Main

GRB 050904 was a long burst (duration T90 = 225 s) detected by the Swift γ-ray burst satellite on 4 September 2005, 01:51:44 ut (Universal time; refs 6, 7). Its position was immediately disseminated via the GRB Coordinates Network. Although the optical observations at the Palomar 60″ telescope carried out 3.5 hours after the trigger did not reveal a new source with upper limits of R > 20.8 mag and I > 19.7 mag (ref. 8), a relatively bright near infrared source with J ≈ 17.5 mag was detected three hours after the burst in the Swift X-ray telescope error circle9, which showed a temporal decay with an index of -1.20, fully consistent with being a typical GRB afterglow. Analysis of the near-infrared colours, combined with the non-detection in the optical bands, led to the suggestion that the burst originated at a high redshift10, 5.3 < z < 9.0. A refined photometric redshift of was reported based on European Southern Observatory (ESO) Very Large Telescope (VLT) observations in the J, H, K and I bands11.

We observed the field of GRB 050904 with the Faint Object Camera And Spectrograph (FOCAS)12 on the 8.2-m Subaru Telescope on top of Mauna Kea, Hawaii, starting on the night of 6 September 2005. In the z′ band image (600-s exposure, mid-epoch on 7 September, 8:04 ut), we detected the afterglow at z′(AB) = 23.71 ± 0.14 mag, but we failed to detect it in the IC band even with a longer exposure (900 s), which implied that the Lyman break should be present around λ ≈ 8,500–9,000 Å.

We then obtained a grism spectrum of the afterglow candidate, which exhibited a sharp cut-off at λ ≈ 9,000 Å with strong depletion of the continuum at shorter wavelengths, strikingly similar to the spectra of quasars13 at z > 6 except for the absence of a broad Lyα emission line. The emission is very weak in the wavelength range shorter than 8,900 Å. In particular, the flux is consistent with zero in the ranges 8,500–8,900 Å and 7,000–7,500 Å that extend shortward of the Lyα and Lyβ wavelengths for z ≈ 6.3. This is a clear signature of absorption by neutral hydrogen in the intergalactic medium at z > 6, and marks the first detection of a Gunn–Peterson trough14 from an object other than high-z quasars15. We also find weak emission features at ∼7,500–8,300 Å which are presumably leakage flux from the continuum emission that is also found in quasar spectra at similar redshifts.

At the longer-wavelength end of the spectrum is a flat continuum with a series of absorption lines, which we identify as S ii, Si ii, O i, and C ii lines at a common redshift of z = 6.295 ± 0.002. We believe that this is the redshift of the GRB host galaxy, since no other absorption line system was observed at a redshift consistent with that of the Lyα break. This firm spectroscopic identification of the redshift breaks the previous record of GRB 000131 at z = 4.50 (ref. 5).

From a closer examination of the absorption lines, we find that they are not saturated and can be used to estimate the column densities of the heavy elements as shown in Table 1. Using the standard photospheric solar abundances16 we obtain the metallicity of these elements as [C/H] = -2.4, [O/H] = -2.3, [Si/H] = -2.6, and [S/H] = -1.0, where log[NHi (cm-2)] ≈ 21.3 is assumed from a damped Lyα system model for the Lyα damping wing presented below. These values may not represent the typical abundances in the GRB host galaxy for several reasons. First, they are derived using only a single ionization state for each element. Depletion due to dust condensation may modify the Si abundance in particular. And second, the spatial distribution of the heavy elements may be significantly different from that of hydrogen. It is possible that the heavy elements are distributed only locally around the GRB source in a metal-enriched circumstellar shell, while the neutral hydrogen is distributed on a larger scale in or outside the host galaxy.

Table 1 Absorption lines detected in the spectrum of the optical afterglow of GRB 050904
Full size table

Further analysis of the Si lines allows us to constrain the scale of the absorbing metals. Using the equivalent width ratio of the fine-structure transition lines Si ii* λ = 1,264.7 Å and Si ii λ = 1,260.4 Å, the electron density ne can be constrained17 as log[ne(cm-3)] = 2.3 ± 0.7 for a reasonable temperature range of 103 K < T < 105 K. Combined with the column density and the abundance of Si derived above and assuming a hydrogen ionization fraction of 0.1, we obtain the physical depth of the absorbing system to be 0.4 pc with an uncertainty of a factor of ∼10, reflecting the statistical errors and the possible temperature range. These fine-structure lines have been found in GRB afterglow spectra18,19,20, whereas they have never been clearly detected in quasar damped Lyα systems18. This is consistent with a local origin for the absorption such as a metal-enriched molecular cloud in the star-forming region or a dense metal-enriched shell nebula swept-up by a progenitor wind prior to the GRB onset suggested for GRB 021004 (refs 21, 22) and GRB 030226 (ref. 23). The column density of C ii is also consistent with the calculation for a carbon-rich Wolf–Rayet wind24.

As shown in Fig. 1, the Lyα cut-off exhibits the signature of a damping wing redward of the Lyα wavelength. To our knowledge, this is the first detection of significant neutral hydrogen absorption at z≳6, allowing us to explore the distribution of neutral hydrogen in the vicinity of a GRB, in the host galaxy, and/or in intergalactic space at very high redshifts. Such a study is difficult with high-z quasars owing to their enormous ultraviolet flux, which ionizes the surrounding environment, and owing to the presence of a strong Lyα emission line.

Figure 1: The spectrum of the afterglow of GRB 050904.
figure 1

It covers the wavelengths 7,000–10,000 Å with a resolution R = λ/Δλ ≈ 1,000 at 9,000 Å. It was taken with Subaru/FOCAS at mid-epoch on 7 September, 12:05 ut, 3.4 days after the burst, for a total exposure of 4.0 hours. The abscissa is the observed wavelength converted to that in vacuum. The spectrum is smoothed to a resolution R = λ/Δλ ≈ 600 at 9,000 Å. The locations of the identified absorption lines (see Table 1) at z = 4.840 and z = 6.295, as well as the wavelengths of Lyα and Lyβ at z = 6.295, are shown with vertical dotted lines. The one-sigma errors are shown with an offset of -1.0 at the bottom of the panel. In the inset, the solid line shows a model for the damping wing of Lyα absorption with a neutral hydrogen column density log[NHi (cm-2)] = 21.3 at a redshift of z = 6.3, overlaid on the observed spectrum in the wavelength range of 8,700–9,500 Å. The dotted line indicates the unabsorbed continuum model following a power-law (fν∝ν-1) function as typically observed for GRB afterglows. We note that only the red wing is relevant to the fit, because the emission blueward of Lyα is absorbed by the IGM.

Full size image

There are two possibilities for the nature of the absorber. It may be a damped Lyα system associated with the host galaxy, which has been observed in the afterglows of several GRBs at lower redshifts18,19,20. The other possibility is the neutral hydrogen in the intergalactic medium (IGM) left over from the pre-reionization era2. If the latter is the case, we can now measure the neutral fraction of the IGM at z≳6, giving important information on the reionization history of the Universe. We find that the wing shape can be reproduced either by a damped Lyα system (see inset of Fig. 1) or by the IGM. A comprehensive spectral fitting analysis is necessary to examine these possibilities, which is beyond the scope of this Letter.

With the detection of metal absorption lines, we have shown that GRBs are found in metal-enriched regions even at such an early phase of the Universe as z > 6. It is, therefore, possible that we would detect the metal absorption lines even from GRBs originating from the metal-free first-generation stars, because their environment may be self-polluted by pre-burst winds, as suggested by the present observation. We can expect to obtain afterglow spectra with much higher quality for GRBs at even higher redshifts in the immediate future, considering that Swift is constantly localizing faint GRBs, and that our spectrum was taken when the afterglow had faded by more than an order of magnitude since its first detection in the J band. Such future data will give us even better opportunities to probe the formation of stars and galaxies in the early Universe.