A PHOTOMETRIC REDSHIFT OF z ∼ 9.4 FOR GRB 090429B

The Burst Alert Telescope (BAT; Barthelmy et al. 2005) aboard the Swift satellite triggered on GRB 090429B at T₀ = 05:30:03 UT. The 15–350 keV light curve is composed of three distinct peaks with a total duration T₉₀ = 5.5 s, and the time-integrated spectrum can be fitted by a single power law with an exponential cutoff. The derived total fluence in the 15– 150 keV band is 3.1 × 10⁻⁷ erg cm⁻², with E_peak = 49 keV. This peak energy is among the few detected by Swift within the BAT bandpass.

After 106 s, the narrowfield instruments began their standard burst-response observation sequence. The X-ray Telescope (XRT; Burrows et al. 2005) identified an uncataloged fading source at R.A.(J2000) = 14^h02^m40$\mbox{$.\!\!^{\mathrm s}$}10$, decl.(J2000) = +32°10'146; no optical/UV counterpart was seen in the UV–Optical Telescope (Roming et al. 2005) data. The X-ray data has been characterized using standard routines in Heasoft, Xspec, and QDP, with the light curve fitting process as described in Evans et al. (2009). For some analyses, we have used the automatic data products produced by the UK Swift Science Data Centre (Evans et al. 2007, 2009). Our presentation of parameters derived from the Swift data follows the convention of quoting errors at the 90% level. The time-averaged 0.3–10 keV X-ray spectrum from 97 to 29893 s after the burst is best fit by a power law with photon spectral index Γ_X = 2.01^+0.16 _{− 0.24} and with a total absorption column density of $\mbox{$N_{\rm H}$}= 10.1^{+4.6}_{-5.3} \times 10^{20}$ cm⁻², mildly (2.7σ) in excess of the Galactic absorption of 1.2 × 10²⁰ cm⁻²; we discuss the possible significance of this finding in Section 3.5.

The X-ray light curve, given in Table 1 and illustrated in Figure 1, is adequately fit by a combination of brightening and fading temporal power laws: initially, the X-ray flux rises with temporal index α_X1 = −0.96^+0.43 _{− 0.52}, referenced to the burst time; following the peak time T_X = 589⁺¹⁴⁶ _{− 80} s, the light curve then breaks to a power-law decay with α_X2 = 1.20^+0.08 _{− 0.07}.

Zoom In Zoom Out Reset image size

Basic reduction steps for all optical and NIR photometry were performed using IRAF software.²⁵ Photometric analysis used both IRAF and the Starlink GAIA software, as well as our own custom scripts. Errors in the sky-subtraction step are estimated from multiple apertures of size equal to that of the source aperture, placed around the field of the GRB.

Optical images were calibrated using field stars from the Sloan Digital Sky Survey (SDSS) Data-Release 7 (DR7) catalog (Abazajian et al. 2009) and NIR images were provisionally calibrated directly to the Two Micron All Sky Survey (2MASS) catalog, but subsequently refined as described below. Detections and limits on the brightness of any associated source are presented in Table 2.

2.2.1. ESO2.2m/GROND Observations

2.2.2. VLT Observations

2.2.3. Gemini-North Observations

Beginning roughly 2.5 hr after the burst trigger, we carried out a series of observations from Gemini-North. We gathered optical i'z' imaging with the Gemini Multi-Object Spectrograph (GMOS; Hook et al. 2004) and NIR JHK imaging with the Near-Infared Imager (NIRI; Hodapp et al. 2003). GMOS observations consisted of five exposures of 3 minutes each, per filter; NIRI observations consisted of eight dithered positions of 60 s each. The Gemini GMOS and NIRI packages under the IRAF environment were used to sky-subtract, align, and combine the images. The NIRI images were also corrected for the small nonlinearity effect seen in the detectors.²⁶ Photometry was performed relative to SDSS stars for the GMOS data and relative to secondary calibrators from GROND for the NIRI data (see Section 2.3). Our photometry is presented in Table 2.

While no optical counterpart was present in our i' or z' images, we did identify a source within the X-ray localization in our NIR observations. The position of the source was R.A. = 14^h02^m4010, decl. = +32°10'1420. Following this discovery we attempted spectroscopic observations from Gemini-North; however, increasing summit winds forced the closure of the telescope and meant that these were aborted with <10 minutes of useable exposure time and no trace is visible in the observations. We obtained a second epoch of K-band observations on April 30 UT revealed a clear fading of ≈1.2 mag of the identified source, confirming its transient nature (and corresponding to a power-law index of α_K = 0.53 ± 0.10, shallower than the X-ray decay at that time). Figure 2 presents our Gemini imaging data, while the lower panel of Figure 1 shows our optical/NIR light curve. The resulting SED, from X-ray to IR, is shown in Figure 3.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

No evidence of a host galaxy is present in our images. A deep r'-band image of the field, taken again with GMOS under good conditions (04 seeing) at 14 days after the GRB, is shown in Figure 4. This allows us to place a 3σ upper limit on the host galaxy apparent magnitude of r' > 27.07 mag. We also note in these images the presence of a massive elliptical galaxy, offset roughly 45'' from the GRB location. This galaxy has absolute magnitude M_r ≈ −21.6 and M_K ≈ −24.5 (∼L*; Jones et al. 2006). It appears to be the central galaxy of a modest cluster at z = 0.079.²⁷ It is likely that this foreground structure provides some lensing boost to the observed flux of the burst, although the relatively large impact parameter suggests it will not be a major factor.

Zoom In Zoom Out Reset image size

2.2.4. HST Observations

2.2.5. UKIRT Observations

Since our photometric redshift analysis will depend critically on the accuracy of our photometry, we took particular care in both the calibration and estimates of photometric uncertainties.

The Gemini-North/NIRI detections are crucial, but also difficult to analyze since the field of view is small (2 arcmin on a side) and there is only one 2MASS star (namely, star B in Table 4, and Figure 4) that is in all the sub-exposures of the nine-point dither pattern. There is another 2MASS star (A in Table 4) which appears on two of the sub-exposures, and we used this as a cross check on the derived photometry. Both these stars are toward the faint end of the 2MASS catalog and have relatively large photometric uncertainties. To overcome this we used the wide-field UKIRT/WFCAM and ESO2.2 m/GROND JHK images (both of which were obtained close in time to the Gemini observations), which were very precisely calibrated using many bright 2MASS stars, to obtain more accurate magnitudes for these reference stars. The two independent determinations were consistent with each other within their respective calculated errors (typically 0.01–0.02 mag), and we therefore formed a weighted average to obtain our best estimates of the NIR magnitudes, as shown in Table 2.

Magnitudes for the afterglow were measured relative to star B, although this procedure was further complicated by the fact that the point spread function (PSF) was found to change across the frame resulting in the core of the reference star becoming noticeably extended when it was close to the southern edge of the detector, as it was in some sub-exposures. This precluded small aperture (5 pixel radius, ≈06) photometry for these exposures, so in such cases we used a fainter star (C in Table 2) closer to the GRB position as a secondary reference, having determined its magnitude relative to star B using those frames where it was not near the edge. We note that a small aperture was required to maximize the signal-to-noise ratio for the afterglow and that profile-fitting photometry was deemed inappropriate due to the small number of bright stars available to define the PSF.

The magnitudes (and errors) for the afterglow in each band were then determined from an error weighted mean of the different sub-exposures. Finally, we converted to flux density using a recent NIR spectrum of Vega (see Bohlin 2007) which resulted in values that are 2%–3% higher for our passbands than found using the conversion in Table 7 of Hewett et al. (2006). These are the flux densities reported in Table 2, although we note that when we come to the photometric redshift analysis (below) we fit in counts rather than flux, to allow for the different spectral shapes of the afterglow and comparison star.

Since the optical observations provided only upper limits the overall fit is not strongly sensitive to the precision of the optical photometry. However, in this case, the field of GRB 090429B lies fortuitously within the SDSS survey area, and our most constraining optical limits (from GMOS) are obtained in the same filter set. This allows a precise photometric calibration of these images. For our Very Large Telescope (VLT) observations we calibrate the field using SDSS observations and the transforms of Jester et al. (2005). These latter values were confirmed as reasonable using archival zero points.

Since our observations were taken over several hours, temporal variations in the afterglow luminosity could affect our analysis. Unfortunately, we have a rather limited handle on the variability at optical/IR wavelengths. Although most GRBs begin power-law decline in luminosity fairly early, in some cases flat or even increasing luminosity can be seen for a period of time (e.g., Rykoff et al. 2009). A rapidly fading afterglow (similar to those commonly observed) would imply even more stringent upper limits in our (earlier) blue band filters since the extrapolation to a common time would yield a more extreme limit on color index in, e.g., z − H. Alternatively, the more unusual case of a rapidly rising afterglow would yield somewhat weaker constraints since the non-detections in the bluer bands could be ascribed to the brightening of the afterglow in the time frame between the optical and IR observations. However, this is countered by the fact that such a rising afterglow would also imply an even bluer H − K color, which is more difficult to attain with extinction. In this scenario, the rising of the afterglow may offer some support for a high-redshift scenario, since the time dilation at z ∼ 9 would result in a forward shock which takes a factor of ∼10 longer to reach maximum than at z ∼ 0.

In an attempt to constrain the temporal slope of the optical afterglow we first perform photometry on the individual NIRI frames. We find no statistically significant variation over the ∼10 minute time frame of these observations, implying that the afterglow is not varying especially rapidly. Second, we utilize acquisition images taken prior to the aborted NIRI spectroscopic observations. These suggest a minor brightening of the afterglow between 13000 s and 17000 s after the burst (0.3 ± 0.2 mag, corresponding to α ≈ −1.0 ± 0.7). However, these observations were obtained at a single dither position and contained substantial persistence. Hence, we can accurately remove neither sky nor dark current and the resulting observations contain large variations in the sky on relatively short length scales. Thus, we caution against their use for detailed photometric work, aside from noting that the afterglow is neither rising nor falling at an unusually rapid rate. We gain a much better handle on the decay between the first- and second-night observations, which gives α ≈ 0.6 ± 0.1, but of course this may not apply during the first few hours post-burst.

On balance, then, we favor photometric fitting in which the observed magnitudes are assumed to be constant over the period of the early observations (i.e., we assume α = 0). This is consistent with the relatively flat X-ray behavior between 1 and 3 hr post-burst (Figure 1). For completeness, we have included a single power-law temporal decay as a possible parameter within these fits and confirm that for any reasonable slope −1 < α < 1 our results are broadly insensitive to the assumed value of α (see below). To avoid extrapolating over too wide a range of times and to counter against unusual afterglow behavior (which is normally most notable in the first few hundred seconds in the rest frame) we also fit only data taken after 4000 s. The inclusion of earlier data would further strengthen our results if we assumed the afterglow to be decaying, but would produce minimal difference to the fit if we assumed a flat or rising light curve (since the early limits are shallower than those obtained at later times).

Here we attempt to derive the redshift of GRB 090429B via our broadband photometry of its afterglow. The absence of any detections within the optical window, if interpreted as the signature of high redshift, immediately implies z > 6.3. Similarly, if we interpret the red J − H color of the afterglow as indicative of a Lyman-break lying within the J band, the inferred redshift is 8.0 < z < 10.5.

To obtain stronger constraints on the redshift of GRB 090429B we performed the following analysis. We considered just the seven deepest observations, namely, those obtained at the VLT and Gemini-North on the first night with filters redder than 6000 Å. We assume initially that there is no temporal variation over the course of observations, although including plausible variability within our fits also confirms that our results are broadly insensitive to this assumption (see below). After correcting for Galactic foreground extinction (E_{B − V} = 0.015; Schlegel et al. 1998) we fitted these flux density measurements with a grid of simple models for the SED of the afterglow. The errors are likely to be Gaussian distributed, to good approximation, since the uncertainties are dominated by background subtraction (although we also included zero-point calibration uncertainties in the modelling), and therefore used minimum-χ² fitting.

Specifically, the model was a simple power law, with the spectral index, β_O, and overall normalization as free parameters. The grid of models spanned a range in redshift of 0 < z < 12 and rest-frame V-band extinction of 0 < A_V < 12. Beyond z ∼ 7 we are effectively fitting only three data points so there exists a degeneracy between extinction (A_V) and β_O. We therefore include a weak prior for the probability distribution of the value of β_O (see Figure 5) which is modeled as a lognormal with a peak likelihood at β_O = 0.5 and a width such that the relative likelihood is 50% of the maximum from about 0.3 < β_O < 0.85. This is physically motivated since it allows values of β_O over a broad range, comparable to the range usually observed (e.g., Schulze et al. 2011), but in particular prefers β_O = β_X − 0.5 ≈ 0.5, as would be expected if there was a cooling break between the X-ray and optical regimes (e.g., Sari et al. 1998). The plausibility of such a cooling break is clear from Figure 3, while relaxing this assumption does not lead to any viable fits at low redshift. Added to this was absorption due to neutral hydrogen in the IGM (Madau 1995; neutral hydrogen in the host was taken to have a typical column density of 10²¹ cm⁻², although the results are insensitive to the exact number assumed) and extinction due to dust. We experimented with several dust laws, from the Milky Way (MW), Large Magellanic Cloud (LMC) and Small Magellanic Cloud (SMC; Pei 1992), as well as the extinction law of Maiolino et al. (2004).

We also impose a weak prior on the intrinsic luminosity of the optical afterglow (Figure 5). Studies such as that of Kann et al. (2010) indicate that there is an upper envelope to the (broad) distribution of GRB optical afterglow luminosities. We therefore apply a prior which is flat below this envelope and cuts off exponentially at brighter luminosities, although the cutoff is slow enough to allow a reasonable probability that the luminosity could be somewhat higher.

Zoom In Zoom Out Reset image size

In fact, rather than fitting directly to the flux densities, we integrated our model spectra over response functions and compared the counts obtained by integrating an approximate spectrum of the comparison star. The response functions were obtained from the measured filter transmission curves, multiplied by a typical atmospheric absorption curve generated by ATRAN.³⁰ Going to these lengths effectively corrects for the small difference in the SED shape of the afterglow from the reference star, although again the conclusions are not greatly affected. In Figure 6 we show the photometric data points and the best-fit model for the afterglow spectrum assuming that the afterglow did not evolve temporally during the first 3 hr (see below) and that the dust is similar to that in the SMC, which has frequently been found to be a good approximation to the dust laws along many other GRB sight lines (e.g., Schady et al. 2007, 2010). We also show the best-fit low-z model (as it happens z ≈ 0), which is formally ruled out at high confidence. In Figure 7 we plot contours of χ² over a grid of models spanning a range of redshift and rest-frame V-band extinction, A_V. The red cross shows the best-fitting model, which has z = 9.36 and extinction A_V = 0.1, although the 99% confidence contour runs as low as z ≈ 7.7 if there is a modest amount of dust (rest-A_V ∼ 0.5) in the host. Marginalizing the likelihood (which we define $\mathcal {L}\propto {\rm exp}(-\chi ^2/2)$) over A_V (assuming a flat prior) indicates a 90% likelihood range of 9.02 < z < 9.50. There is no solution at lower redshifts (z < 7) which is not ruled out at ≫99.9% level; the best fit at low redshift (z ≈ 0 as it happens, as shown by the blue cross) requires a very high extinction of A_V ≈ 10.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In Figure 8 we show similar likelihood contours for fits spanning a broader range of models with different prior assumptions for the temporal power-law decline index α and commonly used dust laws. Changing α to ±1 makes rather little difference, and in any case, as discussed above, there is evidence to suggest that the luminosity was not changing even as rapidly as this. Varying the dust law does have more effect, largely due to the 2175 Å feature in the MW, LMC (Pei 1992), and Maiolino et al. (2004) laws producing the blue H − K color even at slightly lower redshifts, although in most cases the best fit remains z ≳ 9. The Maiolino et al. (2004) dust law was determined from observations of a quasar at z = 6.2 and is argued to be consistent with dust produced largely from early supernovae (note that this law is only defined up to ∼3200 Å in the rest frame, and we therefore graft it to the SMC law at this point). This case is interesting as it does allow redshifts as low as z ∼ 6.5 at 99% confidence, although to date, only GRB 071025, with a photo-z ∼ 5, has shown evidence of requiring such a dust law (Perley et al. 2010).

Zoom In Zoom Out Reset image size

Finally, in Figure 9 we show the likelihood as a function of redshift for the SMC dust-law models having marginalized over both α (assumed a flat prior between −1 and +1) and A_V (assumed a flat prior between 0 and 12). The maximum likelihood is at z = 9.38, and 90% of the likelihood is between 9.06 < z < 9.52.

Zoom In Zoom Out Reset image size

Our late-time date taken with Gemini and HST are potentially extremely valuable, since we can use the absence of any host galaxy candidates to assess the plausibility of any lower-z solutions to our photometric redshifts (the HST images are shown in Figure 10). The detection of a host galaxy in the optical was used, for example, to show that GRB 060923A was z < 3 despite its afterglow being a K-band dropout (Tanvir et al. 2008). In the case of GRB 090429B, the possible low-redshift scenarios seem to be those with z < 1 and high dust extinction A_V ∼ 10 (although we emphasize that such models remain formally ruled out). The limits these data provide on this are shown graphically in Figure 11, where we plot the absolute inferred magnitude of the host galaxy in the observed V, Y and H bands as a function of redshift. For completeness we cut each line at the point where 1216 Å × (1 + z) passes the central wavelength of the band. At z = 0.1, close to the minimum of our lower redshift solution, we obtain inferred absolute magnitude limits of M_V > −10.6, M_Y > −9.9, M_H > −10.5, these exceptionally deep limits are comparable to the luminosities of bright globular clusters, and significantly fainter than any known GRB or supernova host galaxy; indeed, they place limits of ⩽10⁻⁴ L* (Blanton et al. 2003). Even at z = 1 the observed Y-band limits would imply M_B > −15.1, or ⩽0.001L* (Ryan et al. 2007).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In this regard it is worth noting that the lower redshift solutions are only viable in cases where the host galaxy extinction is high, whereas such faint galaxies typically have low metallicity and little dust, and it is therefore extremely unlikely that one could create the extinction (A_V ∼ 10) necessary to explain GRB 090429B. Any z < 3 solution would require the host of GRB 090429B to be fainter than the large majority of GRB hosts currently known. Furthermore, our wide wavelength coverage would also allow us to uncover any very red dusty host galaxies, which would provide the necessary extinction, but would be missing from optical only searches (e.g., Levan et al. 2006; Svensson et al. 2010).

This offers strong support for our high-z model, In these cases only the F160W observation yields potential information as to the magnitude of the host galaxy. The inferred 1500 Å absolute AB magnitude at z ∼ 9.4 is −19.95. This lies roughly in the middle of the observed absolute magnitude distribution of z ∼ 8 candidate galaxies found in deep HST ACS and WFC3 imaging (Bouwens et al. 2010), and thus the non-detection of a galaxy at z ∼ 9.4 is not unexpected in observations of this depth.

In addition to the above discussion there are additional lines of evidence which offer support for the high-z interpretation of GRB 090429B. The best z ≲ 5 solution is actually at z ∼ 0 (although it remains a very poor fit to the available optical data) and requires A_V ∼ 10, for an SMC extinction law, which would correspond to a foreground N_H ∼ 10²³ cm⁻², nearly two orders of magnitude larger than is observed. While the dust-to-gas ratios observed through the MW can show moderately large variations, such a large offset would be unheard of, particularly in GRB afterglows where typically the ratio of dust extinction to X-ray determined gas column is actually less than is seen locally (Schady et al. 2010). Hence, the observed X-ray spectrum seems to rule out any low-redshift (z ≲ 1), high-extinction scenario. This is illustrated in Figure 7 which shows that the best low-redshift solutions are well above the contours of A_V inferred from the excess N_H assuming typical GRB dust-to-gas ratios.

A second line of evidence comes from observed high-energy correlations seen in many GRBs, in particular, in the relation between the peak energy of the νF_ν spectrum (E_p) and the burst isotropic energy (Amati et al. 2008). Although this relation has a significant scatter, it can also be used to place some constraints on the burst redshift under the assumption that all long bursts should follow the relation. For GRB 090429B, the burst is only consistent with this relation at better than 3σ if z > 1, implying that it also disfavors very low redshift models for the origin of GRB 090429B. This result couples with the lag–luminosity relation, which would imply an isotropic luminosity of L_iso ∼ 10⁵³ erg s⁻¹, similar to other long GRBs in the "silver sample" (Ukwatta et al. 2010).

The observed fluence of GRB 090429B is 3.1 × 10⁻⁷ erg cm⁻², comparable to that observed for GRB 090423, and implies an isotropic energy release in the 15–150 keV band of E_iso = 3.5 × 10⁵² erg at z ∼ 9.4. Its absolute X-ray (rest frame ∼3–100 keV) brightness of ∼2 × 10⁴⁹ erg s⁻¹ at ∼1000 s, and K-band luminosity of M_{λ/(1 + z)} of ∼ − 26.2(AB) are also similar both to GRB 090423 and to the bulk of the long GRB population.

The rest-frame duration at first sight seems surprisingly short, with T₉₀/(1 + z) ∼ 0.5 s. Interestingly, three of the four highest redshift GRBs discovered prior to this one have also had rather low values of T₉₀/(1 + z) around 1–4 s (Ruiz-Velasco et al. 2007; Greiner et al. 2009; Tanvir et al. 2009; Salvaterra et al. 2009). A possible explanation for this tendency may be that for high-redshift sources the BAT is observing at rest-frame MeV energies, where the light curves tend to be more rapidly variable and of shorter duration than at lower energies, and thus it is plausible that just a single peak of emission is detected rising up above the noise in these cases.

An analysis obtained using the technique discussed in Ukwatta et al. (2010) reveals that this GRB presents a small value of lag, 31/(1 + z) = 58 ± 27 ms between the low-energy (15–25 keV) and the high-energy (50–100 keV) channels. The rest-frame correlations found by Ukwatta et al. (2010) indicate this corresponds to a peak isotropic luminosity L_iso > 10⁵² erg s⁻¹, consistent with the observed fluence and short duration.

Another interesting issue is that of the absorption inferred from the X-ray spectrum. Although the measurement is not highly significant, if taken at face value, the rest-frame column density is N_H = 1.4^+1.0 _{− 1.0} × 10²³ cm⁻² (90% confidence range). This would be already high compared to most other Swift observed GRBs and would be higher still if, as is very likely, the metallicity is substantially less than solar (which by convention is often assumed in calculating N_H). Such a high-column density is no doubt surprising, although a similar value was found for GRB 090423 (Tanvir et al. 2009; Salvaterra et al. 2009). As in that case, it also raises the question of whether a high gas column would be compatible with the low extinction indicated by the NIR afterglow. GRBs are expected to be able to destroy dust to fairly large distances from their birth sites (e.g., Waxman & Draine 2000; Fruchter et al. 2001), but the good fit for many afterglow SEDs with an SMC extinction law suggests they have not generally been highly modified by dust destruction, which would tend to produce "gray" extinction laws (e.g., Schady et al. 2010). In any event, the large error bar on this measurement and potential systematic uncertainties in calibrating the soft response of the XRT, taken together with the wide range of dust-to-gas ratios seen to other GRB sight lines, makes the significance of this conflict hard to assess at the present time.

On balance, it seems that in most respects the general properties of GRB 090429B do not stand far apart from the population of long-duration GRBs, even at the inferred redshift of z ∼ 9.4. In particular, it shows no evidence that its progenitor is distinct from those of GRBs seen in the more local universe. This is of particular importance at z ∼ 9.4, since it is close to the redshift where WMAP observations imply the bulk of reionization of the universe occurred (z = 10.6 ± 1.2; Komatsu et al. 2011). This reionization process is likely to have been driven by the first generations of star formation, including Population III stars, whose pristine H+He composition is expected to lead to generally more massive stars. It has been proposed that a consequence of this could be that Population III stars produce particularly long duration and energetic GRBs (e.g., Mészáros & Rees 2010). This is clearly not the case for GRB 090429B, and hence we conclude that its progenitor was more likely to be a high-mass second-generation (Population II) star.