Component separation for cosmic microwave background studies

Thesis

The detection of primordial B-mode polarization in the cosmic microwave background (CMB) remains the major outstanding goal of CMB cosmology. However, current upper limits on the strength of the primordial signal mean that it will be sub-dominant to astrophysical foreground emission at all frequencies, over the entire sky. Detecting primordial B-modes therefore becomes a problem in component separation.

In this thesis I address this problem from two angles. First, I present my work analysing data from the C-Band All-Sky Survey (C-BASS), a 4.76 GHz survey covering the whole sky in total intensity and polarization. I describe the point-source detection algorithm developed for C-BASS, and the northern sky point-source catalogue that has been produced using it. This catalogue has allowed us to confirm the accuracy of the C-BASS pointing and flux-density scale, and will form the basis of any C-BASS point source masks. I also present an analysis of the synchrotron power spectra, using C-BASS, WMAP and Planck data. From this it is found the minimum synchrotron contamination to CMB B-modes corresponds to a tensor-to-scalar ratio of r~0.001, with a typical contamination at the level of r~0.01.

Alongside the analysis of C-BASS data, I present a novel implementation of Bayesian parametric component separation. This uses the No-U-Turn Sampler (NUTS) to explore the posterior distribution. NUTS is a gradient-based sampling algorithm with excellent scaling to high dimensions, and also contains important self-diagnostics of potential failures in geometric ergodicity. This is particularly important for the hierarchical foreground model introduced here, which can exhibit highly complex posterior geometries. The hierarchical foreground model fits for hyper-distributions over large sky regions, from which foreground spectral parameters are drawn. This is compared to a complete pooling model, where foreground spectral parameters are assumed to be constant in each sky region. The hierarchical model is able to remove artefacts from the recovered CMB maps without inflating parameter uncertainties, translating through to reduced biases on cosmological parameters.

Authors: Grumitt, RDP

• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English