NASA's Swift satellite and AI reveal distance of farthest gamma-ray burst

The advent of AI has been hailed by many as a society-changer, as it opens up a world of possibilities for improving almost every aspect of our lives.

Astronomers are now literally using AI to measure the expansion of the universe.

Two recent studies led by Maria Dainotti, visiting professor at UNLV’s Nevada Center for Astrophysics and assistant professor at the National Astronomical Observatory of Japan (NAOJ), incorporated multiple machine learning models to bring new levels of precision to distance measurements of gamma-ray bursts (GRBs), the most luminous and violent explosions in the universe.

In just a few seconds, GRBs release as much energy as the Sun does in its entire life. Because they are so bright, GRBs can be seen at a wide range of distances, including to the edge of the visible universe, helping astronomers track the oldest and most distant stars. However, due to limitations in current technology, only a few known GRBs have all the observational characteristics that astronomers need to calculate the distance at which they originate.

Dainotti and her team combined GRB data from NASA's Neil Gehrels Swift Observatory with multiple machine learning models to overcome the limitations of current observational techniques and more precisely estimate the proximity of GRBs whose distances are unknown. Because GRBs can be observed from far away or relatively close by, knowing where they occur can help scientists understand how stars evolve over time and how many GRBs occur in a given space and time.

“This work advances the state of the art in both gamma ray astronomy and machine learning,” Dainotti said. “Subsequent research and innovation will enable even more reliable results, helping us answer some of the most pressing cosmological questions, including the earliest processes in the universe and how it has evolved over time.”

AI Pushes the Limits of Deep Space Observations In one study, Dainotti and Aditya Narendra, a final-year PhD student at Jagiellonian University in Poland, used several machine learning techniques to precisely measure the distances of GRBs observed by ground-based telescopes, including the space-based Swift Ultraviolet/Visible Optical Telescope (UVOT) and the Subaru Telescope. The measurements were based solely on the properties of the GRBs, which are unrelated to distance. The study was published May 23. Astrophysical Journal Letters.

“The results of this study are so accurate that the predicted distances can be used to determine the number of GRBs (called the rate) within a particular volume and time, which is very close to the actual observed estimates,” Narendra said.

Another study led by Dainotti and international collaborators used data from NASA's Swift X-ray Telescope (XRT) on afterglows called long GRBs to successfully measure the distance of GRBs using machine learning. GRBs are thought to arise in different ways. Long GRBs occur when massive stars reach the end of their lives and explode in spectacular supernovae. Another type, called short GRBs, occurs when the remnants of dead stars, such as neutron stars, gravitationally fuse and collide into each other.

Dainotti says what's novel about this approach is that it combines multiple machine learning techniques to improve their overall predictive power. The method, called Super Learner, assigns each algorithm a weight that ranges from 0 to 1, with each weight corresponding to the predictive power of that technique.

“The advantage of the Super Learner is that the final predictions will always outperform any single model,” Dainotti said. “The Super Learner is also used to discard the algorithms with the poorest prediction accuracy.”

The study was published on February 26th. The Astrophysical Journal, Supplementary Seriesreliably estimated the distances of 154 long GRBs with unknown distances, significantly increasing the number of this type of burst with known distances.

Answering difficult questions about GRB formation

A third study, published Feb. 21, Astrophysical Journal Letters A team led by Stanford astrophysicist Vahe Petrosian and Dainotti used Swift's X-ray data to answer a puzzling question by showing that the rate of GRBs doesn't follow the rate of star formation, at least at relatively small distances.

“This suggests that nearby, long-distance GRBs may be produced not by the collapse of a massive star, but by the nuclear fusion of a very dense object, like a neutron star,” Petrosian said.

With support from NASA's Swift Observatory Guest Researcher Program (Cycle 19), Dainotti and his colleagues are now working to make their machine learning tools available to the public through an interactive web application.