Description
Large mock galaxy catalogues are essential to our understanding of rare environments, galaxy clustering, and cosmic variance. However, large simulations are prohibitively expensive to run without making compromises on the resolution and/or complexity of physics. By learning the galaxy-halo connection in zoom simulations, we can map galaxies onto large N-body simulations at low computational expense. The deterministic methods commonly applied to this task have consistently failed at recovering the tails of the simulated distributions. This is because stochasticity in the subgrid models drives chaotic behaviour that, for any given dark matter halo, can result in markedly different galaxies. This motivates the use of probabilistic methods that can learn the probability distribution of galaxy properties conditioned on their dark matter halos.
