Accuracy of power spectra in dissipationless cosmological simulations

Abstract We exploit a suite of large N-body simulations (up to N = 40963) performed with abacus, scale-free models range spectral indices n, better understand and quantify convergence the matter power spectrum. Using self-similarity identify converged regions, we show that maximal wavenumber resolved at given level accuracy increases monotonically as function time. At 1 per cent it starts early times from fraction $k_\Lambda$, Nyquist initial grid, reaches most, if force softening is sufficiently small, ${\sim}2{-}3 k_\Lambda$ very latest evolve to. $5{{\ \rm per\ cent}}$ level, extends up wavenumbers order $5k_\Lambda$ late times. Expressed suitable scale-factor, shows simple n-dependence, allowing extrapolation place conservative bounds on non-scale-free like LCDM. note deviations due discretization in are not well modelled by shot noise, subtracting fact degrades accuracy. Quantitatively our findings broadly line assumptions about resolution adopted recent studies using cosmological (e.g. Euclid Flagship) aiming constrain mildly non-linear regime. On other hand, remark conclusions small-scale clustering concerning validity stable clustering) obtained PS data larger than few $k_\Lambda$ may need revision light analysis.