Optimizing Kilonova Searches: A Case Study of the Type IIb SN 2025ulz in the Localization Volume of the Low-Significance Gravitational Wave Event S250818k

The Big Picture

Imagine receiving a text message that might be the most important news of your lifetime, but the signal is so garbled it could be noise. That’s what astronomers faced on August 18, 2025, when gravitational wave detectors picked up a faint ripple in spacetime: S250818k, a low-significance signal possibly produced by two colliding neutron stars.

When two neutron stars merge, they trigger a cosmic fireworks show called a kilonova: a brief, brilliant flash of ultraviolet, optical, and infrared light powered by the radioactive decay of newly formed heavy elements like gold and platinum. The 2017 detection of kilonova AT 2017gfo alongside GW170817 was a landmark, the first time anyone watched a neutron star merger in both gravitational waves and light.

But GW170817 was a best-case scenario. Most gravitational wave alerts are fuzzier. The sky regions they point to are enormous, and finding the actual kilonova among hundreds of unrelated stellar explosions is like finding a specific firefly in a field of fireworks.

In a new paper, a team led by Noah Franz and Bhagya Subrayan at the University of Arizona details their pursuit of a kilonova following S250818k, what they found instead, and a new scoring algorithm that could change how these searches are conducted.

A supernova caught masquerading as a kilonova prompted the team to develop a quantitative scoring tool that, applied in real time, could help astronomers stop chasing false leads and focus follow-up resources where they matter most.

How It Works

The search began immediately after S250818k was flagged. The alert came with a localization volume, a three-dimensional region of sky consistent with the signal’s source. Teams worldwide began scanning for any new optical transient that could be the kilonova.

One candidate stood out early: SN 2025ulz. It was bright, appeared to fade quickly, and sat within the GW localization region. Rapid fading is a kilonova hallmark. Unlike supernovae that stay bright for weeks, kilonovae typically dim within days. SN 2025ulz ticked that box.

The team didn’t stop at first impressions. They mobilized telescopes across ultraviolet, optical, infrared, and radio bands to build a multi-wavelength dataset over multiple epochs. That kind of coverage is what separates a confident identification from a lucky guess.

The data told a different story. SN 2025ulz’s spectrum showed classic signatures of a Type IIb supernova, a core-collapse explosion from a massive star that had shed most of its hydrogen envelope. These events can briefly display an early, rapidly fading shock cooling emission phase as the blast wave breaks through the thin outer envelope, before settling into a slower decline. That early drop can look deceptively like a kilonova.

Hydrogen and helium absorption lines, combined with a brightness curve that didn’t match kilonova models, let the team conclusively rule out the kilonova interpretation.

The Scoring Algorithm

Rather than just reporting that SN 2025ulz fooled them, the team built a quantitative scoring algorithm to evaluate every transient candidate in a GW localization volume. It combines four factors into a single score:

• 3D spatial probability: How well does the candidate’s position align with the gravitational wave localization volume, including the distance estimate?
• Light curve evolution: Does brightness evolve on kilonova timescales (fast and fading) or slower, like a supernova?
• Host galaxy properties: Is the candidate near a galaxy of the right type and at the right redshift?
• Multi-wavelength behavior: Does color evolution match what we’d expect from a kilonova’s neutron-rich ejecta?

Applied retroactively to all candidates from the S250818k campaign, the numbers were clear: at every moment after SN 2025ulz was discovered, at least four other candidates had comparable or higher scores. The observing community had heavily prioritized SN 2025ulz, but it never clearly dominated the field. Resources spent characterizing it might have been better distributed.

Telescope time during GW follow-up campaigns is genuinely precious. Observers race against the clock, juggling instruments across hemispheres, making judgment calls under pressure. A tool that turns expert intuition into a reproducible number can make those calls faster and fairer.

Why It Matters

The LIGO-Virgo-KAGRA network has been running its fourth observing run, generating gravitational wave alerts at a higher rate than ever. Each alert potentially represents a neutron star merger: a natural laboratory for nuclear physics, a factory for heavy elements, a test of general relativity. With dozens of candidates per year, the community cannot afford to send its most powerful telescopes chasing every lead.

The scoring algorithm points toward a future where machine learning could automate real-time triage. The current system uses well-understood physical priors, but neural networks trained on archival multi-wavelength data could eventually do a better job of distinguishing early supernovae from true kilonovae. IAIFI co-authors Alexander Gagliano and Harsh Kumar are well-positioned to pursue those extensions.

A deeper open question remains: how common are genuine kilonovae relative to the supernovae, active galactic nuclei, and other transients that contaminate the search? Each follow-up campaign, whether it ends in a detection or not, adds data to the empirical picture. Scoring algorithms like this one get better calibrated with use.

SN 2025ulz was a cosmic impersonator, a supernova that briefly mimicked the rarest of events. But the real discovery is a smarter way to search: a scoring algorithm that could prevent future teams from burning all their telescope hours on a single red herring.