Improving statistical and computational techniques for supernova cosmology

Abstract

Modern cosmological analysis is focused on investigating the mechanics of the accelerated expansion of our Universe and uncovering the nature of the mysterious dark energy which drives this acceleration. Accelerated expansion was first discovered in the late 1990s using type Ia supernovae (SNe Ia) as standard candles to probe the evolution of our Universe over time. Since this discovery, significant effort has been devoted to improving the statistical and systematic uncertainties present in supernova cosmology to improve the constraining power of these analyses. Statistical uncertainties were improved by observing orders of magnitude more SNe Ia, whilst improvements in systematic uncertainty were driven by developments in every step of the analysis.

These efforts have culminated in the Dark Energy Survey (DES) supernova program. With 1635 SNe Ia, this represents the largest sample of SNe Ia, and thus the greatest constraining power of any supernova sample to date. DES recently released the results of their five-year supernova cosmology program. When combining constraints on a Flat wCDM cosmology from SNe Ia with other cosmological probes, DES find \(\left(\Om{},\w{}\right)=\left(0.321\pm0.007,-0.941\pm0.026\right)\), the first possible evidence of a \(>2\sigma\) divergence from dark energy as a cosmological constant. This has been the prevailing theory of dark energy since its discovery, leading to the \(\Lambda{}\)CDM concordance model of our Universe. However, with the improved constraints provided by the DES analysis, this model is thrown into question for the first time.

To achieve the constraining power required to obtain these results, improving the statistical and systematic uncertainties throughout the entire DES cosmological pipeline is required. These improvements are the result of the work of dozens of researchers around the world. In this thesis, I will focus on my contributions improving Pippin, the overarching pipeline which enables the DES analysis, and developing a new method for evaluating the consistency of the DES results.

The results from the DES analysis are far from conclusive, so if we are to determine the legitimacy of the cosmological constant, further improvements to the statistical and systematic uncertainties are required. Upcoming surveys, such as the Legacy Survey of Space and Time (LSST), aim to observe orders of magnitudes more SNe Ia than DES; however, with these improvements to the sample size come a host of challenges that must be addressed. As part of this thesis, I discuss these challenges and my efforts to develop new statistical techniques to address these challenges. In particular, I investigate likelihood-free, forward-modelled techniques which provide significantly greater control over systematic uncertainties with fewer assumptions, at the cost of much greater computational complexity.