2024 workshop program
How DREAMS are made: Emulating subhalo populations under alternative dark matter scenarios with Variational Diffusion Models
The work
Abstract
In this talk, I introduce a novel framework based on normalizing flows and variational diffusion models to emulate populations of subhalos and galaxies. This approach models each population as a point cloud, which preserves maximal information from the data without the need for binning or voxelization. As a result, it can resolve scales as small as those limited only by the spatial resolution of the simulations. The framework is trained on the new DREAMS suite of simulations, consisting of Milky Way-mass simulations under alternative dark matter models and baryonic physics. I will show validation results that our framework can correctly capture key summary statistics such as subhalo mass functions and two-point correlation functions. I will then use the framework to explore how these summary statistics vary with changes in warm dark matter mass, baryonic physics, and halo structural properties. These results underscore the capability of our emulator to provide insightful predictions about the influence of underlying dark matter models and physical processes on galaxy formation and dark matter structure.