Improving Accuracy of Precessing Binary Black Hole Waveform Templates Using Gaussian Process Regression

With the first data run for advanced LIGO (aLIGO) in Fall 2015 we are quickly approaching a regime in which detection of gravitational waves is highly plausible . Not only do the advanced detectors represent the most promising experimental verification of Einstein’s theory of general relativity, but they additionally usher in the next era of astronomy, bringing in a new way of exploring our universe and complimenting existing astronomical observations. These observations have the potential to transform our understanding of astrophysical objects, but only if the source parameters can be accurately estimated. Such parameter estimation is currently dependent on the usage of waveform templates. Accurate templates can be computed using numerical methods, but the prohibitive cost of these simulations means this can only be done for a small handful of parameters. Inference will therefore be reliant on approximate, analytical waveforms. In [?] a novel method to fold model uncertainties into data analysis is proposed; the waveform uncertainty is analytically marginalized over using a prior distribution constructed by using Gaussian process regression to interpolate the waveform difference from a small training set of accurate templates. We present the first steps to applying this method using the latest binary black hole (BBH) Numerical Relativity (NR) catalog produced by the GAtech NR group, and analytic waveforms from the model PhenomP in the LAL database. In addition, we present the results of using Gaussian process regression to produce error estimates in the catalog’s seven dimensional parameter space with the aim that such information can guide the NR community to regions of parameter space where new simulations are most needed.