Parameter estimation for LIGO’s third observing run

The third observing run (O3) of the advanced LIGO and Virgo detectors ran from April 2019 to March 2020, and represented a dramatic step forward. By the time O3 ended, the total catalogue of confident gravitational-wave detections had grown from 10 to 90 — nearly an order of magnitude increase over the first two observing runs.

Extracting the astrophysical information from these signals — the masses, spins, and distances of the colliding compact objects — requires Bayesian parameter estimation (PE): an intensive computation that can take days per event on a computing cluster, and requires careful setup, monitoring, and post-processing.

During O3, I coordinated and carried out the PE analyses for 88 of the 90 events that appeared in the three major catalogues: GWTC-2 (covering the first half of O3), GWTC-2.1 (a reanalysis of all events from O1, O2, and O3a with updated methods), and GWTC-3 (the complete O3 dataset). Each event was typically analysed with two independent waveform models to check for consistency, and results were reviewed before inclusion in the catalogue.

Managing PE at this scale required new tools and processes. Keeping track of hundreds of simultaneous analyses, restarting failed jobs, and coordinating post-processing across a large team was far beyond what could be handled manually. The coordination work during O3 directly motivated the development of asimov, a toolkit for automating and managing large-scale gravitational-wave analyses. Asimov has since been used for the O4 catalogues and released as open-source software for the wider community.