Files
Abstract
Modern cosmological surveys, from cosmic microwave background experiments to optical galaxy surveys, amass data at the petabyte scale. Prior to inference, these data are typically compressed into low-dimensional summary statistics, discarding substantial information, including non-Gaussian features, and complicating multiprobe analyses. I present an emerging opportunity to reimagine cosmological inquiry through field-level inference (FLI). FLI replaces summary statistics with a forward model that directly compares simulated observables to data. This formulation preserves map-level information and naturally unifies map-making, multiprobe combination, and parameter inference, but it leads to high-dimensional, non-Gaussian posteriors whose structure must be understood and controlled. I introduce field-level inference as a practical methodology spanning theory, modeling, and applications. I analyze the mathematical structure of FLI in analytically tractable regimes; extend FLI to weak gravitational lensing with controlled systematics; develop a fast, differentiable forward model that evolves and ray-traces through nonlinear matter structures; and introduce a flexible field-level correction for baryonic feedback effects. I also discuss adjacent topics, including a map-level test of the isotropy of cosmic expansion and the decontamination of atmospheric foregrounds in ground-based CMB data. Collectively, these results demonstrate how the field-level approach can be made accurate, computationally feasible, and robust for cosmological inference.