Robustness of Cosmic Void Properties: A Predictive Approach

Abstract

Cosmic voids are used as a tool to understand cosmology because they are dominated by dark energy. The tracers used to detect cosmic voids are typically derived from simulations that only contain dark matter or observations that only detect galaxies. In this project, we examine the effects of tracer bias---only having galaxies present---and evaluate the robustness of two specific void properties: the abundance and the density profile.

In order to evaluate robustness, we take two approaches to predicting various properties of galaxies within dark matter-only simulations. Both of these approaches use the z=0 data in the Illustris simulation. In the first, we learn the mapping from halo masses to two properties: the luminous mass and stellar formation rates (SFR) of corresponding galaxies. This mapping is established as a series of probability mass functions which are then used to probabilistically assign dark matter halos a luminous mass or SFR. In the second approach, we expand previous work in using a convolutional network, trained on the same Illustris dataset, to predict galaxy density fields based on dark matter density fields.

Both of these learned models are then applied to the halos in the z=0 snapshot of the MassiveNuS simulation (without massive neutrinos present). We evaluate a novel approach of using a CNN on a spatially larger simulation. We then compare the properties of voids present in the resulting galaxy catalogs to those from the original halo catalogs in MassiveNuS. We find that both the abundances and density profiles are robust against tracer bias when assigning probabilistic luminous masses and SFR. However, the convolutional network does not accurately reproduce the spatial distribution well enough to conclude on robustness from the second predictive approach.