AI algorithms discovered a new type of supernova triggered by black hole-star mergers. Research on SN 2023zkd reveals how machine learning detects rare cosmic events invisible to traditional observation methods.
For centuries, astronomers have observed stellar explosions through telescopes, cataloging their properties to understand how stars end their lives. Yet despite this extensive observation, cosmic mysteries remain hidden beneath layers of transient phenomena. In July 2023, the Zwicky Transient Facility detected supernova SN 2023zkd, located 730 million light-years from Earth.
Six months later, an artificial intelligence algorithm named Lightcurve Anomaly Identification and Similarity Search (LAISS) flagged this explosion as anomalous—revealing what may be a new type of supernova born from the violent merger of a dying star and its black hole companion. This discovery demonstrates how machine learning transforms astronomy, identifying rare cosmic events that human observation might overlook indefinitely.
For the new type of supernova, The detection and analysis were conducted by a team led by Harvard-Smithsonian Center for Astrophysics (CfA) and MIT as part of the Young Supernova Experiment (YSE). Published in The Astrophysical Journal in August 2025, this research suggests an entirely new class of stellar explosions awaits discovery. As astronomer V. Ashley Villar stated, “We think this might be part of a whole class of hidden explosions that AI will help us discover.”
Machine Learning Reveals New Type of Supernova Through Algorithmic Detection
The story of SN 2023zkd begins not with visual observation, but with algorithmic analysis. The LAISS algorithm was specifically designed to identify unusual transient events within massive astronomical datasets—a task perfectly suited to machine learning’s pattern-recognition capabilities. When the algorithm examined SN 2023zkd in January 2024, approximately six months after initial detection, it flagged the event as anomalous.
This flagging proved crucial. To human observers, SN 2023zkd appeared unusual but not extraordinary, particularly when initial observations seemed to follow typical supernova behavior. However, the LAISS algorithm recognized subtle patterns deviating from established classifications. As lead author Alex Gagliano, an astronomer at the NSF-funded Institute for AI and Fundamental Interactions (IAIFI), explains: “To our eyes, the supernova was not too unusual, but our algorithm was luckily able to pick up on it as something unusual.”
For the new type of supernova, The alert triggered immediate and widespread follow-up observations. This rapid response proved essential, capturing critical archival data that revealed two behaviors departing dramatically from typical supernovae: years-long precursor brightening and a dramatic rebrightening event months after initial explosion. These characteristics ultimately revealed a new type of supernova phenomenon involving black hole-star merger dynamics.
Understanding Stellar Death: Physics of Standard Supernovae
For the new type of supernova, and to appreciate SN 2023zkd’s uniqueness, we must first understand how stars normally end their lives. A star spends its existence in delicate equilibrium, with outward pressure from nuclear fusion in its core pushing against inward gravitational crush. This balance persists as long as hydrogen fuels the fusion process.
As a star ages and exhausts its hydrogen fuel, nuclear fusion shifts toward progressively heavier elements—helium, carbon, oxygen, and beyond. This process continues until fusion reaches iron. Here lies a fundamental threshold: fusing iron consumes energy rather than releasing it. The star’s fusion engine abruptly shuts off, shattering its equilibrium. With nothing halting its collapse, gravity triumphs absolutely.
For sufficiently massive stars, gravitational collapse becomes catastrophic. The core implodes into an ultradense object while outer layers rebound off this newfound barrier, launching a tremendous shock wave outward. This creates what astronomers call a Type II supernova—perhaps the cosmos’s most dramatic event outside black hole physics itself.
SN 2023zkd belongs to a specialized subclass called Type IIn, where “n” signifies the explosion occurs within a dense gas cloud the star previously shed. The supernova’s shock wave slamming into this material creates an exceptionally complex light show. Yet even within this chaotic classification, SN 2023zkd stood apart as decidedly strange.
The Precursor Period: Four Years of Violent Inspiral Dynamics
For the new type of supernova, By analyzing archival data, Gagliano’s team made a remarkable discovery: before SN 2023zkd exploded, it had been brightening for four years—an extraordinarily long precursor period. The team divided this activity into two distinct phases: Precursor A, a long period of steady brightening; and Precursor B, a final accelerated ramp-up in brightness.
This years-long precursor activity represented an “‘aha’ moment,” according to Gagliano. The gradual brightness increase over years before explosive detonation matched precise predictions for a specific scenario: a binary system containing a star and black hole spiraling toward merger. As Gagliano explains: “The years-long ramp-up to the explosion is a very clear prediction of a scenario involving a star and a black hole.”
Concerning the new type of supernova, the team concluded the two objects occupied a “really, really tight binary,” spiraling increasingly inward toward catastrophic collision. As the black hole’s gravity yanked material off the star, some material spiraled toward the black hole, forming an accretion disk wrapped directly around it. For this new type of supernova, This heated material, prior to disappearing into the black hole, generated the Precursor A emission. Simultaneously, other material shed from the star formed a larger disk outside both the black hole and star, creating a dense circumbinary disk enveloping the entire system.
Precursor Activity Timeline:
| Phase | Duration | Characteristics | Physical Process |
| Precursor A | 3+ years | Steady brightening | Accretion disk heating |
| Precursor B | ~1 year | Accelerated ramp-up | Chaotic spiral inward |
| Explosion Event | Days-weeks | Initial brightening | Core-collapse trigger |
| Rebrightening | ~240 days later | Secondary peak | Shock interaction with disk |
The Double-Peaked Explosion: Unraveling the Cosmic Puzzle
The precursor period’s unusual activity represented only half the mystery. After standard supernovae detonate, brightness climbs to a single brilliant peak before beginning steady fade. SN 2023zkd initially followed this script perfectly. But after several months of fading as expected, something extraordinary occurred: the supernova dramatically rebrightened, reaching a second nearly-as-brilliant peak approximately 240 days after the first.
This double-peaked light curve represented an extremely rare occurrence—the second major clue confirming this event defied standard stellar explosion classifications. Using computer models, the team determined that the first peak resulted from the shock wave hitting diffuse polar clouds. The second, more sustained peak was caused by the shock wave slamming into the dense circumbinary disk.
For the new type of supernova, The sequence seems counterintuitive. Why would collision with the circumbinary disk—located inside the polar clouds—create the second peak? The answer involves an observational effect driven by structures’ differing densities. When the supernova’s blast wave moved through the system, low-density polar clouds were swept up rapidly. As the shock wave moved through these clouds and excited their atoms, it created an enormous expanding photosphere—the visible surface from which light escapes—so hot and opaque that internal emissions remained hidden.
Simultaneously, the shock wave slammed into the denser circumbinary disk, but its material was swept up more slowly. “The photosphere very quickly overtakes the disk,” Gagliano explains, “so that the disk is effectively concealed inside the photosphere even though it is producing light.” Only as outer material cooled and the photosphere began receding did the disk become visible, powering the dramatic rebrightening. This transition reveals where the outermost photosphere becomes the disk-explosion interaction, eventually revealing the interaction between disk and explosion that powers the second peak.
Black Hole-Star Merger: Theorized Event Now Observationally Confirmed
The violent inspiral and complex gas structures pointed unambiguously toward one culprit: a black hole companion that triggered the star’s explosive death. In this scenario, stresses induced by the black hole’s intense gravity as the two objects orbited each other increased pressure in the star’s core until it could no longer support itself, inducing core-collapse supernova.
This represents an event theorized for decades but never previously observed with such compelling evidence. It addresses a long-standing astronomical question: whether material coming off stars in the millennia to years before explosion results from internal changes alone, or from internal changes driven by dynamical interactions with companions. “This discovery provides confirmation that companions can indeed play a determining role in the explosion itself,” Gagliano states.
Artificial Intelligence and the Future of Astronomical Discovery
Fittingly, a story that began with automated algorithmic detection points toward astronomy’s future. As next-generation observatories like the Vera C. Rubin Observatory begin generating unprecedented floods of transient data, intelligent systems like LAISS become indispensable. Traditional human observation simply cannot process the volume and complexity of modern astronomical surveys.
For the new type of supernova, The Vera C. Rubin Observatory alone will generate petabytes of imaging data nightly, capturing thousands of transient events far exceeding human review capacity. Machine learning systems will become essential for identifying genuinely anomalous events worthy of intensive follow-up observation. LAISS demonstrated this capability by recognizing SN 2023zkd’s unusual properties when human observers would likely have classified it as an interesting but unremarkable Type IIn supernova.
In SN 2023zkd’s case, the LAISS algorithm’s warning turned the team’s attention to the event long before the rebrightening confirmed its unusual nature. This early alert enabled comprehensive archival analysis and rapid follow-up observations that would have been impossible without algorithmic flagging. The discovery validates the partnership between artificial intelligence and traditional astronomy.
AI’s Role in Modern Astronomy:
- Processes millions of astronomical datasets beyond human capacity
- Identifies subtle pattern anomalies indicating unusual transient events
- Enables rapid follow-up observation prioritization
- Reveals previously hidden event classes within survey data
- Reduces discovery timescales from months to days
- Allows real-time alert systems triggering immediate observations
Building a Population Sample and Future Research Directions
The immediate goal following this discovery involves finding additional examples of this phenomenon through systematic algorithmic searches. With one spectacular example established, building a statistically meaningful sample allows population-level analysis. Machine learning algorithms like LAISS provide the mechanism for conducting this systematic search through historical and future survey data.
Future discoveries will enable astronomers to test predictions emerging from merger-induced explosion models. Different stellar masses, black hole masses, orbital parameters, and initial separation distances should produce variations in observable characteristics—precursor timescales, explosion energies, rebrightening characteristics, and spectral signatures. Building a diverse sample will permit testing these theoretical predictions quantitatively.
Additionally, detailed spectroscopic observations of future merger-induced supernovae will provide elemental abundance information, revealing nucleosynthesis patterns specific to this explosion mechanism. Such data would constrain models of how massive stars with black hole companions explode, potentially illuminating pathways for black hole growth in dense stellar environments. As Gagliano explains, “If, with the help of AI, astronomers can build a sample of these events and capture detailed spectra, then you can start to make population-level statements that we just have never been able to make before.”
The discovery also motivates improved machine learning algorithms specifically trained to identify this explosion category. Algorithms optimized for detecting merger-induced supernova signatures will prove far more efficient than generalized anomaly detection systems. As astronomical survey data volumes increase exponentially, such specialized detection systems become increasingly valuable.
Conclusion
SN 2023zkd represents a watershed moment in astronomical discovery—the first compelling evidence of a new type of supernova born from catastrophic black hole-star merger. Detected not through traditional visual observation but through machine learning pattern recognition, this explosion reveals nature’s capacity for cosmic phenomena remaining hidden until technology and methodology advance sufficiently.
The years-long precursor brightening, the violent inspiral dynamics, and the dramatic double-peaked explosion all bear witness to gravitational violence between black hole and star finally reaching cataclysmic conclusion. As Gagliano eloquently states: “We’re now entering an era where we can automatically catch these rare events as they happen, not just after the fact. That’s incredibly exciting.” The partnership between artificial intelligence and traditional astronomy has opened an entirely new window onto stellar death. To explore more about cosmic explosions and AI-driven astronomy, visit our YouTube channel—join NSN Today.
