An ideal mass assignment scheme for measuring the Power Spectrum with FFTs

Abstract

In measuring the power spectrum of the distribution of large numbers of dark matter particles in simulations, or galaxies in observations, one has to use Fast Fourier Transforms (FFT) for calculational efficiency. However, because of the required mass assignment onto grid points in this method, the measured power spectrum $\la |δ^f(k)|^2\ra$ obtained with an FFT is not the true power spectrum $P(k)$ but instead one that is convolved with a window function $|W(\vec k)|^2$ in Fourier space. In a recent paper, Jing (2005) proposed an elegant algorithm to deconvolve the sampling effects of the window function and to extract the true power spectrum, and tests using N-body simulations show that this algorithm works very well for the three most commonly used mass assignment functions, i.e., the Nearest Grid Point (NGP), the Cloud In Cell (CIC) and the Triangular Shaped Cloud (TSC) methods. In this paper, rather than trying to deconvolve the sampling effects of the window function, we propose to select a particular function in performing the mass assignment that can minimize these effects. An ideal window function should fulfill the following criteria: (i) compact top-hat like support in Fourier space to minimize the sampling effects; (ii) compact support in real space to allow a fast and computationally feasible mass assignment onto grids. We find that the scale functions of Daubechies wavelet transformations are good candidates for such a purpose. Our tests using data from the Millennium Simulation show that the true power spectrum of dark matter can be accurately measured at a level better than 2% up to $k=0.7k_N$, without applying any deconvolution processes. The new scheme is especially valuable for measurements of higher order statistics, e.g. the bi-spectrum,........