Bayesian Inference Analysis of Unmodelled Gravitational-Wave Transients (1807.01939v3)

Abstract: We report the results of an in-depth analysis of the parameter estimation capabilities of BayesWave, an algorithm for the reconstruction of gravitational-wave signals without reference to a specific signal model. Using binary black hole signals, we compare BayesWave's performance to the theoretical best achievable performance in three key areas: sky localisation accuracy, signal/noise discrimination, and waveform reconstruction accuracy. BayesWave is most effective for signals that have very compact time-frequency representations. For binaries, where the signal time-frequency volume decreases with mass, we find that BayesWave's performance reaches or approaches theoretical optimal limits for system masses above approximately 50 M_sun. For such systems BayesWave is able to localise the source on the sky as well as templated Bayesian analyses that rely on a precise signal model, and it is better than timing-only triangulation in all cases. We also show that the discrimination of signals against glitches and noise closely follow analytical predictions, and that only a small fraction of signals are discarded as glitches at a false alarm rate of 1/100 y. Finally, the match between BayesWave- reconstructed signals and injected signals is broadly consistent with first-principles estimates of the maximum possible accuracy, peaking at about 0.95 for high mass systems and decreasing for lower-mass systems. These results demonstrate the potential of unmodelled signal reconstruction techniques for gravitational-wave astronomy.