Statistical re-examination of reported emission lines in the X-ray afterglow of GRB 011211

Abstract

A 0.2–12 keV spectrum obtained with the XMM—Newton EPIC/pn instrument of GRB 011211, taken in the first 5 ks of a 27-ks observation, was found by Reeves et al. in 2002 to contain emission lines that were interpreted to be from Mg xi, Si xiv, S xvi, Ar xviii and Ca xx, at a lower redshift (zobs = 1.88) than that of the host galaxy (zhost = 2.14). We examine the spectrum independently, and find that the claimed lines would not be discovered in a blind search. Specifically, Monte Carlo simulations show that the significance of reported features, individually, is such that they would be observed in 10 per cent of featureless spectra with the same signal-to-noise ratio. Imposing a model in which the two brightest lines would be Si xiv and S xvi Kα emission velocity shifted to between z = 1.88 and 2.40, such features would be found in ∼1.3–1.7 per cent of observed featureless spectra (that is, with 98.3–98.7 per cent confidence). When we account for the number of trials implicit in a search of five energy spectra (as were examined by Reeves et al.), and permit a wider z phase space search (z = 2.14 ± 1.0), the detection confidence of the two-line complex decreases to 77–82 per cent. We find the detection significance to be insufficient to justify the claim of detection and the model put forth to explain them. Kα line complexes are also found at z = 1.2 and 2.75 of significance equal to or greater than that at z = 1.88. Thus, if one adopts the z = 1.88 complex as significant, one must also adopt the other two complexes to be significant. The interpretation of these data in the context of the model proposed by Reeves et al. is therefore degenerate, and cannot be resolved by these data alone. Our conclusions are in conflict with those of Reeves et al., because our statistical significance accounts for the multiple trials required — but not accounted for by Reeves et al. — in a blind search for emission features across a range of energies. In addition, we describe a practical challenge to the reliability of Monte Carlo Δχ2 tests, as employed by Reeves et al.