[Paper Review] A machine learning approach to mapping baryons onto dark matter halos using the EAGLE and C-EAGLE simulations

Christopher C. Lovell, Stephen M. Wilkins+5 more|arXiv (Cornell University)|Jun 9, 2021

Galaxies: Formation, Evolution, Phenomena9 references4 citations

TL;DR

This paper proposes a machine learning framework that predicts baryonic galaxy properties from dark matter halo characteristics using EAGLE and C-EAGLE simulations. By training on both periodic and zoom simulations, the model accurately reproduces key baryonic distribution functions and clustering statistics in large-volume dark matter-only simulations at a tiny fraction of the computational cost of full hydrodynamic 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 $(800 \; \mathrm{Mpc})^3$ P-Millennium simulation, and present predictions for key baryonic distribution functions and clustering statistics from the EAGLE model in this large volume.

Motivation & Objective

• To develop a computationally efficient method for predicting baryonic galaxy properties in large cosmological volumes where full hydrodynamic simulations are infeasible.
• To learn environmental biases in galaxy evolution by training on both periodic EAGLE simulations and high-resolution C-EAGLE zoom simulations of overdense regions.
• To enable application of the EAGLE model to large-scale dark matter-only simulations, such as the (800 Mpc)^3 P-Millennium simulation, for cosmological clustering statistics.
• To reproduce key baryonic distribution functions and clustering properties with minimal computational cost compared to full hydrodynamic simulations.

Proposed method

• A tree-based machine learning model is trained to predict galaxy baryonic properties (e.g., stellar mass, gas content) from host dark matter halo properties (e.g., mass, concentration, formation time).
• Training data is drawn from both the periodic EAGLE simulations and the high-resolution C-EAGLE zoom simulations of galaxy clusters to capture environmental dependencies.
• The model learns the halo-galaxy relationship, including the bias in galaxy evolution across different environments, by leveraging the contrast between underdense and overdense regions.
• The trained model is applied to the (800 Mpc)^3 P-Millennium dark matter-only simulation to generate predictions for baryonic properties and clustering statistics.
• Model performance is validated by comparing predicted distribution functions and clustering statistics to those from the original EAGLE simulations.

Experimental results

Research questions

• RQ1 Can a machine learning model trained on limited hydrodynamic simulations accurately predict baryonic galaxy properties in large-scale dark matter-only simulations?
• RQ2 How does the halo-galaxy relationship vary in underdense versus overdense environments, and can this be captured in a single model?
• RQ3 To what extent can the model reproduce key baryonic distribution functions and clustering statistics in a large-volume simulation?
• RQ4 What is the computational efficiency gain of using the ML model compared to running full hydrodynamic simulations?

Key findings

• The trained machine learning model successfully reproduces key baryonic distribution functions—such as stellar mass functions and gas fraction distributions—across a wide dynamic range with high fidelity.
• The model accurately predicts clustering statistics of galaxies in large-scale structures, matching results from the original EAGLE simulations.
• By incorporating data from both periodic and zoom simulations, the model captures environmental biases in galaxy evolution, improving its predictive power in overdense regions.
• The model enables predictions in the (800 Mpc)^3 P-Millennium dark matter-only simulation at an infinitesimal fraction of the computational cost of a full hydrodynamic simulation.