Simulation-based inference for AGN jet population modelling: Towards more robust comparisons of black hole jet speeds
We present the most complete modelling of the MOJAVE 1.5 Jansky Quarter Century active galactic nuclei (AGN) jet population, using likelihood-free simulation-based inference. Due to the complex impact of a flux-limit on observed AGN data sets, careful modelling of the parent population is required. In particular, when observing and fitting to multiple data distributions likelihoods become non-intuitive. Parameter degeneracies further complicate the problem and make it suitable for likelihood-free, simulation-based inference. This method relies on a normalising flow learning the likelihood surface or posteriors directly. We extensively validate the flow to show that previous parameter estimates for the AGN jet speed distributions underestimated parameter errors significantly and do not capture the non-gaussianity of the parameter posteriors. The new results enable a better statistical comparison to other AGN population studies, but also a more accurate comparison of supermassive black hole jets with their lower mass counterparts, X-ray binaries (XRB). We find that the AGN follow a Lorentz factor distribution of the shape $N(\Gamma) \propto \Gamma^b$ with $b= -1.32_{-0.19}^{+0.20}$. This slope is consistent with the XRB Lorentz factor distribution at $2\sigma$. Simulation-based inference as a method is generally well-suited to many astrophysical problems, and this paper shows the convenient applicability of this methodology to parent population studies of jetted AGN with multiple observables specifically.