Data analysis and scientific exploitation of the CMB B-modes experiment, POLARBEAR

Over the last two decades cosmology has been transformed from a data-starved to a data-driven, high precision science. This transformation happened thanks to improved observational techniques, allowing to collect progressively bigger and more powerful data sets. Studies of the Cosmic Microwave Background (CMB) anisotropies have played, and continue on doing so, a particularly important and impactful role in this process. The huge data sets produced by recent CMB experiments pose new challenges for the field due to their volumes and complexity. Its successful resolution requires combining mathematical, statistical and computational methods all of which form a keystone of the modern CMB data analysis. In this thesis, I describe data analysis of the first data set produced by one of the most advanced, current CMB experiments, Polarbear and the major results it produced. The Polarbear experiment is a leading CMB B-mode polarization experiment aiming at detection and characterization of the so-called B-mode signature of the CMB polarization. This is one of the most exciting topics in the current CMB research, which only just has started yielding new insights onto cosmology in part thanks to the results discussed hereafter. In this thesis I describe first the modern cosmological model, focusing on the physics of the CMB, and in particular its polarization properties, and providing an overview of the past experiments and results. Subsequently, I present the Polarbear instrument, data analysis of its first year data set and the scientific results drawn from it, emphasizing my major contributions to the overall effort. In the last chapter, and in the context of the next generation CMB B-mode experiments, I present a more systematic study of the impact of the presence of the so-called E-to- B leakage on the performance forecasts of CMB B-mode experiments, by comparing several methods including the pure pseudospectrum method and the minimum variance quadratic estimator. In particular, I detail how the minimum variance quadratic estimator in the case of azimuthally symmetric patches can be used to estimate efficiently parameters, and I present an efficient implementation based on existing parallel algorithms for computing Spherical Harmonic Transforms. (author)