tree formats (TreeType)

lhalo_binary, lhalo_hdf5, consistent_trees_ascii, consistent_trees_hdf5, genesis_lhalo_hdf5, gadget4_hdf5

---

Tests

cd tests && make test               # build and run all suites
cd tests && make test_conservation  # conservation tests only (fastest)
cd tests && make quick              # single fastest check
bash tests/run_integration_tests.sh # full integration test (slower)

The regression baseline checks that output is bit-identical across 5444 datasets:

bash tests/regression_baseline.sh

After any physics change, recapture the baseline with:

python3 tests/regression_baseline.py capture input/millennium.par

---

Parameter calibration (SAGE-PSO)

Automated parameter calibration via Particle Swarm Optimization is available as a separate package: SAGE-PSO. It drives libsage.so to evaluate parameter samples against observational constraints (stellar mass functions, star formation rates, etc.) and supports machine-learning emulators to accelerate optimization.

---

Citation

If you use SAGE26 in a publication, please cite:

@article{bradley2026a,
  title = {SAGE26 Paper I: Modelling the Baryon Cycle from Cosmic Dawn to the Present Day},
  author = {Bradley, Michael and Croton, Darren J. and Paun, Robert A. Mostoghiu and Chowdhury, Dhruba Dutta and Willingham, Jayde},
  year = 2026,
  journal = {The Astrophysical Journal Supplement Series},
  publisher = {(In prep.)}
}

and the original SAGE paper:

@article{croton2016sage,
  author  = {Croton, D.~J. and Stevens, A.~R.~H. and Tonini, C. and
             Garel, T. and Bernyk, M. and Bibiano, A. and Hodkinson, L. and
             Mutch, S.~J. and Poole, G.~B. and Shattow, G.~M.},
  title   = {Semi-Analytic Galaxy Evolution (SAGE): Model Calibration and Basic Results},
  journal = {ApJS},
  year    = {2016},
  volume  = {222},
  pages   = {22},
  doi     = {10.3847/0067-0049/222/2/22},
  eprint  = {1601.04709},
}

Original SAGE is also available on ascl.net.

---

Links

• Documentation
• Parameter reference
• Developer guide
• Changelog
• Contributing
• SAGE-PSO — automated parameter calibration

---

Authors and maintainers

• Michael Bradley (@MBradley1985) — mbradley@swin.edu.au
• Darren Croton (@darrencroton)

Questions and comments welcome via GitHub Issues or email.

---

License

MIT — see LICENSE.

---

Acknowledgement

Claude Code (Anthropic) was used during development for documentation and style-guide work, linting, code-error fixes, building the test suite, and some code restructuring. All model design, physics choices, and results remain the authors' own, and all such changes were reviewed and tested before inclusion.