Bayesian forward modeling of galaxy clustering
Nhat-Minh Nguyen · Elektronische Hochschulschriften der LMU München (Ludwig-Maximilians-Universität München) · 2020
With the future large-scale structure (LSS) surveys being on the horizon, precision cosmology is seeing a unprecedented opportunity to constrain cosmological parameters and differentiate cosmological models. Such opportunities naturally bring also unparalleled challenges -- specifically in the form of understanding, examining and, especially, combining various datasets. Bayesian forward modeling and inference, in this context, provides a consistent and transparent framework to extract information from separate datasets while accounting for multiple systematic sources. This thesis is a dedicated effort to bring this framework one step closer to being ready for the upcoming challenges posed by high-precision Cosmic Microwave Background (CMB) experiments and large-volume galaxy redshift surveys. We systematically examine the constraining power of the Bayesian forward modeling approach to galaxy clustering on both cosmological and astrophysical observables, namely the initial conditions of our Universe, the clustering amplitude of galaxies and the kinematic Sunyaev-Zel'dovich (kSZ) effects of galaxy clusters. While the first two focus only on halo clustering in N-body simulation, the last one brings together observational datasets from separate experiments and surveys: the Planck CMB experiment, the Sloan Digital Sky survey (SDSS) and the maxBCG cluster catalog, which include both common and different sources of systematics. We find in chapter 5 that the Bayesian forward inference approach is able to, on large scales, recover up to ~90% the input initial conditions of the GADGET-2 simulation using halos identified in the same simulation as tracers. The framework is robust regarding to choices of gravitational forward model for the matter density fields and deterministic bias model for tracers. The LSS likelihood, on the other hand, might play an important role for unbiased inference of not only the initial conditions, but also the cosmological parameters. This is demonstrated in chapter 7, where we are able to recover the input $\sigma_8$ of the same simulation with systematic error under ~10%, using a Fourier-space likelihood derived from the effective field theory (EFT) approach to LSS with rigorously controlled theoretical systematics. In chapter 6, we use results from the Bayesian forward reconstruction of the BOSS/SDSS3 volume to measure the large-scale bulk flow and kSZ signal of maxBCG catalog. We find evidence of the kSZ effect at approximately 2-sigma, consistently in individual- as well as multi-scale measurements. Our reported signal-to-noise is the first to include uncertainties from the velocity reconstruction in this type of measurement.