Cosmoglobe: Deconstructing the microwave and sub-mm sky

Abstract

The cosmic microwave background is one of the most prominent sources of cosmological information. Modern cosmological theories rely upon both small- and large-scale observations of the CMB. As a necessity, these observations rely heavily upon the ability to properly model the instrument making the observations as well as emission from the Galaxy. This thesis sets out to describe a broad series of topics with respect to CMB observations, underlining the interplay between the constituent parts. The goal of CMB observations today is to set better constraints on cosmological theories, particularly aiming for a detection or exclusion of direct evidence for inflation. It is vital to introduce several topics in this thesis in order to talk about the details of microwave and sub-mm data analysis. The first part of the manuscript introduces modern cosmology which is the framework for our current understanding of the universe at large physical and temporal scales. This is followed by a primer on the microwave sky, which is full of several non-cosmological signals, commonly referred to in the field as foregrounds. Additionally, several historical, contemporary, and future observatories are introduced, with a few common telescope systematics introduced in the following chapters. Finally, a detailed introduction to the methodologies utilized in this thesis is given. The work outlined here emphasizes the tight coupling between each of these components and highlights the benefits of including the modeling of these within a unified framework. This is directly demonstrated within the papers presented here. Modeling the instrument and Galactic signals together at the time ordered level has made the 44 GHz LFI data, which was previously discarded because of the high level of unmodeled systematics, once again usable for cosmological analysis by the BeyondPlanck project. The Cosmoglobe Data Release 1 (DR1) has similarly recovered the WMAP W-band polarization data for cosmological analysis for the first time, which was made possible due to the common calibration allowed by jointly analyzing LFI and WMAP data. The Commander framework used for the above analyses provided a suite of frequency maps, each of which represent a different realization of what the sky and instrument look like. The frequency maps produced by the BeyondPlanck project were used to set upper bounds for the maximum polarization fraction for large-scale anomalous microwave emission to be under 3.5 %, marginalizing over LFI systematics. Modeling of instrumental systematics is an integral part of CMB data analysis, with a select few instrumental systematics discussed in the latter parts of the manuscript. Nonlinearities in the LFI analog-to-digital converters inspired a new set of corrections to be created. These corrections, which are by nature degenerate with the calibration of the instrument, show that neglecting this systematic effect may obfuscate a weak cosmological signal. Systematic effects such as calibration estimates and instrument bandpasses couple directly to estimates of the microwave sky. This complicates determination of what the sky truly looks like as well as what data the instrument is actually taking. The tight coupling between instrumental and sky estimates has inspired the use of calibration priors for the WMAP K-band in Cosmoglobe DR1. Additionally, knowledge about the interstellar medium has spurred the implementation of informative priors from data outside the microwave/sub-mm regime, such as the application of neutral hydrogen data as a prior for the anomalous microwave emission component within the BeyondPlanck project. The above examples illustrate the importance of bringing not only as much data together as possible, but to also compile broad types of experience to get the best view of the CMB as possible. The research presented here has primarily focused on component separation, coming in the form of the development of an independent pixel-based component separation software, dang, in addition to contributing to and utilizing the Commander software. In an attempt to practice pedagogy, the introduction to Bayesian component separation methods is in depth and explicit, hopefully serving as a starting point for future researchers who wish to implement similar frameworks for their own work.