“The blurry orange doughnut” seen in the first-ever image of a black hole has been reduced to a “skinnier golden ring” with the help of an AI that improves black hole images. Redefinition of this image of the M87 galaxy’s black hole could help understand its characteristics better. The technique could be applied to the Sagittarius A* black hole in the heart of our galaxy, the Milky Way.
PRIMO technique
Historic image of the supermassive black hole M87, known as M87*, was captured by the Event Horizon Telescope (EHT). The data to create the image was collected by the EHT over several days two years prior. The EHT is a network of seven telescopes worldwide that create a telescope the size of Earth. But despite its combined observational power, there are still gaps in the data it collects, just like missing puzzle pieces.
A team of researchers, including Lia Medeiros, a member of the EHT collaboration, used a new machine learning technique. The technique called Principal Component Interferometric Modelling (PRIMO) “fills in the gaps” in the M87* image and maximizes its resolution.
“As we cannot study black holes up close, the details of an image play a critical role in our ability to understand their behavior” Medeiros said in a statement (ref.). “The width of the ring in the image is now smaller by about a factor of two. This will be a powerful constraint for our theoretical models and gravity tests”.
Machine learning
When the image of the M87* supermassive black hole was first revealed, scientists were surprised by its correspondence with Albert Einstein‘s general theory of relativity predictions. This refined PRIMO image of M87* offers scientists the opportunity to better match observations of a real black hole to theoretical predictions.
“PRIMO is a new approach to the difficult task of building images from EHT observations” said EHT member Tod Lauer in the statement. “It provides a way to compensate for missing information on the observed object. We generate an image that would have been seen using a single giant radio telescope the size of Earth”.
The Institute for Advanced Study at Princeton, New Jersey, explained that PRIMO operates using dictionary learning, a branch of machine learning that allows computers to generate rules based on large sets of training material. For example, if a program like this is given a number of pictures of an apple, it can learn to determine whether the image of an unknown object is an apple or not.
To train PRIMO with black holes, the team provided it with 30,000 high-fidelity simulated images of these cosmic titans feeding on surrounding gas, the “accretion” phase. The images covered a wide range of theoretical predictions on how black holes grow matter, allowing PRIMO to search for logical patterns.
Integration of AI in images
Once identified, these patterns were ordered based on how often they were considered in the simulations. The process can be incorporated into EHT images to generate a high-fidelity image of M87* and reveal structures that the telescope cannot perceive.
“We are using physics to fill in missing data regions in a way that has never been done before using machine learning” Medeiros explained. “This could have important implications for interferometry, which plays a role in fields ranging from exoplanets to medicine”.
The resulting image rendered by PRIMO agrees with EHT data and theoretical models of black holes. These models explain that the luminous ring is the result of gas acceleration at speeds near the speed of light from the incredible gravitational influence of the black hole. This causes the gas to heat up and glow as it rushes around the surface that traps the light. The region that forms the outer boundaries of the black hole is called the event horizon.