Extracting astrophysics from long-lived gravitational wave signals with the Laser Interferometer Space Antenna

Chapman-Bird, Christian Edward Anthony (2024) Extracting astrophysics from long-lived gravitational wave signals with the Laser Interferometer Space Antenna. PhD thesis, University of Glasgow.

Full text available as:
[thumbnail of 2024Chapman-BirdPhD.pdf] PDF
Download (35MB)

Abstract

The Laser Interferometer Space Antenna (LISA) mission is a space-based gravitational wave (GW) detector that will operate in the mHz frequency band, and is expected to observe a rich variety of astrophysical GW sources including Galactic binaries (GBs), massive black hole binaries (MBHBs) and extreme-mass-ratio inspirals (EMRIs). While the analysis and astrophysical interpretation of LISA data has been growing as a field over the last few decades, the adoption of LISA by the European Space Agency (ESA) in January 2024 places well-defined goalposts (the launch of the mission in a decade’s time) that present an urgent need for rapid development of waveform models and data analysis methods. This thesis addresses this need, describing tools for the accurate and efficient analysis of LISA data.

The enormous astrophysical potential of LISA observations is only accessible if techniques developed for its extraction can be constructed and extensively validated. In order to ensure that analysis pipelines will perform optimally on the real LISA dataset once the mission flies, they must be tested on realistic simulations of the instrument. However, the infrastructure required for these simulations has only recently been developed. Part of this thesis focuses on the validation of these simulation tools, assessing the accuracy with which injected GW signals are recovered with statistical techniques. The essential capability of time delay interferometry (TDI) to suppress instrumental noise sources by orders of magnitude without degrading GW signals in the data stream is demonstrated for realistic LISA simulations. The results of this work, which were presented to ESA as part of the successful LISA adoption effort, lay the necessary groundwork for future investigations of the performance of the LISA instrument and the algorithms constructed to analyse its observations.

While LISA is expected to measure the parameters of astrophysical GW sources with great precision, this in turn requires waveform models for these signals to be highly accurate. However, these accurate models must also be computationally inexpensive if LISA data analysis is to be a feasible prospect. This is particularly pertinent for EMRIs, which produce complicated and long-lived waveforms that must be carefully modelled to avoid systematic biases. In this thesis, EMRI waveform models are augmented with machine learning (ML )techniques to improve their efficiency. By substituting inefficient operations in the waveform model with a neural network, a significant reduction in computational cost is attained without loss of accuracy. The resulting streamlined EMRI waveform model is an order of magnitude faster than existing techniques and can be used to rapidly infer the parameters of EMRI signals. As EMRI waveform models improve, their complexity will grow; this is reflected in the techniques presented in this thesis, which can be readily extended as EMRI waveform models become increasingly sophisticated.

Similarly to the analysis of individual GW sources, population studies with LISA observations will be computationally expensive. Accurate modelling of selection effects is necessary for unbiased population inferences, but the high computational cost of waveforms for LISA sources makes selection biases prohibitively expensive to correct for with standard techniques. The latter part of this thesis focuses on addressing this limitation by modelling the detectability of GWsources with ML methods. Investigating EMRI population inference as a representative example, selection effects are accurately modelled by neural networks that are orders of magnitude more efficient than standard techniques. Embedding this method in a Bayesian hierarchical inference framework, we rapidly constrain EMRI population parameters from simulated detection catalogues and demonstrate that the resulting inferences are free of systematic bias. The methodology we develop is completely agnostic to the waveform and population models used, and is sufficiently flexible to be applied to the complicated population inference problems that will be encountered in the hierarchical analysis of LISA observations.

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
Funder's Name: Science & Technologies Facilities Council (STFC)
Supervisor's Name: Woan, Professor Graham and Berry, Dr. Christopher
Date of Award: 2024
Depositing User: Theses Team
Unique ID: glathesis:2024-84613
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 09 Oct 2024 13:29
Last Modified: 09 Oct 2024 14:07
Thesis DOI: 10.5525/gla.thesis.84613
URI: https://theses.gla.ac.uk/id/eprint/84613
Related URLs:

Actions (login required)

View Item View Item

Downloads

Downloads per month over past year