Modeling assembly bias with machine learning and symbolic regression: application to cosmic neutral hydrogen

Laurence Perreault Levasseur ( Université de Montréal )

Upcoming 21-cm surveys will map the spatial distribution of cosmic neutral hydrogen (HI) over unprecedented volumes. Mock catalogs are needed to fully exploit the potential of these surveys. Many standard techniques used to create these catalogs, such as the halo occupation distribution (HOD) model, rely on the assumption that the baryonic properties of dark matter halos depend on only their masses. From the hydrodynamic simulation IllustrisTNG we can show, to the contrary, that the HI content of halos strongly depends on their local environment. In this talk, I'll show how this effect can be modeled by machine learning algorithms and parametrized in the form of new analytic equations. From these, we can provide physical explanations for the environmental effect and show that ignoring it leads to ~10% bias in the real-space 21-cm power spectrum, which is larger than the expected precision from upcoming surveys.