Gravitational Waves Detected
LIGO measures gravitational waves from merging black holes for the first time on 14 September 2015; over 200 events are now cataloged.
Open in the interactive tree →Einstein predicted the waves in 1916. LIGO uses 4-km laser interferometers to detect length changes far smaller than a proton. GW150914, a merger of two black holes about 1.3 billion light-years away, was announced in February 2016 and earned the 2017 Nobel Prize (Weiss, Barish, Thorne). In 2017 a neutron-star merger was seen in both gravitational waves and light.
As of October 2026
The GWTC-4.0 catalog (26 August 2025) lists 218 candidate events, 128 of them new from observing run O4a (May 2023 to January 2024). The signal GW250114 (14 January 2025, signal-to-noise about 80, the loudest yet) confirmed Hawking's 1971 area theorem (total horizon area grew from about 240,000 to about 400,000 km^2) and the Kerr black-hole solution (Physical Review Letters, 10 September 2025). The fourth observing run ended on 18 November 2025.
Open steps
- Learned noise control at every site Medium AI leverageMove the reinforcement-learning mirror controller from one LIGO site to all detectors and runs, aimed at the low-frequency band where control noise limits range.
- Waveform models for loud signals High AI leverageMake waveform models accurate enough for events with signal-to-noise near 80 and beyond, where model error rather than detector noise will limit tests of general relativity.
- Seconds-fast inference for alerts High AI leverageRun full parameter estimation for neutron-star mergers in seconds, so telescopes can point at the fireball before it fades; longer signals in upgraded detectors make this harder.
- Next-generation detector design Medium AI leverageOptimize arm length, mirror coatings, cryogenics and site noise models for the next generation; LIGO's A-sharp upgrade after the fifth run is the pathfinder.
Where AI could help
Medium AI leverage. AI already helps detector control and event analysis, but sensitivity gains depend on new hardware, detectors and observing time.
- Learned controllers that reduce control noise and widen the sensitive frequency band
- Neural inference that characterizes events in seconds to alert telescopes before a merger
- Search for weak, overlapping or unmodeled signals with neural filters and anomaly detection
Shown so far
- In September 2025 Science published Deep Loop Shaping, a reinforcement-learning controller that cut mirror control noise at LIGO Livingston by a factor of 30 in the 10 to 30 Hz band (up to 100 in sub-bands). source
- In March 2025 Nature published DINGO-BNS, a neural network that characterizes merging neutron stars in about a second, against about an hour for the fastest conventional methods. source
Prerequisites
- Wave Optics (Young, Fresnel)1801LIGO detects gravitational waves with light interference
- Michelson-Morley Experiment1887LIGO is a giant Michelson interferometer
- Theory of Relativity1905
- Laser1960
- Pulsars1967The Hulse-Taylor binary pulsar gave the first evidence of gravitational waves
- Black Holes (Cygnus X-1)1971-72LIGO's first signal came from merging black holes