Towards improved statistical analyses of the cosmic dawn and epoch of reionisation

Abstract

We are witnessing the evolution of cosmology into a precision science. When the universe cools after the big bang, hydrogen atoms eventually form releasing the cosmic microwave background. Measurements of this, combined with direct observations of fully formed galaxies through telescopes inform us about the seeding of structure across the cosmos. In-between these milestones there is little-to-no observational data, particularly regarding the first stars and their growth into mature galaxies. Primordial hydrogen must eventually progress into everything we see today, and we know from local observations of galaxies that the remaining hydrogen must also become ionised again. A new generation of radio telescopes aim to detect the cosmological 21cm line of these hydrogen atoms in order to observe these events directly. This will provide unmatched insight into the formation of the first stars and their progression into the early galaxies responsible for reionisation. As never seen before, these new instruments will provide tomographical maps of our universe that fill in a significant proportion of the unobserved universe. Scientists across the globe hope to witness the cosmic dawn within the next decade. This work is focused on improving the statistical analyses techniques that probe the first star formation and when the early universe is being reionised - the cosmic dawn and epoch of reionisation respectively. The majority of this work looks at Bayesian model selection, which is a robust method for quantifying how well models fit observations. We apply it within the context of high redshift 21cm power spectrum observations, comparing a variety of scale and morpho- logical implementations of the reionisation process. We also use it to look into distinguishing effects caused by the remnants of the first stars and the statistical benefits of including UV observations. We then turn to improving the likelihood statistic, required for Bayesian analyses. The Morlet transform is implemented as an alternative to the Fourier transform within the power spectrum to enable the whole observational light-cone to be analysed at once. Bayesian analyses are incredibly promising and with this work we are a step closer to statistical machinery that provides objective conclusions about the quality of models when presented with data. Synergy between observational methods is important to maximise the discerning power of any Bayesian method. Now that observations are becoming more precise within cosmology, we can begin to constrain primordial astrophysics.