We study how well the Gaussian approximation is valid for computing the covariance matrices of the convergence power and bispectrum in weak gravitational lensing analyses. We focus on its impact on the cosmological parameter estimations by comparing the results with and without non-Gaussian error contribution in the covariance matrix. We numerically derive the covariance matrix as well as the cosmology dependence of the spectra from a large set of $N$-body simulations performed for various cosmologies and carry out Fisher matrix forecasts for tomographic weak lensing surveys with three source redshifts. After showing the consistency of the power and bispectra measured from our simulations with the state-of-the-art fitting formulas, we investigate the covariance matrix assuming a typical ongoing survey across $1500{\mathrm{deg}}^2$ with the mean source number density of $30{\mathrm{arcmin}}^{-2}$ at the mean redshift $z_s=1.0$. Although the shape noise contributes a significant fraction to the total error budget and it mitigates the impact of the non-Gaussian error for this source number density, we find that the non-Gaussian error degrades the cumulative signal-to-noise ratio up to the maximum multipole of 2000 by a factor of about 2 (3) in the power (bi-) spectrum analysis. Its impact on the final cosmological parameter forecast with 6 parameters can be as large as 15% in the size of the one-dimensional statistical error. This can be a problem in future wide and deep weak lensing surveys for precision cosmology. We also show how much the dark energy figure of merit is affected by the non-Gaussian error contribution and demonstrate an optimal survey design with a fixed observational time.

• Received 16 January 2013

DOI:https://doi.org/10.1103/PhysRevD.87.123538