A machine learning approach to mapping baryons on to dark matter haloes using the eagle and C-EAGLE simulations

Abstract

High-resolution cosmological hydrodynamic simulations are currently limited to relatively small volumes due to their computational expense. However, much larger volumes are required to probe rare, overdense environments, and measure clustering statistics of the large scale structure. Typically, zoom simulations of individual regions are used to study rare environments, and semi-analytic models and halo occupation models applied to dark matter only (DMO) simulations are used to study the Universe in the large-volume regime. We propose a new approach, using a machine learning framework to explore the halo-galaxy relationship in the periodic eagle simulations, and zoom C-EAGLE simulations of galaxy clusters. We train a tree based machine learning method to predict the baryonic properties of galaxies based on their host dark matter halo properties. The trained model successfully reproduces a number of key distribution functions for an infinitesimal fraction of the computational cost of a full hydrodynamic simulation. By training on both periodic simulations as well as zooms of overdense environments, we learn the bias of galaxy evolution in differing environments. This allows us to apply the trained model to a larger DMO volume than would be possible if we only trained on a periodic simulation. We demonstrate this application using the (800 Mpc)3 P-Millennium simulation, and present predictions for key baryonic distribution functions and clustering statistics from the eagle model in this large volume.

Mimicking the halo-galaxy connection using machine learning

“…Overall, our findings are consistent with previous results in the literature, with stellar mass being the most accurately predicted property, with a Pearson correlation coefficient of ∼ 0.98 (previously reported values are typically 0.92 − 0.957, see Kamdar et al 2016;Lovell et al 2022). The second best-predicted property is sSFR, with a correlation coefficient of ∼ 0.8 (previously, 0.745 − 0.794, see Kamdar et al 2016;Agarwal et al 2018;Lovell et al 2022). For size and colour we obtain coefficients in the range 0.7 − 0.8 and 0.59 − 0.71, respectively.…”

“…The main goal in this field is to establish relations between the properties of galaxies and the properties of their hosting haloes, in the cosmological context of the LSS of the Universe. This problem can be treated in ML in terms of an input dataset (halo properties), which is known a priori, and an output dataset, corresponding to the galaxy properties that we attempt to predict (Kamdar et al 2016;Agarwal et al 2018;Calderon & Berlind 2019;Jo & Kim 2019;Man et al 2019;Kasmanoff et al 2020;Lovell et al 2022;Shao et al 2021).…”

“…The final prediction of the model is the average over the predictions from all individual DTs. ERTs have already been successfully employed in the context of halo-galaxy connection studies (Kamdar et al 2016;Jo & Kim 2019;Lovell et al 2022). Here, we use the sklearn ensemble Extra Trees Regressor library (Pedregosa et al 2011).…”

“…The link between each galaxy and its dark matter halo is known a priori, but the model is complex enough to produce a highly realistic dataset. ML techniques have been employed in the context of the halo-galaxy connection before (e.g., Kamdar et al 2016;Agarwal et al 2018;Calderon & Berlind 2019;Jo & Kim 2019;Man et al 2019;Yip et al 2019;Zhang et al 2019;Jo & Kim 2019;Kasmanoff et al 2020;Delgado et al 2021;McGibbon & Khochfar 2021;Lovell et al 2022;Shao et al 2021). What has generally transpired from these analyses is that stellar mass is by far the galaxy property that is easier to predict, as it is known to display a strong connection with halo mass.…”

“…Other properties, such as colour or star formation rate (SFR), which are severely influenced by several secondary halo properties in non-trivial ways, are still challenging to reproduce even by the most sophisticated ML methods. To give some numbers, the Pearson correlation coefficient between the true and the predicted values for stellar mass are often above 0.9 in these analyses (e.g., Kamdar et al 2016;Lovell et al 2022), whereas the same metric yields values typically below 0.8 for other properties such as SFR (e.g., Kamdar et al 2016;Agarwal et al 2018;Lovell et al 2022).…”