The impact of non-Gaussian errors on weak lensing surveys

The weak lensing power spectrum carries cosmological information via its dependence on the growth of structure and on geometric factors. Since much of the cosmological information comes from scales affected by non-linear clustering, measurements of the lensing power spectrum can be degraded by non-Gaussian covariances. Recently, there have been conflicting studies about the level of this degradation. We use the halo model to estimate it and include new contributions related to the finite size of lensing surveys, following Rimes and Hamilton's study of three-dimensional simulations. We find that non-Gaussian correlations between different multipoles can degrade the cumulative signal-to-noise ratio (S/N) for the power spectrum amplitude by up to a factor of 2 (or 5 for a worst-case model that exceeds current N-body simulation predictions). However, using an eight-parameter Fisher analysis, we find that the marginalized errors on individual parameters are degraded by less than 10 per cent (or 20 per cent for the worst-case model). The smaller degradation in parameter accuracy is primarily because: individual parameters in a high-dimensional parameter space are degraded much less than the volume of the full Fisher ellipsoid; lensing involves projections along the line of sight, which reduce the non-Gaussian effect; some of the cosmological information comes from geometric factors which are not degraded at all. We contrast our findings with those of Lee and Pen who suggested a much larger degradation in information content. Finally, our results give a useful guide for exploring survey design by giving the cosmological information returns for varying survey area, depth and the level of some systematic errors.