From Localization to Discovery: Bayesian Ranking of Electromagnetic Counterparts to Gravitational-Wave Events

TL;DR

We present a Bayesian framework to identify and rank electromagnetic counterparts to gravitational-wave events using only localisation information. The method fuses three-dimensional GW skymaps with host-galaxy data through a joint distance–sky overlap, incorporating a morphology-aware host association via the directional light radius and a gamma-kick inspired intrinsic offset prior, all while accounting for peculiar velocities and catalog incompleteness. Applied to GW170817, the approach correctly ranks AT2017gfo as the top candidate and recovers NGC 4993 as the host, with distance consistency ($\mathcal{I}_{D_L}$) dominating the discrimination over sky position alone. The framework is adaptable to current and future GW observing runs, capable of operating on transient lists with only location information, and provides a practical, probabilistic pathway to prioritise follow-up and improve counterpart identification amid large sky localizations and many unrelated transients.

Abstract

The robust association of electromagnetic candidates discovered during follow-up of gravitational-wave alerts is challenging, not only due to the large sky areas and broad distance uncertainties, but also due to the tens to hundreds of unrelated optical transients that are observed per event. We present a Bayesian ranking method to identify electromagnetic counterparts to GW events using only location information. The framework combines three-dimensional gravitational wave skymaps with host-galaxy information, a morphology-aware host association, empirical offset priors, and peculiar velocity corrections. We apply the method to GW170817 where it ranks AT2017gfo as the top candidate and correctly selects NGC\,4993 as the host. The approach is directly applicable to transient candidates with only location information and enables more efficient follow-up with prioritized candidates and leads to more reliable counterpart identification in current and future observing runs.

Figures (4)

• Figure 1: GW170817 luminosity distance posteriors evaluated along the line of sight $\hat{\Omega}$ for AT2017gfo. The blue curve is the GW conditional luminosity distance from the public skymap (solid) and its KDE evaluation (dotted). The EM components from the host $z_{\rm phot}$ mixture for core (orange dashed), outlier (green dot dashed), and catastrophic outlier (red dotted) each mapped to $D_L$ with peculiar velocity broadening. The purple solid curve is the resulting EM distance posterior for the full mixture. The black curve shows the normalized product proportional to the distance-overlap integrand of $\mathcal{I}_{D_{L}}$ with the area equal to $\mathcal{I}_{D_{L}}$.
• Figure 2: Host association results from LS-DR10 for two GW170817 transients. The galaxies are those which like within a projected physical offset $\delta R \leq 70\,{\rm kpc}$ from the transient. Left: AT2017gfo. The yellow star marks the transient and the magenta ellipse shows the correctly matched host (NGC 4993). The shape parameters of the ellipse use the morphology parameters given in the catalog. The dotted vector points from the host centroid to the transient. Right: AT2017bej. Blue ellipses mark all catalog galaxies within the projected physical offset. The magenta ellipse indicates the most probable host. The lower-left zoom inset shows a clearer image of the geometry and offset of the preferred association.
• Figure 3: Marginal posterior distributions for RA (left), Declination (center) and luminosity distance $D_L$ (right) obtained from the initial probability skymap for GW170817 (bayestar-HLV.fits.gz). Filled histograms show the posterior samples from the FITS skymap and solid stepped curves are the one-dimensional (marginalized) KDE evaluations. Dashed vertical lines mark the coordinates (RA/Dec) and host-inferred distance of EM candidates within $D_L \leq 300\,{\rm Mpc}$. Dashed line color encodes the ranking statistic $\mathcal{B}_{\mathcal{C}/SS} \approx \mathcal{I}_{{\Omega},D_{L}}$ Eq. \ref{['eq:overlap']} with values below the threshold ($\mathcal{I}_{{\Omega},D_{L}}\leq 10^2$) shown in gray.
• Figure 4: DESI Legacy Survey DR10 cutout centered on a transient candidate (yellow star). The ellipses are cataloged galaxy model fits: DEV (de Vaucouleurs, green), REX (round exponential, blue), SER (Sérsic, magenta), and EXP (exponential, orange). Text labels give the cataloged photometric redshift for each entry. The bright central galaxy is represented by multiple deblended model components with inconsistent $z_{\rm phot}$ values ($z=0.037$, $0.176$, $0.322$, $0.685$, $0.214$), which illustrates how wide-field catalogs can return several overlapping candidates around a single host. Naively choosing the ellipse containing the transient (green, $z=0.685$) would mis-associate the transient.