A case study of hurdle and generalized additive models in astronomy: the escape of ionizing radiation

TL;DR

This paper introduces a novel statistical model using hurdle and generalized additive models to analyze the complex, non-linear relationship between galaxy properties and the escape fraction of ionizing radiation during the Universe's reionization epoch.

Contribution

It develops a new statistical approach that accounts for non-linearities and non-Gaussian features in modeling the escape fraction of ionizing photons in early galaxies.

Findings

Baryonic fraction and ionizing photon rate significantly influence escape fraction.

The model reveals larger effects of certain properties than naive analyses suggest.

Accounting for non-linearities reduces the perceived importance of specific star formation rate.

Abstract

The dark ages of the Universe end with the formation of the first generation of stars residing in primeval galaxies. These objects were the first to produce ultraviolet ionizing photons in a period when the cosmic gas changed from a neutral state to an ionized one, known as Epoch of Reionization (EoR). A pivotal aspect to comprehend the EoR is to probe the intertwined relationship between the fraction of ionizing photons capable to escape dark haloes, also known as the escape fraction ($f_{esc}$), and the physical properties of the galaxy. This work develops a sound statistical model suitable to account for such non-linear relationships and the non-Gaussian nature of $f_{esc}$. This model simultaneously estimates the probability that a given primordial galaxy starts the ionizing photon production and estimates the mean level of the $f_{esc}$ once it is triggered. The model was employed in the First Billion Years simulation suite, from which we show that the baryonic fraction and the rate of ionizing photons appear to have a larger impact on $f_{esc}$ than previously thought. A naive univariate analysis of the same problem would suggest smaller effects for these properties and a much larger impact for the specific star formation rate, which is lessened after accounting for other galaxy properties and non-linearities in the statistical model.