Caltech astronomers use new algorithm to capture black hole jet aimed at Earth

Caltech astronomers have produced the first high-resolution video of a supermassive black hole firing a jet of plasma directly toward Earth, using not a new telescope but a new way of seeing. The black hole sits in blazar 3C 345, located 5.5 billion light-years away in the constellation Hercules. What makes this achievement remarkable is not the discovery itself—astronomers have known about this object for decades—but the method. The team combined 27 years of telescope observations, 116 separate measurements taken between 1995 and 2022, and fed them through an algorithm called kine, a video-reconstruction tool that uses neural fields to stitch disparate images into a coherent, detailed sequence.

A blazar is what happens when a supermassive black hole's ejected plasma happens to point in our direction. The jet moves at nearly the speed of light, and from our vantage point, we see it head-on. Before kine, researchers could measure the overall velocity of the plasma stream but could not distinguish how different sections of the jet moved at different speeds. The new algorithm changed that. It allowed the Caltech team to see, for the first time, the internal structure and motion of the blazar's plasma with unprecedented clarity.

Dr. Marianna Foschi, the lead researcher, emphasized the practical significance of the breakthrough. In astrophysics, better images have traditionally required better hardware—new telescopes with improved lenses or sensors, projects that take years to fund and build. The kine algorithm, by contrast, extracted new information from data already in hand. "We get this basically for free," Foschi said, describing the resolution and contrast gains the algorithm delivered. The work is detailed in a scientific journal that explains how kine was constructed using neural fields and documents the 116 observations that went into the final video.

The two-view presentation of the data shows the difference starkly. One view displays what the blazar looks like through conventional video reconstruction. The other, processed through kine, reveals far greater detail and contrast. The algorithm essentially teaches itself to recognize patterns across decades of observations and synthesizes them into a coherent moving image. This is not simulation or interpolation in the traditional sense; it is a computational method for extracting signal from noise and time.

The implications extend beyond this single object. If algorithmic processing can unlock this level of detail from existing data, it suggests a new path for astrophysical discovery. Rather than waiting for the next generation of instruments, researchers might accelerate understanding by developing smarter ways to analyze what they already have. For a field accustomed to planning decades ahead for hardware upgrades, the prospect of algorithmic leaps offers a different kind of momentum. The question now is what other cosmic phenomena, already observed but not yet fully understood, might yield their secrets to similar computational approaches.