Multi-Epoch Machine Learning 2 - Identifying physical drivers of galaxy properties in simulations

Aug 4, 2026

Upload by Himbo

Using a novel machine learning approach, the study analyzes how galaxy properties build up across different simulations and across varying environments within a single simulation. Feature importance from a trained model is used to track how stellar mass depends on other galaxy and halo properties at multiple times. Training on IllustrisTNG shows earlier star production in high-density regions than in low-density regions. The method is applied to Illustris, EAGLE, CAMELS, revealing the role of supernova feedback efficiency and a stronger impact from Simba black hole feedback. Principal component analysis identifies halo potential- and formation-time-related components, with wind speed emerging as a key subgrid parameter compared to energy per unit star formation.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages16

Reading Time40 min

AccessFree to download

Abstract

Introduction

Hydrodynamical cosmological simulations and sub-grid physics

Motivation: feedback and differing assembly histories

arXiv :2306 .07728v1 [ astro-ph .GA] 13 Jun 2023 Multi-Epoch Machine Learning 2: Identifying physical drivers of galaxy properties in simulations Robert J. McGibbon, 1 ★ Sadegh Khochfar 1 1 Institute for Astronomy, University of Edinburgh, Royal Observatory, Edinburgh EH9 3HJ Accepted XXX. Received YYY; in original form ZZZ ABSTRACT Using a novel machine learning method, we investigate the buildup of galaxy properties in different simulations, and in various environments within a single simulation. The aim of this work is to show the power of this approach at identifying the physical drivers of galaxy properties within simulations. We compare how the stellar mass is dependent on the value of other galaxy and halo properties at different points in time by examining the feature importance values of a machine learning model. By training the model on IllustrisTNG we show that stars are produced at earlier times in higher density regions of the universe than they are in low density regions. We also apply the technique to the Illustris, EAGLE, and CAMELS simulations. We find that stellar mass is built up in a similar way in EAGLE and IllustrisTNG, but significantly differently in the original Illustris, suggesting that subgrid model physics is more important than the choice of hydrodynamics method. These differences are driven by the efficiency of supernova feedback. Applying principal component analysis to the CAMELS simulations allows us to identify a component associated with the importance of a halo’s gravitational potential and another component representing the time at which galaxies form. We discover that the speed of galactic winds is a more critical subgrid parameter than the total energy per unit star formation. Finally we find that the Simba black hole feedback model has a larger effect on galaxy formation than the IllustrisTNG black hole feedback model. Key words: galaxies:evolution – galaxies:formation – galaxies:halo – methods:data analysis – hydrodynamics 1 INTRODUCTION Hydrodynamical cosmological simulations have become key tools for helping us to understand both cosmology and galaxy formation. These simulations include baryons alongside dark matter, and incorporate relevant physical processes such as gas cooling, star formation, and feedback. Prominent examples of cosmological simulations include Illustris (Vogelsberger et al. 2014a,b; Genel et al. 2014; Sĳacki et al. 2015), IllustrisTNG (Springel et al. 2018; Pillepich et al. 2018b; Naiman et al. 2018; Nelson et al. 2018; Marinacci et al. 2018), Simba (Davé et al. 2019), EAGLE (Schaye et al. 2015; McAlpine et al. 2016), HorizonAGN (Dubois et al. 2014), and FiBY (Johnson et al. 2013) . They have been successful at reproducing a large number of observables, such as the column density distribution of the Lyman-􀁕 forest (Hernquist et al. 1996), stellar mass and luminosity functions (Schaye et al. 2015; Pillepich et al. 2018b; Cullen et al. 2017), galaxy clustering (Springel et al. 2018), the galaxy color bimodality (Kavirajet al. 2017; Trayford et al. 2015; Nelson et al. 2018), and properties of the circumgalactic medium (Appleby et al. 2021) . For recent reviews of cosmological simulations see Somerville & Davé (2015) and Vogelsberger et al. (2020) . There are various methods for simulating gas, including particle based methods (e.g. Springel 2005), and grid based approaches, using both structured and unstructured meshes (e.g. Springel 2010; Morton et al. 2023), with some codes utilising adaptive mesh refine- ★ E-mail: [rob.mcgibbon@ed.ac.uk](rob.mcgibbon@ed.ac.uk) ment (e.g. Bryan et al. 2014) . Each comes with its own strengthsand weakness, and numerical effects from the implementations of the various methods can affect the hydrodynamics of the gas in different ways. More variation in simulations comes from the fact that many of the relevant physical processes that need to be modelled occur below the typical resolution limits of cosmological

FAQs

What is the main goal of Multi-Epoch Machine Learning 2 in this work?

To identify the physical drivers behind how galaxy properties build up in simulations by using a machine learning method and analyzing feature importance over time and environments.

How do the results from IllustrisTNG compare across environments?

Stellar mass forms earlier in higher-density regions of the universe than in lower-density regions when the model is trained on IllustrisTNG.

Which feedback processes are found to drive differences between simulations?

Differences are linked to the efficiency of supernova feedback, while for black hole feedback the Simba model shows a larger effect on galaxy formation than the IllustrisTNG black hole feedback model.

Aug 4, 2026

Upload by Himbo