Painting galaxies into dark matter haloes using machine learning

Abstract

We develop a machine learning (ML) framework to populate large dark matter-only simulations with baryonic galaxies. Our ML framework takes input halo properties including halo mass, environment, spin, and recent growth history, and outputs central galaxy and halo baryonic properties including stellar mass (M * ), star formation rate (SFR), metallicity (Z), neutral (H i) and molecular (H 2 ) hydrogen mass. We apply this to the Mufasa cosmological hydrodynamic simulation, and show that it recovers the mean trends of output quantities with halo mass highly accurately, including following the sharp drop in SFR and gas in quenched massive galaxies. However, the scatter around the mean relations is under-predicted. Examining galaxies individually, at z = 0 the stellar mass and metallicity are accurately recovered (σ 0.2 dex), but SFR and H i show larger scatter (σ 0.3 dex); these values improve somewhat at z = 1, 2. Remarkably, ML quantitatively recovers second parameter trends in galaxy properties, e.g. that galaxies with higher gas content and lower metallicity have higher SFR at a given M * . Testing various ML algorithms, we find that none perform significantly better than the others, nor does ensembling improve performance, likely because none of the algorithms reproduce the large observed scatter around the mean properties. For the random forest algorithm, we find that halo mass and nearby (∼ 200 kpc) environment are the most important predictive variables followed by growth history, while halo spin and ∼Mpc scale environment are not important. Finally we study the impact of additionally inputting key baryonic properties M * , SFR and Z, as would be available e.g. from an equilibrium model, and show that particularly providing the SFR enables H i to be recovered substantially more accurately.

Mergers, starbursts, and quenching in the simba simulation

“…This is expected given a reasonably tight relation between stellar mass and halo mass produced in simulations (e.g. Agarwal et al 2018), and the fact that the halo merger rate increase with halo mass (e.g. Genel et al 2009) owing to hierarchical structure formation.…”

AI-assisted superresolution cosmological simulations

“…SR mock catalogs could then be made that are more complex-for example, including reionization (27). This type of mock making would have some similarities with the "painting" of galaxies onto darkmatter simulations by Agarwal et al (28), except that, unlike that paper, our method is conditioned on the entire density distribution rather than a few halo properties and would also add SR structure to the model.…”

Surveying the reach and maturity of machine learning and artificial intelligence in astronomy

“…Finally, although a single catch-all name does not do such a diverse field justice, simulation will be used to describe the data products from any numerical or computational method. For example, cosmological simulations (e.g., Agarwal, Davé, & Bassett, 2018;Hui, Aragon, Cui, & Flegal, 2018;Lucie-Smith, Peiris, Pontzen, & Lochner, 2018;Nadler, Mao, Wechsler, Garrison-Kimmel, & Wetzel, 2018;Rodríguez et al, 2018) follow the gravity-induced formation and growth of structures, requiring approximations to various physical mechanisms, a suitable choice of initial conditions, and a strategy for time-based evolution (down to some minimum level of accuracy).…”

“…ML is providing new methods for examining the outputs of cosmological simulations, leading to new insights about the connections between physical properties of galaxies, dark matter halos and the cosmic environment. Examples include the use of an ANN to aid in determining the total mass of the Milky Way and the Andromeda Galaxy from the Small MultiDark simulation (McLeod, Libeskind, Lahav, & Hoffman, 2017), and both classification of sub-halos (Nadler et al, 2018) and assignment of galaxies to halos (Agarwal et al, 2018) in dark matter-only simulations.…”