Can We Discover Physical Models Using Machine Learning? A Case Study of Galaxy Sizes

Authors

Festa Buçinca-Çupallari, CUNY Graduate Center
Ariyeh Maller, CUNY New York City College of TechnologyFollow
Viviana Acquaviva, CUNY New York City College of TechnologyFollow
Austen Gabrielpillai, CUNY Graduate Center
Rachel S. Somerville, Center for Computational Astrophysics, Flatiron Institute

Document Type

Article

Publication Date

7-10-2025

Abstract

We explore the ability of machine learning methods to discover underlying equations of physics by searching for the equations governing galaxy size in a semianalytic model. This case study allows us to evaluate the process as we know the ground truth. We find that we fail to find an equation to predict galaxy size on the entire data set, but are successful when we separate out disk galaxies where we expect the physics driving galaxy size to be different than in bulge-dominated systems. We are also able to find an equation for bulge size, but not without adding an additional feature based on our knowledge of elliptical galaxy scaling relations.

Comments

This article was originally published in The Astrophysical Journal, available at https://doi.org/10.3847/1538-4357/addc75

This work is distributed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).