Optimizing simulation parameters for weak lensing analyses involving non-Gaussian observables

TL;DR

This study investigates how simulation parameters like lens plane thickness and mass resolution affect the accuracy of weak lensing statistics, providing guidelines for efficient and precise modeling for upcoming surveys.

Contribution

It offers a systematic analysis of hyper-parameter impacts on weak lensing predictions, identifying optimal simulation settings for non-Gaussian observables.

Findings

Thin lens planes (< 60 h^{-1} Mpc) suppress power spectrum accuracy.

A mass resolution of 7.2×10^{11} h^{-1} M_sun per particle suffices for LSST-like analyses.

Using these parameters maintains accuracy for scales down to 1 arcmin.

Abstract

We performed a series of numerical experiments to quantify the sensitivity of the predictions for weak lensing statistics obtained in raytracing DM-only simulations, to two hyper-parameters that influence the accuracy as well as the computational cost of the predictions: the thickness of the lens planes used to build past light-cones and the mass resolution of the underlying DM simulation. The statistics considered are the power spectrum and a series of non-Gaussian observables, including the one-point probability density function, lensing peaks, and Minkowski functionals. Counter-intuitively, we find that using thin lens planes ($< 60~h^{-1}$Mpc on a $240~h^{-1}$Mpc simulation box) suppresses the power spectrum over a broad range of scales beyond what would be acceptable for an LSST-type survey. A mass resolution of $7.2\times 10^{11}~h^{-1}\,M_{\odot}$ per DM particle (or 256$^3$ particles in a ($240~h^{-1}$Mpc)$^3$ box) is sufficient to extract information using the power spectrum and non-Gaussian statistics from weak lensing data at angular scales down to 1 arcmin with LSST-like levels of shape noise.