Momentum space sampling of neutrinos in N-body simulations

Including massive neutrinos in N-body simulations is a challenging task due to the large thermal velocities of the neutrinos. In particle based codes this leads to problems of shot-noise due to insufficient sampling of the neutrino momentum distribution function. In this paper we investigate the benefits and drawbacks of a scheme first suggested in a paper by Banerjee et al. [1] in which the initial neutrino distribution is symmetrised in momentum space. Using a somewhat different, but qualitatively similar implementation, we confirm that this method reduces shot-noise significantly, but we also find that it generates some spurious power in the neutrino power spectrum at intermediate and small scales. We speculate that this happens because many neutrinos in the simulation sample the same underlying dark matter structures while moving through the simulation. By carefully tuning the number of directions in momentum space of the initial neutrino distribution we show that some improvements can be made over the case where initial neutrino directions are purely random. At redshifts z $\gtrsim$ 3 the method works very well, but at smaller redshifts significant improvements are not possible due to the spurious power generation.