Enhancing searches for gravitational waves from short transient bursts

Smith, Leigh (2025) Enhancing searches for gravitational waves from short transient bursts. PhD thesis, University of Glasgow.

Abstract

The advanced detector era of gravitational wave (GW) searches has been highly successful in the detection of compact binary coalescence (CBC) events throughout three complete observing runs, while more detections are expected with the fourth run currently underway. As the search for GWs continues with increased detector sensitivities, it is expected for sources beyond CBCs to be detected. One example of an expected yet so far undetected source are short transient GW bursts; this signal type encapsulates a wide range of astrophysical sources with varying signal morphologies. The detection of transient bursts often relies upon search algorithms which hold minimal assumptions on a given signals morphology, referred to as un-modelled searches. One particular un-modelled search is the coherent WaveBurst (cWB) algorithm, which bases the detection of GWs upon excess coherent energy across a network of detectors. The weakly-modelled nature of cWB makes it sensitive to a wide range of transient GW signals, however also makes it highly susceptible to spurious transient noise artefacts known as glitches. Glitches directly effect the detection capabilities of searches, and the employment of noise mitigation techniques within searches is required to separate them from GW signals. The development of such noise mitigation techniques is crucial in optimising the sensitivities of searches as new detector upgrades introduce new sources of glitches into the data. The work presented here explores the enhancement of Gaussian mixture modelling (GMM) as a noise mitigation tool to the cWB algorithm in the search for short-duration GW transients. GMM allows for the populations of noise and signal to be modelled over a set of representative attributes, aiding in the classification of GW signals against glitches.
Investigations into various aspects of model training are executed in order to increase the reliability of the GMM methodology. Specifically, the analysis is made robust to a wide range of signal morphologies by removing a bias in the training set, and new approaches to optimise models are chosen to increase the accuracy of the analysis. Through initial tests, we show that the modification the GMM methodology also results in increased sensitivity to a selection of expected burst sources. Following this result, the performance of the enhanced cWB+GMM algorithm is further evaluated through extensive testing with data from the third LIGO-Virgo-KAGRA (LVK) observing run. For both 2- and 3-detector networks, we show that the GMM methodology obtains significant sensitivity improvements for Gaussian pulse and cosmic string waveforms compared to those obtained by previous cWB post-production methodologies. Thus, we demonstrate the ability of GMM to effectively mitigate the dominant source of noise for un-modelled searches: blip glitches. Through these tests it also shown that with the GMM methodology, the 3-detector network achieves sensitivities similar to those of the 2-detector network, which has not previously been possible with the cWB algorithm due to high glitch rates. Further comparison of the GMM method with other post-production pipelines highlight that it is competitively sensitive to wide range of expected burst sources while making minimal assumptions on morphology, proving it is beneficial to apply it in the generic search for GW transients.
We employ the cWB+GMM pipeline in the offline all-sky search for short-duration, low-frequency GWs in the LVK fourth observing run, presenting details on search configuration investigations and results with increased sensitivities. The newly enhanced GMM-based search detects 13 confident CBC events. The loudest non-CBC event is observed with significance of inverse false alarm rate (iFAR) equal to 1.75 years, however initial investigations indicate that this is likely an artefact of noise. Despite no confident burst events being detected, we show that the GMM observes significant sensitivity improvements across the short-duration low-frequency parameter space compared to those obtained in O3. Additionally, the implementation as an offline search highlighted many avenues for future improvements, with insight into GMM behaviour and initial investigations into dominant sources of glitches.
We extend the use of GMM as a noise mitigation tool in GW searches by applying it in a targeted sensitivity study for parabolic radiation-driven capture systems. The approach of a targeted GMM model is introduced, and performance is tested against against standard and generic GMM post-production approaches. We demonstrate that the use of a targeted GMM has the potential to significantly increase sensitivities, however further investigations into the construction of an optimised training set will refine this application further. Through this study we also conclude that high mass radiation-driven black hole capture systems with equal mass ratio and initial angular momentum of 0.9 may be detected with sensitive distance of 1.74Gpc, however upper limits of rates we can achieve with current sensitivities are not competitive with literature.
Overall, the development and extensive testing of the GMM post-production methodology to cWB proves that it is an effective technique in the mitigation of the dominant source of transient noise in un-modelled searches, increasing search sensitivities to expected burst sources across the parameters space. Furthermore, it is shown that the application to a 3-detector network and targeted searches for transient burst sources are promising implementations for the future.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Additional Information: Supported by funding from the Science and Technologies Facilities Council (STFC).
• 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 and Technology Facilities Council (STFC)
• Supervisor's Name: Heng, Professor Ik Siong
• Date of Award: 2025
• Depositing User: Theses Team
• Unique ID: glathesis:2025-84905
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 17 Feb 2025 13:39
• Last Modified: 20 Mar 2026 11:07
• Thesis DOI: 10.5525/gla.thesis.84905
• URI: https://theses.gla.ac.uk/id/eprint/84905

Related URLs

• Article DOI
• Article DOI