DARK SAGE: Next-generation semi-analytic galaxy evolution with multidimensional structure and minimal free parameters

Adam R.H. Stevens, Manodeep Sinha, Alexander Rohl, Mawson W. Sammons, Boryana Hadzhiyska, César Hernández-Aguayo, Lars Hernquist

Research output: Contribution to journalArticlepeer-review


After more than five years of development, we present a new version of Dark Sage, a semi-analytic model (SAM) of galaxy formation that breaks the mould for models of its kind. Included among the major changes is an overhauled treatment of stellar feedback that is derived from energy conservation, operates on local scales, affects gas gradually over time rather than instantaneously, and predicts a mass-loading factor for every galaxy. Building on the model's resolved angularmomentum structure of galaxies, we now consider the heating of stellar discs, delivering predictions for disc structure both radially and vertically. We add a further dimension to stellar discs by tracking the distribution of stellar ages in each annulus. Each annulus-age bin has its own velocity dispersion and metallicity evolved in the model. This allows Dark Sage to make structural predictions for galaxies that previously only hydrodynamic simulations could. We present the model as run on the merger trees of the highest-resolution gravity-only simulation of the MillenniumTNG suite. Despite its additional complexity relative to other SAMs, Dark Sage only has three free parameters, the least of any SAM, which we calibrate exclusively against the cosmic star formation history and the z=0 stellar and Hi mass functions using a particle-swarm optimisation method. The Dark Sage codebase, written in C and python, is publicly available at https://github.com/arhstevens/DarkSage.

Original languageEnglish
JournalPublications of the Astronomical Society of Australia
Publication statusAccepted/In press - 2024


Dive into the research topics of 'DARK SAGE: Next-generation semi-analytic galaxy evolution with multidimensional structure and minimal free parameters'. Together they form a unique fingerprint.

Cite this