On fast generation of cosmological random fields

The statistical translation invariance of cosmological random fields is broken by a finite survey boundary, correlating the observable Fourier modes. Standard methods for generating Gaussian fields either neglect these correlations, or are costly, or both. Here, we report on a fast and exact simulation method applicable to a wide class of two-point statistics that requires the simulation of a periodic grid of only twice the survey side with fast Fourier transforms. Supersurvey modes, dominating the covariance of power spectra beyond linear scales in galaxy surveys and causing the correlation of large and small scales, 'beat coupling', or 'super-sample' covariance, are precisely accounted for in non-linear transformations of the Gaussian field. As an application, we simulate the CFHTLS (Canada-France-Hawaii Telescope Large Survey) similar to 7 degrees x 7 degrees W1 galaxy density field, modelled as a Poisson sampling of a lognormal density field. We show that our simulations produce power spectra, A(star)-power spectra, counts-in-cells probability distributions as well as covariances perfectly consistent with the data. In addition, our technique reproduces the information plateau beyond linear scales as observed previously in Sloan Digital Sky Survey galaxy catalogues and in N-body simulations. Our method is thus an efficient yet powerful simulation and prediction tool for galaxy survey data and covariances.