A New Search Pipeline for Short Gamma Ray Bursts in Fermi/GBM Data -- A 50% Increase in the Number of Detections

TL;DR

The paper introduces a fully automated, Poisson-based search pipeline for short gamma-ray bursts in Fermi/GBM data, leveraging Neyman–Pearson hypothesis testing, template banks, and time-slid significance to improve detection sensitivity beyond onboard triggers. It combines coherent template banks across sky position and spectral shapes with a robust background estimation and drift-correction scheme, plus a suite of vetoes and Bayesian discriminants to classify triggers by origin. Applied to 2014 GBM data, the pipeline recovers most catalog GRBs and yields about 27 new sGRB candidates with $p_{\text{astro}}=1$, alongside many other transient detections, and achieves SNR improvements of $2$–$15\times$, effectively increasing the detectable volume for sGRBs. The approach enhances the prospects for joint sGRB–GW searches and motivates applying the pipeline to the full GBM archive to further expand the sGRB–GW detection horizon.

Abstract

In this paper, we present the development and the results of a new search pipeline for short gamma-ray bursts (sGRBs) in the publicly available data from the Gamma-Ray Burst Monitor (GBM) on board the Fermi satellite. This pipeline uses rigorous statistical methods that are designed to maximize the information extracted from the Fermi/GBM detectors. Our approach differs substantially from existing search efforts in several aspects: The pipeline includes the construction of template banks, Poisson matched filtering, background estimation, background misestimation correction, automatic routines to filter contaminants, statistical estimation of the signal location and a quantitative estimator of the signal probability to be of a cosmological, terrestrial, or solar origin. Our analysis also includes operating the pipeline on "time-slided" copies of the data, which allows exact significance assessment and $p_{\text{astro}}$ computation, akin to the state-of-the-art gravitational waves (GW) data analysis pipelines. Depending on the spectral properties of the bursts, our pipeline achieves a signal-to-noise ratio (SNR) improvement by a factor of 2 to 15 over the onboard GBM triggering algorithm. This enhancement increases the detectable volume for sGRBs and results in an approximate 50% increase in sGRB detections in the 2014 GBM dataset. As a further consequence of the sensitivity increase, we detect hundreds of soft gamma-ray flares of galactic origin. This improved sensitivity enhances the chances of detecting fainter, off-axis GRBs that would likely fall below the standard triggering thresholds. Applying this pipeline to the full GBM archive is expected to expand further the joint sGRB-GW detection volume.

Figures (19)

• Figure 1: A flowchart describing the pipeline structure.
• Figure 2: Comparison of templates generated with different software at different times. We matched (as in Eq. \ref{['eq: match']}) every template in the coarse bank, generated using the response generator gbm_drm_gen at met=654393605.0 sec (2021-09-27 00:00:00), with templates having the same spectral parameters and sky positions but generated using GBM's response generator at the specified dates, which were randomly chosen. One can see that the maximal mismatch is only 15%, and that for the most part, the time in which the response was generated does not affect the template significantly. On 2018-01-17 at 16:51:09 (brown line), Fermi was in an offset-pointing position due to a Target of Opportunity (ToO) observation (Obs ID 100601-1-1), which resulted in a slightly higher mismatch. A full table describing the observing modes during these times is presented in Appendix \ref{['appendix: tables']}.
• Figure 3: Visualization of the constructed template bank. The small blue dots represent the dense bank template distribution over the sky and the larger points represent the coarse bank created by applying the random placement algorithm. Each point in this plot represents a single point in the five-dimensional parameter space $(\alpha, \beta, E_{\text{peak}}, \text{RA}, \text{DEC})$. The coarse bank points are colored by one of the spectral parameters - $E_{\text{peak}}$.
• Figure 4: Template space coverage checks. Match of each template in the reference template bank and each template in the GBM-like bank with all the templates in the coarse bank, keeping the maximal value. The y-axis is the cumulative fraction of templates with a maximal match larger than the number specified by the x-axis. The reference bank is a larger bank, containing $5\times10^4$ templates. We see that only a small fraction, $\sim 0.02$, has a match less than $0.95$. Compared with the GBM-like templates, our $\sim 500$ templates cover more than $90\%$ of the parameter space covered by the GBM search bank containing $3$ spectral shapes for each of the $10^4$ sky positions required for $\sim 4$ deg${}^2$ resolution.
• Figure 5: Local Background Estimation Example. The figure shows a simulated signal with noise and the local background estimation using a standard rolling mean, a rolling mean with a gap, and a second-order filter. It is clear that when a burst occurs, the gapped version captures the real background more accurately at the time of the burst. Furthermore, by applying the second-order filter (see main text and Appendix \ref{['appendix: bkg']}), we improve the estimation, especially where the background varies in a non-linear fashion.