The control of systematic effects when measuring galaxy shapes is one of the main challenges for cosmic shear analyses. In this context, we study the fundamental limitations on shear accuracy due to the measurement of the Point Spread Function (PSF)
from the finite number of stars. In order to do that, we translate the accuracy required for cosmological parameter estimation to the minimum number of stars over which the PSF must be calibrated. We first derive our results analytically in the case of infinitely small pixels (i.e. infinitely high resolution). Then image simulations are used to validate these results and investigate the effect of finite pixel size in the case of an elliptical gaussian PSF. Our results are expressed in terms of the minimum number of stars required to calibrate the PSF in order to ensure that systematic errors are smaller than statistical errors when estimating the cosmological parameters. On scales smaller than the area containing this minimum number of stars, there is not enough information to model the PSF. In the case of an elliptical gaussian PSF and in the absence of dithering, 2 pixels per PSF Full Width at Half Maximum (FWHM) implies a 20% increase of the minimum number of stars compared to the ideal case of infinitely small pixels; 0.9 pixels per PSF FWHM implies a factor 100 increase. In the case of a good resolution and a typical Signal-to-Noise Ratio distribution of stars, we find that current surveys need the PSF to be calibrated over a few stars, which may explain residual systematics on scales smaller than a few arcmins. Future all-sky cosmic shear surveys require the PSF to be calibrated over a region containing about 50 stars.
