AI video of blazar 3C 345 clocks its bright knots at the speed of the plasma around them
Older pipelines could measure a bright knot's speed but not the flow it moves through. Kine measures both in 3C 345, and the two land within about a tenth of each other, which a shock front should not do.
The Scientist · Science desk
What happened
- Marianna Foschi and colleagues report in Nature, in a study dated Aug. 26, an algorithm that reconstructs scattered VLBI snapshots of blazar 3C 345 into a single continuous polarized video.
- The video draws on 116 observations recorded between 1995 and 2022 by the Very Long Baseline Array, a network of ten radio telescopes across the United States.
- The reconstruction reports about four times the usual resolution limit and roughly 140 times the overall image contrast of traditional imaging methods.
- The jet's brightest components were clocked at 10 to 13 times the speed of light while the gas around them moved at about nine to 12 times the speed of light.
- The bright features also showed no correlation with peaks in fractional polarization, which the authors treat as a second mark against reading them as shocks.
Why it matters
- What's now possible capability Measuring the flow beside a knot, not just the knot, makes the shock model's central prediction something an observation can fail; the comparison speed simply was not available to earlier pipelines.
- What doesn't add up contradiction Phys.org presents the video as challenging shock wave theory while Foschi says the model is questioned only for this source, and which reading you take decides whether this is a retirement or a to-do list.
- What it sets up precedent Because the monitoring archives already exist, motion maps for hundreds of active galactic nuclei become a reprocessing job rather than a new observing campaign, which raises the bar for the next kinematics paper.
- What it limits constraint One object cannot unseat an interpretation built across many, so anyone wanting to generalise has to run the same two-number test source by source.
Next story · Science · 5h ago
Baker lab's from-scratch zinc proteases cut their chosen target in 36% of 135 designsConfirmed2 publishers
The shock reading of superluminal components was always a comparison, and until now only one side of it could be measured. Fit a model component to a bright knot, follow it from epoch to epoch, and you have the knot's apparent speed [19]. You do not have the speed of the plasma it is supposedly compressing [11]. A shock has to travel at a different speed from the flow to be a shock [6], so the standard interpretation sat untested rather than confirmed.
Kine gets the second number by declining to treat the epochs as separate pictures. It fits a single neural representation of the whole video, sharing information across frames and producing a flux density model that is continuous in time and sampleable at any moment, which lets local motion be read out by optical flow instead of by matching components between snapshots [1][19]. The authors note that the same reconstruction structure turns up in MRI, where the target also moves while you measure it [24].
In 3C 345 the bright components come out at 10 to 13 times the speed of light in projection, and the surrounding gas at about 9 to 12 [7]. At the endpoints that is a difference of 8 to 11 percent (13/12 = 1.08, 10/9 = 1.11) [22]: the same order, which is how the authors put it [16], and not the excess a shock front should show over the fluid it drives [8]. The argument has a second leg that does not depend on those margins. Fractional polarization peaks do not track the bright features [17], which fits knots that are locally magnetized, brighter regions with their apparent brightness boosted by their direction of motion rather than compressed fronts [18].
What the published numbers do not give is uncertainty. Neither account quotes error bars on either velocity field, and an 8 percent gap with unknown scatter is a weaker statement than an 8 percent gap with tight ones.
Nor is the video continuous in the way the word suggests. The 116 epochs span 1995 to 2022 [3], an average of one observation every 2.8 months (27 years x 12 months / 116) [21]. Between those epochs the smoothness belongs to the neural model, and the four-fold resolution and roughly 140-fold contrast gains are quoted against frame-by-frame imaging [4][14], which is the right baseline for a claim about sharing information across frames and also means the gain rides on the prior doing the sharing.
The two accounts differ slightly on provenance: phys.org has the 116 observations coming from the MOJAVE monitoring program [2], while Live Science credits the array and names two programs, BEAM-ME and MOJAVE, that between them followed hundreds of blazars [12].
What generalizes here is the measurement, not the verdict. The authors say the result is about 3C 345 and does not automatically extend to other jets [10], and first author Marianna Foschi of Caltech told Live Science the work puts the shock model in question for this source without invalidating it [9][20]. Anyone who has been treating a fitted component's speed as the jet's speed now has a concrete reason to check, in their own source, whether those two numbers differ at all.
What to watch
- Published uncertainties on the two velocity fields; the whole comparison turns on error bars neither account quotes.
- Whether the knot-and-flow match survives in other blazars once kine is run on other sources.
- Validation of the four-fold resolution and 140-fold contrast figures on synthetic jets whose velocity field is known in advance.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence72
- Adoption18
- Hype gap+20
- Incentives52
- Confidence66
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
The team created an algorithm they call kine, described as a video reconstruction algorithm for VLBI observations of variable sources; it uses a neural representation of the video to process images taken at different times simultaneously while learning and leveraging spatiotemporal correlations, producing a smooth continuous model of flux density that can be sampled at any time coordinate.
- [2]
The researchers used 116 VLBI radio observations of blazar 3C 345 from the MOJAVE monitoring program to reconstruct a video of its motion.
- [3]
The video was made from 116 images taken between 1995 and 2022 by the Very Long Baseline Array, a network of 10 telescopes spread across the U.S.
- [4]
The reconstruction achieved a resolution about four times higher than the usual limit and roughly 140 times the overall image contrast of traditional methods.
- [5]
Live Science reports the work as an Aug. 26 study in the journal Nature; the citation lists Foschi, Zhao, Fuentes, Bouman, Gomez and Levis (2026), 'Video reconstruction of variable VLBI observations with neural fields'.
- [6]
Earlier studies tracked broad bright features in 3C 345 that appeared to move faster than light and interpreted them as traveling shock waves caused by disturbances compressing the plasma; for that to be true the shock must travel at a different speed from the relativistic jet.
- [7]
The jet's brightest components were measured flying at 10 to 13 times the speed of light while the surrounding gas moved at about nine to 12 times the speed of light.
- [8]
Foschi: the result is unexpected because the general consensus is that these bright components are shock perturbations moving through the plasma, and as such should have a higher velocity than the surrounding fluid.
- [9]
Foschi told Live Science: '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.'
- [10]
The team notes the results apply specifically to blazar 3C 345 and not automatically to all black hole jets.
- [11]
Earlier methods had not been able to measure the speeds of the plasma surrounding the bright knots; the video allowed the researchers to estimate local plasma motion throughout the jet rather than only track individual bright knots.
- [12]
Live Science reports that the VLBA had two programs that caught views of the jet, BEAM-ME and MOJAVE, which collectively followed hundreds of blazar sources.
- [13]
3C 345 is a blazar in the constellation Hercules, a type of quasar in which a feeding supermassive black hole at a distant galaxy's center emits jets that blaze at nearly the speed of light.
- [14]
Traditional VLBI imaging has poor resolution and treats each observation as a separate snapshot in time, limiting information about how features move.
- [15]
Phys.org headlined its report 'Continuous video of black hole radio observations challenges shock wave theory'.
- [16]
Study authors: 'We found no evidence that traveling bright components are strongly shocked regions, as previously proposed, because the component speeds are of the same order as the average flow speed in the same regions.'
- [17]
Study authors: 'The shock interpretation is also disfavored by the absence of correlation between the bright features and the peaks in the fractional polarization.'
- [18]
Polarization evidence suggests the knots are locally brighter, magnetically energized regions whose brightness is boosted by their direction of motion, and the bright knots moved at speeds similar to the surrounding plasma.
- [19]
Previous methods tracked individual bright knots or model-fitted components; the authors write that from high-resolution time-continuous videos it becomes possible to measure the instantaneous local velocity field, with continuous and local motion analysis by optical flow, and that simultaneous imaging of all observations shares information across frames to improve resolution and dynamic range beyond frame-by-frame imaging.
- [20]
Marianna Foschi, the study's first author, is a postdoctoral researcher at Caltech, and several co-authors had previous experience boosting the resolution of distant black hole images with the Event Horizon Telescope Collaboration.
- [21]
116 observations spread over the 27 years from 1995 to 2022 is an average of one epoch every 2.8 months.
- [22]
At the endpoints of the quoted ranges, the bright components' apparent speeds exceed the surrounding flow's by about 8 to 11 percent.
- [23]
Applying kine across major archival jet-monitoring programs could provide full motion maps for hundreds of active galactic nuclei.
- [24]
The researchers note the underlying reconstruction problem resembles MRI imaging, suggesting possible future improvements for imaging moving organs.
- [25]
Foschi: 'We believe this method will drastically change the way jet dynamics is studied from observations. In fact, our method enables a precise measurement of the projected velocity at any point in the jet.'
Sources
2 independent publishers whose own reporting we read for this story.
- livescience.comSee the highest-resolution footage of a black hole jet ever taken, thanks to AI and 27 years of observations
1 article · September 9, 2026
- phys.orgContinuous video of black hole radio observations challenges shock wave theory
1 article · September 10, 2026
Topics and entities
Topics
- Radio interferometry
- Computational imaging
Entities
- California Institute of Technology
- Event Horizon Telescope Collaboration
- Nature
- Very Long Baseline Array