Estimation of the masses in the Local Group by Gradient Boosted Decision Trees

TL;DR

This paper employs gradient boosted decision trees trained on simulated data to estimate the masses of the Local Group and its primary galaxies, providing new mass estimates with quantified uncertainties.

Contribution

It introduces a machine learning approach using GBDT trained on dark matter simulations to infer galaxy group masses from observational data.

Findings

Total mass of the Local Group: 3.31 (+0.79/-0.67) x 10^12 M_sun

Mass of the Milky Way: 1.15 (+0.25/-0.22) x 10^12 M_sun

Mass of M31: 2.01 (+0.65/-0.39) x 10^12 M_sun

Abstract

Our goal is to estimate the mass of the Local Group (LG) and the individual masses of its primary galaxies, the M31 and the Milky Way (MW). We do this by means of a supervised machine learning algorithm, the gradient boosted decision trees (GBDT) and using the observed distance and relative velocity of the two as input parameters. The GBDT is applied to a sample of 2148 mock LGs drawn from a set of 5 dark matter (DM)-only simulations, ran within the standard $\Lambda$CDM\ cosmological model. The selection of the mock LGs is guided by a LG model, which defines such objects. The role of the observational uncertainties of the input parameters is gauged by applying the model to an ensemble of mock LGs pairs whose observables are these input parameters perturbed by their corresponding observational errors. Finally the observational data of the actual LG is used to infer its relevant masses. Our main results are the sum and the individual masses of the MW and M31: $M_{tot} = 3.31 ^{+0.79}_{-0.67} $, $M_{MW}=1.15^{+0.25}_{-0.22}$ and $M_{M31}=2.01^{+0.65}_{-0.39} \ \ \times 10^{12}M_{\odot}$ (corresponding to the median and the 1st and 3rd quartiles). The ratio of the masses is $M_{M31}/M_{MW}=1.75^{+0.54}_{-0.28}$, where by convention the M31 is defined here to be the more massive of the two halos.