Generating Cosmological Gaussian Random Fields

We present a generic algorithm for generating Gaussian random initial conditions for cosmological simulations on periodic rectangular lattices. We show that imposing periodic boundary conditions on the real-space correlator and choosing initial conditions by convolving a white noise random field results in a significantly smaller error than the traditional procedure of using the power spectrum. This convolution picture produces exact correlation functions out to separations of L/2, where L is the box size, which is the maximum theoretically allowed. This method also produces tophat sphere fluctuations which are exact at radii R ≤ L/4. It is equivalent to windowing the power spectrum with the simulation volume before discretizing, thus bypassing sparse sampling problems. The mean density perturbation in the volume is no longer constrained to be zero, allowing one to assemble a large simulation using a series of smaller ones. This is especially important for simulations of Lyman-α systems where small boxes with steep power spectra are routinely used. We also present an extension of this procedure which generates exact initial conditions for hierarchical grids at negligible cost. Subject headings: methods: numerical — large-scale structure of universe