Abstract:
Dark Matter, the most abundant matter in the universe, has eluded our understanding for decades. From rotation curves, CMB, and gravitational lensing, we see that structure formation in the universe is driven by dark matter, not baryons. Dark matter halos are some of the densest structures in the universe, making them the best objects for studying the microphysics of dark matter, like annihilation and reaction rates, which depend on phase space number density. Simulations go a long way in helping ascertain the best dark matter models and their properties. Comparing simulations to real data lets us constrain various parameters of these models. While halo finders do an amazing job of finding structures like halos and subhalos in simulations, they fall short when it comes to finding elongated structures like streams, which occupy a distinct phase space region when compared to halos and subhalos. To identify such structures, especially the elongated structures, in simulations, we use data from a LCDM zoom—in simulation and implement a non-linear dimension reduction algorithm, namely UMAP. We focus on a 1Mpc h1 box around the MW. We use 6D phase space information of all the particles in the box as our input data. We reduce the 6D information to a 2D representation using UMAP. UMAP separates the largest halos in the box, MW and four massive infalling halos in output space. Within the virial boundary of the MW, particles are segregated based on velocity and dynamics. Infalling streams are separated from the intact core of infalling subhalos. Infalling subhalo particles at their pericentre are separated from the rest of the subhalo. We can use these separations to identify streams and other substructures within the virial boundaries of halos. Which in turn will help us constrain various microphysical properties. This also shows that topological methods like UMAP and GNNs are viable options for data analysis in cosmology and simulations.
