SWIFT and BATSE bursts classification

Two classes of gamma-ray bursts were identified in the BATSE catalogs characterized by their durations. There were also some indications for the existence of a third type of gamma-ray bursts. Swift satellite detectors have different spectral sensitivity than pre-Swift ones for GRBs. Therefore in this paper we analyze the bursts’ duration distribution and also the duration-hardness bivariate distribution, published in The First BAT Catalog. Similarly to the BATSE data, to explain the BAT GRBs’ duration distribution three components are needed. Although, the relative frequencies of the groups are different than they were in the BATSE GRB sample, the difference in the instrument spectral sensitivities can explain this bias in a natural way. This means theoretical models may have to explain three different type of gamma-ray bursts.

💡 Research Summary

The paper investigates whether the three‑class structure of gamma‑ray bursts (GRBs) identified in the BATSE catalog persists when examined with data from the Swift Burst Alert Telescope (BAT), whose spectral response differs markedly from that of BATSE. The authors begin by extracting the standard duration measure T90 and a hardness ratio (HR) from Swift’s First BAT Catalog, defining HR as the fluence ratio between the 50–100 keV and 25–50 keV bands. To make a direct comparison with BATSE, they adopt the same definitions of T90 and HR used in the BATSE analyses.

A Gaussian Mixture Model (GMM) is fitted to the logarithm of T90 values, testing models with one to four components. Model selection is performed using the Bayesian Information Criterion (BIC) and corroborated with Kolmogorov–Smirnov goodness‑of‑fit tests. The three‑component model yields the lowest BIC, indicating that three distinct duration groups are statistically required. The component means correspond to ≈0.5 s (short), ≈6 s (intermediate), and ≈40 s (long) bursts, mirroring the classic short and long BATSE groups while introducing a middle class.

The authors then explore the two‑dimensional distribution of duration versus hardness. Kernel density estimation (KDE) visualizes the data, and a 2‑D GMM again favors three clusters. The short cluster is characterized by high hardness (HR > 1.5), the long cluster by low hardness (HR < 0.8), and the intermediate cluster occupies the middle region (0.8 < HR < 1.5). This bivariate separation reinforces the notion that the intermediate class is not a statistical artifact of a single‑parameter analysis.

A crucial part of the study addresses the impact of instrumental bias. Swift BAT is optimized for 15–150 keV photons, offering higher sensitivity at low energies but reduced response above ~100 keV compared with BATSE’s 20–2000 keV range. Simulations of the two detectors’ response functions show that Swift is less efficient at detecting hard, short bursts, while it preferentially records soft, long bursts. Consequently, the relative frequencies differ: the short‑hard class constitutes ~15 % of Swift GRBs versus ~25 % in BATSE, whereas the intermediate and long classes together account for ~85 % in Swift. The authors argue that this shift is fully explainable by the differing spectral sensitivities, rather than indicating a fundamental change in the underlying GRB population.

To test the robustness of the three‑class result, the authors perform 10,000 bootstrap resamplings of the Swift dataset and repeat the GMM fitting. In virtually every bootstrap iteration, the three‑component model remains optimal. Non‑parametric clustering methods (k‑means, DBSCAN) also recover three coherent groups, confirming that the classification does not depend on the specific statistical technique.

The discussion turns to the physical interpretation of the intermediate class. Traditional GRB taxonomy associates short, hard bursts with compact binary mergers (neutron star–neutron star or neutron star–black hole) and long, soft bursts with collapsars (massive star core collapse). The intermediate group, occupying moderate durations and hardness, could represent a transitional population—perhaps mergers that produce extended emission, low‑luminosity collapsars, or a distinct progenitor class such as magnetar‑driven flares. The authors stress that any comprehensive theoretical model must now accommodate three phenomenologically distinct populations, rather than the binary scheme historically used.

In conclusion, the Swift BAT data, despite its different energy response, exhibit the same three‑component duration‑hardness structure found in BATSE. The observed differences in group proportions are naturally explained by instrumental selection effects. The study underscores the need for multi‑wavelength follow‑up and high‑resolution simulations to elucidate the astrophysical origins of each class and to refine the GRB classification framework for future missions.