Accelerating Low Frequency Gravitational Wave Inference for LISA with DINGO

The Laser Interferometer Space Antenna (LISA) will open a new window into low frequency gravitational wave science by observing in the 0.1 mHz to 1Hz range. Among the new sources that LISA will detect are mergers of Massive Black Hole Binary systems (MBHBs), whose signals will be detectable for months to years. The long duration and complexity of these signals, combined with LISA's intricate response function, pose a significant computational challenge for the estimation of binary parameters using traditional Bayesian techniques. To address this challenge we adapt DINGO, a neural-network based approach developed for current ground based detectors, to perform fast and accurate inference on MBHB systems. We train networks using LISA's full response function and validate with injected signals in Gaussian noise. This proof of concept work shows that machine learning approaches can be used to vastly reduce the computational cost of estimating binary parameters for LISA sources, a critical step in maximizing the science from the LISA mission.