A team of astronomers led by the University of California, Santa Cruz has identified an unusual stellar explosion involving a black hole and a massive star, using artificial intelligence to aid in the discovery. The event, designated SN 2023zkd, was first detected in July 2023 through a new AI algorithm that scans for atypical cosmic explosions in real time. This early detection allowed researchers to conduct immediate follow-up observations.
The explosion was observed with various telescopes on Earth and in space, including two at Hawaiʻi’s Haleakalāa Observatory operated by the Young Supernova Experiment (YSE), which is based at UC Santa Cruz. YSE surveys a significant portion of the night sky every three days and has discovered thousands of cosmic explosions soon after they occur.
“Something exactly like this supernova has not been seen before, so it might be very rare,” said Ryan Foley, associate professor of astronomy and astrophysics at UC Santa Cruz. “Humans are reasonably good at finding things that ‘aren’t like the others,’ but the algorithm can flag things earlier than a human may notice. This is critical for these time-sensitive observations.”
Researchers concluded that the most likely scenario involved a collision between the star and its black hole companion as their orbits decayed over time. This interaction appears to have triggered the supernova due to gravitational stress from the black hole partially swallowing its companion.
The findings were published August 13 in the Astrophysical Journal. “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 lead author Alexander Gagliano, fellow at the NSF Institute for Artificial Intelligence and Fundamental Interactions.
An alternative explanation considered by scientists is that the black hole may have destroyed the star completely before it could explode independently, producing bright light as debris collided with surrounding gas. In both interpretations, only one larger black hole remains.
SN 2023zkd occurred about 730 million light-years from Earth. Initially appearing as an ordinary supernova with one burst of light, it unexpectedly brightened again months later. Archival data revealed that its system had been gradually brightening for more than four years prior—an uncommon trait among supernovae.
Analysis conducted partly at UC Santa Cruz showed this pattern resulted from material shed by the dying star interacting with its surroundings: an initial brightening from low-density gas followed by another peak caused by collision with denser material arranged in a disk-like structure. Researchers believe these features point to extreme gravitational stress caused by proximity to a compact object such as a black hole.
Foley explained his role alongside Gagliano’s leadership on interpreting spectral data: “Our team also built the software platform that we use to consolidate data and manage observations. The AI tools used for this study are integrated into this software ecosystem,” Foley said. “Similarly, our research collaboration brings together the variety of expertise necessary to make these discoveries.”
Other key contributors included Enrico Ramirez-Ruiz of UC Santa Cruz leading theoretical work; V. Ashley Villar from Harvard providing AI expertise; and collaborators from institutions such as MIT and Center for Astrophysics | Harvard & Smithsonian working under YSE’s umbrella.
Funding came from organizations including the National Science Foundation, NASA, the Moore Foundation, and the Packard Foundation. Several students involved were supported through NSF graduate research fellowships.
Despite recent progress, Foley expressed concern about uncertain funding impacting future work: “The uncertainty means we are shrinking,” he said, “reducing the number of students who are admitted to our graduate program—many of them being forced out of the field or to take jobs outside the U.S.”
Foley added that while predicting future applications for AI-driven anomaly detection is challenging, similar techniques could benefit fields ranging from healthcare diagnostics to security monitoring and fraud detection: “Anywhere real-time detection of anomalies could be useful, these techniques will likely eventually play a role.”
