Neural Bayesian updating for sequential population inference on growing gravitational-wave catalogs

As gravitational-wave catalogs grow, they will become increasingly computationally expensive to analyze in their entirety, especially when inferring astrophysical source populations with high-dimensional, flexible models. Bayesian statistics offers a natural remedy, letting us update our knowledge of physical models as new data arrive, without re-analyzing existing data. However, doing so requires the posterior probability of model parameters, which may be computationally intractable in high-dimensional spaces or when likelihoods are computed over large datasets. Here, we use variational neural posterior estimation to rapidly update the inferred population of binary black holes as data are observed in gravitational-wave detectors. We demonstrate Bayesian sequential updates of the population distribution between each LIGO-Virgo-Kagra observing run and as each event arrives, while employing both low- and high-dimensional population models. Then, we evaluate the computational savings — in terms of time and memory — provided by the reuse of prior population analyses. Finally, we outline potential scientific applications enabled by Bayesian updating, such as identifying informative events in the binary black hole population.