Newswise — KINGSTON, R.I. – March 6, 2025 – Observations of the merger of binary neutron stars are vital to the growing field multi-messenger astronomy. The collision of these massive star remnants, occurring millions of light years from Earth, emit gravitational waves followed by light. They provide unique opportunities for the study of gravity and matter under extreme conditions, with exciting implications for nuclear physics and cosmology.
But traditional data-analysis methods are too slow to fully interpret gravitational waves from these multi-messenger signals, while fast analysis is critical in directing time-sensitive electromagnetic observations because of the length and complexity of the signals.
In a in the journal Nature, an interdisciplinary team of researchers presents a novel approach that uses a neural network to analyze gravitational waves from binary neutron star mergers that is significantly faster and more accurate than traditional methods. So, fast it can be done in the blink of an eye – even before the merger of the neutron stars is fully observed.
“Rapid and accurate analysis of gravitational wave data is important in locating the source of the merger and providing that information to observatories so they can target their telescopes to observe all accompanying signals,” said Maximillian Dax, a Ph.D. student at Max Planck Institute for Intelligent Systems in Germany who led the study.
“With this process, we are able to analyze gravitational waves several thousand times faster than the best traditional method. What used to take about an hour, we can do in a second,” says , a University of Rhode Island adjunct professor in and one of 10 contributing authors on the study.
The first detection of gravitational waves – from the merger of black holes — occurred less than a decade ago with the observation by the NSF-funded Laser Interferometer Gravitational-Wave Observatory (LIGO), leading to the Nobel Prize in Physics being awarded to the founders of the (LSC). Two years later, the LSC made the first observation of gravitational waves produced by the merger of two binary neutron stars.
“The amazing thing is that if you have a merger of two neutron stars, they also emit radiation in the electromagnetic spectrum,” said Pürrer, a member of the LIGO Scientific Collaboration since 2013. URI has been a member of the LSC since 2017 and is the only member in Rhode Island and one of very few in New England.
“That allows you to learn a lot of very interesting things, not just properties such as the masses of the neutron stars, how squishy the neutron stars are, how fast they were spinning and how far away the binary was,” he said about the multiple signals. “But also, we know the universe is expanding and observing neutron star mergers provide an alternative way of measuring that expansion.”
But to be able to capture the data from electromagnetic signals, observatories must act fast and have accurate findings.
In their study, “Real-time inference for binary neutron star mergers using machine learning,” the researchers present a machine learning algorithm – DINGO-BNS – that saves valuable time in analyzing gravitational waves from the merger of binary neutron stars. (DINGO-BNS stands for Deep INference for Gravitational-wave Observations for Binary Neutron Stars.) The algorithm can fully characterize systems of merging neutron stars in about a second. The fastest traditional methods take about an hour.
“We have to tell our electromagnetic partner observatories in less than a minute, a few seconds if we can,” added Pürrer, who contributed code for the algorithm and helped shape the paper. “That’s really the clincher. If we wait too long, the signal from the kilonova explosion after the merger might already have decayed and they may not be able to find it. Detecting both signals is the holy grail of multi-messenger astronomy. Kilonovas, which are less bright than supernovas, confirm that binary neutron star mergers are a key site where heavy elements like gold and platinum are made in the universe.”
The real-time method—occurring even before the binary neutron stars collide—is also more accurate than current rapid-analysis algorithms. The machine learning framework fully characterizes neutron star mergers in just one second without making approximations that can be less accurate, including quickly determining sky position 30% more precisely, according to the study.
The study brought together a team of researchers from Germany, the United Kingdom, and the U.S. Along with Dax and Pürrer, the team consisted of Stephen R. Green of the University of Nottingham; Jakob H. Macke and Bernard Schölkopf, both of Max Planck Institute for Intelligent Systems; Alessandra Buonanno, Jonathan Gair, and Nihar Gupte, all of the Max Planck Institute of Gravitational Physics; Vivien Raymond of the Gravity Exploration Institute at Cardiff University; and Jonas Wildberger of ELLIS Institute Tübingen.
“The team consisted of machine learning experts who regularly apply cutting-edge methods from computer science to various application domains, and experts steeped in gravitational wave astronomy and data analysis. This study showcases the effectiveness of a successful marriage between modern deep learning methods and physical domain knowledge,” said Pürrer, who is also a computational scientist in in ITS. Pürrer is a member of the collaboration, a URI and University of Massachusetts collaboration focused on gravitational waves research with dozens of researchers.
Pürrer is co-organizing a at the Institute for Computational and Experimental Research in Mathematics this June in Providence, where this line of work will be continued.
“DINGO-BNS could one day help to observe electromagnetic signals before and during the collision of the two neutron stars and be instrumental in preparing the field for the next generation of observatories, such as in the U.S. and in Europe to illuminate unknown physics in the state of matter at ultra-high densities,” Pürrer said.
RSS