An 11-Year Catalog of Gamma-Ray Transients: A Comprehensive Search with Fermi Gamma-ray Burst Monitor Data

TL;DR

This study performs a comprehensive, blind search of 11 years of Fermi-GBM continuous data to identify untriggered gamma-ray transients across a broad energy and time-scale space. By combining four search modes with three statistical methods (SNR, Poisson, Bayesian Blocks) and a two-step classification pipeline (known-event filtering and Bayesian unknown-event classification), the authors assemble a public, multi-class transient catalog that substantially expands subthreshold gamma-ray populations, including magnetar bursts, X-ray bursts, TGFs, solar flares, and other transients. The work provides rigorous false-positive controls, robust event-flagging, and probabilistic classifications, with the event database and lightcurves openly accessible to the community via a dedicated web portal. Overall, the methodology and catalog significantly enrich the GBM transient landscape, enabling broader cross-correlation with multiwavelength and multimessenger observations, as well as statistical studies of subthreshold gamma-ray phenomena. Key mathematical formulations include the SNR definition $\mathrm{SNR}=(R_T-R_B)/\sigma$ with $\sigma=\sqrt{N_B}/\Delta t$, the Bayesian block likelihood $\ln L(k)=N(k)\ln\lambda(k)-\lambda(k)T(k)$, and the posterior $P(C_i|D_j)=\frac{P(C_i)P(D_j|C_i)}{P(D_j)}$ used for classifying unknown events.

Abstract

The Gamma-ray Burst Monitor (GBM) on board Fermi Gamma-ray Space Telescope has produced the largest database of all-sky observations in gamma rays with its continuous data with high time and energy resolutions. These data contain a wealth of unidentified transient events that did not trigger the detectors for various reasons. We conducted extensive searches to identify such untriggered transient events observed with GBM in 11 years (July 2010 - June 2021). In particular, we employed four different search modes with various energy ranges (mainly below 300 keV) and time resolutions (from 8 ms to 2 s), utilizing three statistical methods (signal-to-noise ratio, Poisson, and Bayesian statistics), each with different effectiveness in identifying specific classes of transients. Moreover, we developed algorithms for known-event flagging as well as unknown-event classification for our candidate events found in the searches. In this paper, we present our search methodologies, event flagging and classification algorithms and the resulting comprehensive event catalog. The catalog contains more than a million events in total, including known events such as gamma-ray bursts, soft-gamma repeater bursts, galactic X-ray source activities, terrestrial gamma flashes, and solar flares. For each candidate event, the catalog presents the event time, detection significance, event duration, hardness ratios, known-event flagging results, and classification probabilities. Our short-transient catalog significantly expands the currently-existing list of known events and complements the GBM trigger catalog. The event database with filtering capabilities is also publicly available at https://magnetars.sabanciuniv.edu/gbm, which allows users to retrieve event information based on their input queries along with the event lightcurves.

Figures (7)

• Figure 1: Trigger history of Fermi-GBM up to June 30, 2021 (i.e., our search period). Note that the number of triggers in 2008 and 2021 is not for a complete year. The "OTHERS" category includes UNCERT and Galactic Binary (GALBIN, only three events classified as such).
• Figure 2: A burst flagged in the SNR search (likely from a magnetar, Swift J1818.0--1607). [Left panel] The event is detected in three detectors (NaI 9, 10, 11) in Mode-1 energy range (10--100 keV), with SNR = 5.0$\sigma$, 11.1$\sigma$ and 4.9$\sigma$, respectively. The numbers at the top left of each panel are the significance in $\sigma$. [Right panel] The corresponding energy-dependent ligthcurves for the brightest detector, NaI 10.
• Figure 3: Calculated Poisson probabilities of all 12 detectors as a function of SNR for 220 events detected by the search with SNR method. The searched data were binned to 8-ms in the energy range of 10–100 keV (i.e., Mode 1). $P \approx 10^{-4}$ corresponds to SNR = 4.5$\sigma$.
• Figure 4: A burst flagged in the Poisson search (likely from a magnetar, PSR J1846.4--0258). [Left panel] The event is detected in two detectors (NaI 10 & 11) in the Mode-1 energy range (10--100 keV), with $P$ = 3.9E--17 and 1.6E--21, respectively. The numbers at the top left of each panel are the Poisson probability. [Right panel] The corresponding energy-dependent ligthcurves for the brightest detector, NaI 11.
• Figure 5: A burst flagged only with the Bayesian block search (likely from a magnetar, SGR J1935+2154). The event is detected in three detectors (NaI 6, 7 & 9) in the Mode-1 energy range (10--100 keV). The red lines show the Bayesian-block representation of each lightcurve, plotted over the 8-ms lightcurve.