Gravitational-wave “dark sirens” are events detected by gravitational waves that lack a clear flash of light. This paper presents a way to turn those dark sirens into an astrophysical probe. The authors build a statistical method that jointly asks two questions: which galaxy hosted a merger, and how mergers prefer certain types of galaxies. The output is a short, probabilistic ranking of candidate host galaxies for each event. That ranking can guide follow-up observations.

The approach uses hierarchical Bayesian inference to combine the full gravitational-wave data and a galaxy catalog in a single model. Two discrete variables are treated as unknowns for each event: whether the true host is in the catalog, and which catalog galaxy it is if present. The method encodes a simple model for how mergers trace galaxies by weighting galaxies by their luminosity L with a power law w(L) = (L/L*)^α, where L* is a pivot and α is a population parameter learned from the data. The authors implement the sampler in PyMC, using a combination of the No‑U‑Turn Sampler for continuous parameters and Gibbs‑Metropolis updates for the discrete host labels. They also account for catalog incompleteness by including a homogeneous “out‑of‑catalog” population.

The paper tests the method in two ways. First, they apply it blindly to the well known binary‑neutron‑star event GW170817 and recover NGC 4993 as the top-ranked host galaxy. The analysis also yields a constraint on the luminosity weighting α, mildly favoring non‑uniform host weights (that is, some preference for brighter or dimmer galaxies rather than all galaxies being equally likely). Second, they validate the framework on simulated binary‑black‑hole populations placed into the MICE mock galaxy catalog and analyzed with a simulated LIGO–Virgo–KAGRA–LIGO India (LVKI) network at planned A+ sensitivity.