AI Enhances Black Hole Jet Footage to Record Resolution

Astronomers have unveiled an unprecedentedly detailed animation of a black hole jet, created using data spanning 27 years and enhanced by artificial intelligence. The video, composed of 116 images captured between 1995 and 2022, depicts the blazar 3C 345, located in the Hercules constellation. Blazars are quasars—supermassive black holes at the heart of distant galaxies actively consuming matter—that launch powerful jets of gas at near-light speeds, emitting intense X-rays and gamma rays.

The groundbreaking visualization was made possible by observations from the Very Long Baseline Array (VLBA), a network of 10 radio telescopes distributed across the United States. These observations were part of two programs, BEAM-ME and MOJAVE, which collectively monitored hundreds of blazar sources. The research, detailed in a study published on August 26 in the journal Nature, focused on transforming these individual snapshots into a dynamic video.

AI Model Boosts Resolution, Reveals Surprising Speeds

To achieve this remarkable clarity, scientists employed a sophisticated AI neural network named Kine. This advanced algorithm processed the decades of observational data, yielding a resolution four times greater than any single image could provide. This significant enhancement allowed researchers to map the velocity of the black hole jet with unparalleled precision. Notably, several researchers involved in this project had previously contributed to high-resolution imaging efforts for the Event Horizon Telescope Collaboration, which produced the first-ever image of a black hole.

Marianna Foschi, a postdoctoral researcher at Caltech and lead author of the study, explained the impact of the improved footage. "The higher quality of our video reconstruction enabled a detailed measurement of the plasma velocity in the jet," Foschi stated in an email. The team was particularly surprised by their findings regarding the jet's speed. The brightest components of the jet were observed moving at speeds 10 to 13 times the speed of light, while the surrounding gas streamed at approximately nine to 12 times the speed of light.

"This is unexpected because the general consensus is that these bright components are shock perturbations moving through the plasma, and as such they should have a higher velocity compared to the surrounding fluid," Foschi elaborated. "Our work does not invalidate the shock model in general, but it puts it into question, at least in the case of this specific source." This suggests a potential refinement or challenge to existing theoretical models of blazar jet behavior.

The VLBA's configuration, utilizing radio antennae across the continental U.S., provides a broad view of the cosmos but traditionally faced limitations in capturing fine details. The array generates two-dimensional images based on measurements between pairs of telescopes. However, with only 10 telescopes, the resulting pixel resolution was inherently constrained.

The Kine AI model directly addresses this limitation. Developed to create videos from observations taken at different times, Kine focuses on astronomical sources exhibiting variable brightness. As a type of machine learning, neural networks like Kine learn from vast datasets, mimicking aspects of human brain function. "Kine can process observations at different times, while learning and leveraging the spatio-temporal correlations present in the data," the research team explained in their paper. This technique effectively reconstructs a more coherent and detailed temporal sequence of events.

Foschi expressed enthusiasm for applying this AI-driven approach to other astronomical research. "We believe this method will drastically change the way jet dynamics is studied from observations," she said. "In fact, our method enables a precise measurement of the projected velocity at any point in the black hole jet." The advancement promises new insights into the complex physics governing these powerful cosmic phenomena.