“`html

The US National Science Foundation LIGO (Laser Interferometer Gravitational-wave Observatory) has been referred to as the world’s most accurate measuring device due to its capacity to detect movements smaller than 1/10,000 the diameter of a proton. Through these exceedingly accurate measurements, LIGO, featuring two locations—one in Washington and another in Louisiana—can sense ripples in space-time known as gravitational waves that emanate from colliding celestial objects like black holes.

LIGO initiated the domain of gravitational-wave astronomy in 2015 when it achieved the first direct observation of these disturbances, a finding that subsequently awarded three of its creators the Nobel Prize in Physics in 2017. Enhancements to LIGO’s interferometers now allow it to identify an average of approximately one black hole merger every three days during its ongoing science run. In collaboration with its associates, the Virgo gravitational-wave detector located in Italy and KAGRA in Japan, the observatory has collectively identified hundreds of potential black hole mergers, alongside a few involving at least one neutron star.

Scientists aim to further enhance LIGO’s capabilities, enabling the detection of a wider array of black-hole mergers, including more massive mergers that may belong to a proposed intermediate-mass category bridging the divide between stellar-mass black holes and considerably larger supermassive black holes located at galaxy centers. They also aim to facilitate the detection of black holes with eccentric or elongated orbits and to identify mergers earlier in the coalescing sequence, when the compact bodies spiral closer together.

To accomplish this, scientists at Caltech and Gran Sasso Science Institute in Italy collaborated with Google DeepMind to create a novel AI technique—termed Deep Loop Shaping—that can more effectively minimize undesired noise in LIGO’s sensors. The term “noise” may pertain to a variety of bothersome background disturbances that disrupt data acquisition. This noise can be literal, such as sound waves, but in LIGO’s context, it frequently refers to a minuscule amount of vibration in the gigantic mirrors at the core of LIGO. Excessive vibration can obscure gravitational-wave signals.

Currently, reporting in Science, the researchers demonstrate that this new AI algorithm, although still a proof-of-concept, attenuated motions of the LIGO mirrors by 30 to 100 times more than achievable using conventional noise-reduction techniques alone.

“We were already leading in innovation, performing the most accurate measurements globally, but with AI, we can elevate LIGO’s efficiency to perceive larger black holes,” remarks co-author Rana Adhikari, professor of physics at Caltech. “This advancement will not only enhance LIGO but also aid in constructing LIGO India and even more expansive gravitational-wave observatories.”

This approach might also enhance technologies employing control systems. “In the future, Deep Loop Shaping could also be utilized in numerous engineering challenges related to vibration suppression, noise cancellation, and highly dynamic or unstable systems critical in aerospace, robotics, and structural engineering,” state study co-authors Brendan Tracey and Jonas Buchli, an engineer and scientist, respectively, at Google DeepMind, in a blog post discussing the study.

The Stillest Mirrors

Both the Louisiana and Washington LIGO facilities are designed in vast “L” shapes, where each arm of the L comprises a vacuum tube that contains sophisticated laser apparatus. Within the 4-kilometer-long tubes, lasers oscillate back and forth with the support of enormous 40-kilogram suspended mirrors at each terminus. As gravitational waves arrive on Earth from the cosmos, they distort space-time in such a manner that the length of one arm alters relative to the other by extraordinarily minute amounts. LIGO’s laser system identifies these insignificant, subatomic-length variations in the arms, recording gravitational waves.

However, to reach this degree of accuracy, engineers at LIGO must ensure that background noises are minimized. This study concentrated specifically on unwanted noises, or displacements, in LIGO’s mirrors that arise when the mirrors deviate from the desired orientation by very slight amounts. Despite both LIGO facilities being relatively distanced from the coastline, one prominent source of these mirror vibrations is ocean waves.

“It’s as if the LIGO detectors are positioned at the beach,” elaborates co-author Christopher Wipf, a gravitational-wave interferometer research scientist at Caltech. “Water is moving around on Earth, and the ocean waves generate these low-frequency, slow vibrations that both LIGO facilities are significantly disrupted by.”

The resolution to this challenge functions similarly to noise-canceling headphones, Wipf clarifies. “Envision you are seated on the beach with noise-canceling headphones. A microphone picks up the ocean sounds, followed by a controller sending a signal to your speakers to counteract the wave noise,” he explains. “This mirrors how we manage oceanic and other seismic ground-shaking noise at LIGO.”

However, as is typical with noise-canceling headphones, there is a trade-off. “If you’ve ever used these headphones in a serene environment, you might detect a faint hiss. The microphone generates its own inherent noise. This self-generated noise is what we aim to eliminate in LIGO,” states Wipf.

LIGO already addresses this challenge exceedingly well through a conventional feedback control system. The controller detects the tremor in the mirrors caused by seismic noise and subsequently mitigates these vibrations, although in a manner that introduces a new higher-frequency quiver in the mirrors—akin to the hiss in the headphones. The controller also senses the hiss and continuously adjusts to both types of disturbances to maintain the mirrors as stable as possible. This system is sometimes likened to a waterbed: attempting to calm waves at one frequency results in extra movement at another frequency. Controllers can autonomously detect the disruptions and stabilize the system.

Researchers aspire to further enhance the LIGO control system by diminishing this controller-induced hiss, which disrupts gravitational-wave signals in the lower-frequency segment of the observatory’s spectrum. LIGO detects gravitational waves with frequencies ranging from 10 to 5,000 Hertz (humans perceive sound waves with frequencies from 20 to 20,000 Hertz). The undesirable hiss falls within the 10 to 30 Hertz range—and this is where more massive black hole mergers would be identified, as well as where black holes would be detected near the onset of their final death spirals (for example, the renowned “chirps” perceived by LIGO commence in lower frequencies and then ascend to a higher pitch.)

Approximately four years ago, Jan Harms, a former Caltech research assistant professor who is currently a professor at Gran Sasso Science Institute, contacted specialists at Google DeepMind to investigate whether they could assist in developing
“““html

an AI approach to enhance regulation of vibrations in LIGO’s mirrors. At that moment, Adhikari became engaged, and the researchers initiated collaboration with Google DeepMind to explore various AI techniques. Ultimately, they implemented a method known as reinforcement learning, which effectively instructed the AI algorithm on how to more effectively manage the noise.

“This technique necessitates considerable training,” Adhikari states. “We provided the training data, and Google DeepMind executed the simulations. Essentially, they were operating numerous simulated LIGOs concurrently. You can envision the training as participating in a game. You earn points for diminishing the noise and are penalized for increasing it. The successful ‘participants’ continue striving to excel in the game of LIGO. The outcome is remarkable—the algorithm functions to diminish mirror noise.”

Richard Murray (BS ‘85), the Thomas E. and Doris Everhart Professor of Control and Dynamical Systems and Bioengineering at Caltech, describes that in the absence of AI, scientists and engineers mathematically represent a system they wish to manage in precise detail. “Yet with AI, if you educate it on a model of adequate detail, it can leverage characteristics in the system that you might not have contemplated employing classical methodologies,” he remarks. An authority in control theory for intricate systems, Murray (who is not a co-author on the present research) creates AI instruments for specific control mechanisms, such as those utilized in autonomous vehicles.

“We believe this research will motivate more students to aspire to work at LIGO and become part of this extraordinary advancement,” Adhikari expresses. “We are at the forefront of what can be achieved in measuring minuscule, quantum distances.”

Thus far, the novel AI approach has been evaluated on LIGO for only one hour to confirm that it is functional. The team is eager to perform longer tests and ultimately apply the method across several LIGO systems. “This is an instrument that transforms our perspective on what terrestrial detectors can accomplish,” Wipf remarks. “It renders an extraordinarily challenging issue less intimidating.”

The Science publication entitled “Enhancing cosmological reach of LIGO utilizing Deep Loop Shaping” was partly endorsed by the National Science Foundation, which finances LIGO.

“`