Understanding the regulation of star formation within TNG100 galaxies on kpc-scales using machine learning I: Global versus local Journal: Astronomy & \& Computing

Bryanne McDonough, Sathvika S. Iyengar, Ansa Brew-Smith, Asa F. L. Bluck, Joanna M. Piotrowska · 2026

Read Understanding the regulation of star formation within TNG100 galaxies on kpc-scales using machine learning I: Global versus local Journal: Astronomy & \& Computing by Bryanne McDonough, Sathvika S. Iyengar, Ansa Brew-Smith, Asa F. L. Bluck, Joanna M. Piotrowska (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-15182.txt.

Introduction

McDonough and colleagues use machine learning to ask what drives star formation and quenching in TNG100 galaxies. From about 63000 annular bins across 6189 simulated galaxies, Random Forest and XGBoost models predict quenching states and star-formation rate densities. Both methods rank features consistently, with XGBoost spreading importance more evenly across correlated features. For central and massive satellite galaxies, black hole mass dominates quenching, pointing to AGN feedback, while halo mass rules low-mass satellites.

Catalog record

CREATORBryanne McDonough, Sathvika S. Iyengar, Ansa Brew-Smith, Asa F. L. Bluck, Joanna M. Piotrowska
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE64951
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.15182

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Understanding the regulation of star formation within TNG100 galaxies on kpc-scales using machine learning I: Global versus local Journal: Astronomy & \& Computing Bryanne McDonough, Sathvika S. Iyengar, Ansa Brew-Smith, Asa F. L. Bluck, Joanna M. Piotrowska Abstract We apply Random Forest and XGBoost machine learning algorithms to determine which galaxy properties most effectively predict star formation and quenching in simulated galaxies. Using spatially-resolved data from approximately 63,000 annular bins across 6,189 6,189 TNG100 galaxies, we train classification models to predict quenching states and regression models to predict star formation rate surface densities. Despite their different algorithmic approaches, both methods produce consistent feature importance rankings, with XGBoost distributing importance more evenly among correlated features. For central galaxies and high-mass satellites, black hole mass dominates quenching predictions, consistent with quenching via active galactic nuclei (AGN) feedback. Classification of low-mass satellites shows overwhelming importance for halo mass, indicating environmental quenching. Star formation predictions are dominated by local stellar mass surface density across all star-forming galaxy types, confirming that active star formation is a local process while quenching is driven by global properties. 1 Introduction There exists a bimodality in the star formation properties of galaxies at low-redshift, with most galaxies either forming stars at a rate expected for their mass or lacking significant levels of star formation ( 74 ; 14 ; 1 ; 59 , e.g.) . Galaxies in the latter category are often referred to as quenched or quiescent galaxies. The quenched fraction has been shown to depend on various intrinsic and environmental properties associated with galaxies, especially mass, environment, and morphology ( 1 ; 23 ; 45 ; 15 ; 59 ; 60 ; 8 ; 57 ; 39 , e.g.,) . Galaxies will continue to form stars as long as they have a supply of cool, dense molecular gas that can collapse into protostellar objects. Molecular gas in the interstellar medium (ISM) is constantly depleted as it is formed into stars. In high-mass star-forming galaxies, the ISM can be continually replenished via accretion from the circumgalactic medium (CGM). The ISM of low-mass galaxies can be replenished via streams of cold gas accreted from the intergalactic medium ( 22 ; 72 ; 21 ; 32 ; 33 ; 43 , e.g.,) . There are several avenues through which galaxies may lose access to their reservoirs of star formation fuel. Winds from supernovae and/or active galactic nuclei (AGN) can eject gas from the ISM and into the CGM ( 71 ; 52 ; 78 , e.g.,) . Accretion of fresh gas onto the ISM from the CGM can be prevented via heating of the circumgalactic gas or removal of the CGM via stripping and/or tidal interactions ( 25 ; 20 ; 67 ; 80 ; 77 ; 69 , e.g.,) . More recent studies have also connected turbulence in the ISM to quenching ( 81 ; 83 ; 62 ; 63 , e.g.,) . Alternative quenching pathways proposed in the literature include halo mass quenching through virial shock heating of infalling gas ( 2 ; 84 , e.g.,) , supernovae feedback ( 17 ; 27 , e.g.,) , morpholgoical quenching via dynamical stabilization ( 45 ; 40 , e.g.,) , and environmental quenching ( 60 ; 12 ; 75 ; 26 ; 70 ; 12 ; 48 ; 30 , e.g.,) . Whether, and to what extent, these processes operate on galaxies and ultimately drive the cessation of star formation is still under debate. Several studies have indicated different quenching pathways for different galaxy populations. In particular, different pathways to quenching have been identified for central galaxies and low-mass satellite galaxies ( 60 ; 4 ; 29 ; 30 , e.g.,) . Satellite galaxies are more susceptible to environmental processes—such as galaxy-galaxy tidal interactions and ram pressure stripping—as they orbit within the halos of more massive centrals. While it has become clear that a variety of galaxy parameters are correlated with quenching, the relative importance of these parameters in the quenching process is still unclear. For example, the higher quenched fractions of satellite galaxies in massive clusters indicates that the quenching of these systems is likely driven by environmental interactions. Potential drivers of environmental quenching include tidal stripping from galaxy-galaxy interactions ( 38 ; 82 , e.g.,) and ram pressure stripping from galaxy-halo interactions ( 31 ; 36 ; 35 ; 12 , e.g.,) . Which processes dominate the quenching of low-mass satellite galaxies in massive clusters remains an open question. The primary challenge in understanding which features of a galaxy drive quenching is that many of these features are intercorrellated. For example, the mass of the supermassive black hole at the center of a galaxy is correlated with the stellar mass of that galaxy, which is in turn correlated with the mass of the gaseous halo surrounding the galaxy. This complication has been addressed in prior works ( 7 ; 4 ; 62 ; 63 ; 29 ; 30 , e.g.,) with a machine learning tool known as Random Forest ( 13 ; 42 ; 58 , RF; ) . Random Forest classification utilizes an ensemble of decision trees to determine the relative importance of various ‘features’ (i.e., properties) of galaxies for predicting whether a galaxy is quenched or star forming. Using a Random Forest classifier on spatially-resolved observational data, 7 found that quenching of central galaxies was best predicted by global (i.e., integrated) features. In a similar analysis, 4 found that quenching of regions from observed central galaxies and high-mass satellite galaxies was best predicted by the mass of the central supermassive black holes (as inferred by central velocity dispersion) while the influence of environmental parameters, particularly the local galaxy overdensity, was much more pronounced for the quenching of low-mass satellites. 29 found nearly identical results in an analysis of integrated properties of SDSS galaxies. However, they found that halo mass, rather than overdensity, was a better predictor of quenching for simulated low-mass satellite galaxies. Accurately modeling the quenching of galaxies in hydrodynamic simulations of galaxy evolution has historically been a challenge. However, with careful implementation of feedback models, modern simulations are now reproducing observationally-realistic galaxy populations. Of concern is how well these feedback models actually capture real physical processes. Comparing the relative importance of various features for quenching in both observed and simulated galaxies offers another way to analyze the accuracy of feedback models. While 29 compared Random Forest analyses of observed galaxies from SDSS and simulated galaxies from Illustris ( 54 ; 79 ) , TNG100 ( 55 ; 61 ; 73 ; 53 ; 44 ; 56 ) , and EAGLE ( 19 ; 66 ) , this work looked only at properties on a global (i.e., integrated) level. However, quenching does not happen simultaneously at all points within a galaxy, instead some regions can be quenched while others are still actively star-forming. Thus, it is of interest to analyze the importance of various features at a spatially-resolved scale with machine learning methods that allow us to effectively control for spurious inter-correlations. 7 and 4 performed such an analysis on observed galaxies from MaNGA, but a similar analysis has not been performed on spatially-resolved data from simulations. In this work, we perform a machine learning

Read the complete text