Artificial intelligence and physics combine to reveal 3D structure of flares erupting around black holes

Scientists believe that the environment around a black hole is turbulent, characterized by hot, magnetized gas swirling in a disk at extremely high speeds and temperatures. Astronomical observations show that within such a disk, mysterious flares appear several times a day, briefly brighten, and then disappear.

Now, a team led by Caltech scientists has used telescope data and artificial intelligence (AI) computer vision techniques to recover the first three-dimensional video showing what such a flare would look like around the supermassive black hole Sagittarius A* (Sgr A*). The center of our galaxy.

The 3D flare structure features two bright, compact features located about 75 million kilometers (or half the distance between Earth and the Sun) from the center of the black hole. It is based on X-ray data collected by the Atacama Large Millimeter Array (ALMA) in Chile within 100 minutes of the volcanic eruption on April 11, 2017.

“This is the first three-dimensional reconstruction of gas swirling close to a black hole,” said Katie Bouman, assistant professor of computational and mathematical sciences, electrical engineering and astronomy at Caltech. natural astronomy The title is “Orbital polarization tomography of flares near the Sagittarius A* supermassive black hole.”

Arviad Levis, a postdoctoral scholar in Bouman’s group and the paper’s lead author, emphasized that while the film is not a simulation, it is also not a direct record of the events as they occurred. “This is a reconstruction based on our physical models of black holes. There are still a lot of uncertainties associated with it because it relies on the accuracy of those models,” he said.

Use physics-based artificial intelligence to find possible 3D structures

To reconstruct the 3D images, the team had to develop new computational imaging tools that could, for example, account for the bending of light caused by the curvature of space-time around massive gravitational objects like black holes.

In June 2021, the multidisciplinary team first considered whether it would be possible to create 3D movies of flares around black holes. of supermassive black holes, are working to do the same thing using EHT data from Sgr A*.

Pratul Srinivasan of Google Research, a co-author of the new paper, was visiting the team at Caltech. He helped develop a technology called Neural Radiation Fields (NeRF) when it was just beginning to be used by researchers. It has had a huge impact on computer graphics ever since. NeRF uses deep learning to build a 3D representation of a scene from 2D images. It provides a way to view a scene from different angles, even with a limited view of the scene.

The team wanted to know if, through recent advances in neural network representation, they could reconstruct the 3D environment around a black hole. The big challenge they face: From Earth, as everywhere, we can only get a single view of a black hole.

The team thinks they might be able to overcome this problem because the gas behaves in a somewhat predictable way as it moves around the black hole. Consider the analogy of trying to capture a 3D image of a child wearing an inner tube around their waist.

To capture this type of image using traditional NeRF methods, you need to take photos from multiple angles while the child remains still. But in theory, you could have the child spin while the photographer remains still and takes the picture.

Timed snapshots combined with information about the child’s rotation speed allow for an equally good reconstruction of the 3D scene. Likewise, by leveraging knowledge of how gas moves at different distances from the black hole, the researchers aimed to solve the problem of 3D flare reconstructions from measurements taken over time from Earth.

Armed with this insight, the team built a version of NeRF that takes into account how gas moves around a black hole. But it also requires taking into account how light bends around massive objects like black holes. Under the direction of co-author Andrew Char of Princeton University, the team developed a computer model to simulate this bending, also known as gravitational lensing.

Taking these factors into account, the new version of NeRF is able to recover the structure of the bright features surrounding the black hole’s event horizon. In fact, the initial proof-of-concept showed promising results on synthetic data.

Flares available for research around A* Sgt.

But the team needed some real data. This is where ALMA comes into play. But a few days later, on April 11, astronomers noticed an explosive sudden brightening of the surrounding environment.

When team member Maciek Wielgus of the Max Planck Institute for Radio Astronomy in Germany was reviewing the day’s ALMA data, he noticed a signal whose period matched the time it takes for a bright spot within the disk to complete one orbit. * Sgt. The team set out to recover the 3D structure of the bright region around Sgr A*.

ALMA is one of the most powerful radio telescopes in the world. However, due to its great distance from the galactic center (more than 26,000 light-years), even ALMA does not have the ability to see Sgr A*’s immediate surroundings. ALMA measures a light curve, which is essentially a video of a single blinking pixel, created by collecting all the radio wavelength light detected by the telescope at each observation moment.

Recovering 3D volumes from single-pixel movies seems impossible. However, by leveraging additional information about the physical properties of the disk surrounding the black hole, the team was able to address the lack of spatial information in the ALMA data.

Strongly polarized light from flares provides clues

ALMA captures more than just a single light curve. In fact, it provides several such “videos” for each observation, as the telescope records data related to different polarization states of light. Like wavelength and intensity, polarization is a fundamental property of light and represents the direction of the electrical component of a light wave relative to the overall direction of propagation of the wave.

“What we got from ALMA were two polarized single-pixel videos,” said Buman, who is also a Rosenberg Scholar and Institute for Traditional Medicine researcher. “Polarized light is actually very, very abundant.”

Recent theoretical studies have shown that hot spots that form within gases are strongly polarized, meaning that light waves from these hot spots have clear preferred directions. This is in contrast to the rest of the gas, which has a more random or chaotic orientation. By collecting different polarization measurements, ALMA data provides scientists with information that helps localize emission sources in 3D space.

Introduction to Orbital Polarization Tomography

To find possible 3D structures that explain the observations, the team developed an updated version of its method that combines not only the physical principles of light bending and dynamics around black holes, but also the polarization expected in hot spots orbiting black holes. emission. In this technique, each potential flare structure is represented as a continuous volume using a neural network.

This allowed the researchers to calculate the hotspot’s initial 3D structure over time as it orbited the black hole to create the complete light curve. They can then solve for the best initial 3D structure that matches the ALMA observations when progressing in time according to black hole physics.

The result is a video showing the clockwise motion of two compact, bright regions that follow a path around the black hole. “It’s very exciting,” Bouman said. “It doesn’t have to appear this way. There could be arbitrary brightness scattered throughout the volume. The fact that this looks a lot like the flares predicted by computer simulations of black holes is very exciting.”

Levis says the work is uniquely interdisciplinary: “It’s a unique synergy between computer scientists and astrophysicists. Together we develop something that’s at the forefront of both fields – both A digital code that simulates how light travels.

Scientists point out that this is just the beginning of this exciting technology. “This is a very interesting application of how artificial intelligence and physics can come together to reveal something that would otherwise be invisible,” Levis said. “We hope that astronomers can use this with other rich time series data to elucidate the complex dynamics of other such events and draw new conclusions.”