Cosmological parameter inference using gravitational waves and machine learning

Stachurski, Federico (2024) Cosmological parameter inference using gravitational waves and machine learning. PhD thesis, University of Glasgow.

Abstract

In 1929, Edwin Hubble’s discovery of the relationship between galaxy distances and their recession velocities unveiled the universe’s expansion, laying the foundation for modern cosmology and the task to measure the Hubble constant, H0. In 1986, Bernard F. Schutz proposed using gravitational waves from compact binary mergers, such as neutron stars and black holes, as a novel method to estimate H0. This innovative approach marked the beginning of a new era in cosmological research, significantly advanced by the advent of gravitational wave detection through the Laser Interferometer Gravitational Wave Observatory (LIGO). The field has since evolved, employing advanced Bayesian techniques and extensive galaxy catalogues to improve the precision of H0 measurements. However, as the sensitivity of detectors increases and the rate of gravitational wave observations grows, computational challenges—particularly in hierarchical Bayesian analysis—pose significant hurdles due to the intensive and time-consuming nature of traditional methods. In response to these challenges, this thesis, under the supervision of Dr. Christopher Messenger and Prof. Martin Hendry, explores the integration of machine learning into cosmological research, specifically focusing on a novel approach called CosmoFlow to extract cosmological information from gravitational waves. CosmoFlow uses Normalising Flows, machine learning models capable of efficiently estimating probability distribution functions of complex datasets, providing a faster and and potentially advantageous approach to hierarchical Bayesian inference of the Hubble constant. Our work demonstrates how CosmoFlow can significantly accelerate the process compared to existing methodologies. Throughout this thesis, we rigorously compare the results of CosmoFlow with those obtained using gwcosmo, a well-established tool in gravitational wave cosmology. By contrasting CosmoFlow with gwcosmo results, we highlight the strengths and limitations of each method, emphasising the potential of machine learning to address existing computational bottlenecks in cosmological analyses. This comparative study aims to contribute to the ongoing efforts to resolve the current 4.4σ tension between different H0 measurement techniques, paving the way for more efficient and accurate future analyses in this rapidly evolving field.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Subjects: Q Science > QB Astronomy; Q Science > QC Physics
• Colleges/Schools: College of Science and Engineering > School of Physics and Astronomy
• Supervisor's Name: Messenger, Dr. Christopher and Hendry, Professor Martin
• Date of Award: 2024
• Depositing User: Theses Team
• Unique ID: glathesis:2024-84817
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 16 Jan 2025 11:31
• Last Modified: 16 Jan 2025 11:32
• Thesis DOI: 10.5525/gla.thesis.84817
• URI: https://theses.gla.ac.uk/id/eprint/84817