LISA Data Analysis using MCMC methods

TL;DR

This paper demonstrates the successful application of MCMC methods to analyze simulated LISA gravitational wave data, effectively identifying signals and estimating parameters in a complex, high-dimensional space.

Contribution

It introduces the first use of MCMC techniques for LISA data analysis, showcasing their efficiency in handling large parameter spaces and overlapping signals.

Findings

MCMC accurately identifies the number of signals.

Parameter estimates are consistent with Fisher matrix predictions.

The approach effectively explores a 50,000-dimensional parameter space.

Abstract

The Laser Interferometer Space Antenna (LISA) is expected to simultaneously detect many thousands of low frequency gravitational wave signals. This presents a data analysis challenge that is very different to the one encountered in ground based gravitational wave astronomy. LISA data analysis requires the identification of individual signals from a data stream containing an unknown number of overlapping signals. Because of the signal overlaps, a global fit to all the signals has to be performed in order to avoid biasing the solution. However, performing such a global fit requires the exploration of an enormous parameter space with a dimension upwards of 50,000. Markov Chain Monte Carlo (MCMC) methods offer a very promising solution to the LISA data analysis problem. MCMC algorithms are able to efficiently explore large parameter spaces, simultaneously providing parameter estimates, error analyses and even model selection. Here we present the first application of MCMC methods to simulated LISA data and demonstrate the great potential of the MCMC approach. Our implementation uses a generalized F-statistic to evaluate the likelihoods, and simulated annealing to speed convergence of the Markov chains. As a final step we super-cool the chains to extract maximum likelihood estimates, and estimates of the Bayes factors for competing models. We find that the MCMC approach is able to correctly identify the number of signals present, extract the source parameters, and return error estimates consistent with Fisher information matrix predictions.