Gravitational wave inference with marginalization over waveform uncertainty

Errors in modeling of gravitational waves (GW) may lead to systematic errors in Bayesian parameter estimation of compact binary coalescences, in particular for next-generation GW detectors. While numerical relativity (NR) simulations provide the most accurate waveforms for binary black hole mergers, they are expensive to compute and provide limited coverage of the parameter space. Traditionally, efforts to improve waveform models have been focused on accuracy and evaluation speed. Moreover, uncertainties in the waveform models arising from the modeling process and inputs are usually ignored, but systematic biases in parameter estimation can also be reduced by marginalizing over waveform uncertainties as proposed by Moore & Gair. Effective-one-body (EOB) models tune internal calibration parameters against NR simulations. We introduce an SEOBNRv4-based EOB model which incorporates uncertainty from calibration against NR via Gaussian process regression. Using this model, we perform Bayesian inference on a set of synthetic GW signals in zero noise, for a range of mass ratios and effective spins while marginalizing over waveform model uncertainty. We observe that in many cases, incorporating uncertainty into the waveform results in more accurate binary parameter estimates, at the expense of less precise posterior distributions.