Distinguishing signal from noise : new techniques for gravitational wave data analysis

Abstract

The principal problem of gravitational wave detection is distinguishing true gravitational wave signals from non-Gaussian noise artifacts. We describe two methods to deal with the problem of non-Gaussian noise in the Laser Interferometer Gravitational Observatory (LIGO). Perturbed black holes (BH) are known to vibrate at determinable quasi-normal mode frequencies. These vibrational modes are strongly excited during the inspiral and merger of binary BH systems. We will develop a template based search for gravitational waves from black hole ringdowns: the final stage of binary merger. Past searches for gravitational waves developed ad hoc detection statistics in an attempt to separate the expected gravitational wave signals from noise. We show how using the output of a multi-variate statistical classifier trained to directly probe the high dimensional parameter space of gravitational waves can improve a search over more traditional means. We conclude by placing preliminary upper limits on the rate of ringdown producing binary BH mergers. LIGO data contains frequent, non-Gaussian, instrument artifacts or glitches. Current LIGO searches for un-modeled gravitational wave bursts are primarily limited by the presence of glitches in analyzed data. We describe the BayesWave algorithm, wherein we model gravitational wave signals and detector glitches simultaneously in the wavelet domain. Using bayesian model selection techniques and a reversible jump Markov chain Monte Carlo, we are able determine whether data is consistent with the presence of gravitational waves, detector glitches, or both. We demonstrate BayesWave's utility as a data quality tool by fitting glitches non-Gaussian LIGO data. Finally, we discuss how BayesWave can be extended into a full-fledged search for gravitational wave bursts.