In a paper published in Science on September 4, 2025, Brendan Tracey and Jonas Buchli introduced a novel method called Deep Loop Shaping, leveraging AI to improve control of gravitational wave observatories, aiding astronomers in understanding the dynamics and formation of the universe. Our team has focused on stabilizing one of the most sensitive observational instruments to support studies of the universe's most powerful processes.
Deep Loop Shaping reduces noise and enhances control in the feedback systems, stabilizing components necessary for measuring gravitational waves—tiny ripples caused by events like neutron star collisions and black hole mergers. This research will assist astronomers in gathering critical data to better test fundamental theories of physics and cosmology.
Developed in collaboration with the Laser Interferometer Gravitational-Wave Observatory (LIGO) operated by Caltech and the Gran Sasso Science Institute (GSSI), we validated our method at the LIGO observatory in Livingston, Louisiana. LIGO measures the properties and origins of gravitational waves with incredible accuracy, but even the slightest vibrations, such as waves crashing 100 miles away, can disrupt its measurements. LIGO relies on thousands of control systems to maintain near-perfect alignment and adapts to environmental disturbances with continuous feedback.
Our Deep Loop Shaping method reduces noise levels in LIGO's most unstable feedback loop by 30 to 100 times, improving the stability of its highly sensitive interferometer mirrors. Applying this method could enable astronomers to detect hundreds more events per year in far greater detail.
LIGO uses laser light interference to measure the properties of gravitational waves. Since first detecting gravitational waves produced by colliding black holes in 2015, LIGO's measurements have significantly changed our understanding of the universe. With LIGO, astronomers have detected hundreds of black hole and neutron star collisions, confirmed the existence of binary black hole systems, and studied the creation of heavy elements like gold.
We aimed to improve the most challenging part of the control system and extend the reach of our observational capabilities. Studying the universe using gravity instead of light is akin to listening rather than looking. Our work allows us to tune into the universe's deeper information.
Deep Loop Shaping employs a reinforcement learning approach using frequency domain rewards that surpass current state-of-the-art feedback control performance. In a simulated LIGO environment, we trained a controller to avoid amplifying noise in the observation band used for measuring gravitational waves. Through repeated interaction, the controller learns to stabilize the mirrors without adding harmful control noise, bringing noise levels down below those caused by quantum fluctuations in the radiation pressure of light.
We tested our controllers on the real LIGO system in Livingston, Louisiana, finding they performed as well in hardware as in simulation. Our results show that Deep Loop Shaping controls noise 30 to 100 times better than existing controllers and has eliminated the most unstable and difficult feedback loop as a meaningful source of noise for the first time.
By applying Deep Loop Shaping to LIGO's entire mirror control system, we have the potential to eliminate noise from the control system itself, paving the way for expanding its cosmological reach. We expect our work to influence the design of future observatories, both on Earth and in space, ultimately helping connect missing links throughout the universe for the first time.
Blogger's Review: The introduction of Deep Loop Shaping not only represents a technological breakthrough in gravitational wave research but also exemplifies the application of AI in scientific inquiry. By effectively controlling noise, this method will greatly enhance observational precision and push the frontiers of astronomy. In the future, similar technologies may play vital roles in other engineering fields.