to create a 3D visualization of these flare-ups using neural networks and black hole physics models. The team’s reconstruction provides a glimpse of the turbulent activity in the vicinity of the supermassive black hole, offering valuable insight into the dynamics of the gas disk surrounding it. This breakthrough has the potential to deepen our understanding of the behavior of black holes and the extreme conditions in their immediate vicinity. The use of neural networks in conjunction with radio telescope data showcases the innovative approach taken by the team to unlock the mysteries of black hole phenomena.and an AI computer vision technique was used to create the first three-dimensional video showing potential flares around Sagittarius A (Sgr A), the supermassive black hole at the center of the Milky Way galaxy.
The 3D flare structure contains two bright, compact features located approximately 75 million kilometers from the black hole, which is about half the distance between Earth and the Sun. This structure was created using data collected by the Atacama Large Millimeter Array (ALMA) in Chile over a period of 100 minutes, following an eruption observed in X-ray data onOn April 11, 2017, Katie Bouman, an assistant professor at Caltech, announced the first three-dimensional reconstruction of gas rotating near a black hole. This groundbreaking effort was led by her group and is detailed in a new paper in Nature Astronomy. Postdoctoral scholar Aviad Levis, the lead author of the paper, explained that the video is not a simulation, nor is it a direct recording of actual events. Instead, it is a reconstruction based on models of black hole physics, with a significant level of uncertainty still associated with it.According to Bouman, “it’s crucial to develop AI models that are accurate because the whole process relies on them.” The team used AI informed by physics to determine potential 3D structures. They developed new computational imaging tools to reconstruct the 3D image, taking into account factors like the bending of light due to space-time curvature around massive gravity objects such as black holes. In June 2021, the multidisciplinary team explored the possibility of creating a 3D video of flares around a black hole. Bouman and Levis, who are members of the Event Horizon Telescope (EHT) Collaboration, had already published their work.The team successfully captured the first image of the supermassive black hole at the core of the M87 galaxy and was also working on doing the same with EHT data from Sgr A. Pratul Srinivasan, a co-author on the new paper from Google Research, was visiting the team at Caltech at that time. He had played a role in developing a technique called neural radiance fields (NeRF), which was just beginning to be used by researchers and has since made a significant impact on computer graphics. NeRF uses deep learning to generate a 3D representation of a scene based on 2D images, allowing for the observation of scenes from different angles even when only limited views of the scene are available.
Researchers were interested in leveraging recent advancements in neural network representations to reconstruct the 3D environment surrounding a black hole. The main obstacle they faced was the fact that Earth, like anywhere else, can only provide a single viewpoint of the black hole.
However, the team believed that they could potentially overcome this challenge because gas tends to follow predictable patterns as it moves around the black hole. They used the analogy of trying to capture a 3D image of a child wearing an inner tube around their waist to illustrate this point. In order to achieve this using the traditional NeRF method, multiple photos taken fThe child was photographed from different angles while staying in one place. However, in theory, the photographer could ask the child to rotate while taking pictures from a stationary position. The timed photos, along with information about the child’s rotation speed, could be used to accurately reconstruct the 3D scene. Similarly, the researchers aimed to solve the 3D flare reconstruction problem by utilizing knowledge of how gas moves at various distances from a black hole, using measurements taken from Earth over time.
With this understanding, the team developed a version of NeRF that considers the movement of gas around black holes. However, it also requires
The researchers focused on understanding how light moves around large objects like black holes. Co-author Andrew Chael from Princeton University helped develop a computer model to simulate this movement, which is also known as gravitational lensing.
Using this model, the new version of NeRF was able to analyze the structure of the bright features orbiting around the event horizon of a black hole. The initial test results with synthetic data were promising.
Studying a flare around Sgr A
However, the team needed real data for their study. This is where ALMA came into play. The famous image from the EHT provided the initial proof-of-concept, showing encouraging results with synthetic data.
The data used to study Sgr A was collected on calm days in April 6-7, 2017, but a sudden brightening was detected just a few days later on April 11. Maciek Wielgus, a member of the team from the Max Planck Institute for Radio Astronomy in Germany, noticed a signal in the ALMA data from that day that matched the period it would take for a bright spot within the disk to orbit Sgr A. The team aimed to reconstruct the 3D structure of the brightening around Sgr A.ALMA is one of the most powerful radio telescopes in the world. However, due to the vast distance to the galactic center (more than 26,000 light-years), even ALMA does not have the resolution to see Sgr A‘s immediate surroundings. Instead, ALMA measures light curves, which are essentially videos of a single flickering pixel. These are created by collecting all of the radio-wavelength light detected by the telescope for each moment of observation.
Recovering a 3D volume from a single-pixel video may seem impossible. However, by using an additional piece of information about the expected physics for the disk around black hole Sgr A*, it becomes achievable.The team found a way to work around the spatial information limitations in the ALMA data by using roles. The strongly polarized light from the flares offered important clues. ALMA doesn’t just capture a single light curve, it actually provides multiple “videos” for each observation because it records data related to different polarization states of light. In addition to wavelength and intensity, polarization is a fundamental property of light that indicates the orientation of the electric component of a light wave in relation to the wave’s general direction of travel. The data from ALMA consists of two polarized single-pixel videos.Bouman, a Rosenberg Scholar and Heritage Medical Research Institute Investigator, emphasizes the significance of polarized light in providing valuable information. Recent research indicates that hot spots in the gas are highly polarized, with distinct orientation direction for the light waves. This is different from the rest of the gas, which has a more random orientation. The ALMA data collected polarization measurements that provided scientists with valuable insights into localizing the emission in three-dimensional space.Orbital Polarimetric Tomography Explained
A team of researchers have developed an improved method for determining a likely 3D structure to explain their observations. This updated method takes into account the physics of light bending and dynamics around a black hole, as well as the polarized emission expected in hot spots orbiting a black hole. The technique uses a neural network to represent each potential flare structure as a continuous volume, allowing the researchers to progress the initial 3D structure of a hotspot over time as it orbits the black hole. This creates a complete light curve, which can then be solved for.The initial 3D structure of the black hole was chosen to match the ALMA observations over time, resulting in a video displaying the clockwise movement of two compact bright regions around the black hole. Bouman expressed excitement at the outcome, noting that it aligns with computer simulations of black holes, which is very promising. Levis highlighted the interdisciplinary nature of the work, involving both computer scientists and astrophysicists.Physicists and computer scientists have collaborated to create innovative technology that combines numerical modeling of light around black holes with computational imaging. The scientists believe that this is just the start of the potential for this exciting technology. They hope that astronomers can use it to uncover new insights from rich time-series data and gain a better understanding of complex dynamics in various events.
Journal Reference:
- Aviad Levis, Andrew A. Chael, Katherine L. Bouman, Maciek Wielgus, Pratul P. Srinivasan. Orbital polarimetric tomography of a flare near the Sagittarius A* supermassive black hole. Nature Astronomy, 2024; DOI: 10.1038/s41550-024-02238-3