Using Artificial Neural Networks to Constrain the Halo Baryon Fraction during Reionization

TL;DR

This paper uses advanced simulations to study how radiative feedback affects baryon content in halos during reionization and introduces an neural network model to predict baryon fractions based on environmental factors.

Contribution

It presents a self-consistent simulation approach to constrain halo baryon fractions during reionization and develops a neural network model for predicting baryon content in halos.

Findings

Halo mass with half the cosmic baryon fraction shows little variation with ionization.

Including metal cooling and self-shielding reduces the characteristic halo mass.

The neural network accurately predicts baryon fractions based on environmental parameters.

Abstract

Radiative feedback from stars and galaxies has been proposed as a potential solution to many of the tensions with simplistic galaxy formation models based on $\Lambda$CDM, such as the faint end of the UV luminosity function. The total energy budget of radiation could exceed that of galactic winds and supernovae combined, which has driven the development of sophisticated algorithms that evolve both the radiation field and the hydrodynamical response of gas simultaneously, in a cosmological context. We probe self-feedback on galactic scales using the adaptive mesh refinement, radiative transfer, hydrodynamics, and $N$-body code. Unlike previous studies which assume a homogeneous UV background, we self-consistently evolve both the radiation field and gas to constrain the halo baryon fraction during cosmic reionization. We demonstrate that the characteristic halo mass with mean baryon fraction half the cosmic mean, $M_{\mathrm{c}}(z)$, shows very little variation as a function of mass-weighted ionization fraction. Furthermore, we find that the inclusion of metal cooling and the ability to resolve scales small enough for self-shielding to become efficient leads to a significant drop in $M_{\mathrm{c}}$ when compared to recent studies. Finally, we develop an Artificial Neural Network that is capable of predicting the baryon fraction of haloes based on recent tidal interactions, gas temperature, and mass-weighted ionization fraction. Such a model can be applied to any reionization history, and trivially incorporated into semi-analytical models of galaxy formation.