Unveiling the spectral morphological division of fast radio bursts with CHIME/FRB Catalog 2

TL;DR

This study uses machine learning to classify FRBs based on spectral morphology, revealing two main clusters with distinct properties and suggesting a physical connection between repeaters and nonrepeaters.

Contribution

The paper introduces an unsupervised clustering approach to classify FRBs by spectral features, providing new insights into their intrinsic population structure.

Findings

FRBs primarily split into two spectral morphology-based clusters.

A stable subclass of repeaters resembles nonrepeaters in properties.

Differences between classes are explained by observational biases and distance effects.

Abstract

Fast radio bursts (FRBs) are commonly classified into repeating and apparently nonrepeating sources, yet whether this distinction reflects intrinsically different physical populations remains uncertain. Using the Second CHIME/FRB Catalog, we apply an unsupervised machine learning framework combining Uniform Manifold Approximation and Projection (UMAP) with density-based clustering to investigate the intrinsic structure of the FRB population in a multi-dimensional parameter space. We find that FRBs are primarily separated into two robust clusters dominated by spectral morphology. One cluster is characterized by narrowband emission and longer durations, while the other exhibits relatively broadband spectra and shorter burst timescales. This classification scheme achieves a recall of 0.94 for known repeaters. Within the repeating population, we further identify a stable subclass of atypical repeaters that are broadband, shorter in duration, and more luminous, resembling nonrepeating bursts. Furthermore, broadband nonrepeaters exhibit systematically higher dispersion measures (by approximately 200 $\text{pc cm}^{-3}$) and isotropic luminosities approximately an order of magnitude larger than those of repeating FRBs. Without invoking catastrophic progenitor scenarios, these differences are naturally explained by instrumental sensitivity limits and distance-dependent selection effects. Our results provide new statistical evidence for a physical connection between repeating and nonrepeating FRBs.