天文学：AI の手法で星の合体をより迅速に検出

A new machine learning approach could speed up the way astronomers pinpoint the location of binary neutron star mergers, according to a study in Nature this week. The approach uses an algorithm to study the gravitational wave emissions from neutron star mergers, taking just one second to detect and localize a merger when a signal hits the Earth. Such an automated detection of gravitational waves from star mergers could offer new insights into these cosmic events.

Gravitational wave signals from binary neutron star mergers can be picked up by detectors on Earth, but rapid follow-up observations are needed to understand the composition of the stars and what happens when they collide. Traditional methods are not always able to assess the data fast enough to point telescopes towards the source in the crucial moments post-merger. Machine-learning-based approaches can speed up the analysis, but challenges associated with the length and complexity of the signals result in low-accuracy predictions of the location of the merger.

Maximillian Dax and colleagues present a machine learning method, named DINGO-BNS, that can characterize and localize binary neutron star mergers in just one second following gravitational wave detection with high accuracy. The algorithm is not only notably faster than earlier iterations but also 30% more accurate in its results. The detailed information obtained by DINGO-BNS can be used to determine which events are most worthy of expensive telescope time, the authors note. They suggest that, in the future, this method could be adapted for other types of gravitational wave sources, improving our understanding of the universe.