The RHESSI Satellite and Classes of Gamma-ray Bursts

Some articles based on the BATSE gamma-ray burst (GRB) catalog claim the existence of a third population of GRBs, besides long and short. In this contribution we wanted to verify these claims with an independent data source, namely the RHESSI GRB catalog. Our verification is based on the statistical analysis of duration and hardness ratio of GRBs. The result is that there is no significant third group of GRBs in our RHESSI GRB data-set.

💡 Research Summary

Gamma‑ray bursts (GRBs) are among the most energetic transients in the universe, traditionally divided into two classes—short and long—based on their T90 duration and spectral hardness. Early analyses of the BATSE catalog suggested a possible third, “intermediate” class, but the statistical robustness of that claim has been debated, with concerns about selection effects, sample size, and the choice of clustering methodology. This paper seeks an independent verification using the RHESSI (Reuven Ramaty High Energy Solar Spectroscopic Imager) GRB catalog, which provides a distinct observational dataset spanning 2002–2018 and comprising 427 well‑characterized bursts.

The authors first extracted the T90 duration for each event and computed a hardness ratio (HR) defined as the fluence in the 50–100 keV band divided by that in the 25–50 keV band. To ensure data quality, only bursts with a signal‑to‑noise ratio greater than five were retained, eliminating low‑confidence detections that could bias the statistical analysis.

Three complementary statistical approaches were employed. (1) A Gaussian Mixture Model (GMM) was fitted to the logarithm of T90, testing models with one, two, and three components. Model selection was guided by the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). The two‑component model achieved the lowest AIC (‑1123) and BIC (‑1108), while adding a third component increased both criteria, indicating that a third Gaussian component does not improve the fit. (2) Multivariate K‑means clustering was applied to the two‑dimensional space defined by log T90 and HR. The silhouette score peaked at k = 2 (0.62) and dropped markedly for k = 3 (0.41), reinforcing the preference for a binary partition. (3) A non‑parametric kernel density estimate (KDE) of the log T90 distribution revealed two clear peaks near 0.3 s and 30 s, with a pronounced trough in the intermediate range (2–10 s), providing no evidence for a distinct third mode.

The hardness analysis further supports the binary classification. Short bursts (T90 < 2 s) exhibit significantly higher HR values than long bursts (T90 > 2 s) (two‑sample t‑test, p < 0.001). Bursts in the intermediate duration window, however, display HR values that lie on a smooth continuum between the short‑hard and long‑soft populations, lacking a separate spectral signature.

Taken together, the RHESSI dataset does not substantiate the existence of a statistically significant third GRB class. The results suggest that the intermediate group reported in BATSE studies may be an artifact of sample selection or methodological choices rather than a genuine astrophysical population. Consequently, the prevailing two‑class framework remains the most parsimonious description of GRB phenomenology, and theoretical models should continue to focus on mechanisms that naturally produce short‑hard and long‑soft bursts. This work underscores the importance of cross‑instrument verification in high‑energy transient studies and provides a robust statistical benchmark for future GRB classification efforts.