An Unambiguous Separation of Gamma-Ray Bursts into Two Classes from Prompt Emission Alone

TL;DR

This paper uses machine learning to classify gamma-ray bursts into two distinct groups based solely on prompt emission data, confirming their association with different astrophysical origins.

Contribution

It introduces a novel machine learning approach, t-SNE, to unambiguously separate GRBs into long and short classes using only prompt emission light curves.

Findings

All GRBs with supernovae are in the long group.

All GRBs with kilonovae are in the short group.

Two long GRBs lack supernovae, possibly indicating direct-collapse black holes.

Abstract

The duration of a gamma-ray burst (GRB) is a key indicator of its physics origin, with long bursts perhaps associated with the collapse of massive stars and short bursts with mergers of neutron stars.However, there is substantial overlap in the properties of both short and long GRBs and neither duration nor any other parameter so far considered completely separates the two groups. Here we unambiguously classify every GRB using a machine-learning, dimensionality-reduction algorithm, t-distributed stochastic neighborhood embedding (t-SNE), providing a catalog separating all Swift GRBs into two groups. Although the classification takes place only using prompt emission light curves,every burst with an associated supernova is found in the longer group and bursts with kilonovae in the short, suggesting along with the duration distributions that these two groups are truly long and short GRBs. Two bursts with a clear absence of a supernova belong to the longer class, indicating that these might have been direct-collapse black holes, a proposed phenomenon that may occur in the deaths of more massive stars.