Getting ready for new Data: Approaches to some Challenges in Cosmology

Abstract

Cosmology is entering an era of data abundance. With numerous experiments coming online and drastically reducing statistical uncertainty, theory needs to follow suit in its ability to model small-scale effects and to make maximum use of the available information. Numerical simulations and machine learning will likely play an important role in this program. At the same time, unresolved cracks in the standard model call for model-building efforts. In Chapter 2, I investigate a theoretically attractive proposal to address one of these cracks, the Hubble tension. I show that a simplified model of baryon clumping prior to recombination (as in primordial magnetic fields scenarios) cannot solve the Hubble tension due to subleading corrections to the small-scale CMB. In Chapters 3 and 4, I argue that the Sunyaev-Zel’dovich effects will enable us to refine our understanding of small-scale energy input (baryonic feedback). In Chapters 5 and 6, I construct deep learning surrogate models for the Sunyaev-Zel’dovich effects that can reduce the need for expensive hydrodynamic simulations. Chapter 7 presents an alternative machine learning method, symbolic regression, applied to the thermal Sunyaev- Zel’dovich effect. The developed machinery will be applicable to upcoming CMB measurements from Simons Observatory, CMB-S4, and balloon-bourne experiments. In the final two chapters, I consider higher- order summary statistics for late-time observables of the large scale structure. First (Chapter 8), I infer a constraint on the matter clustering parameter S8 from the probability distribution function of Hyper Suprime Cam weak lensing convergence maps. This analysis constitutes a pathfinder for non-Gaussian statistics in Rubin/LSST. Second (Chapter 9), I use cosmic voids identified in Sloan Digital Sky Survey data to constrain the sum of neutrino masses. The unknown statistical distribution of the void statistics necessitates neural implicit likelihood estimation. Measurements with DESI are currently underway and will enable us to scale up this type of analysis.