Accelerating global parameter estimation of gravitational waves from Galactic binaries using a genetic algorithm and GPUs

TL;DR

This paper presents a GPU-accelerated genetic algorithm approach to efficiently estimate parameters of Galactic binary signals in LISA gravitational wave data, significantly reducing computation time.

Contribution

It introduces a novel method combining genetic algorithms and GPU computing to rapidly analyze complex LISA data with overlapping signals.

Findings

Posterior distributions computed in 2 seconds per signal

Accelerated analysis enables real-time parameter estimation

Effective separation of overlapping Galactic binary signals

Abstract

The Laser Interferometer Space Antenna (LISA) is a planned space-based gravitational wave telescope with the goal of measuring gravitational waves in the milli-Hertz frequency band, which is dominated by millions of Galactic binaries. While some of these binaries produce signals that are loud enough to stand out and be extracted, most of them blur into a confusion foreground. Current methods for analyzing the full frequency band recorded by LISA to extract as many Galactic binaries as possible and to obtain Bayesian posterior distributions for each of the signals are computationally expensive. We introduce a new approach to accelerate the extraction of the best fitting solutions for Galactic binaries across the entire frequency band from data with multiple overlapping signals. Furthermore, we use these best fitting solutions to omit the burn-in stage of a Markov chain Monte Carlo method and to take full advantage of GPU-accelerated signal simulation, allowing us to compute posterior distributions in 2 seconds per signal on a laptop-grade GPU.