Novel Forward Models and Solver-in-Loop Approaches for Joint Astrophysics and Cosmology

Explore how machine learning techniques can revolutionize cosmological hydrodynamical simulations in this 20-minute conference talk from the Workshop on "Putting the Cosmic Large-scale Structure on the Map: Theory Meets Numerics." Discover the computational challenges facing current hydrodynamical simulations that study dark matter and baryon interactions, and examine how these simulations struggle to reproduce observational data while remaining computationally expensive. Learn about the ongoing tensions in key cosmological parameters like σ₈ and investigate whether new fundamental physics or baryonic feedback effects are responsible for these discrepancies. Delve into innovative approaches for integrating ML techniques into hydrodynamical simulations to enhance cosmological information extraction, while understanding the inherent challenges and limitations of these methods. Gain insights into "solver-in-loop" techniques enabled by end-to-end differentiable simulation pipelines that could dramatically accelerate simulations or potentially discover previously unknown physics and survey systematics. Follow the outlined path toward developing a more efficient and robust framework for large-scale structure inference in modern astrophysics research.