AGN Reverberation Mapping with LITMUS: Fundamental Limits on lag Recovery Rates

Abstract

Reverberation mapping of active galactic nuclei provides one of the most direct probes of the geometry and kinematics of the broad-line region by measuring time delays between continuum and line variability. Modern RM surveys frequently suffer difficulties with lag measurements due to poor signal to noise and aliasing, whereby multimodal lag posterior distributions arise due to seasonal gaps in our data. These challenge the reliability of commonly used fitting tools such as JAVELIN, which can return a high rate of false positives. We implement a new lag measurement package, LITMUS, and introduce a new framework that uses Bayesian evidence to identify false positive lag measurements, as well as examine the question of how many AGN present detectable lags in high redshift industrial scale surveys like OzDES and SDSS. Our analysis differs from previous RM studies in six key respects: (i) our inference is robust to the previously under-diagnosed numerical component of aliasing, (ii) we use a consistent methodology for all sources, (iii) uncertainty in the underlying AGN variability is fully marginalised, (iv) lag significance is assessed via Bayesian model comparison rather than heuristic metrics, (v) false-positive rates are quantified by comparison against random-chance recoveries and (vi) we use marginal likelihoods to distinguish between sources where a lag is not detectable in our data and sources that show no evidence of reverberation. Applied to the OzDES sample, we find that previous RM studies are likely to have overestimated the confidence of recovered lags, and we find a stark contrast between a low reverberation percentage for the MgII line (3-28% depending on assumptions) and much higher percentages in the CIV and especially the H$β$ line, which is consistent with 100%. We also present a re-analysed set of lags from the OzDES sample with better quantified reliabilities