In this talk we present in-depth state-of-the-art parameter estimation studies of several gravitational wave events with open LIGO-Virgo data.
We use the fourth generation of phenomenological waveform models, the frequency domain IMRPhenomX and the time domain IMRPhenomT families which include higher harmonics and precession, and constitute the computationally most efficient inspiral-merger-ringdown frequency and time domain waveform models for binary black holes with are currently available in the LALSuite software framework. We systematically compare the Bilby and LALInference frameworks for parameter estimation, and perform highly parallel simulations in a traditional supercomputing environment. We discuss our automatisation of Bayesian inference runs, and the use of a machine learning model in order to predict the duration of a PE run or its number of likelihood evaluations. Due to the computational efficiency of the waveform models we can perform systematic tests of different priors, sampler settings, waveform models and inference codes. We discuss the computational cost, and the wall-clock time required to get fast parameter estimation results, and aspects of the systematic uncertainties in the waveform models. Regarding results, special emphasis is put on the intermediate mass black hole GW190521 event, which is astrophysically particularly interesting, but also particularly challenging due to the shortness of the signal.