Deciphering dark energy: the adventures of Steve, Barry, and Pippin.

Abstract

The dark universe poses some of the largest unsolved questions in modern astronomy. Investigations into the nature of dark matter and dark energy have sought to discriminate primarily between different physical models or modifications of General Relativity. Unfortunately, obvious candidates for either phenomenon have failed to step forward. In their absence, cosmological surveys are rapidly increasing statistical strength to measure the properties of the dark side of the Universe more precisely. With this increase in precision, observational and theoretical systematics in our cosmological probes becomes our chief concern. In this thesis, I present work which focuses on improved control of systematics in two cosmological probes – Type Ia supernovae and Baryon Acoustic Oscillations. To begin, I provide a general framework for representing experiments that suffer from selection effects, to mimic Malmquist bias. I take this generalised result and apply it to supernova cosmology, implementing Steve, a Hierarchical Bayesian model that incorporates a more flexible and rigorous treatment of systematic uncertainty than the traditional data-correcting methods. I find no significant biases in the model after validating on a set of two hundred fiducial simulations. Additionally, I find increased, but more representative, uncertainty than the other leading method in the field, “Beams with Bias Corrections” (BBC). Evidence of slight bias depending on intrinsic scatter model is found in both methods, and indicates an area requiring improved determination in future surveys. I then apply my new BHM model to the three-year spectroscopically confirmed Type Ia supernova sample from the Dark Energy Survey (DES), which is a five-year survey of the southern sky. DES observed ten supernova fields with high cadence, gathering hundreds of spectroscopically confirmed Type Ia supernovae and thousands of photometrically classified supernova. The results of the spectroscopically confirmed dataset are compared when analysed using both my model and the standard BBC method. I summarise the three-year analysis, providing information on the dataset, analysis pipeline, and global systematics. I present interim work on the five-year analysis, which is a photometric data sample with approximately an order of magnitude more events than the three-year dataset. For this analysis, I wrote an end-to-end pipeline for SN Ia analysis called Pippin. The classifiers implemented within the pipeline are introduced and compared, and I present statistically rigorous investigations implemented with Pippin showing that the cosmological impact of classification uncertainty will be a sub-dominant systematic. Finally, I turn to Baryon Acoustic Oscillations, presenting a new code framework, Barry, created to streamline model implementation, testing and validation for BAO cosmology. I detail the four implemented BAO models within the code framework and verify that, with the statistical power of modern surveys, three out of the four models show no evidence of bias. For the fourth model, I show that the bias found is created due to reduced degrees-of-freedom in the model, coupled with their choice of non-linear model. This framework should be useful for upcoming large-scale-structure surveys such as DESI to provide rapid model evaluation and testing.