During the first phase of this project, the FIRSTDAWN team developed new tools to analyse and interpret observations from upcoming radio telescopes and inform the development of those experiments. These fall into three main areas: the all-sky 21 cm signal, which is being targeted by single dipole experiments, the non-Gaussianity of the 21 cm signal, which encodes information about the nature of reionization and cosmic heating, and numerical simulations, which provide a framework for interpreting upcoming observations.
For the all-sky 21 cm signal, we built an analysis framework to cope with the challenge of separating the signal from galactic and extra-galactic foregrounds. Our Bayesian analysis pipeline, would exploit both frequency and sky location information to disentangle these components. These work was incorporated into a proposal for a lunar orbing radio dipole experiment - the Dark Ages Radio Explorer (DARE), which has been considered by NASA, and is now being redeveloped for work on REACH, a ground based global 21 cm experiment in South Africa.
Radio telescopes targeting the 21 cm signal from the epoch of reionization have tended to focus on the power spectrum of brightness temperature fluctuations. The power spectrum is a very valuable statistic, but is only complete if the 21 cm signal is a Gaussian random field. In practice, the 21 cm signal will contain many features from the percolation of ionized regions that require additional statistical tests to probe. FIRSTDAWN developed machinery for efficiently calculating the bispectrum from mock data and building the theoretical framework to interpret this for different reionization scenarios. A preliminary attempt to measure this signal was made with MWA data and the results suggest significant promise for future data sets from SKA.
All of this work requires numerical simulations to interpret, since the complex interplay of ionization, heating, and illumination of the intergalactic gas by light from the first galaxies is hard to capture in simple analytic models. Alongside work on improving the existing simulations, we adapted tools from the machine learning community to allow rapid emulation of numerical simulations. These allow a wide parameter space to be explored much more rapidly, making a complete analysis of the relevant models more tractable.
