CONSTRAINTS ON THE UNIVERSAL C iv MASS DENSITY AT z ∼ 6 FROM EARLY INFRARED SPECTRA OBTAINED WITH THE MAGELLAN FIRE SPECTROGRAPH^*

In 2010 March, our group commissioned the Folded-Port Infrared Echellette (FIRE), a new IR echelle spectrograph for the 6.5 m Magellan Baade Telescope. FIRE delivers R = 6000 spectra with uninterrupted coverage from 0.82 to 2.5 μm for a 06 slit (Simcoe et al. 2010; Simcoe 2011). One of the major science drivers for its construction was exploration of the high-redshift intergalactic medium (IGM) via absorption-line spectroscopy. Here, we report initial results on the search for high-redshift C iv based on commissioning observations taken during the first year of FIRE's operation.

The existence of heavy elements in the IGM has been known since the earliest detections of C iv associated with z ∼ 3 Lyα forest absorbers in high-resolution optical spectra (Cowie et al. 1995; Songaila & Cowie 1996; Cowie & Songaila 1998). Soon after this discovery, intergalactic C iv lines were detected even in the spectra of quasars with z > 4, establishing a lower limit on the amount of metal production that must have occurred in the early universe (Songaila 2001). A surprising result of these studies was the relatively constant comoving number density of C iv atoms over a wide range in redshift, where this density is smoothed over very large scales. More recently, larger samples of C iv lines both in high-redshift optical spectra (D'Odorico et al. 2010) and in lower redshift UV spectra (Cooksey et al. 2010; Danforth & Shull 2008) point toward a slow rise in the C iv density—expressed as $\Omega _{\rm C\,\mathsc{iv}}$, the C iv contribution to closure—increasing by a factor of three to four between z ∼ 2 and the present day.

The discovery of luminous quasar populations at z > 6 (Fan et al. 2006; Willott et al. 2010; Mortlock et al. 2009) enables studies of the C iv density at higher redshift, but above z ∼ 5.5 observations become challenging as the C iv transition moves into the Y and J bands. For this reason, the earliest pilot studies in this region focused on very small (i.e., N = 2) samples of quasar sight lines. These pilot studies yielded similar values of $\Omega _{\rm C\,\mathsc{iv}}$ as at lower z, but with few actual line detections they were strongly affected by small number statistics (Simcoe 2006; Ryan-Weber et al. 2006; Becker et al. 2009). More recently, Ryan-Weber et al. (2009) compiled a larger set of IR observations for nine QSOs, with VLT/ISAAC and Keck/NIRSPEC. Their data set contains several sight lines with no C iv detections, which points to a decrease in $\Omega _{\rm C\,\mathsc{iv}}$ at z > 5.5.

During FIRE's first year of operation we have obtained spectra of eleven z > 5.5 quasars, of which seven are public domain and have suitable signal-to-noise ratio (S/N) for IGM abundance measurements. This paper presents a first analysis of seven of these spectra, focusing on the search for C iv doublets in particular. We combine our results with data from the literature to derive a new value for $\Omega _{\rm C\,\mathsc{iv}}$ based on 13 unique sight lines. Where relevant throughout the remainder of the paper, we adopt a flat cosmology with Ω_M = 0.3, $\Omega _\Lambda =0.7$, and H₀ = 71 km s⁻¹ Mpc.

During the period from 2010 March 31 to 2011 April 3, we obtained FIRE spectra of 11 QSOs with z_em > 5.5, seven of these are in the public-domain and have high enough S/N to study IGM absorption. Table 1 lists details of the observations for the seven high-redshift targets included in the present analysis. All spectra were obtained with a 06 slit, in seeing conditions ranging from 035 to 09, for a measured resolution of R = 6000, or 50 km s⁻¹. Individual integration times are listed in the table, with typical values of 4–5 hr. Full integrations were broken up into 15–20 minute intervals between which the object was moved on the slit.

We reduced the data using a custom-developed IDL pipeline, evolved from the MASE suite used for optical echelle reduction (Bochanski et al. 2009). There were several important modifications implemented for IR work. The most important of these was for wavelength calibration, which is principally solved using the terrestrial OH lines imprinted upon science spectra. Unlike most IR spectral packages which perform pairwise subtraction of A/B slit positions for sky subtraction, the FIRE pipeline instead calculates a direct B-spline model of the sky for each exposure using the techniques of Kelson (2003). This yields Poisson-limited residuals in the sky-subtracted frame, for a theoretical improvement of $\sqrt{2}$ in S/N relative to pairwise subtraction. This pipeline is being released to the community as part of the instrument package; a full description of its functionality and the efficacy of the sky subtraction will be provided in a separate forthcoming paper.

During observations, we obtained contemporaneous spectra of A0V stars for correction of telluric absorption features using the method of Vacca et al. (2003). We used the xtellcor procedure released with the spextool pipeline (Cushing et al. 2004) to perform telluric correction and relative flux calibration, finally combining the corrected echelle orders from all exposures into a single, one-dimensional spectrum for analysis.

For each sample target, we obtained additional far-optical spectra using the Magellan Echellette spectrometer (MagE; Marshall et al. 2008), for the purpose of augmenting S/N in the region blueward of 9000 Å. These spectra had typical integration times of ∼1 hr and R = 5600, a close match to FIRE. The data were processed in similar fashion as described above. Rather than performing detailed telluric corrections to the MagE data, we instead masked out regions susceptible to telluric contamination when combining with the FIRE spectra, only including the MagE contribution in telluric-free areas. The MagE spectra were smoothed to match the resolution of FIRE and the spectra from the two instruments were combined with inverse variance weighting, to create a single composite for analysis.

3.1. Line Identification

Before searching the data for C iv doublets, we fit a smooth continuum estimate to each quasar spectrum using a slowly varying cubic spline. Figure 1 displays the C iv region for each of the continuum-normalized spectra. These regions were then searched both by hand and also using automated software to identify C iv doublets.

Zoom In Zoom Out Reset image size

The automated search algorithm followed a two step process. First, the inverse spectrum (1 − f/f₀, where f is the flux and f₀ is the continuum) was smoothed with a Gaussian kernel having full width at half-maximum of one spectral resolution element, i.e., 50 km s⁻¹ or four pixels. A peak finding procedure was run on the convolved spectrum to identify all deviations of ⩾2.5σ, which after convolution is equivalent to the S/N per resolution element.

This list of single-line identifications was then parsed to identify pairs at the appropriate velocity spacing for C iv doublets. We further required that the ratio of the absorption amplitudes between the 1548 Å and 1551 Å lines be consistent with the ratio in oscillator strengths, i.e., between 2.0 (for an unsaturated doublet) and 1.05 (completely saturated). Each candidate whose doublet ratio fell within ±1σ of this range was inspected visually to reject obvious anomalies, for example, from poor sky subtraction near OH sky lines. We also verified that candidate C iv pairs were not Si ii/O i doublets at higher redshift, by inspecting the corresponding C iv and Mg ii wavelengths, as well as Si ii λ1526 for each system that could be misidentified in this way.

3.2. Completeness Estimates

At z ≲ 4, the C iv column density distribution has a power law form extending at least to $N_{\rm C\,\mathsc{iv}}= 10^{12.5\!-\!13.0}$ and possibly lower, as measured from high S/N, high-resolution optical spectra (Ellison et al. 2000; Songaila 2001, 2005). Our high-redshift sample is incomplete at these lower column densities, because of our ∼10 × lower resolution (e.g., compared to High Resolution Echelle Spectrometer (HIRES)) and ∼5 × lower S/N when compared with the best-optical QSO spectra in the literature.

To quantify this incompleteness, we ran a simple set of Monte Carlo tests where a simulated C iv signal was injected into our actual QSO spectra. Our line finding algorithm was then run to determine the recovery fraction as a function of redshift, column density, and line width b.

The intrinsic properties of z ∼ 6 C iv lines are not yet well characterized because the largest existing surveys are likely limited by instrumental resolution rather than intrinsic line widths.¹¹ At lower redshift, typical b parameters take values in the range b ∼ 8–20 km s⁻¹, with a slight preference for larger b at higher column densities. For the completeness tests, we took a forward-modeling approach by drawing b parameters at random from a distribution matched to HIRES C iv fits at z ∼ 2–3 (Rauch et al. 1996). This distribution is Gaussian in shape and centered at 10 km s⁻¹ for $N_{\rm C\,\mathsc{iv}}=10^{13}$. It has a cutoff below 5 km s⁻¹ and a steady rise in the mean b to 15 km s⁻¹ at $N_{\rm C\,\mathsc{iv}}=10^{14}$ but again cuts off above b ∼ 25 km s⁻¹.

Over a grid of column density, we drew redshifts at random, with a distribution matched to the path-length-weighted redshift distribution of the survey sight lines. We then calculated Voigt profiles directly for each artificial line, convolved these with the instrumental response function, and inserted them to the survey spectra, with one artificial system, per spectrum, per trial.

We then re-ran the line search algorithm described above on each trial spectrum and examined the output to see if the line was recovered. Each successful recovery was inspected by eye to ensure that our completeness was not being overestimated due to chance correlations in noise being picked up as "detections" for intrinsically weak lines. This procedure was repeated 5000 times, with 500 trials at each point in a grid of $N_{\rm C\,\mathsc{iv}}$ between 10¹³ and 10¹⁵ cm⁻².

Figure 2 shows the total completeness distribution of the survey; the left panel includes all z while the right panel shows the completeness map by both redshift and $N_{\rm C\,\mathsc{iv}}$. Broadly speaking, we recover most lines (>80%) above $N_{\rm C\,\mathsc{iv}}= 10^{14}$ and relatively few (<10%) below $N_{\rm C\,\mathsc{iv}}= 10^{13.2}$, with the 50% level occurring near 10^13.7. Thus our survey sits between the parameter space explored by Becker et al. (2009), who are complete to lower $N_{\rm C\,\mathsc{iv}}$ over a significantly smaller path length, and Ryan-Weber et al. (2009) who have a comparable path length but slightly lower completeness.

Zoom In Zoom Out Reset image size

From the right panel of Figure 2 it is also apparent that the completeness varies with search redshift, with our areas of highest sensitivity occurring at z < 4.8 and z > 5.2. The high completeness at z > 5.7 reflects the rising sensitivity of FIRE toward these wavelengths, while at z < 4.8 the MagE spectra contribute significantly to the S/N. The relatively low completeness at z ∼ 5 results from decreased flux and/or imperfect correction in the vicinity of telluric absorption bands, which can be seen as increased noise around λ = 0.94 μm in Figure 1.

In statistical calculations below, we correct for incompleteness by scaling each detected line by the inverse of its estimated completeness fraction. It is worth noting that the C iv lines reported in the literature to date nearly all have $N_{\rm C\,\mathsc{iv}}> 10^{14}$ cm⁻²; we should detect the large majority of these over our entire survey volume, so the correction to $\Omega _{\rm C\,\mathsc{iv}}$ from this source of incompleteness is only ∼25% or 0.1 dex—smaller than the Poisson errors from the limited number of systems in the sample. At smaller column density the correction is more severe, but these systems make a smaller fractional contribution to the total value of $\Omega _{\rm C\,\mathsc{iv}}$.

Using the criteria described above, we identified 19 unique C iv systems ranging in redshift from z = 4.61 to z = 5.92 in the six survey sight lines. Properties of the individual systems are listed in Table 2, and plots of the individual absorbers are shown in Figure 3. We excluded regions within 3000 km s⁻¹ of the QSO emission redshift from the search; however, no lines were excluded by this criterion, except for the one BAL absorber in SDSS1411. Table 2 contains two additional systems in SDSS0818+1722 at z = 5.789 and z = 5.876; these systems were not flagged as C iv in our data but have been reported previously in the literature and are discussed in Section 4.8.1.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 2. C iv Detections

Object	zabs	Wr(1548)a	Wr(1550)a	$N_{\rm C\,\mathsc{iv}}({\rm AODM})$b	$N_{\rm C\,\mathsc{iv}}({\rm VPFIT})$c	$b_{\rm C\,\mathsc{iv}}({\rm VPFIT})$d	log10(wi)e
SDSS0818+1722	4.69193	115 ± 23	63 ± 32	13.3 ± 0.20	13.43 ± 0.09	26 ± 13	0.49
...	4.72586	462 ± 25	290 ± 17	>14.1 ± 0.03	13.96 ± 0.04	43 ± 8	0.05
...	4.72737	...	...	...	13.67 ± 0.15	10	0.19
...	5.78917f	37 ± 20	54	...	⩽13.5	...	...
...	5.87689f	47 ± 13	33 ± 13	13.3 ± 0.08	⩽13.41	6.7 ± 8.0	...
SDSS0836+0054	4.99623	163 ± 17	73 ± 17	13.6 ± 0.06	13.70 ± 0.05	71.1 ± 11.5	0.63
SDSS1030+0524	4.94814	109 ± 27	237 ± 27	13.9 ± 0.07	13.76 ± 0.11	33.8 ± 23.1	0.57
...	5.51694	331 ± 31	256 ± 23	>14.0 ± 0.04	14.02 ± 0.04	72.9 ± 12.0	0.07
...	5.72438	698 ± 17	482 ± 18	>14.6 ± 0.03	14.59 ± 0.04	50.3 ± 3.0	0.04
...	5.74399	273 ± 15	172 ± 13	13.87 ± 0.05	14.00 ± 0.04	40.2 ± 5.7	0.05
SDSS1306+0356	4.61499	320 ± 13	184 ± 14	13.9 ± 0.03	14.09 ± 0.15	64.4 ± 4.7	0.02
...	4.66825	292 ± 31	198 ± 22	>14.0 ± 0.05	14.07 ± 0.03	58.1 ± 5.8	0.02
...	4.86591	1744 ± 50	901 ± 41	>14.7 ± 0.03	14.80 ± 0.01	80.6 ± 32.3	0.05
...	4.87965	514 ± 32	354 ± 30	>14.4 ± 0.10	14.30 ± 0.20	55.6 ± 19.4	0.06
ULAS1319+0905	5.57415	392 ± 44	196 ± 51	>14.1 ± 0.10	14.14 ± 0.04	65.5 ± 9.1	0.04
...	5.26490	133 ± 33	118 ± 27	13.6 ± 0.15	13.66 ± 0.10	62.9 ± 21.7	0.32
SDSS1411+1217	5.24988	734 ± 26	321 ± 22	>14.3 ± 0.04	14.14 ± 0.07	30.0g	0.03
...	5.24788	...	...	...	13.66 ± 0.10	30.0g	0.32
...	5.25151	...	...	...	13.95 ± 0.07	30.0g	0.11
...	5.786h	...	...	...	...	...
CFHQS1509-1749	4.61078	26 ± 7	11 ± 8	12.8 ± 0.16	13.23 ± 0.28	5.0 ± 6.0	0.87
...	4.64143	71 ± 10	60 ± 10	13.4 ± 0.10	13.44 ± 0.04	48.8 ± 7.0	0.43
...	4.66598	89 ± 14	35 ± 11	13.3 ± 0.10	13.41 ± 0.05	40.3 ± 7.7	0.50
...	4.81552	308 ± 22	206 ± 19	>14.0 ± 0.10	14.01 ± 0.04	60.0 ± 5.2	0.06
...	4.82492	...	33 ± 12	13.2 ± 0.20	13.45 ± 0.09	26.0 ± 10.0	0.63
...	5.91578	331 ± 39	183 ± 25	>14.1 ± 0.07	14.02 ± 0.03	48.8 ± 5.5	0.03

Notes. ^aRest-frame equivalent width (mÅ). ^bColumn density (cm⁻²) determined via the apparent optical depth method (Savage & Sembach 1991). The AODM is formally a lower bound on the column density for saturated doublets, so systems with $N_{\rm C\,\mathsc{iv}}>10^{14}$ cm⁻² are listed as limits in the table. ^cVoigt profile column density (cm⁻²). ^dVoigt profile doppler parameter (km s⁻¹). ^eCompleteness correction (dex). ^fThis system is not a statistically significant C iv system in our spectrum, but we confirm the presence of Si ii and other neutral atoms as reported by D'Odorico et al. (2011) and Ryan-Weber et al. (2009). It is not included in the $\Omega _{\rm C\,\mathsc{iv}}$ calculation. ^gThis system is blended such that the uncertainty in b was quite high; this has minimal effect on the total column density. ^hIntrinsic BAL.

Download table as:  ASCIITypeset image

We characterized the absorption strength for each system using a variety of different methods and to test the robustness of our column density determinations. At a resolution of 50 km s⁻¹, we almost certainly fail to resolve the detailed velocity structure seen in C iv systems at lower redshift, and we must also be attuned to the possibility of unresolved saturated components.

For each system, we list the rest-frame equivalent width (W_r) of both transitions of the C iv doublet, determined by direct summation of the continuum-normalized spectral pixels. The sums were performed over a region 50 km s⁻¹ above and below where the normalized profile returns to unity. We then list a C iv column density as determined using the "apparent optical depth" method of Savage & Sembach (1991), and finally the column density and b parameter estimates from direct Voigt profile fitting using the vpfit package.

Inspection of the Voigt profile Doppler parameters in Table 2 shows a high average b, similar to what was described in Ryan-Weber et al. (2009). This width is more likely to represent the full velocity spread of a suite of narrow components, rather than a large, single component. However, in such cases the total column density of the system is represented surprisingly well in a profile fit, since the ratio of the absorption depths reflects the degree of saturation (i.e., the column density for a single component) and the b parameter then scales with the number of individual sub-clumps in the system.

4.1. SDSS0818+1722, z_em = 6.00

4.2. SDSS0836+0054

4.3. SDSS1030+0524

4.4. SDSS1306+0305

4.5. ULAS1319+0950

4.6. SDSS1411+1217

4.7. CFHQS1509–1749

4.8. Special Cases

4.8.1. The z = 5.7899 Absorber in SDSS0818

Ryan-Weber et al. (2009) reported a possible C iv system in SDSS0818+1722 at z = 5.7899; this system was neither flagged in our visual inspection of the FIRE data nor found by our automated line search. However, as also reported by D'Odorico et al. (2011) and Becker et al. (2011), we do find evidence of low-ionization absorption offset from the reported C iv redshift by ∼75 km s⁻¹. A plot of several absorption species is shown in Figure 4, along with a shaded band showing the ±1σ contours about the continuum.

Zoom In Zoom Out Reset image size

Inspection of the error contour shows that both members of the C iv doublet are in regions of higher noise; this is because they are both blended with bright telluric OH emission lines at this redshift. There is weak evidence for absorption at the 1550 Å component, but the 1548 Å component appears to suffer from systematic sky subtraction effects. The data for this object were obtained on a single night (2010 March 5) when the heliocentric velocity amplitude was 28 km s⁻¹, i.e., nearly radial toward SDSS0818. A revisit of this object with seasonal offset of ∼1/2 year would effectively shift the telluric lines by one resolution element, providing an improved view of the absorber.

Using the 1550 Å line alone, we fit a Voigt profile model to check for consistency in column density with the value reported by Ryan-Weber et al. (2009); the result is shown in the top panel of Figure 4. Based on this component, we verified that a $\log (N_{\rm C\,\mathsc{iv}})\approx 13.5$ C iv line represents our data quite well. Our fits converged to lower values of the b parameter, though these may be unphysical since the values are well below one resolution element for FIRE. While the velocity width is uncertain, the total C iv column is roughly constant for any choice of between 10 < b < 60 km s⁻¹; this is certainly the case as well for b = 39 km s⁻¹ (the value measured by Ryan-Weber et al. 2009).

This absorption system is intriguingly different from typical absorption-line systems at lower redshift, where the neutral phase traced by O i, Fe ii, Si ii, and C ii is usually accompanied by strong absorption from high-ionization species. In fact the presence of O i absorption suggests this system may be a damped Lyα absorber or super-Lyman-limit system whose normal highly ionized gas phase is substantially suppressed.

4.8.2. Re-identifying a Reported High-z C iv System as Mg ii

Another finding with implications for the existing literature is our improved spectrum of SDSS 1030+0524, where a strong C iv system at z = 5.8288 was reported by both Simcoe (2006) and Ryan-Weber et al. (2009). This system was a marginal detection in the ISAAC spectrum, and Simcoe (2006) noted the presence of an unidentified, blended line in his GNIRS spectrum. In the FIRE spectrum, we clearly identify this absorber as a Mg ii system at z = 2.780, based on the detailed match of the Mg ii 2796 and 2803 Å velocity profiles and the weak but detectable presence of Fe ii absorption at coincident redshift. Figure 5 shows a stacked velocity plot of three transitions from the system along with a Voigt profile model fit for Mg ii.

Zoom In Zoom Out Reset image size

We detect individual Mg ii components¹² spanning a remarkable range of ±400 km s⁻¹ and a total equivalent width in excess of 3 Å. This places the system in the strongest 4% of all Mg ii absorbers found in the SDSS DR4 (>17, 000 systems at 0.36 < z < 2.28; Quider et al. 2011). It also has a velocity width larger than any of the 22 Mg ii systems studied at high resolution in Prochter et al. (2006).

The equivalent width for the z = 2.78 absorber qualifies it as a strong candidate damped Lyα system (Rao & Turnshek 2000), although the Lyα line itself is inaccessible in the saturated forest. No coincident Mg i absorption is seen, so it is uncertain whether there is truly neutral gas present. This leaves the physical picture for this system uncertain. It may represent the chance interception of an unusually violent galactic outflow (Bond et al. 2001), as has been seen in both QSO absorbers (Nestor et al. 2011) and galaxy spectra (Weiner et al. 2009). Alternatively if a gravitational interpretation is invoked to explain the velocity spread, the associated potential must be somewhat large, up to galaxy group scales, unless the system is undergoing a merger or is otherwise out of dynamical equilibrium. If a large galaxy or small cluster potential is present at small enough impact parameter to be seen in Mg ii absorption (generally ≲ 100 h⁻¹ physical kpc from galaxies; Chen et al. 2010), it is possible that this mass could act as a gravitational lens for SDSS1030, explaining in part its large apparent luminosity. A full analysis of this possibility is beyond the scope of the present paper.

4.9. Effects on Past Determinations of $\Omega _{\rm C\,\mathsc{iv}}$

Using the C iv identifications listed in Table 2, we constructed an estimate of $\Omega _{\rm C\,\mathsc{iv}}$ using the customary form:

$$\begin{equation} \Omega _{\rm C\,\mathsc{iv}}= {{1}\over {\rho _c}}m_{\rm C\,\mathsc{iv}}{{\sum w_i N_{{\rm C\,\mathsc{iv}},i}}\over {(c/H_0)\sum {\Delta X}}}. \end{equation} \tag{ 1 }$$

Here, ρ_c represents the (current) critical density, $m_{\rm C\,\mathsc{iv}}$ is the mass of the carbon atom, and w_i is the completeness correction applied for each system, as a function of absorber $N_{\rm C\,\mathsc{iv}}$ and z (see Figure 2). ΔX represents the comoving absorption path length summed over all sight lines, where for each sight line ΔX = X(z_max) − X(z_min) with

$$\begin{equation} X(z) = {{2}\over {3\Omega _M}}\sqrt{\Omega _M(1+z)^3+\Omega _\Lambda }. \end{equation} \tag{ 2 }$$

The path length for each sight line is listed in Table 1; the integrated path for all FIRE spectra is ΔX = 44.93, of which ΔX = 21.92 is at z > 5.3. This is approximately a factor of four larger in path than the survey of Becker et al. and 78% larger than the survey of (Ryan-Weber et al. 2009). Many sight lines studied by these authors are in the north and not accessible with FIRE. So, to maximize survey volume we combine separately our FIRE measurements with (primarily NIRSPEC) measurements for northern targets to obtain a total path length of ΔX = 62.77, of which ΔX = 37.73 is at z > 5.3. Figure 6 shows the resulting calculation of $\Omega _{\rm C\,\mathsc{iv}}$, along with independent results at other redshifts, gathered from the literature as listed in the figure caption. We have corrected the point from Ryan-Weber et al. (2009) to remove the Mg ii system in SDSS1030+0524.

Zoom In Zoom Out Reset image size

Several adjustments must be made to put data from different studies together on the same plot, since each survey quotes different levels of completeness, and in some cases different cosmological parameters are used to calculate $\Omega _{\rm C\,\mathsc{iv}}$ (Ryan-Weber et al. 2009; Cooksey et al. 2010). For older studies which use an Einstein–deSitter cosmology (Ω = 1) we multiply results by $\sqrt{\Omega _M}\approx 0.55$ to (approximately) translate to currently favored concordance models. A more subtle correction must also be made for weak C iv lines not included in the sum from Equation (1) because of survey incompleteness, which varies from study to study.

This effect may be accounted given knowledge of the column density distribution function $f(N_{\rm C\,\mathsc{iv}}) = d\mathcal {N}/dN_{\rm C\,\mathsc{iv}}$ for each survey, since $\Omega _{\rm C\,\mathsc{iv}}$ may also be expressed as a weighted integral of f(N). The power-law slope of f is shallower than $N_{\rm C\,\mathsc{iv}}^{-2}$ in all lower redshift surveys, so the effect of missed systems at low $N_{\rm C\,\mathsc{iv}}$ is subdominant, but could still represent a factor of ∼2 or more. We follow the procedures outlined in Ryan-Weber et al. (2009) and Cooksey et al. (2010) to determine the fraction of each survey's systems that would be detected in our sample, appropriately integrating their fits to $f(N_{\rm C\,\mathsc{iv}})$ over the range where we are complete. We chose integration limits of $10^{13.4}<N_{\rm C\,\mathsc{iv}}<10^{15}$, since we are sensitive to many systems in this range and our completeness corrections are robust. The fraction of each prior survey's systems that would be detected in the FIRE spectra varies; the specific values we used are 0.70 for Songaila (2001), 0.89 for Songaila (2005), 1.12 for Ryan-Weber et al. (2009), 0.86 for Pettini et al. (2003), and 0.93 for Cooksey et al. (2010). The correction for Ryan-Weber's data relies on the unknown power-law slope of f(N) at z > 5 which we assumed to run as N^−1.5; if the slope is as steep as N^−1.8 then the correction becomes a factor of 1.22.

After all corrections are applied, our principal result is confirmation of a marked decline in $\Omega _{\rm C\,\mathsc{iv}}$ at z > 4.5, as initially reported by Ryan-Weber et al. (2009) and Becker et al. (2009). We divided up our sample into two redshift bins with z > 5.3 and z < 5.3, with the lower redshift end extending as low as z = 4.35 and the high-redshift end to z = 6.4. The low-redshift bin contains 14 distinct FIRE detections (where we count multi-component systems as single detections), while the high-redshift bin contains five FIRE detections plus the two likely systems discussed in Section 4.8.1. Our $\Omega _{\rm C\,\mathsc{iv}}$ does not include these systems that are not formal detections with FIRE, although the presence of low-ionization absorption suggests they are probably real. If we treat our fits to these systems as upper limits to the C iv column density, the amount by which $\Omega _{\rm C\,\mathsc{iv}}$ could increase is bounded at 20% (0.08 dex). The total result for the seven FIRE sight lines is $\Omega _{\rm C\,\mathsc{iv}}= 6.8\;\pm\, 3.3 \times 10^{-9}$ at 〈z〉 = 5.66, for the column density range $13.4 < \log (N_{\rm C\,\mathsc{iv}})<15.0$. Similarly the value at 〈z〉 = 4.92 for the FIRE sample alone is 2.0 ± 0.5 × 10⁻⁸.

Given the low apparent density of C iv systems, we also re-calculated $\Omega _{\rm C\,\mathsc{iv}}$ including six sight lines from the literature that were not observed by FIRE. The additional sight lines used include SDSS0002+2550 and SDSS1148+5251, which yielded no detections in high-resolution (23 km s⁻¹) NIRSPEC data (Becker et al. 2009), and SDSS0840+3549, SDSS1137+3549, SDSS1604+4244, and SDSS2054+0054, which were observed at low resolution (185 km s⁻¹) with NIRSPEC by Ryan-Weber et al. (2009). This expanded set of sight lines yields just one additional C iv system, at z = 5.738 in SDSS1137+3529. At this system's $\log (N_{\rm C\,\mathsc{iv}})=14.2$, Ryan-Weber's data should be highly complete so we apply no completeness correction to this system in our modified sum for $\Omega _{\rm C\,\mathsc{iv}}$. Becker's spectra, while yielding no detections, should also be highly complete over our full range of column densities given his higher resolution.

The net effect of these largely null results is to reduce the derived density, with our combined value for 13 sight lines measured at

$$\begin{equation} \Omega _{\rm C\,\mathsc{iv}}= \left\lbrace {\begin{array}{@{}ll}(1.87\pm 0.50) \times 10^{-8} &\quad \langle z\rangle =4.95 \\\ms (0.46\pm 0.20) \times 10^{-8} &\quad \langle z\rangle =5.66. \\ \end{array}}\right. \end{equation} \tag{ 3 }$$

The redshifts quoted above reflect the path-length-weighted mean for each bin, where the lower bin includes all systems from 4.35 < z < 5.3 and the higher includes all systems with 5.3 < z < 6.4. While lower than the value from the FIRE spectra alone, the different methods agree to well within 1σ. The values in Equation (3) represent a steady decline in $\Omega _{\rm C\,\mathsc{iv}}$ by a factor of 4.1 ± 2.1 over the 4.5 < z < 5.6 interval. This decline is measured using counts of C iv systems drawn at different redshifts, but from the same spectra.

Our data support the contention that the C iv density is in decline at z > 4.5. The exact redshift at which this downturn begins is not strongly constrained by present data, although it appears sometime between z = 4 and z = 5. When plotted in the context of all surveys including ones in the local universe, several interpretations are possible regarding evolution. Ground-based studies to date have focused on the relative constancy of $\Omega _{\rm C\,\mathsc{iv}}$ from 2 < z < 5, followed by a steep drop at higher z. As plotted here, an equally plausible interpretation could be $\Omega _{\rm C\,\mathsc{iv}}$ decreasing linearly with increasing redshift from z = 0, at constant slope until z ∼ 4.5 after which the decline quickens (as advocated by D'Odorico et al. 2010). Yet another interpretation would hold that the C iv mass density is on relatively constant decline over all redshifts, but is enhanced briefly at z ∼ 4. Coincidentally this is the redshift where the H i cross-section-weighted C iv ionization fraction peaks for a Haardt & Madau (2001) prescription of the ionizing background spectrum. At present the data do not distinguish strongly between these cases; revisiting the 1 < z < 2 and z ∼ 4 regions with larger samples of optical spectra should in time be able to settle the question.

Over the past five years, C iv observations at z > 5 have evolved from samples of two sight lines which were strongly affected by sample variance (Simcoe 2006; Ryan-Weber et al. 2006), to the present sample of 13 sight lines. Yet the number of strong detections has grown slowly, while several sight lines nearly devoid of C iv have now been observed. At a total detection count of six sure and two likely systems at z > 5.3, it is far from clear that we have sampled enough cosmic volume to perform robust statistical analysis.

Yet the data in hand point strongly toward a lower C iv mass density in our high-redshift search region. A critical question is to what extent this trend reflects a change in the total carbon abundance as opposed to a change in its ionization balance. Rapid changes such as this over a short timescale are more easily achieved through a combination of radiative and ionization effects rather than chemical feedback, but the period before z ∼ 3 was also one of stellar mass buildup in the universe (Bouwens et al. 2007). Some of the byproducts of this process will leak into the IGM, and indeed recent estimates of evolution in the ionization-corrected carbon abundance at z = 2–4 suggest that the C iv mass in the IGM doubled over this period (Simcoe 2011), although this represents a somewhat longer stretch of time.

The C iv ionization balance is governed principally by photons with energies of 3.5 Ry (C iii to C iv), 4.74 Ry (C iv to C v), and >20 Ry (C v to C vi). These high-energy photons are produced predominantly in active galactic nuclei (AGNs), but are also strongly attenuated by intergalactic He ii at z > 4, since He ii reionization likely does not occur until z ≲ 3.5, and perhaps even lower redshift (Shull et al. 2010; Worseck et al. 2011). C iv systems—particularly the strong ones which dominate $\Omega _{\rm C\,\mathsc{iv}}$—may be preferentially present in He ii-ionized bubbles surrounding sites of galaxy formation and/or AGN activity, where the radiation field and abundances are both enhanced.

If the evolution in the C iv abundance is driven by ionization rather than abundance evolution, one might expect to see an increase in the density of low-ionization C ii, Si ii, or O i systems. These systems would arise in regions where abundances are high but the ionizing field is dominated by stellar photons rather than harder AGN radiation.

There are a few lines of evidence hinting at this interpretation for some absorbers, though it is still too early to draw definitive conclusions about the general population. Becker et al. (2006) initially reported a strong enhancement in O i absorption toward the high-redshift QSO SDSS1148+5251, and subsequent observations have not yet yielded corresponding C iv absorption coincident with these O i lines. More recent work by Becker et al. (2011) compiles statistics on C ii, O i, and Si ii absorbers from HIRES and Echellette Spectrograph and Imager spectra of a large sample of high-redshift objects. This study suggests that the evolution of low-ionization absorbers broadly follows the extrapolation of trends established at lower z, with the highly ionized gas phase dropping out at z ⩾ 5. However these studies treat the evolution of low- and high-ionization systems independently, since examples were not available where both ionization phases were detected in a single z > 5.5 system.

Two of the systems studied in Becker et al. (2009) are covered by our data on SDSS0818; one of these was first reported as a tentative C iv detection by (Ryan-Weber et al. 2009) and has since also been discussed by D'Odorico et al. (2011). Our data rule out strong C iv absorption in these two systems but admit the possibility of weak lines, while we confirm evidence of low-ionization absorption. A small handful of systems are present in our other FIRE spectra with similar low-ionization absorption, often identified via Mg ii in the H band. Analysis of these systems is outside the scope of this paper, although it is notable that some systems contain strong C iv (and are included in the present sample) and some do not. At the resolution and S/N of our FIRE spectra we do not see evidence for the emergence of a strong "forest" of C ii, Si ii, or O i lines at z > 5.5, although higher quality spectra will be needed to assess this possibility in detail.

Ultimately a combination of abundance and ionization evolution must be at play, suggesting that forward modeling using numerical simulations will be required (Oppenheimer & Davé 2006, 2008; Oppenheimer et al. 2009; Wiersma et al. 2010). If the C iv systems reside in non-overlapping, He ii-ionized bubbles, radiative transfer effects, and local enrichment are likely to become important. In this case, the C iv systems seen at z > 5.5 are only "intergalactic" in the sense that they are selected at random from observations of background QSOs. In fact they are likely to be influenced strongly by their environments in both radiation and metal enhancement, and may reside quite close to the early galaxies responsible for reionizing the universe.¹³

Using the newly commissioned FIRE spectrograph, we have observed a sample of seven QSOs at z > 5.5 to derive improved constraints on the C iv abundance at early times. Using both manual and automated searches, we identified a sample of 19 C iv absorption systems in our FIRE spectra with C iv column densities in the range $13.2 \le \log (N_{\rm C\,\mathsc{iv}})\le 14.6$ and redshifts ranging from z = 4.61 to 5.91. Monte Carlo tests indicate that our line sample is highly complete at $N_{\rm C\,\mathsc{iv}}> 10^{14}$ cm⁻²; it is 50% complete near 10^13.7 cm⁻² over most survey areas not strongly affected by telluric absorption.

Our conclusions may be summarized as follows.

• 1.  
  Our spectra show evidence of a decline in the C iv abundance at z > 5, consistent with the findings of previous studies not strongly affected by cosmic variance. Expressed in terms of the closure density, we find $\Omega _{\rm C\,\mathsc{iv}}=4.6\pm 2.0\times 10^{-9}$ at 〈z〉 = 5.66, a fourfold decline relative to what is observed at 〈z〉 = 4.95 and lower redshifts.
• 2.  
  A strong system in the spectrum of SDSS1030+0524 previously identified in the literature as C iv at z = 5.82 is re-identified as Mg ii at z = 2.78. This system comprised a significant portion of the total C iv mass measured in all prior surveys, so these literature results must accordingly be revised downward.
• 3.  
  We present new limits on the C iv column density for several systems previously identified based on low-ionization lines and/or weak C iv. The emergence of these systems may reflect an overall change in the ionization balance of absorption systems at z > 5, where high-ionization species common at low redshift are less prevalent. Larger statistical samples of these systems are needed to compare the relative importance of the ionizing background field and metallicity distribution when interpreting the observed evolution in C iv.

Finally, we caution that as the C iv density drops, even though the number of well-observed sight lines grows our results may still be subject to fluctuations from shot noise or sample variance (only 5–7 systems are now known at z > 5.5). The present work summarizes just the first year of an ongoing program to observe high-redshift QSOs. Planned observations should cover the complete catalog of bright, southern-accessible targets, while ongoing and planned surveys such as UKIDSS (Warren et al. 2007), SkyMapper (Keller et al. 2007), and VISTA (McPherson et al. 2006) should substantially increase the number of known high-z quasars in the Southern sky.

It is a pleasure to thank engineers and staff at MIT and Magellan who contributed to FIRE's successful construction and commissioning. We gratefully acknowledge support for FIRE's construction through the NSF under MRI Grant AST-0619490 and also by the MIT Department of Physics and Curtis Marble. Support for science observations and analysis was provided by NSF Grant AST-0908920 and also by the Alfred P. Sloan Foundation. R.A.S. enthusiastically recognizes generous lumbar support from the Adam J. Burgasser Endowed Chair in Astrophysics. A.J.B. acknowledges financial support from the Chris and Warren Hellman Fellowship Program.

Facility: Magellan:Baade (FIRE) - Magellan I Walter Baade Telescope