Gravitational-wave astronomers are building an AI-powered system to find up to 10% more ripples in spacetime from colliding black holes and neutron stars. The problem is that the next generation of detectors will see thousands of events, but the data is messy. Transient noise bursts called "glitches" can mimic or mask real signals, and current analysis software struggles to handle them. The researchers at Royal Holloway, University of London aim to solve this by combining traditional algorithms with machine learning. They will equip the flagship Bilby software—used to measure every confirmed gravitational-wave signal—with the ability to handle glitches, non-stationary noise, and unknown background conditions. They will also introduce a statistical technique called conformal prediction to give reliable uncertainty estimates for signal detections, and build a meta-pipeline that merges multiple search pipelines to maximise efficiency. This is fundamental science. Success would mean more accurate maps of the dark Universe—the black holes and neutron stars that emit no light. It would also help the UK maintain its leading role in gravitational-wave astronomy. Past fundamental research in this field has already yielded unexpected spin-offs in precision timing and data analysis techniques used elsewhere in science and industry.
View original technical description
We are in the midst of a revolution in astronomy. In 2015, the LIGO detectors first observed ripples in the fabric of spacetime caused by two colliding black holes. We call these ripples gravitational waves. This ground-breaking result opened a new form of astronomy that allows us to observe the dark Universe for the first time. The maturing field is entering a new era in the coming years: an improved international network of detectors, including Virgo and KAGRA, will see further, observing thousands of events with ever-greater clarity. However, the marked increase in data quality poses a significant challenge for analysing compact binary coalescence (CBC) signals. This proposal outlines the vision and approach that the newly established RHUL GW group will take to address key issues in the field and contribute to the field's observational science. The RHUL group was established in 2021 by the project lead, Greg Ashton, who has played a central role in the collaboration, co-leading the CBC observational science working group (one of the largest and most active working groups) and delivering the flagship Bilby software used to measure the properties of all CBC signals during O3 and O4. In this Consolidated Grant, the RHUL group aims to cement its place in the field by developing cutting-edge new approaches to the search for and parameter estimation of CBC signals, melding together the best of traditional algorithms with the latest advances in Artificial Intelligence. Specifically, we aim to: A-1 Enhance the search for signals by providing the first quantified and calibrated uncertainties to significance estimates for individual search pipelines. Introducing conformal prediction (a distribution-free approach to uncertainty quantification) to the field of gravitational-wave astronomy, we will provide the first robust uncertainty quantification to the false alarm rate and astrophysical probabilities produced by search pipelines. A-2 Maximise search efficiency by delivering a meta-pipeline that optimally combines independent search pipelines, leveraging their strengths to find up to 10% more signals. Building on A-1, we will use conformal prediction to combine multiple pipelines, increasing their shared efficiency and hence finding more signals. A-3 Deliver fully-robust parameter estimation algorithms that can rapidly measure the properties of signals in the face of the real non-Gaussian non-stationary noise. Non-Gaussian noise poses significant risks to the robust analysis of CBC signals (for example, current low-latency analyses are not robust to the frequent transient noise known as glitches). We will use modelled and machine-learning-based approaches to equip the flagship Bilby inference software with the capacity to handle overlapping glitches, non-stationary noise, and an unknown power spectral density. Thus enabling fully-robust parameter estimation in the observational era. A-4 Detect the stochastic gravitational-wave background arising from the “popcorn” of sub-threshold CBC signals beyond the horizon of the detectors. Using our advances from A-3, we will make leading contributions to the effort to measure the astrophysical stochastic background from unresolved binary black holes. A-5 Use gravitational wave astronomy to educate and enthuse the general public about AI, statistics, and programming Completing these aims, we will maintain and extend the cutting-edge analysis of the LIGO, Virgo, and KAGRA data and cement the newly established gravitational-wave group at RHUL as leaders in the observational science of CBCs. This, in turn, will help to build capacity for gravitational-wave astrophysics within the UK and, through outreach, deliver societal benefits through education.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know