Classification of Gamma-Ray Burst durations using robust model-comparison techniques

TL;DR

This study applies robust model comparison techniques to classify Gamma-Ray Bursts into two or three categories, analyzing multiple datasets and finding strong evidence for three classes only in Swift GRBs.

Contribution

The paper introduces a comprehensive model comparison approach using likelihood, AIC, and BIC to evaluate the existence of multiple GRB classes across various datasets.

Findings

Swift GRBs show evidence for three categories at about 2.4σ significance.

Other datasets do not strongly support the three-class hypothesis.

Model comparison techniques effectively differentiate between two and three class models.

Abstract

Gamma-Ray Bursts (GRBs) have been conventionally bifurcated into two distinct categories dubbed "short" and "long", depending on whether their durations are less than or greater than two seconds respectively. However, many authors have pointed to the existence of a third class of GRBs with mean durations intermediate between the short and long GRBs. Here, we apply multiple model comparison techniques to verify these claims. For each category, we obtain the best-fit parameters by maximizing a likelihood function based on a weighted superposition of lognormal distributions. We then do model-comparison between each of these hypotheses by comparing the chi-square probabilities, Akaike Information criterion (AIC), and Bayesian Information criterion (BIC). We uniformly apply these techniques to GRBs from Swift (both observer and intrinsic frame), BATSE, BeppoSAX, and Fermi-GBM. We find that the Swift GRB distributions (in the observer frame) show evidence for three categories at about $2.4\sigma$ from difference in chi-squares and show decisive evidence in favor of the two components using both AIC and BIC. For all the other datasets, evidence for three components is either very marginal or not favored.