Information content of the non-linear power spectrum: the effect of beat-coupling to large scales

We measure the covariance of the non-linear matter power spectrum from N-body simulations using two methods. In the first case, the covariance of power is estimated from the scatter over many random realizations of the density field. In the second, we use a novel technique to measure the covariance matrix from each simulation individually by re-weighting the density field with a carefully chosen set of functions. The two methods agree at linear scales, but unexpectedly they disagree substantially at increasingly non-linear scales. Moreover, the covariance of non-linear power measured using the re-weightings method changes with box size. The numerical results are consistent with an explanation given in a companion paper, which argues that the cause of the discrepancy is beat-coupling, in which products of Fourier modes separated by a small wavevector couple by gravitational growth to the large-scale beat mode between them. We calculate the information content of the non-linear power spectrum (about the amplitude of the initial, linear power spectrum) using both methods and confirm the result of a previous paper, that at translinear scales the power spectrum contains little information over and above that in the linear power spectrum, but that there is a marked increase in information at non-linear scales. We suggest that, in real galaxy surveys, the covariance of power at non-linear scales is likely to be dominated by beat-coupling to the largest scales of the survey and that, as a result, only part of the information potentially available at non-linear scales is actually measurable from real galaxy surveys.