HIGH-REDSHIFT METALS. II. PROBING REIONIZATION GALAXIES WITH LOW-IONIZATION ABSORPTION LINES AT REDSHIFT SIX^*$^,$^†

The universe at redshift six remains one of the most challenging observational frontiers. The high opacity of the Lyα forest inhibits detailed measurements of the intergalactic medium (IGM), which hosts the vast majority of baryons in the early universe. At the same time, extreme luminosity distances make it difficult to assemble large samples of galaxies or to study individual objects in detail. Despite these challenges, however, significant observational progress is being made at z ∼ 6. Statistics of the scarce transmitted flux in the Lyα forest are being combined with insights from quasar proximity zones to learn about the very high redshift IGM (e.g., Fan et al. 2006; Lidz et al. 2006; Becker et al. 2007; Gallerani et al. 2008; Maselli et al. 2009; Bolton et al. 2010; Calverley et al. 2011; Carilli et al. 2010; Mesinger 2010). Deep galaxy surveys from the ground and from space are also beginning to yield considerable information on the properties of galaxies at z ∼ 6 and beyond (e.g., Bouwens et al. 2007, 2010; Stanway et al. 2007; Martin et al. 2008; Ouchi et al. 2008, 2009; Bunker et al. 2010; Hu et al. 2010; Labbé et al. 2010; McLure et al. 2010; Oesch et al. 2010; Stark et al. 2010). A more complete understanding of the universe at such early times, however, will require insight from a wide variety of observations.

Metal absorption lines offer a unique probe into the high-redshift universe. Transitions with rest wavelengths longer than H i Lyα will remain visible in QSO and gamma-ray burst spectra even when the forest becomes saturated. While absorption line studies at z ∼ 6 are still in their early days, metal lines are expected to provide insight into past and ongoing star formation, the action of galactic winds, as well as conditions in the interstellar media of galaxies, just as they do at lower redshifts (e.g., Songaila 2001; Simcoe et al. 2002; Schaye et al. 2003; Adelberger et al. 2005; Wolfe et al. 2005; Scannapieco et al. 2006; Fox et al. 2007). At high redshifts, the role of metal absorption lines in studying galaxy evolution becomes increasingly significant as galaxies become more difficult to study in emission.

Metal lines play another unique role at z ≳ 6 as potential probes of hydrogen reionization. In a scenario where galaxies provide most of the ionizing photons, dense regions of the IGM may become chemically enriched early on, yet remain significantly neutral until the end of reionization due to the short recombination times at high densities. High-density peaks are commonly predicted to host the last pockets of neutral gas following the growth and overlap of large H ii regions (Miralda-Escudé et al. 2000; Ciardi et al. 2003; Furlanetto & Oh 2005; Gnedin & Fan 2006). If these regions are metal-enriched, then forests of low-ionization absorption lines such as O i, Si ii, and C ii may be observable during this last phase of reionization (Oh 2002; Furlanetto & Loeb 2003).

We have conducted a two-part survey for metals out to z ∼ 6. In the first paper of this series (Becker et al. 2009, hereafter Paper I) we used high-resolution, near-infrared spectra to search for C iv over z = 5.3–6.0. C iv is a widely used tracer of highly ionized metals in the IGM at z < 5 (e.g., Songaila 2001; D'Odorico et al. 2010), and initial studies had indicated that C iv may remain abundant out to z ∼ 6 (Simcoe 2006; Ryan-Weber et al. 2006). In contrast, the longer path length and greater sensitivity of our data allowed us to demonstrate that the number density of C iv systems declines by at least a factor of four from z ∼ 4.5 to z > 5.3. This downturn was also seen by Ryan-Weber et al. (2009) using a larger sample of low-resolution spectra. Since C iv is expected to be an effective tracer of metals in the IGM at z ∼ 5–6 (Oppenheimer & Davé 2006; Oppenheimer et al. 2009), the decline suggests that there are fewer metals in the IGM toward higher redshifts, consistent with gradual enrichment of the IGM by star-forming galaxies. The apparent rapid evolution of C iv near z ∼ 5 may also point to changes in the ionization state of the metal-enriched gas, perhaps due to a hardening of the UV background associated with the beginning of He ii reionization (Madau & Haardt 2009).

In this paper, we present a complementary search for low-ionization metal absorbers over 5.3 < z < 6.4. This is an extension of an earlier work (Becker et al. 2006), in which we used high-resolution spectra of nine QSOs spanning 4.9 < z_em < 6.4, including five at z_em > 5.8, to search for "O i absorbers," so named due to the presence of O i λ1302. Although we identified a possible excess of these systems at z ≳ 6, the significance of that result remained unclear due to the relatively short survey path length and the fact that the z ∼ 6 systems occurred almost entirely along a single sight line. Here we increase our sample of z_em ∼ 6 QSOs to 17. The expanded survey offers us the opportunity to gather a more representative sample of low-ionization systems, providing insight into early galaxies and possibly the tail end of hydrogen reionization.

The remainder of the paper is organized as follows. The data are described in Section 2. We present our detections and the measurements for individual systems in Section 3. In Section 4 we analyze a number of ensemble properties, including the number density, ionic mass densities, relative abundance patterns, lack of strong high-ionization lines, and velocity width distribution. We also present evidence that the mass–metallicity relation of the host galaxies of low-ionization absorbers evolves out to z ∼ 6. Possible causes for the apparent lack of evolution in the number density of low-ionization absorbers are discussed in Section 5, where we also consider the likely hosts of these systems. Finally, we summarize our conclusions in Section 6. Throughout this paper we assume (Ω_m, Ω_Λ, h) = (0.274, 0.726, 0.705) (Komatsu et al. 2009).

We have obtained high- or moderate-resolution spectra of 17 QSOs with emission redshifts z_em = 5.8–6.4. This sample contains more than three times the number of z ⩾ 5.8 QSO sight lines than were included in Becker et al. (2006). Nine of our objects (generally the brightest) were observed at high resolution with either the High Resolution Echelle Spectrometer (HIRES; Vogt et al. 1994) on Keck I or the Magellan Inamori Kyocera Echelle (MIKE) spectrograph (Bernstein et al. 2002) on Magellan II (clay). This sample includes the five z ⩾ 5.8 QSOs from Becker et al. (2006), although additional data have since been acquired for some of these. We complement our high-resolution sample with a set of eight QSO spectra taken at moderate resolution with the Echelle Spectrograph and Imager (ESI; Sheinis et al. 2002) on Keck II. The data are summarized in Table 1.

All data were reduced using custom sets of IDL routines that included optimal sky-subtraction algorithms (Kelson 2003). In order to efficiently reject cosmic rays and bad pixels, a single one-dimensional spectrum was optimally extracted (Horne 1986) from all two-dimensional sky-subtracted frames for a given object simultaneously. Prior to extraction, relative flux calibration across the orders was performed using response functions generated from standard stars. Continua redward of the Lyα forest were fit by hand using a cubic spline.

The HIRES spectra were taken using the upgraded detector. Two grating settings with the red cross disperser provided continuous spectral coverage up to nearly 1 μm. The spectra were taken using a 086 slit, which provides a resolution FWHM of 6.7 km s⁻¹. The final one-dimensional spectra were binned using 2.1 km s⁻¹ pixels. Telluric absorption corrections based on a standard star were also applied to the individual two-dimensional frames prior to final extraction. For objects observed at multiple times during the year, changes in heliocentric velocity allowed telluric features to be efficiently excluded from the final spectra. Data from Becker et al. (2006) were re-reduced to take advantage of the improvements in the reduction pipeline. Only data taken with the upgraded detector were included.

MIKE spectra were taken using a single setting that gave continuous coverage up to 9400 Å. We used a 100 slit, which gives 13.6 km s⁻¹ resolution in the red. The one-dimensional spectra were binned using 5.0 km s⁻¹ pixels. Telluric corrections were performed after extracting the one-dimensional spectra using a high-resolution, high signal-to-noise transmission spectrum derived from solar observations (Hinkle et al. 2003), which was shifted, smoothed, and scaled in optical depth to match the QSO spectra.

ESI provides complete wavelength coverage up to 1 μm in a single setting. We used both the 075 and 100 slits, depending on the seeing. Our combined resolution was 50 km s⁻¹and was roughly consistent across all objects. The one-dimensional spectra were binned using 15.0 km s⁻¹ pixels. Telluric corrections were performed after extraction using the transmission spectrum from Hinkle et al. (2003).

Finally, we include Keck NIRSPEC echelle spectra (FWHM = 23 km s⁻¹) covering high-ionization lines (Si iv and C iv) for SDSS J0818+1722 and SDSS J1148+5251. These data are described in Paper I.

At z < 5, low-ionization absorption systems are typically identified by their strong H i absorption, particularly the extended damping wings of damped Lyα systems (DLAs; e.g., Wolfe et al. 2005). At higher redshifts, however, absorption features in the Lyα forest become sufficiently blended that identifying individual absorbers becomes difficult. To identify low-ionization systems, we therefore searched redward of the forest for groups of metal lines including Si ii λ1260, Si ii λ1304, O i λ1302, and C ii λ1334. A detection required that two or more lines occur at the same redshift. The lines were required to have similar velocity profiles, but the relative strengths of the different elements were not constrained. For each sight line we searched between the QSO redshift and the redshift at which O i enters the Lyα forest. The minimum and maximum redshifts covered are 5.33 and 6.42.

In total we detected ten systems, five of which were reported in Becker et al. (2006). Each contains Si ii and C ii, and all but one contains detectable O i. Seven of the systems occur along sight lines for which we have high-resolution data. The remainder fall toward QSOs we observed with ESI. Five of the seven systems in the high-resolution sample were reported in Becker et al. (2006), while the two additional high-resolution systems occur along a single sight line, SDSS J0818+1722. These two systems were also noted by D'Odorico et al. (2011). The three systems in the ESI data occur along separate sight lines.

The systems are shown in Figures 1–10. In addition to the low-ionization lines (Si ii, O i, and C ii), we plot the wavelength regions covering Si iv and C iv in the cases where we have data. Si iv may fall either in the optical or near-infrared, depending on the redshift of the system. For C iv, our coverage comes entirely from our NIRSPEC data (Paper I). The systems from Becker et al. (2006) are reproduced primarily to demonstrate the apparent lack of high-ionization absorption lines, a point to which we will return later. In the next sections we describe our basic measurements and completeness estimates.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

3.1. Equivalent Widths

We measured the rest-frame equivalent widths, W₀, for each available transition. The results are given in Table 2. For detected low-ionization lines, the wavelength intervals over which the equivalent widths were measured are shown as shaded regions in Figures 1–10. For non-detected low-ionization lines, upper limits are measured over the same intervals. We define our upper limits to be the sum of any net absorption, as in the case of blends with unrelated lines, and the uncertainty in the normalized flux integrated over the chosen window.

Table 2. Equivalent Widths

QSO	zabsa	Si ii λ1260	Si ii λ1304	O i λ1302	C ii λ1334	Si iv λ1394b	C iv λ1548c
SDSS J1630+4012	5.8865	0.235	<0.076	<0.031	0.240	<0.362	⋅⋅⋅
SDSS J2054−0005	5.9776	0.198	0.072	0.124	0.221	<0.096	⋅⋅⋅
SDSS J2315−0023	5.7529	⋅⋅⋅	<0.195	0.238	0.240	<0.069	⋅⋅⋅
SDSS J0818+1722	5.7911	0.267	0.056	0.182	0.230	<0.105	<0.072
SDSS J0818+1722	5.8765	0.064	<0.008	0.058	0.056	<0.044	<0.038
SDSS J1623+3112	5.8415	⋅⋅⋅	0.135	0.391	0.278	<0.060	⋅⋅⋅
SDSS J1148+5251	6.0115	⋅⋅⋅	0.038	0.162	0.154	<0.063	<0.160
SDSS J1148+5251	6.1312	⋅⋅⋅	0.016	0.079	0.067	<0.015	<0.040
SDSS J1148+5251	6.1988	0.020	<0.028	0.020	0.015	<0.072	<0.113
SDSS J1148+5251	6.2575	0.045	<0.008	0.036	0.047	<0.030	⋅⋅⋅

Notes. Rest-frame equivalent widths are given in angstroms. Upper limits are 1σ. ^aMinor differences from redshifts given in Becker et al. (2006) are due to an improvement in the HIRES wavelength solution. ^bThis gives the smaller of the upper limit on Si iv λ1394 and twice the upper limit on Si iv λ1403. ^cThis gives the smaller of the upper limit on C iv λ1548 and twice the upper limit on C iv λ1551.

Download table as:  ASCIITypeset image

Table 3. Column Densities for High-resolution Systems

QSO	zabs	$\log {N_{{\rm Si \, \mathsc{ii}}}}$	$\log {N_{{\rm O \, \mathsc{i}}}}$	$\log {N_{{\rm C \, \mathsc{ii}}}}$	$\log {N_{{\rm Si \, \mathsc{iv}}}}$	$\log {N_{{\rm C \, \mathsc{iv}}}}$
SDSS J0818+1722	5.7911	13.41 ± 0.03	14.54 ± 0.03	14.19 ± 0.03	<12.9	<13.4
SDSS J0818+1722	5.8765	12.78 ± 0.05	14.04 ± 0.05	13.62 ± 0.07	<12.6	<13.0
SDSS J1623+3112	5.8415	14.09 ± 0.02	>15.0	>14.3	<13.0	⋅⋅⋅
SDSS J1148+5251	6.0115	13.51 ± 0.03	14.65 ± 0.02	14.14 ± 0.06	<12.7	<13.8
SDSS J1148+5251	6.1312	13.29 ± 0.06	14.79 ± 0.23	13.88 ± 0.15	<12.3	<13.2
SDSS J1148+5251	6.1988	12.12 ± 0.05	13.49 ± 0.13	12.90 ± 0.11	<12.9	<13.6
SDSS J1148+5251	6.2575	12.92 ± 0.08	14.12 ± 0.15	13.78 ± 0.21	<12.5	⋅⋅⋅

Notes. Column densities were computed using the apparent optical depth method in all cases except the z = 6.1312 and 6.2575 systems toward SDSS J1148+5251. For these system, column densities were measured using Voigt profile fits, for which Doppler parameters of b = 5.7 ± 1.0 km s⁻¹ and 3.8 ± 0.3 km s⁻¹, respectively, were returned.

Download table as:  ASCIITypeset image

High-ionization lines are often kinematically distinct from low-ionization lines in lower-redshift DLAs and sub-DLAs (e.g., Fox et al. 2007). We therefore place conservative upper limits on the equivalent widths of Si iv and C iv by integrating over a ±100 km s⁻¹ window around the nominal system redshift. We measure limits for both lines in the doublet separately, but take advantage of the fact that W₀ for the stronger transition should be no more than twice that of the weaker transition. For Si iv, the smaller of $W_{0}({\mbox{Si\,{\sc {iv}}}}\,\lambda 1394)$ and $2W_{0}({\mbox{Si\,{\sc {iv}}}}\,\lambda 1403)$ is given in Table 2. For C iv, we give the smaller of $W_{0}({\mbox{C\,{\sc {iv}}}}\,\lambda 1548)$ and $2W_{0}({\mbox{C\,{\sc {iv}}}}\,\lambda 1551)$.

3.2. Column Densities

Column densities of each ion were measured for systems in the high-resolution sample, where the velocity structure is most likely to be resolved. For five out of seven of these, column densities were measured using the apparent optical depth method (Savage & Sembach 1991). Here the optical depths of individual pixels, τ_i, are measured, and the total column density in cm⁻² is given by

$$\begin{equation} N = \frac{3.768 \times 10^{14}}{f\,\lambda _{0}} \sum _{i}{\tau _{i} \delta v_{i}} \,, \end{equation} \tag{ 1 }$$

where f is the oscillator strength, λ₀ is the rest wavelength in angstroms, and δv_i is the velocity width of a pixel in km s⁻¹. This approach has the advantage that it does not require the absorption to be decomposed into individual components. We integrated the optical depths over the shaded regions in Figures 1–7. Prior to measuring the low-ionization lines, we smoothed our spectra by convolving the flux with a Gaussian kernel whose FWHM is equal to one-half of the instrumental resolution. This takes advantage of the fact that the spectra are known to have a finite resolution, which is roughly three pixels for our choice of binning. Mild smoothing prevents individual noisy pixels with flux near or below zero from dominating the summed optical depths, while only slightly degrading the spectral resolution (here by ∼12%). In the cases where the spectra had high signal-to-noise ratio (S/N), or for transitions which contained no pixels near zero, smoothing had a negligible effect on the measured column density. We note, however, that smoothing should be used cautiously where the flux may genuinely go to zero. For example, in the z = 6.1312 absorber toward SDSS J1148+5251, O i λ1302 is marginally saturated. We therefore fit Voigt profiles rather than apply the apparent optical depth method in this case. This system appears to have a single, narrow component, and so by requiring that each transition have the same redshift and Doppler parameter (i.e., assuming turbulent broadening), we can obtain a rough column density estimate for O i even though the central two pixels show nearly zero flux. Voigt profile fitting was also used for the z = 6.2575 system toward SDSS J1148+5251, which appears to have a single, narrow component.

Upper limits on the column densities were computed for non-detected species. For low-ionization lines, the optical depths were integrated over the same velocity interval where other low-ionization species were detected. Limits on C iv and Si iv were set by integrating the apparent optical depths over an interval −100 km s⁻¹ to +100 km s⁻¹ from the optical depth-weighted mean redshift of the low-ionization lines. For the high-ionization doublets, the column density in the ith velocity bin was taken to be the inverse variance-weighted mean of the column densities measured from both transitions in the cases where the measurements were consistent within the errors. In the cases where they were not consistent, as would be expected if one transition was blended with an unrelated line, the smaller of the two values was taken.

We used our optical depth measurements of Si ii to infer another quantity used to classify low-ionization systems, the equivalent width of Si ii λ1526, W₁₅₂₆ (Prochaska et al. 2008). Since we do not cover Si ii λ1526 with our current data, the line profile for this transition was reconstructed from the optical depths of Si ii λ1260 and/or λ1304. In all cases where this was done, at least one of these transitions was covered and not saturated. Since the optical depth should scale linearly with the oscillator strength, we therefore expect our estimates of W₁₅₂₆ to be reasonably accurate in all cases. The results are given in Table 4.

Table 4. Line Widths for High-resolution Systems

QSO	zabs	Δv90a	Transitionb	W1526c
		(km s−1)		(Å)
SDSS J0818+1722	5.7911	134	C ii λ1334	0.066
SDSS J0818+1722	5.8765	25	O i λ1302	0.016
SDSS J1623+3112	5.8415	223	Si ii λ1304	0.249
SDSS J1148+5251	6.0115	41	O i λ1302	0.074
SDSS J1148+5251	6.1312	17d	C ii λ1334	0.028
SDSS J1148+5251	6.1988	17	Si ii λ1260	0.004
SDSS J1148+5251	6.2575	13	Si ii λ1260	0.012

Notes. ^aVelocity width spanning 90% of the optical depth in the low-ionization lines. ^bTransition used to measure Δv_90. ^cRest equivalent width of Si ii λ1526 estimated from optical depth measurements of available Si ii lines. ^dMeasured from Voigt profile fit.

Download table as:  ASCIITypeset image

3.3. Velocity Widths

3.4. Completeness

We estimated our completeness by attempting to recover artificial absorption systems inserted into the data. The synthetic lines were modeled as Voigt profiles with a Doppler parameter of 10 km s⁻¹ (Δv₉₀ = 23 km s⁻¹). This choice focuses on the detectability of the narrower, weaker lines in our sample. Artificial systems containing O i, C ii, and Si ii were placed at random into the data, assuming column density ratios equal to the mean values measured in the data (Section 4.3). We then used an automatic routine to determine whether these systems would be detected. The detection criteria were tuned to mimic the results of a by-eye identification. A detection first required that a peak flux decrement greater than 1σ be present at the same redshift for at least two of Si ii λ1260, O i λ1302, and C ii λ1334 after the flux was smoothed by convolving with a Gaussian kernel of FWHM equal to the instrumental resolution. For candidate pairs, the unsmoothed data were then fit using Gaussian profiles. A detection required that the amplitude and FWHM for both lines were constrained at greater than 3σ confidence, and that the measured redshifts and velocity widths agreed with high confidence. The ratio of the maximum optical depths in the two lines was also required to be within a factor of 0.4 to 2.0 of the ratio expected from the assumed relative abundances.

Each line of sight was tested with 10⁴ artificial systems over a range of column densities and covering the entire redshift interval used to search for low-ionization systems. The path-length-weighted mean results for the full sample are shown in Figure 11, along with separate results for the high-resolution and ESI data. We estimate that we are 50% complete at $\log {N(\mbox{C\,{\sc {ii}}})} \simeq 13.3$ in the high-resolution data and $\log {N(\mbox{C\,{\sc {ii}}})} \simeq 13.7$ in the ESI data. Note that the completeness estimates do not go to 100% for large column densities due to the fact that parts of the spectra are unusable due to strong atmospheric absorption or the presence of other absorption lines. This is particularly an issue at redshifts where Si ii λ1260 is in the Lyα forest, and the system must be identified from O i λ1302 and C ii λ1334 alone. Realistically, however, strong systems tend to have multiple components, making them easier to detect, and often have significant Si ii λ1304 absorption.

Zoom In Zoom Out Reset image size

4.1. Number Density

We now address the number density of low-ionization absorption systems at z ∼ 6. We compute the number density of absorbers per unit absorption path length interval, $\ell (X) \equiv d\mathcal {N}/dX$. Here, X is defined as

$$\begin{equation} X(z) = \int _{0}^{z} (1+z^{\prime })^2 \frac{H_{0}}{H(z^{\prime })} dz^{\prime } \,, \end{equation} \tag{ 2 }$$

where H(z) is the Hubble constant at a given redshift (Bahcall & Peebles 1969). A non-evolving population of absorbers will have a constant number density per unit X. Our total path length is ΔX = 39.5 (Δz = 8.0, or Δl = 3.6 comoving Gpc). The high-resolution data cover ΔX = 21.0, while the ESI data cover ΔX = 18.5. Uncorrected for completeness, the line-of-sight number density in the entire sample is ℓ(X) = 0.25^+0.21_{− 0.13} (95% confidence), where the error bars include Poisson uncertainties only.

Corrections to ℓ(X) for completeness are small compared to the errors, except when considering our weakest system. The z = 6.1988 system toward SDSS J1148+5251, which has $\log {N({\mbox{C\,{\sc {ii}}}})} = 12.9$, would have been detected along only 9% of our path length. Few systems that are similar are known at lower redshifts. The only comparable system of which we are aware is the sub-DLA at z = 3.7 found by Péroux et al. (2007), which has $\log {N({\mbox{C\,{\sc {ii}}}})} = 13.1$ yet still contains O i. It is intriguing to consider whether more weak systems may be found at z ∼ 6 in future, more sensitive surveys.

Seven out of our ten systems are detected in the high-resolution data, which suggests that the two lines of sight with multiple systems may be unusual. At a given S/N, however, HIRES and MIKE are more sensitive to narrow absorption systems than ESI, which has considerably lower resolution. In order to determine whether the two data sets are consistent, we first take the ratio of the completeness estimates in Figure 11 at the column densities of the seven systems in our high-resolution sample. Significantly, the weakest system ($\log {N({\mbox{C\,{\sc {ii}}}})} = 12.90$) would be nearly undetectable in the ESI data, while the completeness of the ESI data for the two systems with $\log {N({\mbox{C\,{\sc {ii}}}})} \simeq 13.6$–13.8 would be roughly 60% that of the high-resolution data. If we also take into account the slightly shorter path length of the ESI data, and the fact that the ESI data are somewhat less sensitive to even the stronger systems, then we would expect to detect around 4 systems in the ESI data if the high-resolution systems represent a fair sample. The fact that we detected three systems with ESI suggests that the two data sets are broadly consistent.

It is not entirely straightforward to compare the number density of low-ionization absorption systems at z ∼ 6 to lower redshifts. At z < 5, low-ionization systems are generally identified by their strong H i absorption (e.g., DLAs). A systematic search for low-ionization lines independent of H i column density, as we are forced to do here, has not been performed at lower redshifts. We can estimate the number density of lower-redshift low-ionization systems, however, by identifying the H i column density where low-ionization lines, particularly O i, start to appear. O i is ubiquitous among DLAs ($\log {N({\mbox{H\,{\sc {i}}}})} \ge 20.3$; e.g., Wolfe et al. 2005) and is also present in sub-DLAs down to $\log {N({\mbox{H\,{\sc {i}}}})} \simeq 19.0$ (Dessauges-Zavadsky et al. 2003; Péroux et al. 2007). Low-ionization lines are also seen in lower column density systems. However, these tend to show very strong Si iv and C iv absorption relative to Si ii and C ii (e.g., Prochaska & Burles 1999; Prochter et al. 2010), which is not seen in our systems (except, perhaps, for the z = 5.8865 system toward SDSS J1630+4012, which may contain strong Si iv). We therefore conclude that our z ∼ 6 low-ionization systems are most naturally compared to lower-redshift DLAs and sub-DLAs with $\log {N({\mbox{H\,{\sc {i}}}})} \ge 19.0$. A more objective comparison must await either a blind survey for low-ionization metal lines at lower redshifts, or a more complete characterization of the metal line properties of sub-DLAs and lower column density systems. In the following sections we generally compare our z ∼ 6 systems to lower-redshift sub-DLAs, since they are more common than DLAs and therefore more likely to make up the majority of our high-redshift sample.

The number density of DLAs has been well established using the large path length available from the Sloan Digital Sky Survey (Prochaska & Wolfe 2009; Noterdaeme et al. 2009). Surveys using higher-resolution spectra have meanwhile been used to determine the number density of sub-DLAs (Péroux et al. 2005; O'Meara et al. 2007; Guimarães et al. 2009). The net result is that the total number of systems with $\log {N({\mbox{H\,{\sc {i}}}})} \ge 19.0$ is ℓ(X) ≃ 0.3 over 3 < z < 5. This is very similar to the number density of low-ionization systems we find over 5.3 < z < 6.4, which suggests that the number density of low-ionization systems may be roughly constant over 3 < z < 6. This contrasts to the total number density of optically thick Lyman limits systems ($\log {N({\mbox{H\,{\sc {i}}}})} \ge 17.2$), whose number density per unit X increases by a factor of ∼ 2 from z ∼ 3 to 6 (Songaila & Cowie 2010). It also contrasts to the number density of highly ionized metal absorption systems seen in C iv, which declines by at least a factor of four from z ∼ 3 to 6 (Paper I).

4.1.1. Possible Redshift Evolution

A schematic representation of the redshifts of the detected systems is shown in Figure 12. We also plot the cumulative probability density function of the survey path length, ΔX(z) (not adjusted for completeness), and for the redshifts of the low-ionization systems. One peculiar feature is that all ten of our systems lie at z ⩾ 5.75, even though nearly 40% of our path length is at lower redshifts. We performed a Kolmogorov–Smirnov (K-S) test to determine whether the distribution of absorption redshifts is consistent with our path-length-weighted survey redshift distribution function. The test returns a D statistic of 0.39, which would be larger in a random sample of ten redshifts drawn from our survey path only 7% of the time. The distribution appears more unusual if we consider that, for a uniform distribution, the probability that all ten systems would randomly occur at z > 5.75 is only 0.7%. This ignores clustering, however. If we count only one of the systems toward both SDSS J1148+5251 and SDSS J0818+1722 (i.e., seven "groups" of systems, the maximum effect of clustering), then the probability increases to 3%. We note that completeness should not be a significant factor, as our completeness is somewhat better below z = 5.75 than above. Thus, there is evidence that the number density of low-ionization systems may be increasing with redshift over z ∼ 5 to 6. An expanded search at z < 5.7 would help to clarify this.

Zoom In Zoom Out Reset image size

4.2. Mass Density

The comoving mass density of various ions can be characterized as a fraction of the critical density at z = 0, ρ_crit, as

$$\begin{equation} \Omega _{\rm ion} = \frac{H_0\,m_{\rm ion}}{c\,\rho _{\rm crit}} \int {N_{\rm ion} \, f(N_{\rm ion},X) \, {\it dN}_{\rm ion}} \,, \end{equation} \tag{ 3 }$$

where $f(N_{\rm ion},X) \equiv \partial ^2{\mathcal N}/\partial N_{\rm ion}\partial X$. The integral in Equation (3) is often approximated using the observed column densities within a sample as

$$\begin{equation} \int {N_{\rm ion} \, f(N_{\rm ion},X) \, {\it dN}_{\rm ion}} \simeq \frac{\sum {N^{\rm obs}_{\rm ion}}}{\Delta X} \,. \end{equation} \tag{ 4 }$$

The validity of this approximation, however, is highly model dependent. A power law f(N_ion, X), for example, may hide most of the mass in rare, high column density systems. It is unclear what shape the distribution of low ion column densities should have, as it will depend on the underlying hydrogen column density distribution, the distribution of metallicities, and ionization corrections that are potentially important for sub-DLAs or somewhat weaker systems. The distribution may be log-normal, as there are several multiplicative factors at work. The present sample is too small, however, to characterize the distribution with any confidence. Given that much of the metal mass in the low-ionization phase may reside in rare systems that have both high H i column densities and higher than average metallicities, a conservative approach is to use the present sample to set lower limits on Ω_ion using the approximation in Equation (4). The values for Si ii, O i, and C ii are given in Table 5. We also convert these values into minimum mass-averaged global metallicities by dividing the ionic densities by the total mass density of hydrogen at z = 6, assuming Asplund et al. (2009) photospheric solar abundances. Notably, $\Omega ({\mbox{C\,{\sc {ii}}}})$ is twice as high as $\Omega ({\mbox{C\,{\sc {iv}}}})$ measured by Ryan-Weber et al. (2009), although both were derived from small samples of lines. The true values of Ω_ion may be factors of several higher in both cases.

Table 5. Mass Densities over 5.3 < z < 6.4

Ion	Ωion	[M/H]global
Si ii	⩾0.4 × 10−8	⩾ − 3.9
O i	⩾3.7 × 10−8	⩾ − 3.9
C ii	⩾0.9 × 10−8	⩾ − 4.1

Download table as:  ASCIITypeset image

4.3. Relative Abundances

4.4. High-ionization Lines

A notable feature of our low-ionization systems is the lack of strong absorption by high-ionization species (Si iv and C iv). At z ∼ 6, Si iv and C iv are shifted into the far-red and near-infrared, and we stress that our spectra at these wavelengths are generally of modest quality. In most cases, however, the high-ionization lines must be significantly weaker than the low-ionization lines.

Significant C iv absorption is seen for nearly all lower-redshift DLAs and sub-DLAs. It is kinematically distinct from the low-ionization lines (Wolfe & Prochaska 2000) and has been interpreted as evidence for galactic winds (Fox et al. 2007). We therefore investigate whether the non-detections of Si iv and C iv infer that the host galaxies of our z ∼ 6 systems lack strong winds or other nearby, highly ionized gas. The column densities of high-ionization lines are plotted verses the column densities of low-ionization lines in Figure 13 for our z ∼ 6 systems and a sample of DLAs and sub-DLAs at z < 4.2 drawn from the literature (Dessauges-Zavadsky et al. 2003; Péroux et al. 2007; Fox et al. 2007). For the literature data, direct measurements of $N({\mbox{Si\,{\sc {ii}}}})$ are used when available. Otherwise, $N({\mbox{Si\,{\sc {ii}}}})$ was estimated from the H i column density and metallicity measurements, assuming solar abundances and that all of the silicon in the low-ionization phase is in Si ii (i.e., no ionization corrections). For the 19 sub-DLAs with Si ii measurements, which had $19.1 < \log {N({\mbox{H\,{\sc {i}}}})} < 20.2$, this procedure reproduced the published $N({\mbox{Si\,{\sc {ii}}}})$ to within 0.03 dex in all cases. This is not surprising, since the metallicity estimates were often based on Si ii. Column densities were estimated for C ii using the same procedure. For one sub-DLA, a limit on C iv that was found to be unusually low was re-estimated from the published spectrum. Otherwise, the Si iv and C iv measurements are as originally reported.

Zoom In Zoom Out Reset image size

The systems at z ∼ 6 tend to show lower high-ionization column densities than DLAs and sub-DLAs at lower redshifts. However, the upper limits on Si iv and C iv are consistent with a trend of declining high-ionization line strength with declining low-ionization column density. This is reminiscent of the correlation between C iv strength and the metallicity of the neutral phase identified by Fox et al. (2007) for DLAs and sub-DLAs at z < 3.6. The trend suggests that the z ∼ 6 systems may lack strong high-ionization lines because they have overall low metallicities. Highly ionized envelopes or winds may still be present, but with too few metals to produce absorption that can be detected in present-quality data.

Ryan-Weber et al. (2009) and D'Odorico et al. (2011) noted a possible C iv doublet at z ≃ 5.79 toward SDSS J0818+1722, which would coincide with one of our low-ionization systems. There is possible C iv absorption in our NIRSPEC data near Δv = −100 km s⁻¹, where there is also Si ii and C ii absorption but no O i. The lack of O i suggests that the gas at this velocity may be ionized, and so some C iv might be expected. Confirmation must await higher-quality data.

4.5. Velocity Width Distribution

The absorbers in our sample show a range of velocity profiles, from single, narrow lines to broad, multi-component systems. The distribution of Δv₉₀ values among the seven systems in our high-resolution sample is plotted in Figure 14. We also show the distribution of Δv₉₀ for 28 sub-DLAs in the literature (Dessauges-Zavadsky et al. 2003; Péroux et al. 2007; Fox et al. 2007). For the systems in Dessauges-Zavadsky et al. (2003) and Péroux et al. (2007), the velocity widths were calculated from published Voigt profile fits. There is some indication that the low-ionization systems at z ∼ 6 are more likely to be narrow. A K-S test, however, indicates that there is a reasonable likelihood (27%) that the two samples were drawn from the same parent distribution.

Zoom In Zoom Out Reset image size

4.6. Mass–Metallicity Relation

The high opacity of the Lyα forest at z ∼ 6 prevents us from measuring H i column densities for these systems. We are therefore unable to obtain metallicities directly. Previous works have noted correlations between metallicity and either the velocity width (Ledoux et al. 2006) or the rest equivalent width of Si ii λ1526 (W₁₅₂₆; Prochaska et al. 2008). These relationships have only been calibrated for DLAs, however, while our systems potentially have lower H i column densities. Some insight can be gained, however, by comparing the relationship between Δv₉₀ and W₁₅₂₆ in our study to systems at lower redshifts.

Figure 15 shows the relationship between Δv₉₀ and W₁₅₂₆ for DLAs and sub-DLAs at z < 4.5 and for our low-ionization systems at z ∼ 6. The DLA data are from Prochaska et al. (2008). Values of Δv₉₀ and W₁₅₂₆ for sub-DLAs were computed from published Voigt profile fits from Dessauges-Zavadsky et al. (2003) and Péroux et al. (2007). We also include the three metal-poor DLAs and one metal-poor sub-DLA from Pettini et al. (2008). The z ∼ 6 sample includes the seven systems in our high-resolution sample.

Zoom In Zoom Out Reset image size

Following Prochaska et al. (2008), we plot the expected relation for fully saturated absorption, W^sat₁₅₂₆ = (1526.7 Å)Δv₉₀/c, as a dashed line in each panel. We also show this relation scaled by a factor such that mean residual in log W₁₅₂₆ is zero for each data set. The DLAs roughly follow the relation expected for saturated absorption, while W₁₅₂₆ for the sub-DLAs is lower by 0.44 dex. For the z ∼ 6 systems, the offset is 0.81 dex or a factor of ∼2 lower than the sub-DLAs. At a given velocity width, therefore, the low-ionization systems have significantly weaker metal lines than either DLAs or sub-DLAs at z < 4.5.

We can translate the offset in W₁₅₂₆ into an approximate difference in metallicity. Prochaska et al. (2008) fit a relation for DLAs of the form [M/H] = a + blog W₁₅₂₆, finding (a, b)_DLA = (− 0.92, 1.41) with an rms scatter of 0.25 dex. We have fit a similar relation for the sub-DLAs using the published metallicities, finding (a, b)_sub-DLA = (− 0.28 ± 0.15, 1.19 ± 0.18), with a scatter of 0.33 dex. It is apparent, therefore, that the metallicity estimate will depend on the H i column density of the absorber. As argued in Section 4.1, we expect that most of our z ∼ 6 systems are DLAs or sub-DLAs. If this is correct, then the decline in W₁₅₂₆ suggests that the z ∼ 6 systems have metallicities at a given Δv₉₀ that are lower by ≳0.4 dex than those of their lower-redshift counterparts. Ionization effects could also potentially decrease W₁₅₂₆; however, the column density ratios of O i to Si ii and C ii, which have higher ionization potentials than O i, are similar in our z ∼ 6 systems to those of metal-poor DLAs, for which ionization corrections are negligible (e.g., Pettini et al. 2008). Together with the lack of high-ionization lines, this suggests that ionization effects are probably not large for our systems.

The velocity width is expected to correlate with the mass of the dark matter halo (e.g., Haehnelt et al. 1998; Maller et al. 2001; Prochaska et al. 2008; Pontzen et al. 2008). This decline in W₁₅₂₆ at a given Δv₉₀ can therefore be interpreted as evolution in the mass–metallicity relation for the host galaxies of low-ionization absorbers. Ledoux et al. (2006) noted a similar redshift evolution for DLAs in their [M/H]–Δv₉₀ relation, in which the metallicity at a given Δv₉₀ declined by ∼0.3 dex from 1.7 < z < 2.43 to 2.43 < z < 4.3. Our results indicate that this decline in the mass–metallicity relation potentially extends to galaxies near the reionization epoch.

We now consider whether the number density of low-ionization systems is expected to remain roughly constant with redshift, as observed. Previous works have shown that the number density (per unit X) of both DLAs and sub-DLAs shows little evolution over 3 ⩽ z ⩽ 5 (Péroux et al. 2005; Prochaska et al. 2005; Prochaska & Wolfe 2009; Noterdaeme et al. 2009; Guimarães et al. 2009). Our observations suggest that the number density remains roughly constant out to z ∼ 6, but why should this happen? The number density of a population of absorbers will depend on their comoving number density, ϕ_c, and their mean physical absorption cross section, 〈σ_abs〉, as

$$\begin{equation} \ell (X) = \frac{c}{H_0} \, \phi _{\rm c} \, \langle \sigma _{\rm abs} \rangle \,. \end{equation} \tag{ 5 }$$

Observational (e.g., Cooke et al. 2006a, 2006b) and theoretical (e.g., Gardner et al. 1997; Haehnelt et al. 1998; Nagamine et al. 2004) arguments generally suggest that DLAs are hosted by dark matter halos of mass M ≳ 10⁸–10⁹  M_☉, and may be largely associated with halos in the range 10⁹ < M/M_☉ < 10¹¹ (Pontzen et al. 2008). For a Sheth–Tormen halo mass function, ϕ_c for 10¹⁰  M_☉ halos, for example, will be lower at z = 6 than at z = 3 by a factor of ∼4 (Sheth et al. 2001; Reed et al. 2007). It is possible that the host halos of low-ionization systems are somewhat less massive at higher redshifts, which would increase their space density. The similarity of the velocity width distributions of the z ∼ 6 systems and low-redshift sub-DLAs suggests that the halo masses are similar, although the velocity width may only be moderately sensitive to the mass. For a Navarro–Frenk–White (NFW) dark matter profile (Navarro et al. 1997), the virial velocity scales with the halo mass as v_vir∝M^1/3_virH(z)^1/3, where H(z) is the Hubble parameter at redshift z. If Δv₉₀ scales with the virial velocity (e.g., Haehnelt et al. 1998), then a decrease in the typical halo mass by a factor of 10 from z = 3 to 6 would only produce a decline in the typical Δv₉₀ of ∼ 0.2 dex. Such a decline is consistent with the present data, but as noted in Section 4.5, the distribution of Δv₉₀ values for the z ∼ 6 sample is formally consistent with that of lower-redshift sub-DLAs. A larger sample is needed to determine whether there is a genuine decrease in the mean velocity width.

Multiple factors may increase the absorption cross section with redshift. At a given mass, an NFW profile will have a higher density, particularly in the inner regions. The external UV background also decreases from z ∼ 3 to 6 (Fan et al. 2006; Bolton & Haehnelt 2007; Wyithe & Bolton 2011; Calverley et al. 2011). Both of these factors should increase the radius out to which a halo of a given mass can become self-shielded.

We explored simple models using NFW profiles to predict how the cross section of low-ionization gas should evolve with redshift. Assuming that the gas traces the dark matter and is in photoionization equilibrium, it is straightforward to predict the radius at which the gas becomes self-shielded and to estimate the cross section over which the projected H i column density exceeds a given value. For this we assumed a gas temperature of 20,000 K and a metagalactic H i ionization rate Γ = 10^−11.9 s⁻¹ (10^−12.8 s⁻¹) at z = 3 (6). This simple model suggests that the cross section over which a halo of a given mass would be a DLA should scale roughly linearly with the halo mass, which agrees broadly with recent theoretical models (Nagamine et al. 2004; Pontzen et al. 2008; Tescari et al. 2009). It also predicts that the DLA cross section for a given mass halo should increase by a factor of ∼ 4 from z = 3 to 6. If DLAs are mainly hosted by dark matter halos with masses M ⩾ 10⁹ M_☉, then this increase in the cross section will offset the decline in the comoving spatial density to produce a roughly constant number density of DLAs with redshift. The same is true for sub-DLAs in this model. We stress that this is an extremely simple model, and it is unlikely to reproduce detailed observations such as the column density or velocity width distributions of DLAs. Nevertheless, it gives some assurance that the apparent lack of evolution in the number density of low-ionization absorbers can be plausibly explained by a decline in the spatial density of dark matter halos combined with an increase in the mean absorption cross section. More detailed analysis must be left for future work.

The high number density of low-ionization systems makes it likely that they arise from galaxies below the detection limits of current galaxy surveys. Integrating the z ∼ 6 luminosity function of i-dropout galaxies from Bouwens et al. (2007) down to the limit of the Hubble Ultra Deep Field (∼0.04 L*_{z = 3}) gives a space density ϕ(M_AB ⩽ −18) = 5 × 10⁻³ per comoving Mpc³. For ℓ(X) ≃ 0.25, this implies a mean absorption cross section 〈σ_abs〉 ≃ 1 × 10⁴ kpc⁻², or 〈R_abs〉 ≃ 60 kpc for a circular projected geometry. The Lyα-emitting galaxies (LAEs) at z ∼ 6 included in current surveys would need an even larger cross section. Ouchi et al. (2008) found a density of LAEs at z = 5.7 of ∼7 × 10⁻⁴ per Mpc⁻³ down to a Lyα luminosity of 2.5 × 10⁴² erg s⁻¹. These LAEs would require 〈R_abs〉 ≃ 160 kpc to account for the observed number density of low-ionization absorption systems. In contrast, constraints from paired QSO lines of sight suggest a typical DLA size at z = 1–2 of R_DLA ≃ 5–10 kpc (Monier et al. 2009; Cooke et al. 2010), although in some cases they may be larger (e.g., Briggs et al. 1989). The sizes of sub-DLAs are less well constrained, but they may plausibly be a factor of two larger than DLAs (see, for example, the scaling relation in Monier et al. 2009, or Figure 2 in Pontzen et al. 2008). In that case, they would still need to be significantly larger at z = 6 to arise solely from galaxies similar to those in current surveys. It seems more plausible, therefore, that these low-ionization systems trace fainter, more numerous sources. This would be consistent with the results of Rauch et al. (2008), who find a likely connection between DLAs and a population of ultra-faint LAEs at z ∼ 3.

We note that these absorption systems are potentially our first direct probe of the "typical" galaxies that are responsible for hydrogen reionization. It is widely believed that in order for reionization to complete by z = 6, the majority of ionizing photons must come from galaxies below the limits of current imaging surveys (e.g., Bouwens et al. 2007; Bolton & Haehnelt 2007; Ouchi et al. 2009; Bunker et al. 2010; McLure et al. 2010; Oesch et al. 2010). Metal absorption lines may therefore be the most efficient means of detecting these sources, as directly detecting them in emission must await facilities such as ALMA and JWST. This study already delivers insights into the properties of reionization galaxies, including the suggestion that they are likely to be more metal-poor than their lower-redshift counterparts, at least in the gas phase. If their stars also have low metallicities, then they may be highly efficient in producing ionizing photons (Schaerer 2003; Venkatesan & Truran 2003). As noted by others (e.g., Ouchi et al. 2009), this would potentially help to explain how reionization is completed by z ∼ 6. We add that, in addition to their role in reionization, these low-mass galaxies may contribute significantly to the chemical enrichment of the IGM (e.g., Booth et al. 2010).

We have conducted a search for low-ionization metal absorption systems spanning 5.3 < z < 6.4. Our survey includes high- and moderate-resolution spectra of 17 QSOs with emission redshifts z_em = 5.8–6.4. The total survey path length is ΔX = 39.5 (Δz = 8.0 or Δl = 3.6 comoving Gpc), of which roughly half is covered by high-resolution data. We searched for low-ionization systems by looking for coincidences in redshift between Si ii, C ii, and O i. In total we detect ten systems, five of which were previously reported by Becker et al. (2006). Each contains Si ii and C ii, and all but one contains O i. The majority are detected in our high-resolution data, which is consistent with the fact that these data are more sensitive to narrow absorption lines.

The line-of-sight number density of absorption systems, uncorrected for completeness, is ℓ(X) = 0.25^+0.21_{− 0.13} (95% confidence). This is similar to the number density over 3 < z < 5 of DLAs and sub-DLAs ($\log {N({\mbox{H\,{\sc {i}}}})} \ge 19.0$), which constitute the main population of low-ionization absorbers at those redshifts. The fact that the number density of low-ionization absorbers is roughly constant out to z ∼ 6 is in sharp contrast with the evolution of high-ionization systems traced by C iv, which show a marked decline at z > 5.3 (Paper I; Ryan-Weber et al. 2009). At z ∼ 6, low-ionization systems with $\log {N({\mbox{C\,{\sc {ii}}}})} \gtrsim 13$ are more common than high-ionization systems with $\log {N({\mbox{C\,{\sc {iv}}}})} \gtrsim 13$, a reversal from lower redshifts.

The roughly constant number density of low-ionization systems over 3 ≲ z ≲ 6 may be explained if they are hosted by lower-mass dark matter halos at higher redshifts. Alternatively, if the systems at z ∼ 6 are hosted by halos with masses similar to those that host DLAs and sub-DLAs at lower redshifts, then the apparent lack of evolution may occur if dark matter halos of a given mass have a larger mean cross section of low-ionization gas at higher redshifts due to the higher gas densities and weaker UV background.

Although the H i column densities cannot be measured for these systems, we are able to infer some of their properties by comparing the velocity widths and metal line strengths to samples at lower redshifts. For the seven systems with high-resolution data, the velocity widths span a similar range as sub-DLAs at 2 ≲ z ≲ 4, although there is some indication that the z ∼ 6 systems tend to be narrower. The lines in the z ∼ 6 systems are also weaker in the sense that, at a given velocity width, the inferred equivalent width of Si ii λ1526 is lower than for z ≲ 4 sub-DLAs by a factor of two, and by a factor of six compared to z ≲ 4 DLAs. This suggests that the mass–metallicity relation of the host galaxies evolves toward lower metallicities at higher redshifts, a trend that has also been noted over 2 ≲ z ≲ 4 (Ledoux et al. 2006). Assuming the z ∼ 6 systems span a similar range in $\log {N({\mbox{H\,{\sc {i}}}})}$ as the low-ionization systems at z < 4, the strength of the metal lines implies a decline in the gas-phase metallicity from 2 ≲ z ≲ 4 to z ∼ 6 of at least ∼ 0.4 dex.

The mean relative abundances are [Si/O] = 0.00 ± 0.03 and [C/O] = −0.19 ± 0.03 (1σ uncertainties), assuming no dust depletion and no ionization corrections. These are consistent with the values found for metal-poor DLAs by Pettini et al. (2008) and for low-metallicity halo stars by Akerman et al. (2004). As noted by Becker et al. (2006), the relative abundances are broadly consistent with enrichment from Type II supernovae of low-metallicity massive stars.

Our z ∼ 6 systems are also notable in that they lack strong high-ionization lines (Si iv and C iv), which are ubiquitous among lower-redshift DLAs and sub-DLAs (Fox et al. 2007). The absence of these lines is consistent with a similar fractional decline in the metallicity of the low- and high-ionization phases, however, and does not necessarily indicate that these systems lack halos of highly ionized gas. Deeper spectra will help to determine whether the z ∼ 6 absorbers have significantly lower $N({\mbox{Si\,{\sc {iv}}}})/N({\mbox{Si\,{\sc {ii}}}})$ and $N({\mbox{C\,{\sc {iv}}}})/N({\mbox{C\,{\sc {ii}}}})$ ratios than lower-redshift systems.

The overall consistency between the properties of the z ∼ 6 systems and those of lower-redshift DLAs and sub-DLAs suggests that the z ∼ 6 absorbers arise from galaxy halos, rather than from remnants of neutral IGM at the tail end of hydrogen reionization. One notable aspect of our survey, however, is that all ten of our systems occur at z > 5.75, while roughly 40% of our path length is at lower redshifts. Expanded surveys over 4.5 < z < 5.7 will help to determine whether the number density of low-ionization systems truly remains constant out to z ∼ 6, or whether there is a decline up to z ∼ 5.7 followed by an increase at higher redshifts. The latter scenario would potentially indicate a strong evolution in the UV background near z ∼ 6.

Finally, the high number density of low-ionization systems at z ∼ 6 suggests that we are probing galaxies below the detection limits of current i-dropout and Lyα-emission galaxy surveys. As such, these absorption systems are potentially the first observations of "typical" galaxies responsible for hydrogen reionization. The low metallicities we infer suggest that these galaxies may be highly efficient at producing ionizing radiation, a fact which would help to explain how the IGM becomes fully ionized by z ∼ 6. As more and higher-redshift QSOs are discovered, absorption lines will continue to provide a unique probe of the reionization era that will complement studies with ALMA, JWST, and other next-generation facilities.

We thank Bob Carswell, Andrew Fox, Max Pettini, and Andrew Pontzen for many useful discussions during the course of this work, and the anonymous referee for their helpful comments. We also recognize and acknowledge the very significant cultural role and reverence that the summit of Mauna Kea has always had within the indigenous Hawaiian community. We are most fortunate to have the opportunity to conduct observations from this mountain. G.B. has been supported by the Kavli Foundation. W.S. received support from the National Science Foundation through grant AST 06-06868. M.R. received support from the National Science Foundation through grant AST 05-06845. A.C. has been supported by the UK Science and Technology Facilities Council.