Institute for Advanced Study/Princeton University Early Universe/Cosmology Lunch Discussion
Topic 1: Forward modeling in the era of cosmological surveys Topic 2:Multi-wavelength cluster mass estimation with machine learning
Abstract 1: Upcoming cosmological surveys will measure the large-scale distribution of galaxies at the subpercent level. In order to extract unbiased cosmological data while retaining valuable small-scale information, we need highly accurate models of the connection between galaxies and (dark) matter. While cosmological hydrodynamical simulations are too small and computationally expensive to directly use in the analysis of galaxy observations, they provide a detailed probe of the galaxy-halo link (under the assumptions of a particular, plausible galaxy formation model). We show that the simplest galaxy-halo model, the mass-only halo occupation distribution (HOD), fails to capture the galaxy clustering at the 15% level, which is well beyond the 1% requirement set by current and future experiments. We develop augmented models which reproduce multiple galaxy distribution statistics by the hydro simulation. We develop a pipeline for applying these models to observational data and show that in their crudest form, they manage to alleviate existing tensions (e.g., Lensing is low).
Abstract 2: Accurately estimating masses of galaxy clusters is crucial in order to extract cosmological information from them. In the past, CMB surveys have relied on a scaling relation based on the integrated electron pressure of clusters to estimate their masses. I will present a more accurate proxy for estimating masses which combines observables from SZ and X-ray surveys, and which was obtained using the machine learning tool called symbolic regression. More generally, I will show how machine learning tools can be used to augment traditional astrophysical scaling relations in order to make them more precise.