Detecting compact galactic binaries using a hybrid swarm-based algorithm

TL;DR

This paper introduces a hybrid swarm-based algorithm combining Particle Swarm Optimization and Differential Evolution to improve detection and parameter estimation of galactic binary sources of gravitational waves in noisy data for the eLISA mission.

Contribution

It presents the first application of a hybrid swarm algorithm to gravitational wave data analysis, demonstrating faster convergence and successful detection of multiple sources.

Findings

Successfully detected single and multiple binaries in simulated data.

Recovered source parameters within 99% credible intervals.

Outperformed traditional stochastic methods in convergence speed.

Abstract

Compact binaries in our galaxy are expected to be one of the main sources of gravitational waves for the future eLISA mission. During the mission lifetime, many thousands of galactic binaries should be individually resolved. However, the identification of the sources, and the extraction of the signal parameters in a noisy environment are real challenges for data analysis. So far, stochastic searches have proven to be the most successful for this problem. In this work we present the first application of a swarm-based algorithm combining Particle Swarm Optimization and Differential Evolution. These algorithms have been shown to converge faster to global solutions on complicated likelihood surfaces than other stochastic methods. We first demonstrate the effectiveness of the algorithm for the case of a single binary in a 1 mHz search bandwidth. This interesting problem gave the algorithm plenty of opportunity to fail, as it can be easier to find a strong noise peak rather than the signal itself. After a successful detection of a fictitious low-frequency source, as well as the verification binary RXJ0806.3+1527, we then applied the algorithm to the detection of multiple binaries, over different search bandwidths, in the cases of low and mild source confusion. In all cases, we show that we can successfully identify the sources, and recover the true parameters within a 99\% credible interval.