Astronomers have used artificial intelligence to reconstruct a 3D video that shows a hot pocket of gas orbiting a stone’s throw away from our galaxy’s central black hole.

For all its “supermassive” status, our resident behemoth, dubbed Sgr A*, is quiet as such black holes go. Perhaps due to strong magnetic fields that constrict its diet, it only picks at the gaseous buffet around it. Even so, it emits minor flares up to several times a day at X-ray, infrared, and radio wavelengths.

These flares were a nuisance to the Event Horizon Telescope (EHT) team, which produced an image of Sgr A*’s dark silhouette, based off of data collected in 2017. Now, Aviad Levis (Caltech) and collaborators, including several members of the EHT team, have utilized some of this 2017 data for a new project: following the motion of a flaring hotspot in the tumultuous gas flowing around the black hole.

Astronomers made this 3D model of the flare in the disk of gas around Sgr A* by using ALMA data as input to a type of computer algorithm known as a neural network. The hotspot and its fainter companion orbit the black hole in an hour, distorting in shape over time.
Aviad Levis

The Atacama Large Millimeter/submillimeter Array (ALMA), high in the Atacama Desert in Chile, was imaging Sgr A* on April 11, 2017, as part of the EHT project. While data taken earlier in the month was ultimately used for the EHT image, because the black hole was quieter then, the radio-wave data collected on April 11th was interesting in its own right, as it coincided with an X-ray flare.

However, while the combined EHT data revealed an exquisitely high-resolution image of Sgr A*, the ALMA data taken alone is 100,000 times fuzzier. Basically, all of the information that the team was interested in was packed into a single pixel.

The researchers started with the idea that flares come from hotspots within gas flowing around the central maw. Magnetic fields snap-crackle-popping very close to the black hole can heat and energize this gas.

If a flare comes from such a hotspot, then information about the flare’s origin is encoded in the pixel’s changing brightness: Gas flowing into the black hole won’t change much over a few hours, but hot pockets of gas will shear out and fade more quickly. The radio waves’ polarization, which tells us at what angle the light waves are oscillating, is also expected to be higher when the light is coming from a hot bubble of plasma.

“What we get from ALMA is two polarized single-pixel videos,” summarizes Katie Bouman (Caltech), whose group led the study published in Nature Astronomy. Each video shows the changing brightness for light of a given polarization.

black hole surrounded by a glowing orange ring
The Event Horizon Telescope's work resulted in this image of Sgr A*, the supermassive black hole at the center of our galaxy. The black hole's silhouette at center is only 52 microarcseconds across. ALMA observations were among several data sets used to make this image. In the new research, only ALMA data are used, and they are much fuzzier than the high-resolution image shown here.
EHT Collaboration

To make sense of all this information, the team employed a relatively new form of artificial intelligence, known as neural radiance fields, developed in part by team member Pratul Srinivasan (Google Research). This technique transforms 2D information into a 3D rendering, so that viewers can see a scene from a different viewpoint than was originally recorded. It’s used everywhere from video-game graphics to medical imaging.

In this case, though, the neural network wasn’t just given the information that ALMA had collected; it was also given rules of sorts, according to the physics that the astronomers know or assume to be operating near Sgr A*.

For example, the researchers had to account for the light-bending effects of the supermassive black hole’s gravity. The team also made assumptions that aren’t guaranteed, such as that the bright spot belongs to gas flowing around the black hole — and not gas escaping in a wind or jet.

But the new technique actually allowed the team to make fewer assumptions than before: “In the previous analyses, we interpreted these data with a simple geometric model of an orbiting blob (hot spot),” says team member Maciek Wielgus (Max Planck Institute for Radio Astronomy, Germany). “By utilizing the methods of artificial intelligence (neural networks), we were able to interpret observations with a more general and flexible model, without strong assumptions on the geometry of the hot spot and its evolution in time.”

The result of this analysis is a 3D video showing two hot pockets of gas moving around the black hole, shearing and elongating as they go. The flow of gas is clockwise and nearly edge-on, at an inclination of 18 degrees to Earth, in agreement with previous studies conducted using the EHT and the Very Large Telescope Interferometer’s Gravity instrument.  

“This is the first three-dimensional reconstruction of gas rotating close to a black hole,” Bouman says.

The bright spot looks likely to dissipate within a few orbits, within hours. The brightest part is very close to the black hole, at a distance only six times the size of the event horizon. Put in other terms, the brighter hotspot is closer to Sgr A* than Earth is to the Sun, at a distance of only 0.5 a.u. A fainter hotspot lags behind at somewhat greater distance.

This video isn’t a simulation, but it isn’t a direct recording of events, either. “It is a reconstruction based on our models of black hole physics,” Levis says. “There is still a lot of uncertainty associated with it, because it relies on these models being accurate.”


Sgr A*


You must be logged in to post a comment.