Systematic Effects of Waveform Models Using DINGO

As ground based gravitational wave detectors increase in sensitivity, the distribution of signal-to-noise ratios for events is expected to increase. Consequently, discrepancies between the waveform model and the true signal will lead to larger disagreement between posteriors obtained via Bayesian inference and the true posterior. This study leverages the Deep Inference for Gravitational Wave Observation (DINGO) code to rapidly estimate posterior distributions for synthetic gravitational wave events using normalizing flow neural networks. We analyze mock precessing binary black hole mergers while varying key intrinsic parameters using models trained on phenomenological, effective-one-body, and NR surrogate waveform families. We inject identical NR surrogate signals into each model to identify systematic effects of waveform models on binary black hole parameter estimation. We assess the similarity of the mock posterior distributions and introduce a novel multivariate definition of bias based on the Mahalanobis distance. As expected from waveform mismatches we find large discrepancies in the high spin and precession regime.