BATSE Evidence for GRB Spectral Features

We have developed an automatic search procedure to identify lowenergy spectral features in GRBs. We have searched 133,000 spectra from 117 bright bursts and have identified 12 candidate features with significances ranging from our threshold of P = 5E−5 to P = 1E−7. Several of the candidates have been examined in detail, including some with data from more than one BATSE spectroscopy detector. The evidence for spectral features appears good; however, the features have not conclusively been shown to be narrow lines. LINE SEARCH AND RESULTS Narrow, low-energy (< 100 keV) spectral lines in GRBs have been reported using the data of several instruments (e.g., see review [4]). Based on these reports, the BATSE team expected lines to be easy to find and looked for them manually [7]. The reality was different: lines were not obvious in burst spectra. In order to be sure that the manual search had not missed any lines, we implemented a comprehensive, automatic computer search [5]. Bursts with at least one spectrum with a normed signal-to-noise ratio (SNR) [2] near 40 keV above 5.0 are searched. For each burst, we form spectra from each individual spectral record, every pair, triple, etc. The spectra so formed overlap in many cases. Once a burst is selected for the search, spectra are searched regardless of the presence of burst flux—low SNR spectra serve as controls. Each spectrum is fit with a continuum model and then a series of fits are made adding narrow lines at a closely spaced grid of line centroids extending to 100 keV. Line candidates are identified by a χ change ∆χ of more than 20, corresponding to a chance probability in a single spectrum of P = 5× 10−5. The probability is calculated for two line parameters, intensity and centroid, since the intrinsic width is assumed narrow compared to the detector resolution. So far, 133,000 spectra formed from 12,000 spectral records from 117 bursts have been searched. Most of these spectra have SNR too low to support the detection of a line. Only 16,000 have normed SNR > 5, which our simulations show is needed to have a reasonable sensitivity to lines similar to those found in the Ginga data [1]. The search identified several cases with ∆χ values exceeding our significance threshold, but which we rejected either because they occur in background intervals or because they become insignificant when a better fit requiring a non-negative flux is made. These cases were located in spectra of low SNR, and because of the large number of low-SNR spectra, they are consistent with statistical fluctuations. Several possible candidates with angles between the burst direction and detector normal exceeding 70◦ are inconsistent with the data of other detectors. We believe these cases to be due to inadequacies in the detector model and we have set them aside. After these exclusions, our candidate list has 12 members with ∆χ values ranging from the threshold of 20 to 31.8 (P = 1 × 10−7). The normed SNRs of the spectra in which these candidates are most significant range from 2.1 to 18.4, with 10 normed SNR values exceeding 5. These SNR values are reasonable for real features. The number of independent trials is a nebulous concept. While there are roughly 16,000 spectra sufficiently bright for there to be a high probability of detecting a real line, many of these spectra overlap, e.g., starting a few records earlier or later, or differing in length by a few records. Consequently the number of bright, independent spectra is one or more orders of magnitude below 16,000. The number of independent energy resolution elements averages about 5 per spectrum. We estimate that the ensemble chance probability of the most-significant candidate is below 1 × 10−3, and probably much lower than this value. We therefore believe that few, if any, of these candidates are statistical fluctuations. Of the 12 candidates, in all but one the best line fit is an emission feature with a centroid near 40 keV. The twelfth candidate is an absorption feature at 60 keV. Typically the lowest energy of the data is about 20 keV, so we are unable to find lines much below 40 keV.