A new probabilistic framework, named GENOVA, allows for the prediction of kilonova spectra and light curves directly from gravitational wave posterior samples of binary neutron star mergers. Kilonovae, electromagnetic events associated with these mergers, are crucial for understanding the system's properties and the r-process nucleosynthesis of heavy elements. To date, only one kilonova candidate, AT2017gfo, has been associated with a gravitational wave event, GW170817, highlighting the need for predictive tools.

GENOVA employs a conditional normalizing flow to learn the distribution of rest-frame spectra, conditioned on source-frame component masses, tidal deformabilities, viewing angle, and time since merger. Other kilonova model parameters, such as ejecta opacities, are marginalized over during training, allowing their effects to propagate into the predicted spectra as predictive uncertainty. Self-consistency tests show that the model reproduces median light curves with residuals typically below 0.1 magnitudes, and the ratio of the predicted 68% confidence intervals remains predominantly between 0.8 and 1.4 over a range of 0.4 to 8.0 days.

Applying GENOVA to the GW170817/AT2017gfo event, using multi-band observations including newly re-reduced Y, J, and Ks-band photometry from the Visible and Infrared Survey Telescope for Astronomy, demonstrates that the resulting predictive intervals broadly encompass the observations. This not only captures gravitational wave posterior uncertainty but also the variation induced by marginalized kilonova model parameters. The method's ability to remain informative beyond the training model suggests its robustness and potential for future predictions.