On 8 May 1997 UT, a moderate-fluence classical γ-ray burst (GRB970508) was detected by instruments aboard the Italian–Dutch satellite BeppoSAX9; the burst was localized initially to an error region of 5-arcmin radius10 and later to a region of 3-arcmin radius11. A potential optical counterpart was identified within two days7,8, which we refer to as OT J065349+79163 (here OT stands for optical transient). Interest in this object was heightened when the appearance of a bright X-ray source was reported12 that was not seen in a previous X-ray all-sky survey, and included OT J065349+79163 in the 45-arcsec-radius error circle. The presence of a new and bright X-ray source in a γ-ray burst (GRB) error circle has been seen in the recent three GRBs observed by BeppoSAX. In the first such case, a decaying optical source5 was also seen and these authors suggested that such fading (X-ray and optical) sources are the afterglow of γ-ray bursts. It has been suggested13 that because OT J065349+79163 exhibits unusual variability at optical and X-ray wavelengths, it represents a similar optical afterglow to GRB970508.

We obtained spectra of OT J065349+79163 with the Keck II 10-m telescope on 11 May 1997 UT, using the Low Resolution Imaging Spectrograph14 (LRIS) with a 2,048 × 2,048 pixel CCD (charge-coupled device). Starting at 05:44 UTC, a sequence of three 10-minute spectra were obtained with a 1.0-arcsecond-wide slit oriented along the direction of the atmospheric dispersion. The spectrograph was configured with a 300 lines mm−1 grating blazed at 5,000 Å, covering the region 3,850–8,550 Å. We obtained two further spectra of 10 minutes each on 12 May 1997 UT. Calibration Hg–Kr–Ar and flat lamp spectra were taken at the end of each exposure sequence. Owing to the extremely low elevation angle of the telescope during the observations (25°) and a problem with the Keck II lower shutter, approximately half of the telescope aperture was blocked during the exposures. Instrumental sensitivity was calibrated by an observation of the standard star15 BD+284211, but owing to an approximate correction for the occulted aperture the absolute calibration is only rough.

The resulting combined spectra from 11 May is shown in Fig. 1a. It is customary to represent the spectral flux density (Fν) of a non-thermal source by a power law, Fν να; here ν is the frequency. The optical index computed from the spectrum is αO = −0.9 ± 0.3. The large uncertainty is due to the uncertain correction for atmospheric extinction in the blue region of the spectrum at the large zenith angles of our observations.

Figure 1: The spectrum of the optical variable.
figure 1

a, Full spectrum; b, expansion of a limited region, with strong absorption lines and identifications indicated. The lines marked with an asterisk are identified with an absorption system at redshift z = 0.835, the others at z = 0.767. The spectrum has been smoothed with a three-pixel boxcar filter. A few additional weak features (not shown) have also been tentatively identified with the z = 0.767 system. Fν is the flux density, and d is the wavelength in Å.

Several absorption features are evident; the strongest, near 7,600 and 6,870 Å, are due to telluric O2. In the region between 4,300 and 5,300 Å (Fig. 1b), there are several significant absorption features16 that we identify. The identifications were made based on Mg IIdoublet (5,129 and 5,143 Å) line ratios, and assigning further rough identification of other metal lines based on wavelength ratios between these and the Mg doublet. Table 1shows the lines identified in the spectrum; independent redshifts are computed from each line. This reveals a relatively strong17 metal line absorption system at z = 0.8349 ± 0.0002, and a weaker Mg IIsystem at z = 0.768. The eight lines present in the strong absorption system make the redshift assignment unambiguous. The continuum source is either more distant and absorbed by a gas cloud at this redshift, or perhaps is located physically within the cloud, but the absorption places a firm lower limit to the redshift of the source, z 0.835.

Table 1 OT J065349+79163 absorption lines

Such absorption systems are commonly seen in the spectra of high-redshift quasi-stellar objects (QSOs)17. An imaging study of such systems18 reveals that most are associated with normal galaxies close to the line of sight to the QSO. An analysis of these systems19 at redshift similar to the system we identify in OT J065349+79163 indicates a correlation (with significant scatter) between line equivalent width and impact parameter. As the absorption we see should be similar to QSO systems, we expect that deep images (perhaps taken after the transient fades) would reveal a galaxy responsible for this absorption system, though it is difficult to predict its brightness or separation from the transient. A hint of such an object has already been suggested20. Note that as the OT was far brighter than any other nearby object, any contamination of the spectrum is negligible and thus the OT features were a physical absorption.

At these redshifts, the number of Mg IIabsorption systems with rest equivalent widths Wλ > 0.3 Å per unit redshift is of the order of unity17. Detection of one or two such absorption systems in our spectrum is thus not unusual. However, the ratio of line strengths (Mg I/Mg II) seems unusually high, and combined with the high strength of the Mg IIabsorption system provides some evidence for a dense foreground interstellar medium. This implies either a small impact parameter19, or, more likely, that the z = 0.835 system is due to the GRB host galaxy itself. We can also place an approximate upper limit to the source redshift from the absence of apparent Lyman-α absorption features in our spectra. The short-wavelength limit of our data corresponds to zLyα ≈ 2.3. In addition to the lack of individual lines, the mean observed continuum decrement at this redshift is21,22DA ≈ 0.1–0.2, and it increases with redshift. If present, such a continuum drop should be detected in our data for wavelength λ> 4,000 Å. We can thus place an approximate upper limit to the source redshift of z 2.3.

One might ask whether from current observations we should expect to see a host galaxy for the burst, if such a galaxy were present. If we assume a minimum redshift of z = 0.835 in a standard Friedmann cosmology with H0 = 70 km s−1 Mpc−1 and Ω0 = 0.2, the luminosity distance is 1.49 × 1028 cm. The B band would be redshifted just slightly past the Gunn i band, and for observations13 made on 10 May UT, the observed flux in the redshifted B band is 39 μJy. For the assumed redshift and cosmology, this implies an absolute magnitude of MB ≈ −22.6 mag, or a lower limit for LB ≈ 7 L*, where L* is a characteristic galaxy luminosity33. Thus, OT J065349+79163 is still significantly outshining any host galaxy for the most probable host luminosity range. The properties of the Mg absorption lead us to expect that once the OT has faded, the galaxy could be identified optically.

Taken together, the source's compact optical appearance, a featureless continuum, X-ray emission and high redshift suggest a possible classification (independent of a burst event) as a BL Lac object23. One of the known characteristics of BL Lac objects is their variability from radio to γ-ray wavelengths. We now evaluate the a posteriori probability that we might be seeing a BL Lac object by random coincidence with the γ-ray error box. The surface density of BL Lac objects with Rosat X-ray flux fX 10−12 erg cm−2 s−1 is not very well known, but there are indications that this distribution is quite flat at low flux densities. A simple extrapolation for the expected number of BL Lacs with fX > 6 × 10−13 erg cm−2 s−1 is24 0.03 per square degree. Thus the probability of finding a BL Lac object within the 3-arcmin-radius localization region is 2 × 10−4. The amplitude of the variability detected in the counterpart13 over a few days is also larger than has been observed in studies of BL Lac object variability25,26. Although we cannot completely exclude the possibility that OT J065349+79163 is a chance coincidence of a BL Lac object with the GRB error circle, the probability of finding a random BL Lac object which also exhibits variability that is temporally correlated with a γ-ray burst is quite small. Thus we conclude that the OT is probably associated with GRB970508, regardless of classification, though the strongest constraints naturally come from higher-energy emission.

The high redshift of OT J065349+79163, its featureless spectrum and slowly decaying optical flux are consistent with the so-called fireball models for cosmological bursts27,28,29, which are efficient at emitting γ-rays and produce power-law spectral energy distributions. The fluence30 of GRB970508 in the energy range 20–1,000 keV was 3× 10−6 erg cm−2, and at the minimum redshift implied for OT J065349+79163, this burst would have a total γ-ray energy of 7× 1051 erg (assuming isotropic emission). This falls in the general range of typical γ-ray burst energies from various cosmologicalr models31,32.

The remarkable progress in detecting X-ray and optical counterparts to GRBs has been made possible only by rapid localization of the burst by BeppoSAX and prompt dissemination of the coordinates by the BeppoSAX team. Further progress in understanding GRBs requires many more optical counterparts to be identified. It is clear from experience of the first two optical counterparts that, in order to obtain the critical data, the counterparts must be discovered and followed up spectroscopically within a few days. It now seems that an understanding of the physical mechanisms behind γ-ray bursts is within reach.