Phase-space structure of dark matter haloes: scale-invariant probability density function driven by substructure

We present a method for computing the six-dimensional coarse-grained phase-space density f (x, v) in an N-body system, and derive its distribution function v (f). The method is based on Delaunay tessellation, where v ( f) is obtained with an effective fixed smoothing window over a wide f range. The errors are estimated, and v (f) is found to be insensitive to the sampling resolution or the simulation technique. We find that in gravitationally relaxed haloes built by hierarchical clustering, v (f) is well approximated by a robust power law, v (f) proportional to f(-2.5 +/- 0.05), over more than four decades in f, from its virial level to the numerical resolution limit. This is tested to be valid in the Lambda cold dark matter cosmology for haloes with masses 10(9) -10(15) M-., indicating insensitivity to the slope of the initial fluctuation power spectrum. By mapping the phase-space density in position space, we find that the high-f end of v (f) is dominated by the 'cold' subhaloes rather than the parent-halo central region and its global spherical profile. The value of f in subhaloes near the virial radius is typically > 100 times higher than its value at the halo centre, and it decreases gradually from the outside in toward its value at the halo centre. This seems to reflect phase mixing due to mergers and tidal effects involving puffing up and heating. The phase-space density can thus provide a sensitive tool for studying the evolution of subhaloes during the hierarchical build-up of haloes. It remains to be understood why the evolved substructure adds up to the actual universal power law of v (f) proportional to f(-5/2). It seems that this behaviour results from the hierarchical clustering process and is not a general result of violent relaxation.