AI Spots Star’s Fatal Spiral Into a Black Hole, Then It Explodes

The blast, named SN 2023zkd, was first discovered in July 2023 by the Zwicky Transient Facility. A new artificial intelligence algorithm designed to scan for unusual explosions in real time first detected the explosion, and that early alert allowed astronomers to begin follow-up observations immediately - an essential step in capturing the full story of the explosion. By the time the explosion was over, it had been observed by a large set of telescopes, both on the ground and from space. 

The scientists think the most likely interpretation is that the massive star was locked in a deadly orbit with the black hole. As energy was lost from the orbit, their separation decreased until the supernova was triggered by the star's gravitational stress as it partially swallowed the black hole.

"Our analysis shows that the blast was sparked by a catastrophic encounter with a black hole companion, and is the strongest evidence to date that such close interactions can actually detonate a star," said Alexander Gagliano, lead author of the study and fellow at the NSF Institute for Artificial Intelligence and Fundamental Interactions. "Our machine learning system flagged SN 2023zkd months before its most unusual behavior, which gave us ample time to secure the critical observations needed to unravel this extraordinary explosion."

An alternative interpretation considered by the team is that the black hole completely tore the star apart before it could explode on its own. In that case, the black hole quickly pulled in the star's debris and supernova emission was generated when the debris crashed into the gas surrounding it. In both cases, a single, heavier black hole is left behind.

Located about 730 million light-years from Earth, SN 2023zkd initially looked like a typical supernova, with a single burst of light. But as the scientists tracked its decline over several months, it did something unexpected: it brightened again. To understand this unusual behavior, the scientists analyzed archival data, which showed something even more unusual: the system had been slowly brightening for more than four years before the explosion. That kind of long-term activity before the explosion is rarely seen in supernovae.

Detailed analysis revealed that the explosion's light was shaped by material the star had shed in the years before it died. The early brightening came from the supernova's blast wave hitting low-density gas. The second, delayed peak was caused by a slower but sustained collision with a thick, disk-like cloud. This structure-and the star's erratic pre-explosion behavior-suggest that the dying star was under extreme gravitational stress, likely from a nearby, compact companion such as a black hole.

grz-band color image of the explosion site of SN 2023zkd from the Dark Energy Camera Legacy Survey (DECaLS) Data Release 9 (A. Dey et al. 2019). The SN location is denoted by the red circle with radius 1″, and the image scale is given in the bottom left. The host galaxy at image center is PS0 J154847.495+091200.90, with a spectroscopic redshift of z = 0.0560 ± 0.0002.

"2023zkd shows some of the clearest signs we've seen of a massive star interacting with a companion in the years before explosion," said V. Ashley Villar, a CfA assistant professor of astronomy in the Harvard Faculty of Arts and Sciences and a co-author on the study. "We think this might be part of a whole class of hidden explosions that AI will help us discover."

"This discovery shows how important it is to study how massive stars interact with companions as they approach the end of their lives," said Gagliano. "We've known for some time that most massive stars are in binaries, but catching one in the act of exchanging mass shortly before it explodes is incredibly rare."

With the Vera C. Rubin Observatory recently unveiling its first images and preparing to survey the entire sky every few nights, this discovery marks a glimpse of what's to come. Powerful new observatories, combined with real-time AI systems, will soon allow astronomers to uncover many more rare and complex explosions and begin to map how massive stars live and die in binary systems. 

The Young Supernova Experiment will continue to complement Rubin by using the Pan-STARRS1 and Pan-STARRS2 telescopes to identify supernovae shortly after explosion. This approach offers a cost-effective way to study the dynamic nearby universe.

"We're now entering an era where we can automatically catch these rare events as they happen, not just after the fact," said Gagliano. "That means we can finally start connecting the dots between how a star lives and how it dies, and that's incredibly exciting."

The authors used data from a large number of telescopes including NASA's Neil Gehrels Swift Observatory, the Panoramic Survey Telescope and Rapid Response System, the Asteroid Terrestrial-impact Last Alert System, and a suite of telescopes across the Magellan, MMT, and Las Cumbres Observatories. The Young Supernova Experiment is a collaboration among University of California Santa Cruz, DARK Cosmology Centre, University of Illinois, and principal investigators Vivienne Baldassare (Washington State University), Maria Drout (University of Toronto), Kaisey Mandel (Cambridge University), and V. Ashley Villar (Harvard).

These results are being published in the latest issue of the Astrophysical Journal.