Kilonova modelling and parameter inference: Understanding uncertainties and evaluating compatibility between observations and models

TL;DR

This paper examines uncertainties in kilonova modelling and parameter inference, evaluates model-data compatibility, and demonstrates how Bayesian methods can improve understanding of kilonova observations, especially in multi-messenger contexts.

Contribution

It provides a detailed analysis of systematic uncertainties in kilonova modelling and introduces Bayesian metrics for assessing model fit and inference performance.

Findings

Systematic error margins can serve as goodness-of-fit metrics.

Bayesian inference quantifies information gain in kilonova analysis.

Application to AT2017gfo validates the approach and highlights multi-messenger complementarity.

Abstract

In the study of optical transients, parameter inference is the process of extracting physical information, i.e. constraints on the source's characteristics, by comparing the observed lightcurves to the predictions of different models and finding the model and parameter combination that make the closest match. In the developing field of the study of kilonovae (KNe), systematic uncertainties in modelling are still very large, and many models struggle to fit satisfactorily the whole multi-wavelength dataset of the AT2017gfo kilonova, associated to the Binary Neutron Star (BNS) merger GW170817. In a multi-messenger context, we sometime observe tensions between KN-only inference results and constraints from other messengers. In order to discuss the compatibility of KN models with observations and with the information derived from other messengers, we detail the process of Bayesian parameter inference, identifying the many sources of uncertainty embedded in KN analyses. We highlight the systematic error margin hyperparameter $\sigma_{\rm sys}$, which can be exploited as a metric for a model's goodness-of-fit. We then discuss how to assess the performance of parameter inference analyses by quantifying the information gain using the Kullback-Leibler divergence between prior and posterior. Using the example of the Bu2019lm model with the NMMA Bayesian inference framework, we showcase the expected performance that dedicated KN follow-ups with telescope networks could reasonably reach, highlighting the different factors (observational cadence, error margins) that influence such inference performances. We finally apply our KN analysis to the dataset of AT2017gfo to validate our performance predictions and discuss the complementarity of multi-messenger approaches.