A Comparison of Void-Finding Algorithms using Crossing Numbers

TL;DR

This study compares void-finding algorithms in cosmology, revealing that VoidFinder more reliably identifies dynamically distinct low-crossing regions than V2, especially beyond 0.25 times the effective radius.

Contribution

It introduces a method to compare voids identified by different algorithms using crossing numbers and assesses their effectiveness in capturing dynamical properties.

Findings

VoidFinder better identifies low-crossing, dynamically distinct regions.

Beyond 0.25 effective radius, VoidFinder's voids show fewer high-crossing particles.

Using distance from void edge partially improves V2's performance.

Abstract

We study how well void-finding algorithms identify cosmic void regions and whether we can quantitatively and qualitatively compare the voids they find with dynamical information from the underlying matter distribution. Using the ORIGAMI algorithm to determine the number of dimensions along which dark matter particles have undergone shell-crossing (crossing number) in N-body simulations from the AbacusSummit simulation suite, we identify dark matter particles that have undergone no shell crossing as belonging to voids. We then find voids in the corresponding halo distribution using two different void-finding algorithms: VoidFinder and Voronoi Voids (V2), a ZOBOV-based algorithm. The resulting void catalogs are compared to the distribution of dark matter particles to examine how their crossing numbers depend on void proximity. While both algorithms' voids have a similar distribution of crossing numbers near their centers, we find that beyond 0.25 times the effective void radius, voids found by VoidFinder exhibit a stronger preference for particles with low crossing numbers than those found by V2. We examine two possible methods of mitigating this difference in efficacy between the algorithms. While we are able to partially mitigate the ineffectiveness of V2 by using the distance from the void edge as a measure of centrality, we conclude that VoidFinder more reliably identifies dynamically distinct regions of low crossing number.