Total: 1
Reverberation mapping of active galactic nuclei uses the time taken for variability to travel through the system as a measure of system size, with this size being used to inform estimates of the mass of the central black hole, a technique that has graduated to an industrial-scale effort in recent decades. Such large target-count, multi-year surveys have been consistently undercut by poor light curve quality and the aliasing problems of seasonal observations. These impacts compound with statistical approaches that are not consistent between works and sometimes combine inconsistent statistical frameworks. Here, we present a new end-to-end Bayesian framework for fitting lags at scale in large surveys, simultaneous with the fitting of population models such as the radius-luminosity relation, and formulated in a numerically tractable way. This procedure is a statistically and numerically robust generalisation of the stacking and quality cut procedures of past works, fully marginalising over data uncertainty and data quality. Our approach remains numerically stable when scaling to larger surveys, and can act as a basis for any Bayesian population modelling of large sets of AGN time series observations. Applied to the OzDES sample, we demonstrate that this new approach yields more than double the constraining power on population-level properties like the R-L relationship, and an order-of-magnitude increase in the number of recovered lags and masses. With this method, we go from 24 high quality lag recoveries with traditional methods to a new effective sample size of ~297, all while being robust against random chance false positive measurements.