First-ever black hole image receives an impressive AI upgrade.

 A supercomputer has given a distant supermassive black hole a makeover, making it look crisp.

 

With the use of machine learning, the "fuzzy orange donut" visible in the first black hole image ever obtained has shrunk to a slimmer "skinny golden ring."

 

The supermassive black hole in the centre of the galaxy Messier 87 (M87) could be more clearly understood by redefining this image, which could also be applied to the black hole at the centre of our own galaxy, the Milky Way.

The Event Horizon Telescope (EHT) captured the first-ever picture of the M87 supermassive black hole, also known as M87*, and released it to the public in 2019. The EHT gathered the information across a number of days in 2017 in order to produce the image.

 

At left is the famous image of the M87 supermassive black hole originally published by the Event  Horizon Telescope collaboration in 2019. At right is a new image of the black hole generated by the PRIMO algorithm using the same data set. (Image credit: L. Medeiros (Institute for Advanced Study), D. Psaltis (Georgia Tech), T. Lauer (NSF’s NOIRLab), and F. Ozel (Georgia Tech))

The EHT is a network of seven telescopes located all over the world that together form an Earth-sized telescope, yet despite their combined viewing capability, the data they gather still has gaps, much like a jigsaw puzzle with missing pieces.

 

In order to "fill in the gaps" in the M87 image and raise the EHT array to its highest resolution for the first time, a team of researchers led by EHT collaboration member and astrophysics postdoctoral fellow Lia Medeiros used a novel machine learning technique called principal-component interferometric modelling, or "PRIMO".

"Since we cannot study black holes up close, the detail of an image plays a critical role in our ability to understand its behaviour," said study lead author Medeiros in a statement(opens in new tab). Our theoretical models and testing of gravity will be strongly constrained by the fact that the breadth of the ring in the photograph has shrunk by a factor of around two.

 

Scientists were astonished by how closely the M87 supermassive black hole (M87*), which is 55 million light-years away from Earth and has a mass equal to six and a half billion suns, matched predictions made by Albert Einstein's 1915 general theory of relativity.

Scientists now have the opportunity to more closely compare observations of a real black hole to theoretical predictions thanks to this PRIMO-refined image of M87*.

 

According to EHT member and NOIRLab researcher Tod Lauer, "PRIMO is a new approach to the challenging task of constructing images from EHT observations." It offers a means of making up for the information about the object being viewed that is needed to produce the image that would have been seen using a single enormous radio telescope the size of the Earth.

Training PRIMO to build a better black hole

According to the Princeton, New Jersey-based Institute for Advanced Study, PRIMO uses dictionary learning, a subfield of machine learning that enables computers to produce rules based on enormous amounts of training data. So, for instance, a programme like this can learn to recognise whether an image of an unknown object is a banana or not if it is fed a number of photographs of a banana.

 

The scientists provided PRIMO 30,000 high-fidelity simulated photos of these cosmic titans as they "accreted" gas from their surroundings in order to educate PRIMO to do the same thing with black holes. The photos allowed PRIMO to look for trends by displaying a variety of theoretical assumptions about how black holes accrete stuff.

Following their discovery, these patterns were arranged according to how frequently they appeared in simulations. As a result, a high-fidelity image of M87* might be produced using EHT photos, revealing any structures the telescope array may have missed.

 

Through the use of machine learning, we are "using physics to fill in regions of missing data in a way that has never been done," according to Medeiros. This may have significant effects on interferometry, which is used in everything from astronomy to medicine.

EHT measurements and theoretical black hole models are consistent with the final image produced by PRIMO. According to these simulations, M87*'s dazzling ring is the result of gas being driven to near-light speeds by the black hole's powerful gravitational pull. This causes the gas to flare and heat up as it whirls around the event horizon, the light-trapping boundary that defines the black hole's outermost region.

"We have skillful another landmark, producing an image that uses the entire determination of the display for the first time," said Psaltis. "Approximately four years after the first horizon-scale image of a black hole was unveiled by EHT in 2019, we have marked another milestone. We now have a fantastic opportunity to comprehend black hole physics because to the new machine learning tools we have created.

Now, the image of the Milky Way's supermassive black hole might be processed using the PRIMO approach. In May 2022, a photograph of the Sagittarius A* (Sgr A*) supermassive black hole, which is smaller but much closer, was made public by the EHT. While Sgr A*, a four million solar mass black hole located 26,000 light-years from Earth, was generated using data from the EHT also obtained in 2017, the lower size of Sgr A* made the data more challenging to refine.

 

Estimates of the properties of both supermassive black holes, such as their mass, size, and rate of matter consumption, may be improved by using PRIMO to increase the resolution of EHT images.

Reference: Space.com