AI stabilizes LIGO to detect gravitational waves

Artificial intelligence is already helping LIGO measure gravitational waves with less interference. The Deep Loop Shaping method, developed by Google DeepMind together with LIGO, Caltech, and GSSI, reduced the control system's noise by 30 to 100 times in the observatory's most unstable control loop.

The advance was tested at LIGO's facility in Livingston, Louisiana, and is described in a paper published in Science. The AI does not interpret data from the universe directly. It keeps the mirrors still so the detector can collect that data.

The problem: measuring movements that are almost impossible to see

LIGO detects gravitational waves, tiny distortions in space-time produced by events such as merging black holes and colliding neutron stars.

The observatory uses two 4-kilometer arms and lasers that bounce between mirrors. When a wave passes, it slightly changes the distance between those mirrors. The variation being measured is just 10^-19 meters, close to one ten-thousandth of the size of a proton.

Any vibration can ruin the measurement. Vehicle traffic, weather conditions, and even waves in the Gulf of Mexico, about 160 kilometers away, can affect the instrument.

LIGO has thousands of systems that continuously correct the position of its components. But the control system can also generate noise: if it acts too forcefully, it amplifies the vibrations it was supposed to remove.

How AI gets involved

Deep Loop Shaping uses reinforcement learning, a technique in which a system learns through repeated attempts and receives higher scores when it gets closer to its goal. Here, the goal was to reduce noise precisely at the frequencies where LIGO searches for gravitational waves.

The controller was first trained in a simulation of the observatory. It was then run on the real equipment in Livingston, where it maintained similar performance and remained stable during extended experiments.

The main result affects LIGO's most difficult feedback loop to control. A 30 to 100 times reduction in noise means that this loop was no longer a significant source of interference for the detector.

What could change for astronomy

The team does not yet claim that all of LIGO's systems have been upgraded with this method. However, applying the technique to all the circuits that control the mirrors could allow hundreds of additional events to be detected each year and produce more detailed data about them.

That would be especially useful for studying intermediate-mass black holes, a population about which scientists still have little data and which could help explain how galaxies evolve.

Since its first detection of gravitational waves in 2015, LIGO has observed hundreds of collisions involving black holes and neutron stars. These measurements have helped test predictions from the theory of general relativity and study phenomena such as the formation of heavy elements, including gold.

For you, the effect will not be a visible feature on your phone or a new everyday tool. The change is behind the scenes: a more stable detector can extend the distance at which astronomers observe faint events and improve the precision with which they reconstruct what happened.

The same technique could later be tested on other problems where vibrations and noise make control difficult, including robotics, structural engineering, and the aerospace industry. For now, the key question is whether the result obtained in one specific circuit can be maintained when the method is extended to the entire observatory and to future ground- or space-based detectors.