Fisher vs. Bayes : A comparison of parameter estimation techniques for massive black hole binaries to high redshifts with eLISA

Massive black hole binaries are the primary source of gravitational waves (GW) for the future eLISA observatory. The detection and parameter estimation of these sources to high redshift would provide invaluable information on the formation mechanisms of seed black holes, and on the evolution of massive black holes and their host galaxies through cosmic time. The Fisher information matrix has been the standard tool for GW parameter estimation in the last two decades. However, recent studies have questioned the validity of using the Fisher matrix approach. For example, the Fisher matrix approach sometimes predicts errors of $\geq100\%$ in the estimation of parameters such as the luminosity distance and sky position. With advances in computing power, Bayesian inference is beginning to replace the Fisher matrix approximation in parameter estimation studies. In this work, we conduct a Bayesian inference analysis for 120 sources situated at redshifts of between $0.1\leq z\leq 13.2$, and compare the results with those from a Fisher matrix analysis. The Fisher matrix results suggest that for this particular selection of sources, eLISA would be unable to localize sources at redshifts of $z\lesssim6$. In contrast, Bayesian inference provides finite error estimations for all sources in the study, and shows that we can establish minimum closest distances for all sources. The study further predicts that we should be capable with eLISA, out to a redshift of at least $z\leq13$, of predicting a maximum error in the chirp mass of $\lesssim 1\%$, the reduced mass of $\lesssim20\%$, the time to coalescence of 2 hours, and to a redshift of $z\sim5$, the inclination of the source with a maximum error of $\sim60$ degrees.