Simulations of Wide-Field Weak Lensing Surveys I: Basic Statistics and Non-Gaussian Effects

TL;DR

This paper uses extensive simulations to analyze the convergence power spectrum in weak lensing surveys, revealing non-Gaussian effects that impact the accuracy of statistical estimates and highlighting the importance of accounting for non-Gaussian covariances.

Contribution

It provides a detailed comparison between simulation results and analytic models, demonstrating the significance of non-Gaussian covariances in weak lensing analyses.

Findings

Semi-analytic models underestimate power by ~30% at small scales.

Halo model accurately reproduces covariance across scales.

Non-Gaussian covariances significantly reduce signal-to-noise ratio.

Abstract

We study the lensing convergence power spectrum and its covariance for a standard LCDM cosmology. We run 400 cosmological N-body simulations and use the outputs to perform a total of 1000 independent ray-tracing simulations. We compare the simulation results with analytic model predictions. The semi-analytic model based on Smith et al.(2003) fitting formula underestimates the convergence power by ~30% at arc-minute angular scales. For the convergence power spectrum covariance, the halo model reproduces the simulation results remarkably well over a wide range of angular scales and source redshifts. The dominant contribution at small angular scales comes from the sample variance due to the number fluctuations of halos in a finite survey volume. The signal-to-noise ratio for the convergence power spectrum is degraded by the non-Gaussian covariances by up to a factor 5 for a weak lensing survey to z_s ~1. The probability distribution of the convergence power spectrum estimators, among the realizations, is well approximated by a chi-square distribution with broadened variance given by the non-Gaussian covariance, but has a larger positive tail. The skewness and kurtosis have non-negligible values especially for a shallow survey. We argue that a prior knowledge on the full distribution may be needed to obtain an unbiased estimate on the ensemble averaged band power at each angular scale from a finite volume survey.