Modelling the BOSS void-galaxy cross-correlation function using a neural-network emulator

TL;DR

This paper presents a neural-network emulator for modeling the void-galaxy cross-correlation function, improving parameter constraints and accounting for effects like RSD and AP, with implications for future large surveys.

Contribution

We develop a neural network emulator that enhances modeling of void-galaxy correlations, leading to better cosmological parameter constraints compared to previous template-based methods.

Findings

28% reduction in errors for Ω_m

More accurate AP measurements

Improved constraints on σ_8

Abstract

We introduce an emulator-based method to model the cross-correlation between cosmological voids and galaxies. This allows us to model the effect of cosmology on void finding and on the shape of the void-galaxy cross-correlation function, improving on previous template-based methods. We train a neural network using the AbacusSummit simulation suite and fit to data from the Sloan Digital Sky Survey Baryon Oscillation Spectroscopic Survey sample. We recover information on the growth of structure through redshift-space distortions (RSD), and the geometry of the Universe through the Alcock-Paczy\'nski (AP) effect, measuring $\Omega_{\rm m} = 0.330\pm 0.020$ and $\sigma_8 = 0.777^{+0.047}_{-0.062}$ for a $\Lambda \rm{CDM}$ cosmology. Comparing to results from a template-based method, we find that fitting the shape of the void-galaxy cross-correlation function provides more information and leads to an improvement in constraining power. In contrast, we find that errors on the AP measurements were previously underestimated if void centres were assumed to have the same response to the AP effect as galaxies - a common simplification. Overall, we recover a $28\%$ reduction in errors for $\Omega_{\rm{m}}$ and similar errors on $\sigma_8$ with our new, more comprehensive, method. Given the statistical power of future surveys including DESI and Euclid, we expect the method presented to become the new baseline for the analysis of voids in these data.