Using Unsupervised Machine Learning to Find the Milky Way's Components

Abstract

Understanding the distributions of stars and their metallicities tells us about the formation of the Milky Way and how it has interacted with satellite galaxies in the past. Three distinct sectors lie in the Milky Way—the thin disk, thick disk, and halo. Because there exists significant overlap in various parameter spaces such as velocity and metallicity, it has been difficult to disentangle those components in the past aside from using empirical methods. In this study, unsupervised machine learning techniques were applied to a Gaia + APOGEE dataset and used to identify the components of the Milky Way. The resulting model was compared to prior methods, highlighting the possible difficulties resulting from applying rigid cuts. Regression was used to analyze the trend in metallicity ratios, which suggests that the rate of supernovae in the Milky Way has changed in its history. The Initial Mass Function and the percentage of halo stars generated from the model were used to approximate the number of neutron stars in the Milky Way. Overall, this study shows that unsupervised machine learning techniques enable the discovery of new trends in the Milky Way’s components.