AI-Driven Inference for Gravitational Waves: Accelerating Discoveries in Fundamental Physics

This research combines two timely topics in modern science: gravitational waves and artificial intelligence (AI). It uses AI tools to analyze gravitational-wave data and uncover secrets about our universe.

Gravitational waves are a key prediction of Einstein's revolutionary theory of gravity proposed in 1915. This theory, called general relativity, tells us that gravity is really geometry---a manifestation of the joint curvature of space and time. It describes new phenomena including black holes, neutron stars, and an evolving universe.

Gravitational waves are waves of space and time that are produced by extraordinarily powerful cosmic events, like the merging of black holes and neutron stars. A century after Einstein's prediction, we finally detected these waves on Earth using an experiment called the Laser Interferometer Gravitational-Wave Observatory (LIGO). Gravitational waves carry information about their source.

By analyzing them and comparing to theoretical predictions, we learn about the events that produced them. For black hole mergers, we learn the black hole masses and spins, and also the position and orientation of the system. So far, the LIGO-Virgo-KAGRA (LVK) Collaboration has detected over 90 such events; combining them we learn about the expansion of the universe, the physics of gravity and extreme matter, and the formation of black holes.

This research has three main goals: (1) develop AI tools to analyze gravitational-wave data more quickly and accurately, (2) use these tools to study the data and make new discoveries in astrophysics and cosmology, and (3) use gravitational waves to test our understanding of gravity and fundamental physics.

We have already seen promising results. My collaborators and I have shown that AI tools can reliably reduce analysis times from hours or days to mere seconds while maintaining accuracy. Our approach involves training neural networks to encode theoretical models in such a way that the networks can quickly interpret new data when it arrives.

This approach, known as neural simulation-based inference, is a game-changer for scientific data analysis, and the proposed research aims to develop and apply these methods throughout gravitational wave astronomy. Our plans include continued research on ground-based detectors such as the LVK and the future Einstein Telescope and Cosmic Explorer; new research on space-based detectors such as the Laser Interferometer Space Antenna, which will observe supermassive black holes at the centres of galaxies; research on combining multiple observations using AI tools to learn about populations of black holes and cosmology; AI research adapting state-of-the-art image generation tools to scientific data; and a study to understand in exquisite detail the final stages of a merger-the ringdown-as the combined object settles to a steady-state black hole.

This research will benefit a variety of groups, including the gravitational-wave research community through open-source software development for data analysis. Additionally, the general relativity community will gain from an improved theoretical understanding of black holes. Our work will also benefit the AI for science community by developing new methods and applying them to solve highly challenging problems.

Finally, our project showcases the potential of AI in scientific data analysis and offers a valuable training opportunity for young researchers, contributing to the wider scientific community and UK workforce.