Abstract:
The international astronomy community is building a suite of powerful new telescopes, which over the next decade will map the distribution of galaxies and gas over vast scales, with unprecedented precision. These experiments have the potential to constrain fundamental physics, such as the amount and nature of the mysterious "dark matter", as well as potentially revealing the initial conditions of the Universe. However, we face a challenge that is common to many fields: the properties of the galaxies that we can observe are shaped by physical processes that operate across an immense range of scales, from the sub-lightyear scales of individual stars and black holes to the billion lightyear scales of the 'cosmic web' traced out by dark matter structures. It is impossible to directly solve the relevant physical equations on a computer for this very broad dynamic range. The CCA has been leading an innovative effort to solve this "multiscale" problem in astrophysics using a ladder of multiscale numerical simulations and machine learning techniques. I will describe this work, and how the new Simons Collaboration "Learning the Universe" will make progress towards solving the multiscale problem for galaxy formation, and combine these new galaxy formation models with simulation based inference techniques to learn about both galaxy formation and cosmology.