THE LACK OF INTENSE Lyα IN ULTRADEEP SPECTRA OF z = 7 CANDIDATES IN GOODS-S: IMPRINT OF REIONIZATION?

During the past year, a suite of new near-infrared (NIR) surveys has extended the search for star-forming galaxies to redshift 6.5 ⩽ z ⩽ 10 using the well-proven Lyman-break technique (Giavalisco 2002, and references therein). With respect to lower redshift, the number density of UV-selected galaxies decreases (e.g., Ouchi et al. 2009; McLure et al. 2010; Castellano et al. 2010; Bouwens et al. 2010a; Wilkins et al. 2010a), their UV continuum becomes bluer implying either reduced dust obscuration or poorer metal enrichment or both (e.g., Finkelstein et al. 2010; Bouwens et al. 2010b; Schaerer & de Barros 2010), and their stellar masses are, on average, smaller than those of their lower redshift counterparts (e.g., Labbé et al. 2010). Unfortunately, these results are based on color-selected samples with no spectroscopic validation. At the time of writing, spectroscopic detections of only a few individual objects have been obtained at z>6.6 (Iye et al. 2006; Greiner et al. 2009; Salvaterra et al. 2009; Tanvir et al. 2009).

The lack of knowledge of the true redshifts of the current z ∼ 7 candidates places significant limitations on our ability to robustly measure the properties of the galaxies at this critical cosmic epoch. For example, the fraction of interlopers and the redshift distribution of the sample galaxies are necessary to robustly measure the UV luminosity function (e.g., Reddy & Steidel 2009). Currently, the former remains unknown, and the latter is estimated with Monte Carlo simulations under various assumptions for the intrinsic distributions of UV spectral energy distribution (SED), surface brightness, and morphology, with the result that the measure of the luminosity function remains subject to uncontrolled systematic errors.

In practice, given the marked decrease in sensitivity of current spectroscopic observations at increasing redshift, the spectral confirmation of galaxies at z>5 relies heavily on their Lyα emission line (Stark et al. 2010; Vanzella et al. 2009, S10 and V09 in the following). Indeed, redshifts derived without Lyα typically have lower confidence, although their number may be comparable (Douglas et al. 2010, D10 in the following). The line in itself is an important diagnostic of physical processes at work in the galaxies (e.g., Giavalisco et al. 1996; Pentericci et al. 2010; Shapley et al. 2003), since its strength and velocity profile depend on the instantaneous star formation rate, dust content, metallicity, kinematics, and geometry of the interstellar medium. Particularly relevant here is the evidence that the fraction of Lyα emitters in UV-selected samples increases with redshift (V09; S10; Reddy & Steidel 2009; Stanway et al. 2007, S07 in the following) and that the fraction of galaxies with a large Lyα equivalent width (EW) is substantially larger at fainter UV luminosities.

Finally, the very visibility of the Lyα line during the ending phases of the cosmic re-ionization is subject to the damping effect of an increasing neutral intergalactic medium (IGM; e.g., Zheng et al. 2010; Dayal et al. 2010), expected to attenuate most of its luminosity and make the earliest galaxies consequently more difficult to identify. Hence, the line profile and the evolution of its EW are sensitive diagnostics of the ionization state of the high-redshift IGM.

To address these issues we have started a campaign of spectroscopic follow-up of z ≃ 7 "Z-dropout" candidates, selected from high-quality imaging surveys obtained with Very Large Telescope (VLT)/Hawk-I and HST/WFC3. In this Letter, we present the first results from a sample selected in the GOODS-S field (Castellano et al. 2010, C10 in the following). All magnitudes are in the AB system, and we adopt H₀ = 70 km s⁻¹ Mpc⁻¹, Ω_M = 0.3, and Ω_Λ = 0.7.

This initial spectroscopic sample includes relatively bright Lyman-break galaxy (LBG) candidates at z>6.5 (listed in Table 1), five from the Hawk-I images (4 from C10 and 1 from Hickey et al. 2010) and two from WFC3 (Oesch et al. 2010; Wilkins et al. 2010b), spanning the magnitude range Y ≃ 25.5–27.5. We filled empty slitlets in the multi-object slit masks with other candidates of lower quality and/or at lower redshift, including a candidate brown dwarf (Mannucci et al. 2007, C10) and i-dropouts from the GOODS survey not observed by V09.

2.1. Observations

Observations were taken in service mode with the FORS2 spectrograph on the ESO VLT, between 2009 November 12 and 2010 January 14. We used the 600Z holographic grating, that provides the highest sensitivity in the range 8000–10000 Å with a spectral resolution R ≃ 1390 and a sampling of 1.6 Å per pixel for a 1'' slit. The data presented here come from the co-addition of 75 spectra of 842 s of integration each, on a single mask, for a grand total of 63150 s (17.5 hr), with median seeing around 08. Each slitlet was 1'' wide and 14'' long, to maximize the number of slits available while allowing a careful sky subtraction. All our high priority targets were placed at the center of the slits, and spectra were taken in series of three different positions, offset by ±2'' in the direction perpendicular to the dispersion.

Since our objects are extremely faint, the slit centering was based on the astrometry solution obtained from the Hawk-I images, which is well aligned to the Advanced Camera for Surveys (ACS) one. We have directly verified this by placing a few bright objects from the ACS catalogs in small slits, and ensuring that they were correctly aligned during the observations. It is also reassuring to note that three faint i-dropouts selected from the ACS catalog which were placed in slitlets using the same astrometry have a clear Lyα detection at z ≃ 5.94 (full details will be given in a future paper).

Data reduction was performed using an optimized version of the recipes adopted in V09 and previous papers. After standard flat-fielding and wavelength calibration, we subtracted the sky emission lines with two different procedures. In the first case (Polyn in the following) we fit polynomials of order n (from 1 to 5) to the sky intensity at each pixel position. This procedure in principle ensures the highest signal-to-noise ratio (S/N), but is sensitive to systematics induced by defects in the detector or in the slit manufacturing. A safer but somewhat noisier approach (ABBA in the following) is to subtract the sky background between two consecutive exposures, exploiting the fact that the target spectrum is offset due to dithering. We found the spectra obtained with the two techniques entirely consistent. Finally, spectra were flux-calibrated using the observations of spectrophotometric standards. Slit losses are small, given the extremely compact size of the targets and we neglect them in the subsequent analysis.

The rms of the resulting spectra, which will be used later to determine the probability of our results with a Monte Carlo simulation, has been estimated "by first principles," i.e., computing the absolute rms of each frame from its raw counts c as $\sqrt{(c/g)}$ (where g is the e⁻-ADU conversion factor) and propagating it through all the reduction steps. It turned to be in excellent agreement with the observed rms in the region between 8150 and 8250 Å that is devoid of sky lines and with the predicted efficiencies estimated by the ESO exposure time calculator. The resulting 1σ limiting flux density is shown in the lower panel of Figure 1.

Zoom In Zoom Out Reset image size

To obtain the corresponding limit on the detectable EW for a Lyα line, we have computed three different cases, assuming continuum magnitudes of m = 25.5, 26.5, 27.5, to span the luminosity range of our targets. For the computation we assume that the flux profile is a Gaussian with FWHM =10 Å. The resulting limiting EW is shown in the upper panel of Figure 1, computed at the 10σ level. We could detect weak (EW ≃ 5 Å) Lyα lines in our brightest galaxies, and even for the faintest ones we are able to reach EW ≃ 50 Å over a significant fraction of the redshift interval. This range of sensitivity is similar to that of z ≃ 5–6 surveys (S07, V09, D10, S10)

2.2. Results

We detect only one weak emission line, centered at 9691.5 ± 0.5 Å in the spectrum of the object G2_1408. This galaxy is the brightest candidate identified in the Hubble Ultradeep Field (HUDF) area, and one of the brightest in C10. It was first detected by Bouwens et al. (2004) in the NICMOS HUDF data, and subsequently identified also by C10 and in the HUDF WFC3 data (Bouwens et al. 2010a; Oesch et al. 2010; McLure et al. 2010; Bunker et al. 2009). From the clear elongation observed in the WFC3 images, one can exclude the possibility that it is a brown dwarf. The 2D and 1D spectra of G2_1408 are shown in Figure 2. The spectral feature is extended over 4 pixels in the spatial direction, consistent with the average seeing. The FWHM is ≃10 Å, significantly larger than any feature due to noise. The weak emission line has a total observed flux of 3.4 × 10⁻¹⁸ erg cm⁻² s⁻¹. The formal S/N is 7, but this estimate does not include systematic errors, and should be considered as an upper limit. We made extensive tests to verify the reliability of this detection. We verified that the feature is present both in the Polyn and in the ABBA reductions, as shown in Figure 2. We then inspected all the 75 individual spectra to ensure that the feature is not due to an artifact, and that it is still detected when we separately summed the data in two halves. Because of the large color break (z −  Y>2.1) measured in the HUDF data and the non-detection in the BVI bands, an identification of this line with a lower redshift [O ii] or Hα would imply a very peculiar SED, unlike that of currently known galaxies. This cannot be excluded a priori.

Zoom In Zoom Out Reset image size

We note that there is no evidence of the asymmetry that is expected (but not required, see discussion below) for a z ≃ 7 galaxy, although the S/N is too poor to reach any firm conclusion about this.

Based on these tests, we conclude that the feature is likely real and due to Lyα emission from a z = 6.972 galaxy (z = 6.970 if computed at the blue edge of the line), although this should be validated by independent and possibly deeper observations. No continuum is detected in the spectrum: if we estimate it from the Hawk-I Y-band magnitude (Table 1), the line flux translates into an observed EW of 103 Å, corresponding to 13 Å if placed at z = 6.972.

We do not identify any other emission lines from objects in our sample. We only detect a faint continuum from two objects, namely, G2_2370 (the brightest in our sample) and the brown dwarf candidate of Mannucci et al. (2007). In both cases, the continuum is consistent with the broadband magnitudes but the low S/N prevents us from deriving any robust information about their spectral type or redshift.

The key result of our observations is the lack of prominent Lyα emission lines in our sample, which may imply a rapid evolution in the physical properties of z>6 galaxies and/or in the surrounding IGM. To quantify this issue, we have carried out the following Monte Carlo simulations under the assumptions that (1) all our seven candidates are indeed z ≃ 7 galaxies and (2) the distribution of the Lyα intensity in galaxies as a function of their rest-frame continuum magnitude M_UV does not change significantly from z = 4–6 to z = 7.

For the redshift distribution expected for our sample we use the result by C10 (see their Figure 7), which has a broad maximum from z = 6.4 to z = 7.1 and tails that extend to z = 6 and z = 7.5. The distribution of the Lyα intensity in galaxies at z = 3–6 has been investigated in a number of studies (S07, V09, S10, D10), showing that the intensity of Lyα is anti-correlated with rest-frame UV luminosity. No measure of the dependence of the EW distribution as a function of M_UV has been obtained, however. We model the EW distribution assuming that at EW >0 it is represented by a Gaussian centered on EW = 0 with an additional constant tail up to 150 Å, and at EW <0 by a constant level down to some EW_min value, and null below. We take the width of the Gaussian and the two tails to reproduce the results of V09 and S10 at different rest-frame magnitudes. Specifically, we derive from the bright galaxies in V09 a standard deviation for the Gaussian of 10 Å, and assume that it is constant at all magnitudes. We then divide our sample in two luminosity bins (−20.5 < M_UV and −20.5 < M_UV < −19.5) and adjust the two tails in order to reproduce the fraction of galaxies with EW>50 Å given by S10 and the fraction of galaxies with EW >5 and EW>20 Å (for the two bins, respectively), as given by the V09 data. The resulting distributions are shown in Figure 3 for the two magnitude bins, and are reasonably similar in shape to the EW distribution at z ≃ 5–6 (S07), and show a moderate evolution from the z ≃ 3–5 (Shapley et al. 2003, D10) one.

Zoom In Zoom Out Reset image size

We then compute the probability of detecting N Lyα lines at a given S/N in our sample of seven objects. For each object we randomly extract a redshift from the C10 distribution, we compute the corresponding M_UV from the observed Y-band magnitude (taking into account the IGM absorption at that redshift), and we then randomly extract an EW from the corresponding distribution. If the EW is larger than the minimum detectable EW at the corresponding wavelength (Figure 1) for a given S/N we conclude that the object would be detected. We assume FWHM = 10 Å for the line, as found at z = 6.9 by Iye et al. (2006; see also Figure 1). Clearly, intrinsically broader lines would be harder to detect. We perform this exercise 10⁵ times over the whole sample, requiring S/N >10 for the detection (larger than the S/N of the possible detection in G2_1408), and we finally obtain the probability distribution shown in the lower panel of Figure 3. Under these assumptions, the probability of detecting no Lyα line in our sample is very small, about 2%, while the typical number of Lyα that we should have detected is between two and four. We also find a low probability (4%) of having one detection at S/N >5, as found in our sample. The same probability adopting the S07 distribution would be much smaller (≃10⁻³), because of the substantial tail of objects with large EW. Even using the Shapley et al. (2003) distribution, which has a lower fraction of high EW objects, the probability is still rather low (9%). We conclude that, with all the obvious caveats due to the small size of our sample and to possible observational mishaps, the lack of prominent Lyα lines in our sample is statistically significant.

On a practical level, our results show how challenging it is to obtain large samples of spectroscopically confirmed galaxies at z>6.5 with current instrumentation, especially if one aims at reaching the level of completeness (≫50%) needed to robustly measure the luminosity function. Our observations imply that this goal will have to wait until a future generation of instruments is available, either 8 m telescopes equipped with multi-object spectrographs more efficient in the z and Y bands or, more likely, the new generation of telescopes, such as the James Webb Space Telescope or 20–40 m ground-based facilities. Nonetheless, our analysis appears to show that the failure to detect prominent Lyα in our sample is not only due to the insufficiency of current instrumentation.

One possibility is that a significant fraction of the candidates are lower redshift interlopers. We test this possibility by extrapolating to z ∼ 7 the observed contamination in spectroscopic samples at z ∼ 4, 5, and 6 (V09, Table 4 of B, V, and i dropouts), which increases with redshift. We assume that among our z-band dropout sample the fraction of contaminants could be ∼25%, i.e., two out of seven candidates. This estimate may be pessimistic, given the excellent photometric quality of the Hawk-I and WFC3 data, and the more careful cleaning of lower z interlopers compared to the V09 samples. However, the contaminant population may be changing at higher redshifts, and different and previously unstudied galaxy types may be entering the selection window. Ignoring these uncertainties, we repeated the Monte Carlo simulation for all possible choices of five candidates from our seven, finding that the probability of detecting no Lyα line at S/N >10 is still rather low, being typically 8%, and only in one case reaching 15% (this range depends on which candidates are excluded from the sample).

Another explanation for the paucity of Lyα detections would be physical evolution of either the galaxies or the surrounding IGM. The intrinsic strength of the Lyα emission is expected to increase at higher redshift as galaxies become more metal- and dust-poor. The probability that these photons escape the galaxy and its surroundings, however, depends on a series of complex (and not fully characterized) phenomena in the IGM surrounding the galaxies (Zheng et al. 2010, and references therein), including the relative geometry and dynamics of gas and dust, e.g., backward scattering from wind-driven outflowing shells (which can even boost the strength of the lines) or absorption by the damping wings of in-falling IGM along the line of sight. The presence of H i in proximity to the source can result in an absorption of the intrinsic Lyα by one order of magnitude or more (Zheng et al. 2010), along with a broadening and redshifting of the emerging line profile. Thus, one explanation for the lack of Lyα detections in our sample is a significant increase in the H i fraction of the IGM, $\chi _{\rm H\,{\mathsc {i}}}$, at z ∼ 7, leading to a stronger absorption of the Lyα flux. A similar effect could explain the observed decrease in the number of Lyα emitters at z ≃ 7 (Ota et al. 2010), but results from these surveys are still contradictory (Ouchi et al. 2010; Hu et al. 2010). Detailed simulations (Dayal et al. 2010) show that the IGM absorption increases dramatically when the universe is not fully ionized, leading to a significant absorption of the emerging Lyα. The timescale of this effect around star-forming galaxies is of the order of 100 Myr (Dayal et al. 2010), shorter than the interval of cosmic time between z ≃ 6 and z ≃ 7. An additional prediction is that the asymmetry in the line profile is smoothed by the velocity structure of the infalling IGM (Dayal et al. 2008). Unfortunately, the modest S/N in our only detection is too low to address this effect quantitatively.

In conclusion, this work shows that the spectroscopic confirmation of z ≃ 7 galaxy candidates is a challenging effort. However, these difficulties may not be only due to our current technological limits, but may also reflect the long-sought first evidence of the reionization process in the early universe. Future surveys will definitely solve this fascinating puzzle.

Observations were carried out using the Very Large Telescope at the ESO Paranal Observatory under Programme IDs 084.A-095, 181.A-0717. We thank an anonymous referee for precious comments. We are grateful to P. Møller and the whole ESO staff for their assistance during the execution of service observations. We acknowledge financial contribution from the agreement ASI-INAF 1/009/10/0.