Author : Mehdi Rezaie
Publisher :
Page : 198 pages
File Size : 42,39 MB
Release : 2021
Category : Astronomical surveys
ISBN :
Large-scale structure (LSS) of the Universe traced by galaxies is one of the essential probes of dark energy, dark matter, neutrino masses, nature of gravity, and statistical properties of primordial fluctuations. The clustering of primordial fluctuations from the cosmic microwave background to LSS provides a standard ruler test to study the dynamics of the cosmic expansion and the mysterious dark energy. Additionally, LSS can be utilized to reconstruct the primordial features and the statistical properties of the initial conditions of the Universe. Future galaxy redshift surveys are designed to extend aggressively to higher redshifts to reach a greater cosmic volume for improved precision as well as for studying the dynamics of dark energy further back in time. For instance, the Dark Energy Spectroscopic Instrument (DESI) will observe millions of galaxies and quasars, producing a three–dimensional map probing the Universe across 10 billion light-years or out to the redshift of 3.5. With enormous data volume, we can address fundamental cosmological problems with higher statistical precision, but such analyses demand more advanced methods and theoretical modeling of systematics. Emmission line galaxies (ELGs) are star-forming galaxies that populate the high redshift universe and therefore are promising tracers for LSS that will be targeted in future galaxy surveys. Quasi-stellar objects or quasars (QSOs) are distant galaxies that host massive black holes and the accreting black holes make these objects bright enough to be used as targets for high redshift galaxy surveys. However, the measurements of such targets are subject to various observational systematic effects that are still largely unknown. Mitigating such effects is crucial for deriving unbiased and precise cosmological constraints. This dissertation addresses the challenge of observational systematics by comparing the results of various approaches. My dissertation consists of three parts: In the first part, I establish a deep learning approach to model and mitigate the effects of observational systematics in the large-scale clustering of galaxies. I implement, validate, and apply the method to log-normal mock datasets as well as ELGs from real imaging data that will be used for targeting in DESI. I demonstrate that the nonlinear approach based on neural networks reduces observational systematics more efficiently than conventional linear regression. In the second part, I enhance the methodology for the final sample of quasars from the extended Baryon Acoustic Oscillation Survey (eBOSS). I compare the performance of different mitigation techniques and show that there is no signature of nonlinear systematics in the data. In the third part, I use the resulting improved data to constrain the initial conditions of the Universe. The methods and tools developed in this dissertation pave the path of probing the large-scale structure using data from the upcoming next generation of galaxy redshift surveys such as DESI and Euclid. This thesis presents my contribution as a member of the DESI and eBOSS collaborations.