Improving cosmological reach of LIGO usingDeep Loop Shaping

Abstract

Gravitational wave (GW) detectors have revolutionized astrophysics, but their low-frequency bandremains underexplored. This band is crucial for studying intermediate-mass black holes, binary blackhole eccentricity, and providing early warnings for multi-messenger observations of binary neutron starmergers. We tackle the challenge of disturbance rejection in this band through deep loop shaping, bytraining a reinforcement learned (RL) policy with frequency domain rewards and develop a control systemthat avoids injecting noise. Our RL-based controller, trained in simulation and deployed in real-worldoperation, demonstrates effective handling of complex control problems while maintaining stability androbustness. This approach enhances current GW detectors and holds promise for future observatories.Our results highlight RL’s potential for improving disturbance rejection and noise suppression in GWdetection and other control systems.

Authors

Jonas Buchli, Brendan Tracey, Justin Chiu, Matthias Lochbrunner, Craig Donner, Roland Hafner, Martin Riedmiller

Venue

Science