Mocking the Universe: Accurate Simulations in the Era of Large-survey Precision Cosmology

Efficient generation of accurate mock observations of the sky, tailored specifically to near-future large-scale structure and cosmic microwave background surveys, is a key technical challenge in cosmology. The increasing sensitivity and sky coverage of modern surveys probe unprecedentedly-large cosmological volumes, and when coupled with increases to angular resolution, demand simulations that are accurate to smaller scales, yet cover larger volumes, than ever before. The applications of such simulations are numerous, from performing cosmological parameter analysis, to optimizing survey designs and testing analysis pipelines, to making theoretical predictions, but to adequately perform these tasks requires state-of-the-art accuracy and computational efficiency. In this thesis we develop numerical simulations of cosmological structure formation to create synthetic observations, or mocks, of current and near-future cosmic microwave background and large-scale structure surveys. We first present our development and rigorous validation of the mass-Peak Patch method, a simulation technique to rapidly generate the large-scale structure of cosmological volumes for a fraction of the computational cost of alternative methods. Mass-Peak Patch is then used in union with a number of astrophysical sky models to produce the Websky extragalactic CMB simulations, a publicly released set of synthetic mocks of the extragalactic microwave to infrared sky. Next, we focus on the up-and-coming field of intensity mapping. We construct a suite of carbon monoxide line intensity maps, and use these to forecast the constraints on the CO luminosity function with near-future observations. Both the large data volumes and complicated statistical evaluations involved in standard cosmological analyses of survey data has recently led to the adoption of many machine learning methods throughout the field, and we finish this thesis with an investigation of machine learning methods in cosmology. We first overview machine learning in cosmology and discuss the viability and attractiveness of recently-popularized deep learning techniques, and then construct, train, and validate a convolutional neural network for the purpose of aiding cosmological simulations. Efficient simulation tools such as mass-Peak Patch, accurate mocks such as the Websky extragalactic maps, and machine learning techniques (when applied correctly), will enable precision analysis of the next generation of cosmological data sets.Ph.D