Improving cosmological reach of a gravitational wave observatory using Deep Loop Shaping

Area of Science:

• Gravitational Wave Astronomy
• Astrophysics
• Machine Learning in Scientific Instrumentation

Background:

• Enhanced low-frequency sensitivity in gravitational wave observatories is crucial for studying intermediate-mass black hole mergers, binary black hole eccentricity, and enabling early warnings for multimessenger observations of binary neutron star mergers.
• Current mirror stabilization control systems in observatories like LIGO introduce detrimental noise, hindering significant improvements in detector sensitivity.

Purpose of the Study:

• To eliminate noise injected by mirror stabilization control systems, a major obstacle to improving gravitational wave observatory sensitivity.
• To demonstrate a novel reinforcement learning approach for enhancing gravitational wave detector performance.

Main Methods:

• Implementation of Deep Loop Shaping, a reinforcement learning technique utilizing frequency domain rewards.
• Validation of the Deep Loop Shaping methodology on the LIGO Livingston Observatory (LLO).

Main Results:

• The developed controller significantly reduced control noise in the 10- to 30-hertz band by over 30 times.
• Subband noise reduction reached up to 100 times, exceeding the sensitivity improvement goal motivated by the quantum limit.
• Successful noise elimination in a real-world gravitational wave observatory setting.

Conclusions:

• Deep Loop Shaping offers a powerful solution for mitigating noise in gravitational wave observatories.
• This AI-driven control strategy has the potential to enhance current and future gravitational wave detection systems.
• The methodology's broad applicability extends to improving various instrumentation and control systems beyond gravitational wave astronomy.