Institute for Advanced Study/Princeton University Early Universe/Cosmology Lunch Discussion

Modern cosmological surveys are amassing data in petabytes, ranging from the cosmic microwave background to optical galaxy surveys. Traditionally, these vast data sets are compressed into $O(100-1000)$ correlation statistics to constrain key cosmological parameters. Moreover, extracting cross-information between probes could be cumbersome. In this talk, I will present an emerging opportunity to reimagine cosmological inquiry through field-level inference (FLI). This approach involves forward modeling both the spatial and temporal evolution of the universe as well as astrophysical and systematics effects. It then compares the simulated observables with datasets to constrain both cosmological parameters and, sometimes, the cosmological maps themselves. FLI thus minimizes data compression and effectively extracts joint-probe information, but its likelihood function is often complex. I will illustrate the mathematical intuition behind FLI using simple cases with exact solutions, and then discuss the modeling and computational approximations needed for real data application.