Effects of long-wavelength fluctuations in large galaxy surveys

TL;DR

This paper investigates how finite simulation volumes affect cosmological statistics in large galaxy surveys, concluding that volumes over 1 Gpch^3 are sufficient to minimize significant finite-volume effects.

Contribution

It demonstrates that simulation boxes larger than 1 Gpch^3 effectively mitigate finite-volume effects, reducing the need for extremely large computational simulations.

Findings

Finite volume effects are significant below ~500 Mpc/h boxes.

Power spectra differences are less than 0.1% between 1 Gpch and 4 Gpch boxes.

Volumes over 1 Gpch^3 are sufficient for most survey science requirements.

Abstract

In order to capture as much information as possible large galaxy surveys have been increasing their volume and redshift depth. To face this challenge theory has responded by making cosmological simulations of huge computational volumes with equally increasing the number of dark matter particles and supercomputing resources. Thus, it is taken for granted that the ideal situation is when a single computational box encompasses the whole effective volume of the observational survey, e.g., ~50 Gpch^3 for the DESI and Euclid surveys. Here we study the effects of missing long-waves in a finite volume using several relevant statistics: the abundance of dark matter halos, the PDF, the correlation function and power spectrum, and covariance matrices. Finite volume effects can substantially modify the results if the computational volumes are less than ~(500Mpch)^3. However, the effects become extremely small and practically can be ignored when the box-size exceeds ~1Gpch^3. We find that the average power spectra of dark matter fluctuations show remarkable lack of dependence on the computational box-size with less than 0.1% differences between 1Gpch and 4Gpch boxes. No measurable differences are expected for the halo mass functions for these volumes. The covariance matrices are scaled trivially with volume, and small corrections due to super-sample modes can be added. We conclude that there is no need to make those extremely large simulations when a box-size of 1-1.5Gpch is sufficient to fulfil most of the survey science requirements.