Simulations of weak gravitational lensing - II. Including finite support effects in cosmic shear covariance matrices

Numerical N-body simulations play a central role in the assessment of weak gravitational lensing statistics, residual systematics and error analysis. In this paper, we investigate and quantify the impact of finite simulation volume on weak lensing two- and four-point statistics. These finite support (FS) effects are modelled for several estimators, simulation box sizes and source redshifts, and validated against a new large suite of 500 N-body simulations. The comparison reveals that our theoretical model is accurate to better than 5 per cent for the shear correlation function xi(+)(theta) and its error. We find that the most important quantities for FS modelling are the ratio between the measured angle theta and the angular size of the simulation box at the source redshift, theta(box)(z(s)), or the multipole equivalent a/a(box)(z(s)). When this ratio reaches 0.1, independently of the source redshift, the shear correlation function xi(+) is suppressed by 5, 10, 20 and 25 per cent for L-box = 1000, 500, 250 and 147 h(-1) Mpc, respectively. The same effect is observed in xi(-)(theta), but at much larger angles. This has important consequences for cosmological analyses using N-body simulations and should not be overlooked. We propose simple semi-analytic correction strategies that account for shape noise and survey masks, generalizable to any weak lensing estimator. From the same simulation suite, we revisit the existing non-Gaussian covariance matrix calibration of the shear correlation function, and propose a new one based on the 9-year Wilkinson Microwave Anisotropy Probe)+baryon acoustic oscillations+supernova cosmology. Our calibration matrix is accurate at 20 per cent down to the arcminute scale, for source redshifts in the range 0 < z < 3, even for the far off-diagonal elements. We propose, for the first time, a parametrization for the full xi(-) covariance matrix, also 20 per cent accurate for most elements.