Statistical fluctuations of correlators in the Color Glass Condensate

In the McLerran-Venugopalan model, correlators of Wilson lines are given by an average over a Gaussian ensemble of random color sources. In numerical implementations, these averages are approximated by a Monte-Carlo sampling. In this paper, we study the statistical error made with such a sampling, with emphasis on the momentum dependence of this error. Using the example of the dipole amplitude, we consider various approximants that are all equivalent in the limit of infinite statistics but differ with finite statistics and compare their statistical errors. For correlation functions that are translation invariant, we show that averaging over the barycenter coordinate drastically reduces the statistical error and more importantly modifies its momentum dependence.