A machine learning approach to infer the accreted stellar mass fractions of central galaxies in the TNG100 simulation

Abstract

We propose a random forest (RF) machine learning approach to determine the accreted stellar mass fractions (f acc ) of central galaxies, based on various dark matter halo and galaxy features. The RF is trained and tested using 2,710 galaxies with stellar mass log 10 M * /M > 10.16 from the TNG100 simulation. For galaxies with log 10 M * /M > 10.6, global features such as halo mass, size and stellar mass are more important in determining f acc , whereas for galaxies with log 10 M * /M 10.6, features related to merger histories have higher predictive power. Galaxy size is the most important when calculated in 3-dimensions, which becomes less important after accounting for observational effects. In contrast, the stellar age, galaxy colour and star formation rate carry very limited information about f acc . When an entire set of halo and galaxy features are used, the prediction is almost unbiased, with rootmean-square error (RMSE) of ∼0.068. If only using observable features, the RMSE increases to ∼0.104. Nevertheless, compared with the case when only stellar mass is used, the inclusion of other observable features does help to decrease the RMSE by ∼20%. Lastly, when using galaxy density, velocity and velocity dispersion profiles as features, which represent approximately the maximum amount of information one can extract from galaxy images and velocity maps, the prediction is only slightly improved. Hence, with observable features, the limiting precision of predicting f acc is ∼0.1, and the multi-component decomposition of galaxy images should have similar or larger uncertainties. If the central black hole mass and the spin parameter of galaxies can be accurately measured in future observations, the RMSE is promising to be further decreased by ∼20%.

Mining the information content of member galaxies in the halo mass modelling

“…The black dashed line shows the fraction of halos with more than 7 members as a function of halo mass. Man et al 2019;Petulante et al 2021;Shi et al 2021).…”

What to expect from dynamical modelling of cluster haloes II. Investigating dynamical state indicators with Random Forest

“…However, the default feature importance ranking may suffer from a so-called masking effect, when there are strong correlations amount these features (see e.g. Shi et al 2021). Thus we choose not to use the default importance ranking returned by RFR.…”

“…Albert et al 2008), estimates of cluster and group mass (e.g. Green et al 2019;Man et al 2019) and prediction of accreted stellar mass fractions (Shi et al 2021). It is capable of fitting a nonlinear model to a high dimensional dataset non-parametrically, while also providing an objective way to evaluate the goodness of fit so that different models can be compared for model or feature selection.…”

ERGO-ML I: Inferring the assembly histories of IllustrisTNG galaxies from integral observable properties via invertible neural networks

“…Bottrell et al 2019;Ferreira et al 2020;Bottrell et al 2021;Ćiprijanović et al 2021). Very recently, Shi et al (2021) have also used data from IllustrisTNG to determine, via a Random Forest, the accreted stellar mass fraction.…”