Machine learning and cosmological simulations – I. Semi-analytical models

Abstract

We present a new exploratory framework to model galaxy formation and evolution in a hierarchical universe by using machine learning (ML). Our motivations are twofold: (1) presenting a new, promising technique to study galaxy formation, and (2) quantitatively analyzing the extent of the influence of dark matter halo properties on galaxies in the backdrop of semi-analytical models (SAMs). We use the influential Millennium Simulation and the corresponding Munich SAM to train and test various sophisticated machine learning algorithms (k-Nearest Neighbors, decision trees, random forests and extremely randomized trees). By using only essential dark matter halo physical properties for haloes of M > 10 12 M and a partial merger tree, our model predicts the hot gas mass, cold gas mass, bulge mass, total stellar mass, black hole mass and cooling radius at z = 0 for each central galaxy in a dark matter halo for the Millennium run. Our results provide a unique and powerful phenomenological framework to explore the galaxy-halo connection that is built upon SAMs and demonstrably place ML as a promising and a computationally efficient tool to study small-scale structure formation.

Approximations to galaxy star formation rate histories: properties and uses of two examples

“…All three subsamples have approximately 33000 galaxies, and full and main histories are studied for both. A fourth sample was taken for comparison with machine learning work by Kamdar, Turk & Brunner (2016a), and includes all galaxies with final time halo mass above 10 12 M which are central at both the last and second to last time step (17% were satellites at some point in their histories). There are 386919 galaxies in this sample, so only the main star formation rate history was considered.…”

“…Although main halo histories M h (t) are a key part of the Kamdar, Turk & Brunner (2016a) training set, is it also interesting to understand how well fewer or other inputs recover parameters. This helps to clarify which inputs contain the most predictive power.…”

“…The (fewer than 10% of the) galaxies in the training set are included in the plots and the correlations. The best results came from using the whole (main) halo history, closest to the galaxy information used by Kamdar, Turk & Brunner (2016a). However, many of the variants starting with smaller 26 Changing the random subsample scattered the results around by 3% or so, except for the SF R peak predictions, which sometimes would fluctuate to a very small number, e.g.,∼24%, presumably due to outliers.…”

Inferring physical properties of galaxies from their emission-line spectra

“…For a review of the applications of ML to astrophysical problems we refer the reader to Ball & Brunner (2010) and Ivezić et al (2014). For other practical applications of these methods see also Ball et al (2008); Hoyle et al (2015a); Zitlau et al (2016); Hoyle et al (2015b); Bellinger et al (2016) and Kamdar et al (2016) for a ML framework applied to cosmological simulations and Jensen et al (2016) for a ML approach to measure the escape fraction from galaxies in the EoR (Epoch of Reionization).…”

Multiwavelength cluster mass estimates and machine learning

“…DecisionTreeRegressor is also used inKamdar, Turk & Brunner (2016a)). It fits to a single tree rather than an ensemble.…”