Turning dispersion into signal: density-split analyses of pairwise velocities

Pairwise velocities of the large-scale structure encode valuable information about the growth of structure. They can be observed indirectly through redshift-space distortions and the kinetic Sunyaev-Zeldovich effect. Whether it is Gaussian or non-Gaussian, pairwise velocity has a broad distribution, but the cosmologically useful information lies primarily in the mean - the streaming velocities; the dispersion around the mean is often treated as a nuisance and marginalized over. This conventional approach reduces the constraining power of our observations. Here, we show that this does not have to be the case, provided the physics behind the dispersion is understood. We demonstrate that by splitting the halo/galaxy samples according to their density environments and measuring the streaming velocities separately, the total signal-to-noise is several times greater than in conventional global measurements of the pairwise velocity distribution (PVD). This improvement arises because the global PVD is a composite of a series of near-Gaussian distributions with different means and dispersions, each determined by its local density environment. Around underdense and overdense regions, the mean streaming velocities are positive and negative, respectively. By splitting the data, we avoid cancellation between these opposing velocities, effectively turning what would be considered dispersion in the global PVD into a signal. Our findings indicate substantial potential for improving the analysis of PVD observations using the kinetic Sunyaev-Zeldovich effect and redshift-space distortions.