Unlocking New Paths for Science with Extreme-Mass-Ratio Inspirals: Machine Learning-Enhanced MCMC for Accurate Parameter Inversion

Abstract: The detection of gravitational waves from extreme-mass-ratio inspirals (EMRIs) in space-borne antennas like LISA and Taiji promises deep insights into strong-field gravity and black hole astrophysics. However, the complex, non-convex likelihood landscapes of EMRI signals (compounded by instrumental noises) have long hindered reliable parameter estimation based on traditional Markov Chain Monte Carlo (MCMC) methods, which often fail to escape local optima or require impractical computational costs. To address this critical bottleneck, we introduce Flow-Matching Markov Chain Monte Carlo (FM-MCMC), a pioneering Bayesian framework that synergizes continuous normalizing flows (CNFs) with parallel tempering MCMC (PTMCMC). By leveraging CNFs to rapidly explore high-dimensional parameter spaces and PTMCMC for precise posterior sampling, FM-MCMC achieves unprecedented efficiency and accuracy in recovering EMRI intrinsic parameters. By enabling real-time, unbiased parameter inference, FM-MCMC unlocks the full scientific potential of EMRI observations, and would serve as a scalable pipeline for precision gravitational-wave astronomy.