| Abstract: |
Performing controlled experiments is not a viable approach in cosmology, making simulations and modeling — the ability to create "virtual universes" — one of the key aspects of scientific inference. Sky surveys like DESI effectively pose an inverse problem, where the measurement result is obtained and one tries to infer the values of parameters which characterize the underlying physical system — the evolving Universe. The difficulty here is that producing each virtual universe in principle requires an extremely expensive numerical simulation carried out at high fidelity. In this talk I will review ongoing simulations and modeling efforts in LBL's Computational Cosmology Center which focuses on the intergalactic medium and Lyman alpha forest. Broadly, the effort I will cover in the talk consists of two coordinated components: (1) developing physics-based simulation capability based on our Nyx code, and (2) augmenting physical simulations with data-driven generative models. With the combination of the two -- physics-driven simulation methods and data-driven methods (ML/AI) we can successfully bridge the gap between the required resolution scale needed to accurately resolve Lyman alpha forest and the Gpc-scale simulations needed for sky surveys like DESI. |
