A Study of the Dynamical Structures in a Dark Matter Halo Using UMAP

Abstract We use a dimension reduction algorithm, Uniform Manifold Approximation and Projection (UMAP), to study dynamical structures inside a dark matter halo. We use a zoom-in simulation of a Milky Way mass dark matter halo and apply UMAP to the six-dimensional phase space in the dark matter field at z = 0. We find that particles in the field are mapped to distinct clusters in the lower dimensional space in a way that is closely related to their accretion history. The largest cluster in UMAP space does not contain the entire mass of the Milky Way virial region and neatly separates the older halo from the recently accreted matter. Particles within this cluster, which only comprise ∼70% of the Milky Way particles, have had several pericenter passages and are, therefore, likely to be phase mixed, becoming dynamically uniform. The infall region and recently accreted particles and substructure, even up to splashback, form distinct components in the lower dimensional space; additionally, higher angular momentum particles also take longer times to mix. Our work shows that the current state of the Milky Way halo retains historical information, particularly about the recent accretion history, and even a relatively old structure is not dynamically uniform. We also explore UMAP as a preprocessing step to find coherent subhalos in dark matter simulations.