THE MOST METAL-POOR DAMPED Lyα SYSTEMS: AN INSIGHT INTO DWARF GALAXIES AT HIGH-REDSHIFT^*

The population of dwarf galaxies represents the least luminous, least massive, and perhaps the most numerous galaxies at all redshifts in the observed universe. However, despite their great number, just a handful of these galaxies have been firmly identified, owing to their intrinsically low luminosities (see McConnachie 2012 for a continuously updated list of systems). Typically, these galaxies are very metal-poor and are highly varied in their physical properties (Mateo 1998; Tolstoy et al. 2009). At present, our understanding of the chemical evolution and buildup of stellar mass in these galaxies involves a careful "archaeological"study of their present-day physical and chemical properties.

1.1. The Nearby Universe

1.2. The High-redshift Universe

The high-redshift DLA sample considered here consists of the systems published in our recent surveys (Pettini et al. 2008; Cooke et al. 2011b, 2013, 2014) which focus on DLAs with iron abundance [Fe/H] ⩽−2.0. We omit two systems considered in those works (toward the QSOs J0311−1722 and J1016+4040) where we were unable to measure [Fe/H], but we add one new system (in the spectrum of J1111+1332), and reanalyze another (in J0035−0918; Cooke et al. 2011a). For further details on these two systems, see Appendix B. The data that are presented in this paper were all acquired with high-resolution echelle spectrographs (R ≳ 30, 000), including the W. M. Keck Observatory's High Resolution Echelle Spectrograph (HIRES) and the Very Large Telescope's Ultraviolet and Visual Echelle Spectrograph (UVES). Relevant properties of the DLA sample considered here are collected in Table 1. In Section 5, we refine our sample selection to consider only the DLAs where the contributions of thermal and turbulent broadening to the absorption line profiles could be measured separately with sufficient precision (in many cases turbulent broadening dominates over thermal broadening, so that the latter cannot be determined independently). For further details, see Appendix A.

The kinematic information extracted from the analysis of metal absorption lines in DLAs is one of the key observables that has been used to infer the nature of the DLA host galaxy population. The first detailed investigation of DLA kinematics was presented by Prochaska & Wolfe (1997). These authors introduced a test statistic, known as v₉₀, which measures the velocity interval covering the innermost 90% of the total optical depth of unsaturated metal absorption. In general, DLAs exhibit a broad range in their v₉₀ statistic; the distribution of values peaks near 70 km s⁻¹ with a high-velocity tail that extends out to ∼450 km s⁻¹. Using a suite of Monte Carlo simulations, Prochaska & Wolfe (1997) explored a variety of simple galaxy models to conclude that the kinematics of the DLA population are mostly consistent with thick, rapidly rotating disks. An alternative possibility was proposed shortly after by Haehnelt et al. (1998) using cosmological N-body hydrodynamic simulations. These authors showed that the distribution of asymmetries and velocity widths of DLA metal absorption lines can be naturally explained in the context of hierarchical galaxy formation, where the kinematics of DLAs arise from the combination of ordered and random motions, in addition to infall and merging (see, however, Prochaska & Wolfe 2010). Since these early works, cosmological hydrodynamic simulations have advanced substantially, and yet they still have difficulty in reproducing the large velocity widths exhibited by some DLAs (Pontzen et al. 2008; Tescari et al. 2009; Cen 2012). A potential resolution to this problem has recently been proposed by Bird et al. (2014), who interpret the largest velocity widths as arising from the chance alignment of a DLA host galaxy and a metal-enriched Lyman limit system residing in a filament near the DLA (see, e.g., Pettini et al. 1983).

By combining the measured velocity widths and metallicities of damped Lyα systems, several authors have proposed that DLAs obey a "mass–metallicity" relation (Ledoux et al. 2006; Murphy et al. 2007; Prochaska et al. 2008; Jorgenson et al. 2013; Neeleman et al. 2013). Using the v₉₀ velocity width as a proxy for the mass of the DLA host galaxy, DLAs with the largest velocity widths statistically show the highest metallicities. Thus, it appears that the gas distribution broadly traces the gravitational potential of the host galaxy. In this picture, DLAs that exhibit the lowest metallicities should also exhibit the most quiescent kinematics, and thus be related to the lowest mass galaxies.

Thus, it is of interest to compare the kinematics of the most metal-poor DLAs to those of Local Group dwarf galaxies. The latter are typically characterized by the 1σ line-of-sight dispersion of projected stellar motions or, if the galaxy still bears gas, from the H i 21 cm line emission profile. In both cases, the galaxy's kinematics are quantified by considering the 68 per cent interval of the tracer kinematics.

Therefore, to facilitate a comparison between z ∼ 3 DLAs and Local Group galaxies, we define a new velocity width statistic for DLAs, hereafter denoted v₆₈, which is qualitatively similar to the v₉₀ statistic proposed by Prochaska & Wolfe (1997). Specifically, v₆₈ is the velocity interval containing the innermost 68% of the total optical depth. Although our v₆₈ statistic cannot be directly related to a stellar velocity dispersion, as measured in the Local Group dwarf galaxies, it provides a simple conversion to the velocity dispersion of the gas. Furthermore, compared with the v₉₀ statistic, v₆₈ is much less sensitive to weak absorption that is well separated from the dominant optical depth. In order to measure v₆₈ from the observed profiles of the absorption lines in a DLA, it is necessary to correct for two sources of broadening that are unrelated to the bulk motions of the gas, namely thermal (i.e., microscopic) and instrumental (i.e., due to the spectrograph with which the data were recorded) broadening. In the most quiescent DLAs, where the absorption is confined to a single velocity component, the correction is straightforward (if the thermal and instrumental broadening are known). In this case, $v_{68}\equiv 2\sigma _{\rm los}=\sqrt{2}\,b_{\rm turb}$.

While it is generally the case that the velocity structure of DLAs with [Fe/H] ⩽−2.0 is relatively simple, in many of the DLAs in Table 1 the absorption is spread over a few velocity components. In such cases, we employed the following procedure to determine v₆₈. First, we generated synthetic Si ii (or Fe ii) line profiles for each system using only the column densities and turbulent broadening derived from the cloud model that best fits the absorption lines as described in Appendix A (i.e., we did not include thermal and instrumental broadening for the synthetic profiles). Si ii and Fe ii are the optimum species to use for measuring the velocity fields in these DLAs, since there are generally many Si ii and Fe ii lines of widely different oscillator strengths, allowing the optical depth to be determined for the full range of velocities over which absorption takes place. Furthermore, of the lines that are often seen in metal-poor DLAs, Si ii and Fe ii are the least affected by thermal broadening.⁷ We then converted the synthetic line profiles to an optical depth profile, and calculated the velocity interval containing the innermost 68 per cent of the total integrated optical depth. Finally, we relate our velocity statistic to the 1σ line-of-sight velocity dispersion by the relation σ_los = v₆₈/2. We note that our analysis only considers the velocity width traced by the low ion metal absorption lines (i.e., C⁺, O⁰, Si⁺, Fe⁺), which are known to predominantly trace the H i gas.

The full sample of DLAs for which we have measured v₆₈ is collected in Table 1, which also includes values of [Fe/H] "metallicity" and of [Si/Fe], representative of the ratio of α-capture to Fe-peak elements. The data in the last two columns were culled from the references given in Section 2. In Figure 1, we compare the measured kinematics of the DLAs in our sample to those of Local Group galaxies as compiled by McConnachie (2012). To allow for a closer comparison to the MW dSph population, we also present a zoom-in version of this comparison in Figure 2, where we have placed a red circle around the galaxies that have been discovered since the start of the SDSS survey (i.e., those classified as UFDs). There is evidently a very close correspondence between the kinematics and metallicity for the galaxies displayed in Figures 1 and 2 despite the diversity of techniques employed to measure these two quantities.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

For a given DLA, we are unable to make a direct connection between our v₆₈ statistic and the DLA host potential; since we are unaware of the geometry of the DLA host galaxy, this connection may depend on the inclination of the DLA relative to the sightline probed by the quasar. However, for a sample of systems at a given metallicity, each DLA will be intersected at a random inclination angle, so that statistically we may be justified in considering the range of velocity widths exhibited by DLAs at a given metallicity. From inspection of Figures 1 and 2, it can be seen that systems with higher velocity widths tend to have higher metallicities. Similarly, the lowest velocity width systems are associated with the lowest metallicity DLAs. We therefore conclude that the most metal-poor DLAs are generally associated with the lowest mass galaxies that can potentially still form stars at z ∼ 3.

A clear outlier in the velocity width–metallicity relation for metal-poor DLAs is the DLA toward J0140−0839 (Ellison et al. 2010; included in Table 1 but not shown in Figures 1 and 2 as it is off-scale in those plots). This DLA has one of the highest velocity width measurements of our sample and one of the lowest metallicities. This example highlights the main caveat of our technique; with quasar absorption line spectroscopy, we are unable to distinguish between a DLA in a single halo with a broad velocity distribution or two individual DLAs in close proximity with a small relative velocity along the line of sight. Given that this system exhibits two distinct absorption components, separated by ∼45 km s^–1, each with v₆₈ ≃ 6–10 km s⁻¹, we speculate that its abnormally high velocity width may result from a chance alignment or interaction of two very metal-poor DLAs (which, coincidentally, are also close to the QSO redshift).

Clearly, a direct comparison between the kinematics of redshift z ∼ 3 metal-poor DLAs and Local Group dwarf galaxies is fraught with uncertainty, given the different physics that operates on gas versus stars. Nevertheless, on the basis of this comparison, we conclude that the general trend and range of velocity widths exhibited by the most metal-poor DLAs are similar to those seen in nearby dwarf galaxies.

The chemical composition and relative element abundances of DLAs provide crucial information on the chemical evolution, SFH, and ultimately the nature of the host galaxies that give rise to the absorption. In this section, we present the current complement of data that allows us to infer the most likely SFH of typical DLAs, as traced by the relative abundance of α-capture and Fe-peak elements. At metallicities less than ∼1/100 of solar, the production of α elements in DLAs is best traced by considering the Si ii ion, which has numerous transitions with a variety of oscillator strengths and over a range of wavelengths that are redshifted into a convenient portion of the visible spectrum at redshifts z = 2–3. For the same reasons, the most reliable tracer of the Fe-peak elements at such metallicities is Fe ii.

For DLAs with metallicities greater than ∼1/100 of solar, some α-capture and Fe-peak elements may suffer from depletion onto dust grains (Pettini et al. 1997; Akerman et al. 2005; Vladilo et al. 2011) so that the [Si/Fe] abundance measured in the gas phase may be different from the intrinsic [α/Fe] abundance set by chemical evolution. To avoid the effects of dust depletion, S and Zn are often considered the most appropriate combination to trace the relative abundance of α/Fe in the ISM; both S and Zn have very little affinity for dust grains and, in any case, have an almost identical condensation temperature. Moreover, both S ii and Zn ii offer multiple transitions that are usually straightforward to measure and rarely suffer from saturation.⁸

However, with current telescope facilities, the strongest Zn ii absorption lines become too weak to be reliably measured when the DLA metallicity is ≲ 1/100 of solar. Therefore, in order to track the full behavior of the α/Fe ratio in DLAs, we must combine measurements of [Si/Fe] at low-metallicity with [S/Zn] measures at high metallicity. While using different element ratios in different metallicity regimes is clearly not ideal, we believe that systematic offsets between these two pairs of elements are likely to be small. Combining measurements of Si and S seems justified, since detailed nucleosynthesis calculations of massive stars have illustrated that Si and S are produced in solar relative proportions over a wide range of metallicity and stellar mass (Woosley & Weaver 1995; Chieffi & Limongi 2004). Furthermore, measurements of Zn and Fe in Galactic halo stars with [Fe/H] ≳ −2.0 have shown that Zn/Fe accurately traces the solar ratio (Saito et al. 2009). For these reasons, we propose that the combination of [Si/Fe] at the lowest metallicities and [S/Zn] at the highest metallicities provides the most reliable diagnostic of the chemical evolution of α-capture and Fe-peak elements in DLAs, when drawing comparisons with Galactic halo stars.

Figure 3 shows the distribution of [α/Fe] values in DLAs in 0.25 dex wide bins of "metallicity" over nearly four orders of magnitude in [Fe/H], from [Fe/H] = −3.75 to 0.0, together with the standard deviation appropriate to each bin (represented by the error bars). To account for the uncertainty in both [α/Fe] and [Fe/H], we generated 1000 Monte Carlo realizations of each measurement prior to binning. Therefore, there is a small amount of correlation between neighboring symbols. Red symbols denote values of [S/Zn] and [Zn/H] (as proxies for [α/Fe] and [Fe/H], respectively) from the compilation of such measurements by Vladilo et al. (2011), while the blue circles correspond to the values of [Si/Fe] and [Fe/H] collected in Table 1. Both sets of data are constructed exclusively from measurements using spectra of high signal-to-noise ratio (S/N) and resolution. Note the good agreement between red and blue symbols near [Fe/H] ≃ −2.0, where the two data sets overlap.

Zoom In Zoom Out Reset image size

Also shown in Figure 3 are equivalent data for Galactic halo stars (gray circles; Venn et al. 2004). In this case, the [α/Fe] values were calculated from the average of the ratios Mg/Fe, Si/Fe, and Ca/Fe, depending on the data available for a particular star. This is in line with many previous stellar studies, and is supported by the results of nucleosynthesis calculations showing that the relative yields of Mg, Si, S, and Ca from massive stars are in solar relative proportions when integrated over a Salpeter-like stellar IMF (Woosley & Weaver 1995; Chieffi & Limongi 2004).

In order to demonstrate that a typical DLA experiences a different SFH to the galaxies that built the MW stellar halo, we perform a proof by contradiction, with the starting hypothesis that a typical DLA experiences an identical SFH as the galaxies that predominantly built the stellar halo of the MW. Under this hypothesis, we are justified in our assumption that [S/Si] =0.0 and [Zn/Fe] =0.0 for [Fe/H] ≳ −2.5. As can be appreciated from the comparison presented in Figure 3, halo stars and DLAs agree in exhibiting an [α/Fe] ≃ +0.3 plateau when [Fe/H] ≲ −2.0, as noted previously by Rafelski et al. (2012). A similar plateau is also observed in oxygen: [O/Fe] ≃ +0.4 in DLAs with [Fe/H] ≲ −2.0 (Cooke et al. 2011b). These plateaux represent the enhanced [α/Fe] level set by massive stars.⁹ However, for metallicities [Fe/H] ≳ −2.0, the DLA population exhibits a gentle decline in [α/Fe] toward solar values, presumably marking an increased contribution of Fe-peak elements from Type Ia SNe. In this aspect, DLAs are clearly different from Galactic halo stars which maintain an approximately constant α-enhancement from [Fe/H] =−3.75 to [Fe/H] ∼−0.75. This disagreement contradicts our initial hypothesis, therefore implying that a typical DLA experienced a different chemical enrichment history from that of the MW stellar halo.

Rather, the dependence of the α/Fe ratio on metallicity in high-redshift DLAs seems to be more akin to that seen in local dSph galaxies (Kirby et al. 2011a), as can be realized by considering Figure 4. For clarity, we have divided the dSph into three subsets, as indicated; in each of the three panels we show the same set of DLA measurements as in Figure 3, now uniformly indicated by black symbols, assuming that [Zn/Fe] =0. In general, there appears to be much in common between the chemical evolutions of MW dSph galaxies on the one hand and of DLAs at redshift z_abs ∼ 2–3 on the other (we investigate our assumption that [Zn/Fe] =0 for dSph stars directly below). We also note a similarity in the DLA measures of [α/Fe] and [Fe/H] to those recently determined in the dwarf galaxies associated with M31, in particular AndV (see Figure 4 of Vargas et al. 2014).

Zoom In Zoom Out Reset image size

Regarding the MW systems, there may be some tension toward the lowest metallicities, where dSph stars are somewhat more α-enhanced relative to the DLAs. This could be the result of small number statistics, modeling techniques, or genuine differences between DLAs and the dSphs. We note that several studies working with many fewer stars than Kirby et al. (2011a), but with somewhat higher S/N and resolution, have reported an [α/Fe] plateau for low-metallicity stars in some dSph galaxies at a level that is consistent with that seen in DLAs. Some examples include Fornax (Hendricks et al. 2014), Sculptor (Starkenburg et al. 2013; Hill & DART Collaboration 2012), and Ursa Minor (Cohen & Huang 2010). We also note that our comparison between DLAs and dSph galaxies for metallicities [Fe/H] ≳ −2.0 assumes that S traces Si whilst Zn follows Fe. As argued earlier, S and Si are believed to be produced in solar relative proportions by massive stars, independent of chemical evolution. However, Zn may not always follow Fe in the dSph galaxies; unlike the stars in the MW disk and halo (Saito et al. 2009), the small number of measurements of [Zn/Fe] available in a few dSph galaxies suggests that [Zn/Fe] might vary with metallicity. Specifically, [Zn/Fe] =0.0 near [Fe/H] =−2.0 for dSph galaxies, and gradually decreases to [Zn/Fe] ≃ −0.2 to −0.5 near [Fe/H] =−1.0 (Shetrone et al. 2001; Sbordone et al. 2007; Cohen & Huang 2009, 2010). This difference between the dSph galaxies and the MW halo stars is likely due to their different SFH.¹⁰ Therefore, to enable a fair comparison between DLAs and dwarf galaxies one should shift the DLA points to somewhat lower [α/Fe] and higher [Fe/H] values when [Fe/H] ≳ −2.0. Thus, if the chemical evolution of a typical DLA is similar to that of a dwarf galaxy, such that [Zn/Fe] becomes subsolar for [Fe/H] ≳ −2.0, the DLA "knee" will become even more pronounced.

With these reservations in mind, the similarity in the behavior of [α/Fe] as a function of [Fe/H] in DLAs and dSph galaxies displayed in Figure 4 in the metallicity range −2.0 ⩽ [Fe/H] ⩽−1.0 suggests that the chemical evolution of a typical DLA ([Fe/H] ≃ −1.5) is more akin to that of a dwarf galaxy than the MW. This similarity between the [α/Fe] abundance of DLAs and the MW dSph galaxies was also pointed out by Bonifacio et al. (2004), albeit with many fewer dSph stars and DLAs than in the present study. Our combined sample of 54 DLAs with reliable [α/Fe] measurements covering over three decades in metallicity has allowed us to trace the all-important "knee" in the [α/Fe] ratio. Of course, it is difficult to draw firm conclusions about the entire DLA population; for example, the high-metallicity DLAs considered in Figure 4 are unlikely to evolve as dwarf galaxies. Nevertheless, these data support a picture where a typical DLA likely traces the chemical evolution of a dwarf galaxy. Therefore, by extension, the most metal-poor DLAs likely represent the least chemically evolved dwarf galaxies at redshifts z ∼ 3 that still have the potential to form stars. Future, more extensive measurements of S and Zn in dSph stars and low-metallicity DLAs would allow us to test this picture further.

As explained in the Introduction, the most metal-poor DLAs offer a rare opportunity to study the physical conditions of neutral gas in the metal-poor regime. Such conditions may provide important clues to better understand the physics of star formation in the lowest mass dwarf galaxies at early times. In this section, we consider some of the key physical parameters that can be deduced from the analysis of the absorption lines from the metal-poor DLAs in our sample. While there have been previous studies of this kind (e.g., Srianand et al. 2005; Wolfe et al. 2003), none so far have focused specifically on very metal-poor DLAs.

Our analysis starts with the determination of the kinetic temperature of the gas. As explained below, this is only possible for a subset of the metal-poor DLAs in Table 1. Specifically, only in nine out of the 23 DLAs in our sample, can the thermal (i.e., microscopic) contribution to the line broadening be reliably separated from the turbulent (i.e., macroscopic) broadening; in the other 14 cases either turbulent broadening completely dominates or there is insufficient information in the profiles of the absorption lines covered to decouple the two. This inevitably introduces a bias: The physical properties we deduce below apply to the relatively quiescent DLAs in our survey, and may not be fully representative of, for example, DLAs of a given metallicity. Despite this bias, however, the analysis is still worthwhile, given how little is known at present about some of the most basic physical properties of the most metal-poor DLAs. The sample considered in this section consists of 12 absorption components in nine DLAs whose relevant parameters are collected in Table 2. Formally, the absorption system at z_abs = 2.536509 toward HS 0105+1619 is a sub-DLA, exhibiting a total H i column density less than the DLA threshold of 10^20.3 H i atoms cm⁻². We nevertheless include this system in our sample of DLAs, as it provides a useful indication of the physical properties of neutral gas at somewhat lower H i column density.

Table 2. Measured Physical Properties of Metal-poor DLAs^a

QSO Name	zem	zabs	bturb	Tgas b	$\log \frac{{{N}({\rm H}\,\scriptsize {I})}}{{{\rm cm}^{-2}}}$c	$\log \frac{{{N}({\rm H}\,\scriptsize {I})}}{{{\rm cm}^{-2}}}$d	[O/H]e	[Fe/H]e	[Si/Fe]e
			(km s−1)	(K)
J0035−0918	2.420	2.340097	1.1	7800	20.43	20.43	−2.44	−2.94	+0.37
		±0.000001	±0.4	±1600	±0.04	±0.04	±0.07	±0.06	±0.07
HS 0105+1619	2.652	2.536509	3.0	10800	19.415	19.426	−1.76	−2.10	+0.25
		±0.000001	±0.1	±200	±0.007	±0.006	±0.02	±0.05	±0.06
Q0913+072	2.785	2.618435	4.5	8400	20.09		−2.44	−2.73	+0.28
(comp. A)		±0.000001	±0.2	±1300	±0.02	20.312	±0.03	±0.02	±0.02
Q0913+072		2.618289	3.9	7700	19.84	±0.008	−2.35	−2.85	+0.29
(comp. B)		±0.000001	±0.2	±200	±0.03		±0.02	±0.04	±0.02
J1001+0343	3.198	3.078404	4.6	17000	20.21	20.21	−2.64	−3.48	+0.62
		±0.000002	±0.7	±8000	±0.05	±0.05	±0.06	±0.27	±0.27
J1111+1332	2.420	2.270940	2.2	13000	20.39	20.39	−1.92	−2.27	+0.32
		±0.000001	±0.1	±2000	±0.04	±0.04	±0.08	±0.04	±0.02
J1358+6522	3.173	3.0672594	2.4	5600	20.22	20.495	−2.06	−2.60	+0.31
		±0.0000005	±0.5	±1100	±0.06	±0.008	±0.06	±0.07	±0.03
J1419+0829	3.030	3.049654	2.8	9700	19.93		−1.82	−2.17	+0.23
(comp. A)		±0.000001	±0.2	±100	±0.02	20.392	±0.03	±0.04	±0.03
J1419+0829		3.049843	5.3	11600	20.209	±0.003	−1.99	−2.37	+0.24
(comp. B)		±0.000001	±0.2	±200	±0.008		±0.01	±0.03	±0.03
J1558−0031	2.823	2.702318	3.7	7200	20.27		−1.52	−1.90	+0.22
(comp. A)		±0.000003	±0.2	±1000	±0.08	20.75	±0.09	±0.08	±0.04
J1558−0031		2.702422	1.5	9300	20.29	±0.03	−1.47	−1.81	+0.28
(comp. B)		±0.000002	±0.3	±900	±0.11		±0.12	±0.11	±0.07
Q2206−199f	2.559	2.076229	5.3	12200	20.43	20.43	−2.04	−2.55	+0.30
		±0.000001	±0.2	±3000	±0.04	±0.04	±0.06	±0.04	±0.02

Notes. ^aThe DLAs in this table are a subset of our full sample (see Section 5). ^bOur technique implicitly assumes that the gas is distributed according to a Maxwell–Boltzmann distribution. The uncertainty of the kinetic temperature measurements reported here only reflect the random uncertainty based on the statistical fluctuations in the data. ^cH i column density for a given absorption component of the DLA. ^dTotal H i column density of the DLA. ^eWe have assumed a solar abundance of 12+log (O/H)_☉ = 8.69, 12+log (Si/H)_☉ = 7.51, 12+log (Fe/H)_☉ = 7.47 (Asplund et al. 2009). ^fValues are derived from Pettini et al. (2008) and Carswell et al. (2012).

Download table as:  ASCIITypeset image

For several DLAs, we report the H i column density of individual components. In these rare cases, we have detected multiple neutral deuterium lines. By assuming that each absorption component in these systems has an identical D/H ratio (Cooke et al. 2014), we have derived the H i column density of each component.

5.1. Measured Physical Quantities

DLA absorption line profiles are characterized by a Doppler width for the line broadening, a column density of each absorbing ion, and a redshift. The Doppler broadening further consists of a turbulent and a thermal component, specifically:

$$\begin{equation} b^2 = b_{\rm turb}^2 + \frac{2\,k_{\rm B}\,T_{\rm gas}}{m_{\rm ion}}, \end{equation} \tag{ 1 }$$

where k_B is the Boltzmann constant, T_gas is the gas kinetic temperature, and m_ion is the mass of the ion giving rise to an absorption line. Thus, by considering the absorption lines from multiple ions of widely differing mass, one can decouple the relative contributions of turbulent and thermal broadening (see Appendix A for further details).

In Columns 4 and 5 of Table 2, we list the derived turbulent Doppler parameters and kinetic temperatures in the 12 absorption components of the nine very metal-poor DLAs where this decoupling is possible. For completeness, we also provide the measured H i column density in each component and the corresponding chemical abundances. Our sample contains absorption systems with a range of H i column densities ($19.41\le \log N({{\rm H}\,{\scriptsize I}})/{\rm cm}^{-2}\le 20.43$) and metallicities (−2.64 ⩽ [O/H] ⩽ −1.47). In Figure 5, we show the distributions of turbulence parameter b_turb and of gas kinetic temperature T_gas. The weighted mean values and 1σ dispersion for this sample are 〈b_turb〉 = 3.5 ± 1.1 km s⁻¹ and 〈T_gas〉 = 9800 ± 1200 K, respectively. Despite covering an order of magnitude in both metallicity and H i column density, we found no discernible correlations of the gas turbulence or temperature with either of these parameters.

Zoom In Zoom Out Reset image size

The mean values of b_turb and T_gas reported here for metal-poor DLAs are not dissimilar to those typical of the so-called warm neutral medium (WNM) component of the MW ISM today (Wolfire et al. 1995, 2003), although the temperatures listed in Table 2 tend to be at the upper end of the values found locally. This broad similarity is perhaps surprising, given the difference by 2–3 orders of magnitude in metallicity between the high-z DLAs and the present-day Galactic ISM, and the attendant differences in the dust-to-gas ratio and in the abundance of molecules (Petitjean et al. 2000). Furthermore, the DLA gas is likely to be exposed to a radiation field which is different in both shape and intensity from that irradiating local interstellar clouds. These are all factors that regulate the heating and cooling balance of the ISM; thus, the similarities found here between high-z metal-poor DLAs and the local WNM warrant a re-examination of two-phase models of the ISM under the conditions that apply at z = 2–3.

Only a small number of studies have previously attempted to measure the temperature of DLA gas at high-redshift. In a few favorable cases, a curve-of-growth analysis has been used to estimate the Doppler broadening of unresolved absorption lines from molecular or neutral metal absorption lines (Jorgenson et al. 2009; Tumlinson et al. 2010). In both of these studies, the temperature of the gas was measured to be a few 100 K or less, consistent with the typical values estimated for the "cold neutral medium" (CNM) of the MW ISM. Other investigations have instead used the relative strengths of the fine-structure states of C and Si to estimate the DLA gas kinetic temperature (Howk et al. 2005; Jorgenson et al. 2010). These authors also report temperatures less than a few 100 K. A conclusion that is emphasized by all of the above studies is that the absorption from molecules, neutral species, and the fine structure states provide a probe of the physical properties of the CNM. In our work, we have directly measured the kinetic temperature of the H i gas associated with the first ions, and have found that this DLA gas is much warmer than the CNM and is consistent with a WNM (see also Carswell et al. 2012).

5.2. Derived Physical Quantities

Temperature, turbulence, neutral gas column density, and chemical composition are all parameters that can be measured directly from the absorption lines in our spectra. Armed with this information, we can also deduce several other physical properties relevant to metal-poor DLAs.

To this end, we have performed a suite of Cloudy photoionization simulations (Ferland et al. 2013). We model each DLA in our sample as a constant density plane-parallel slab of gas irradiated by the Haardt & Madau (2012) extragalactic background and the cosmic microwave background at the appropriate redshift. We consider a uniform grid of H volume densities in the range −3.0 ⩽ log n_H/cm⁻³ ⩽ +3.0, with a cosmic abundance of He (i.e., n_He/n_H = 1/12) and globally scale the metal abundance of the slab to the [Si/H] abundance derived for each DLA.¹¹ The calculations are stopped when the H i column density of the slab is equal to that of the DLA, and a selection of metal ion column densities are output. We perform 1000 Monte Carlo realizations of the above slab, each time drawing a new value for the stopping H i column density from the observed value and its associated error. The results from an example calculation are shown in Figure 6, where, for an assumed radiation field, the column density ratio of successive ion stages of a given element is sensitive to the H volume density of the gas, n(H).¹²

Zoom In Zoom Out Reset image size

In metal-poor DLAs, the most commonly observed ion ratios include N ii/N i, Al iii/Al ii, Si iii/Si ii, and Fe iii/Fe ii. As can be seen from Figure 6, the most sensitive probe of the H volume density is the Si iii/Si ii ratio. Which of the four ion ratios is available in a given DLA depends on redshift, metallicity, column density, and blending with other absorption lines. In cases where more than one ion ratio is covered, the values of n_H deduced are generally in good mutual agreement. Column 4 of Table 3 lists the values of DLA volume density so derived; n_H ∼ 0.1 cm⁻³ seems typical.

Table 3. Derived Physical Properties of Metal-poor DLAs^a

QSO Name	zem	zabs	log nH	$r_{\rm H\,\scriptsize{I}}$	MWNM	Pth/kB	Pturb/kB	cs	${\cal M}_{\rm turb}$
			(cm−3)	(pc)	(105 M☉)	(K cm−3)	(K cm−3)	(km s−1)
J0035−0918	2.420	2.340097	−1.3 ± 0.3	$1270^{+930}_{-610}$	$220^{+360}_{-160}$	$700^{+730}_{-360}$	$17^{+24}_{-10}$	$7.2^{+0.7}_{-0.8}$	$0.19^{+0.07}_{-0.07}$
HS 0105+1619	2.652	2.536509	−0.8 ± 0.3	$43^{+42}_{-21}$	⩽0.4	$2790^{+2250}_{-1350}$	$270^{+250}_{-130}$	$8.51^{+0.08}_{-0.08}$	$0.43^{+0.02}_{-0.02}$
Q0913+072	2.785	2.618435	−0.9 ± 0.2	$200^{+120}_{-70}$	$1.6^{+2.5}_{-1.0}$	$1440^{+890}_{-550}$	$380^{+210}_{-140}$	$7.5^{+0.6}_{-0.6}$	$0.73^{+0.07}_{-0.06}$
(comp. A)
Q0913+072		2.618289	−1.12 ± 0.05	$150^{+20}_{-20}$	$0.37^{+0.13}_{-0.10}$	$640^{+80}_{-70}$	$140^{+20}_{-20}$	$7.19^{+0.09}_{-0.09}$	$0.66^{+0.04}_{-0.04}$
(comp. B)
J1001+0343	3.198	3.078404	⩾ − 1.7	⩽1320	⩽63	⩾370	⩾50	$10.7^{+2.2}_{-2.8}$	$0.52^{+0.17}_{-0.11}$
J1111+1332	2.420	2.270940	−1.2 ± 0.3	$1080^{+890}_{-530}$	$180^{+340}_{-130}$	$1440^{+1440}_{-730}$	$56^{+58}_{-28}$	$9.3^{+0.7}_{-0.7}$	$0.28^{+0.04}_{-0.03}$
J1358+6522	3.173	3.0672594	−0.93 ± 0.05	$240^{+50}_{-40}$	$2.4^{+1.5}_{-0.9}$	$750^{+170}_{-160}$	$100^{+40}_{-30}$	$6.1^{+0.6}_{-0.6}$	$0.48^{+0.12}_{-0.11}$
J1419+0829b	3.030	3.049654	−0.35 ± 0.08	$32^{+7}_{-6}$	$0.021^{+0.010}_{-0.007}$	$4850^{+970}_{-820}$	$450^{+120}_{-90}$	$8.07^{+0.04}_{-0.04}$	$0.43^{+0.03}_{-0.03}$
(comp. A)
J1419+0829b		3.049843	⩾ − 1.20	⩽420	⩽6.2	⩾790	⩾210	$8.82^{+0.08}_{-0.08}$	$0.74^{+0.03}_{-0.03}$
(comp. B)
J1558−0031	2.823	2.702318	−0.53 ± 0.11	$110^{+40}_{-30}$	$0.76^{+0.85}_{-0.40}$	$2480^{+820}_{-630}$	$520^{+160}_{-130}$	$6.9^{+0.5}_{-0.5}$	$0.65^{+0.06}_{-0.05}$
(comp. A)
J1558−0031		2.702422	−1.17 ± 0.20	$620^{+420}_{-250}$	$39^{+90}_{-27}$	$850^{+510}_{-320}$	$26^{+20}_{-12}$	$7.9^{+0.4}_{-0.4}$	$0.23^{+0.05}_{-0.05}$
(comp. B)
Q2206−199	2.559	2.076229	⩾ − 1.6	⩽1700	⩽170	⩾330	⩾90	$9.0^{+1.1}_{-1.2}$	$0.72^{+0.11}_{-0.08}$

Notes. Quoted errors are 1σ confidence intervals. Note that the probability distributions show a high degree of asymmetry. ^aThe DLAs in this table are a subset of our full sample (see Section 5). The table columns have the following meaning: Columns 1 and 2: quasar name and redshift; Column 3: absorption redshift of the intervening DLA; Column 4: logarithmic volume number density of the DLA gas; Columns 5 and 6: radial extent and total mass of the DLA's neutral gas; Columns 7 and 8: thermal and turbulent pressure of the gas in units of the Boltzmann constant; Columns 9 and 10: sound speed and turbulent Mach number of the DLA gas. ^bWe caution the reader that this system is a "proximate" DLA, with z_abs ≃ z_em. Our Cloudy modeling does not include the radiation field from the QSO; the effect of the QSO depends on the unknown distance between it and the DLA. If the QSO's radiation field contributes significantly, n_H for Component A should be considered as an upper limit, whereas n_H for Component B would be unconstrained by the present models.

Download table as:  ASCIITypeset image

With these values of n_H, we can straightforwardly calculate the mass density of each DLA

$$\begin{equation} \rho _{\rm gas}=1.33\,m_{\rm p}\,n_{\rm H}, \end{equation} \tag{ 2 }$$

where m_p is the mass of an H atom, and the factor 1.33 accounts for He.

Using the measured kinetic temperature and turbulence of the DLAs, together with our determinations of the number and mass densities of the gas, the thermal and turbulent gas pressures are respectively given by

$$\begin{equation} P_{\rm th}=(n_{\rm H}+n_{\rm He})\,k_{\rm B}\,T_{\rm gas}\simeq 1.083\,n_{\rm H}\,k_{\rm B}\,T_{\rm gas} \end{equation} \tag{ 3 }$$

and

$$\begin{equation} P_{\rm turb}=\frac{\rho _{\rm gas}\,\langle v^{2}\rangle }{2} = \frac{3\rho _{\rm gas}\,b_{\rm turb}^2}{4} \,. \end{equation} \tag{ 4 }$$

We can also convert our measure of the kinetic temperature of the gas into the sound speed of the gas

$$\begin{equation} c_{\rm s} = \sqrt{\frac{P_{\rm th}}{\rho _{\rm gas}}} = \sqrt{\frac{k_{\rm B}\,T_{\rm gas}}{1.23\,m_{\rm p}}} \end{equation} \tag{ 5 }$$

from which we can calculate the turbulent Mach number

$$\begin{equation} {\cal M}_{\rm turb}=\frac{\sqrt{\langle {v^{2}}\rangle }}{c_s}= \frac{\sqrt{3/2}\,b_{\rm turb}}{c_{\rm s}} \,. \end{equation} \tag{ 6 }$$

Note that Equations (5) and (6) are simply conversions of measured physical quantities, and therefore do not depend on the density derived from our Cloudy simulations.

The cloud size (along the line of sight) is obtained directly from the ratio of the column density of H i and the total H volume density:

$$\begin{equation} L_{\rm H\,\scriptsize {I}}=\frac{N({\rm H\,\scriptsize{I}})}{n_{\rm H}}, \end{equation} \tag{ 7 }$$

where we have assumed that $n_{\rm H\,\scriptsize{I}} \simeq n_{\rm H}$. Our Cloudy simulations justify this assumption, since the neutral H volume density is $n_{\rm H\,\scriptsize{I}} \gtrsim 0.9\,n_{\rm H}$ for gas seen in absorption as a DLA.

In order to proceed further, we need to make an assumption about the geometry of the absorbers. One possibility is that, by selecting for analysis the DLAs with the lowest turbulent velocities, we preferentially pick out face-on disks. In such cases, it is hard to extrapolate from the parameters measured along our lines of sight to the global properties of the disk galaxy intersected. On the other hand, if the gas distribution is not highly flattened, then our data may be more representative of the DLA host as a whole. Thus, it is instructive to consider the implications of our measurements for the case of a spherical geometry of the absorber.

If the gas is spherically distributed, and our line of sight is close to the maximum projected H i column density, then the radius of these clouds is approximately given by $r_{\rm H\,\scriptsize{I}}=L_{\rm H\,\scriptsize{I}}/2$. As can be seen from Column 5 of Table 3, the radii of the DLAs sampled are typically less than 1 kpc and range from ∼30 to ∼1300 pc. Radius and density together give the total mass of warm neutral gas (again, assuming a spherical geometry):

$$\begin{equation} M_{\rm WNM}=\rho _{\rm gas}\times \frac{4\pi }{3}\,r_{\rm H\,\scriptsize {I}}^{3}, \end{equation} \tag{ 8 }$$

which is a lower limit if the quasar line of sight does not pass through the center of the metal-poor DLA. All of the quantities derived above are listed in Table 3.

5.3. The Physical Properties of Near-pristine Neutral Gas

The physical properties that we have derived provide new insights into the nature of metal-poor DLAs and of star formation in the lowest metallicity environments. We first investigate the relationship between the kinetic temperature and H volume density, illustrated in Figure 7. We divide our sample equally in two, and color code the symbols according to their [O/H] metallicity; blue symbols represent the relatively metal-poorer sub-sample (with [O/H] ≲ −2.0), while the red symbols are for the relatively metal-richer sub-sample (with [O/H] ≳ −2.0). In Figure 7, we have also drawn a series of solid lines of constant pressure (P_th/k_B) in the T_gas–n_H plane. Although the sample size is still small, these data suggest that metal-richer DLAs have a slight preference to host gas that is at modestly higher densities and temperatures, and therefore higher pressures, compared with the metal-poorer sample.

Zoom In Zoom Out Reset image size

We note that the majority of the DLAs in our sample are at lower pressures than the mean P_th/k_B ≃ 3000 K cm⁻³ determined by Jenkins & Tripp (2001, 2011) for diffuse interstellar clouds within a few kpc of the Sun. Pressures in metal-poor DLAs are also significantly lower than those typical of DLAs with detectable column densities of molecular hydrogen, which tend to be at the metal-rich end of the distribution of DLA metallicities (Srianand et al. 2005).

By inspection of the thermal and turbulent pressures listed in Table 3, we can also conclude that the most metal-poor DLAs are predominantly held up by thermal pressure, as is the case for the WNM in the local ISM (Vallerga 1996; Redfield & Linsky 2004). Similarly, the turbulent Mach number that we derive for the DLAs in our sample (${\cal M}_{\rm turb}\sim 0.5$) is reminiscent of the local ISM values; every DLA that we have analyzed here exhibits a subsonic turbulent Mach number. Such low Mach numbers suggest that the WNM in the most metal-poor DLAs follows a fairly smooth gas density distribution (Kowal et al. 2007; McKee & Ostriker 2007), which is not conducive to gas fragmentation. Presumably, star formation in the most metal-poor DLAs proceeds via a CNM that, based on local studies (e.g., Heiles & Troland 2003), would exhibit a supersonic M_turb. The existence of a two-phase medium with a cold and warm component is supported by H i 21 cm observations of neutral gas in Leo T and Leo P—two nearby low-metallicity dwarf galaxies in the Local Group (Ryan-Weber et al. 2008; Bernstein-Cooper et al. 2014).

Such observations have not yet been conducted on redshift z ∼ 3 very metal-poor DLAs. However, there exists a handful of 21 cm absorption measurements for DLAs in front of radio-bright quasars (Srianand et al. 2012; Kanekar et al. 2014). In this case, the optical depth of 21 cm absorption depends on the H i column density, the spin temperature, and a covering factor that describes how effectively the foreground DLA covers the extended radio emission of the background quasar. For gas in a WNM, Liszt (2001) demonstrated that collisional excitation is unable to efficiently thermalize the H i λ21 cm line, leading to spin temperatures that are somewhat less than the kinetic temperature of the gas. A relationship between the spin and kinetic temperatures is further complicated by the uncertain fraction of gas along the line-of-sight that is in a CNM versus a WNM; a high CNM fraction weights the line-of-sight spin temperature toward lower values. In principle, the fewer coolants that are available in low-metallicity gas imply lower CNM fractions, and therefore higher spin temperatures; presumably this is the origin of the observed anti-correlation between spin temperature and metallicity (Kanekar & Chengalur 2001; Kanekar et al. 2014). By extrapolating the relationship between spin temperature and metallicity of the DLA population, the most metal-poor DLAs in our analysis presumably exhibit relatively high spin temperatures, corresponding to a low fraction of CNM gas.

For convenience, we provide a summary of the key physical properties of the most metal-poor DLAs in Table 4.¹³ In brief, our observations and analysis suggest that the most metal-poor DLAs are clouds that typically contain a reservoir of ∼10^4–7 M_☉ of warm neutral gas with a temperature T_gas ≃ 9600 K. This gas is expected to have a smooth density distribution, and is predominantly supported by thermal pressure. Although our measurement technique is not sensitive to the presence or absence of a CNM (see Appendix A for details), we point out that the properties of the inferred WNM are not conducive to star formation. Given the absence of molecular absorption lines in our DLA sample, considered a tracer of the CNM, we suggest that a CNM could be present in these DLAs with either a small cross-sectional area, or may only exist for a relatively short phase of a DLA's life just prior to a burst of star formation. The absence of a CNM in our DLA sample is consistent with the paucity of strong molecular hydrogen absorption associated with the general DLA population (Jorgenson et al. 2014). In addition, metal-poor DLAs might contain very low CNM fractions, in agreement with recent cosmological hydrodynamic simulations (Maio et al. 2014).

Table 4. Summary of the Physical Properties of Metal-poor DLAs

Property	Confidence Intervals
bturb (km s−1)	$3.3^{+1.8}_{-1.5}$ (1σ) $^{+2.3}_{-2.4}$ (2σ)
Tgas (K)	$9600^{+2500}_{-2600}$ (1σ) $^{+12200}_{-5000}$ (2σ)
log n(H)/cm−3	$-1.0^{+0.4}_{-0.3}$ (1σ) $^{+0.6}_{-0.7}$ (2σ)
$r_{\rm H\,\scriptsize {I}}$ (pc)	$220^{+840}_{-130}$ (1σ) $^{+2200}_{-200}$ (2σ)
log MWNM/M☉	$5.4^{+1.9}_{-0.9}$ (1σ) $^{+2.5}_{-2.5}$ (2σ)
Pth/kB (K cm−3)	$1050^{+1610}_{-450}$ (1σ) $^{+4150}_{-730}$ (2σ)
Pturb/kB (K cm−3)	$130^{+330}_{-100}$ (1σ) $^{+630}_{-120}$ (2σ)
cs (km s−1)	$8.0^{+1.0}_{-1.2}$ (1σ) $^{+4.1}_{-2.5}$ (2σ)
${\cal M}_{\rm turb}$	$0.50^{+0.23}_{-0.24}$ (1σ) $^{+0.34}_{-0.36}$ (2σ)

Download table as:  ASCIITypeset image

In the previous sections, we have assessed the kinematic, chemical, and physical aspects of the metal-poor DLA population. With these observations, we can start building a connection between the metal-poor DLA population at high-redshifts and their modern-day counterparts which are presumably akin to Local Group dwarf galaxies (Salvadori & Ferrara 2012). In this section, we summarize what can be learnt by studying the lowest mass galaxies over cosmic time.

Detailed star formation histories that are extracted from CMDs of Local Group dwarf galaxies (Weisz et al. 2011, 2014a) imply that these galaxies formed a considerable fraction of their present-day stellar population between redshifts z ∼ 2 and 5 (Weisz et al. 2014a), corresponding to the range of redshifts that are probed by our metal-poor DLA sample. Furthermore, the observed kinematics of the metal-poor DLA population show a similar trend to that observed in Local Group galaxies, where the systems with the lowest metallicities also exhibit the lowest velocity dispersions. From this observation, we propose that the most metal-poor DLAs are associated with the lowest mass halos that can potentially form stars at redshift z ∼ 3.

This conclusion receives additional support from the observed chemical evolution of the DLA population. Specifically, the evolution of the α/Fe ratio with metallicity (as measured by Fe/H) for a typical DLA is broadly similar to that seen in the local, low-mass dwarf galaxies, and is clearly different from the behavior of this ratio in MW stars of the halo and thin and thick disks (Venn et al. 2004; Bensby et al. 2005). The chemical evolution of a typical DLA also differs, although to a lesser extent, in the dependence of the α/Fe ratio on metallicity compared to relatively massive dwarf galaxies, such as the Small and Large Magellanic Clouds (Hill et al. 1997; Venn 1999; Pompéia et al. 2008) and the Sagittarius dSph (de Boer et al. 2014). The observations presented herein suggest that the chemical evolution of a typical DLA is similar to galaxies that are somewhat less massive than the Magellanic Clouds. By extension, we propose that the most metal-poor DLAs at redshift z ∼ 3 most likely represent the earliest stages of chemical evolution that were experienced by the lowest mass dwarf galaxies.

Of course, the observed z ∼ 3 population of metal-poor DLAs need not represent the entire Local Group dwarf galaxy population. For example, some of the UFDs appear to have formed most of their stars prior to z ∼ 3 (e.g., Hercules and Leo IV; Weisz et al. 2014a, 2014b). Indeed, some of the UFDs may have experienced "reionization quenching"(e.g., Bullock et al. 2000), and may therefore not be observable as gas-rich metal-poor DLAs at z ∼ 3. Nevertheless, these near-pristine DLAs can still improve our understanding of the physical conditions for star formation in this low-metallicity regime.

Finally, consider the current working definition of a "first galaxy"—that is, in its simplest terms, a long-lived stellar population confined by a dark matter halo (see, e.g., the review by Bromm & Yoshida 2011). Using this definition, we suggest that the most metal-poor DLAs may trace a reservoir of largely primordial neutral gas that is confined to a dark matter halo yet to form its first long-lived stellar population. Indeed, some near-pristine DLAs may contain the hallmark chemical signatures of enrichment by metal-free Population III stars (Cooke et al. 2011a, the chemical abundance pattern of that DLA is updated herein—see Appendix B). It is therefore certainly plausible that some of the most metal-poor DLAs may not even fit the canonical definition of a "galaxy," until they form their second generation of stars.

In this paper, we have provided the first observational links between the local, well-studied population of dwarf galaxies and their most likely progenitors at high-redshift. Our study considers the kinematic, chemical, and physical properties of the metal-poor DLA population, which all favor the interpretation that such systems predominantly trace the early evolution of dwarf galaxies. From the sample of metal-poor DLAs considered here, we draw the following conclusions.

• 1.  
  We measure the kinematics of a sample of 23 metal-poor DLAs with [Fe/H] ≲ −2.0 and conclude that the most metal-poor systems are likely to be associated with the lowest mass galaxies that have the potential to form stars at redshift z ∼ 3. We also find a close correspondence between our DLAs and the kinematics of Local Group dwarf galaxies of similar metallicities.
• 2.  
  To better understand the chemical evolution of the whole DLA population, we have considered the variation in [α/Fe] of DLAs as a function of [Fe/H] metallicity. We find evidence for a "knee" in the evolution of [α/Fe] in DLAs at [Fe/H] ≃ −2.0, which is entirely consistent with the knee that is observed in most of the dSph galaxies near the MW (assuming [Zn/Fe] ≃ 0 when [Fe/H] ≃ −2.0, as supported by observations of dSphs). Conversely, the behavior of [α/Fe] versus [Fe/H] in the DLA population clearly differs from that exhibited by MW halo stars. We conclude that the z ∼ 3 DLA population is not representative of galaxies similar to those that merged to form the bulk of the MW stellar halo.
• 3.  
  We have examined the physical conditions of the gas giving rise to metal-poor DLAs, by compiling a sample of systems with the most well-determined cloud models where it is possible to decouple the turbulent and thermal broadening of the line profiles. From the analysis of nine metal-poor DLAs ([Fe/H] ∼−2.3) with a total of 12 cloud components, we estimate that the typical values of the turbulent Doppler parameter and of the gas kinetic temperature are $b_{\rm turb}=3.3^{+1.8}_{-1.5}$ km s⁻¹ and $T_{\rm gas}=9600^{+2500}_{-2600}$ K, respectively.
• 4.  
  Using Cloudy photoionization calculations, we have also estimated the gas density in each of the 12 DLA absorbers. Taken together, these measures have allowed us to estimate the mass, radial extent, thermal and turbulent pressure, and turbulent Mach number for these near-pristine clouds of neutral gas. We find that metal-poor DLAs contain roughly 10^4–7 M_☉ of neutral gas, and are predominantly supported by thermal pressure. Furthermore, all DLAs in our sample exhibit subsonic Mach numbers, which implies that the distribution of warm neutral gas in these systems is relatively smooth.

Our analysis has provided new insights into the nature of the most metal-poor DLAs, and the first glimpse into the physical conditions of star formation in low-metallicity gas at high-redshift. This now opens up the exciting prospect of studying in depth with future work the early evolution of galaxies analogous to the Local Group dwarfs, more than 10 billion years in the past, at a time when they were building the bulk of their present-day stellar populations.

We are grateful to the staff astronomers at the ESO VLT and Keck Observatory for their assistance with the observations. We thank the Hawaiian people for the opportunity to observe from Mauna Kea; without their hospitality, this work would not have been possible. We are thankful to an anonymous referee for a constructive and thorough report that improved the presentation of our work. We also wish to thank Tom Abel, Thomas de Boer, Mark Krumholz, Marc Rafelski, and Erik Tollerud for valuable discussions and comments. M. P. would like to thank Matteo Monelli and Stefania Salvadori for organizing the Symposium "Local Group, local cosmology" (EWASS 2013), which provided useful insights into some of the properties of Local Group dwarf galaxies. R.J.C. was partially supported by NSF grant AST-1109447 during this work, and is currently supported by NASA through Hubble Fellowship grant HST-HF-51338.001-A, awarded by the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., for NASA, under contract NAS5- 26555. R.A.J. gratefully acknowledges support from the NSF Astronomy and Astrophysics Postdoctoral Fellowship under award AST-1102683.

The cloud models for all of the DLAs considered in this work were analyzed with the new Voigt profile fitting software recently developed for precision absorption line profile analysis (Cooke et al. 2014). This Absorption LIne Software (alis), uses a Levenberg–Marquardt chi-squared minimization algorithm to minimize the residuals between the data and the model, weighted by the error spectrum (e.g., Markwardt 2009). alis uses the atomic data compiled by Morton (2003), with updates from both Jenkins & Tripp (2006) and Murphy & Berengut (2014).

The absorption line profiles for the most metal-poor DLAs typically have a very low velocity dispersion, often with just a single narrow absorption line. In the present work, we have utilized the simplicity of the absorption line profiles exhibited by the most metal-poor DLAs to decouple the relative contributions of turbulent and thermal broadening; thermal broadening depends inversely on the square-root of the atom's mass, whereas turbulent broadening is the same for all gas constituents. Thus, one can measure the kinetic temperature of the absorbing gas cloud directly, provided that:

• 1.  
  The gas in each absorption component is distributed according to a Maxwell–Boltzmann distribution;
• 2.  
  An array of transitions of differing oscillator strengths from a given ion is available to populate the curve-of-growth and thereby measure the total broadening; and
• 3.  
  Absorption lines from at least two atoms of widely differing mass are accessible to measure the thermal contribution to the line broadening. The ideal ions that satisfy these criteria are the light D i with its host of Lyman series transitions in combination with one of the heavier atoms (such as O i, Si ii, or Fe ii) which have a range of transitions with a variety of strengths and are commonly observed in DLAs. In some favorable cases (see Appendix B), it is also possible to solely use a selection of metal absorption lines to obtain a handle on the gas kinetic temperature (Carswell et al. 2012).

In many cases, the combined thermal and turbulent broadening of the metal line profiles is only marginally resolved from the instrumental broadening. An accurate measure of the instrumental profile is therefore required in order to pin down the cloud model of the absorbing gas. Moreover, if the point spread function of the quasar does not uniformly illuminate the spectrograph's slit, the instrumental resolution during the observations will be less than the nominal instrumental resolution measured from the widths of the ThAr arc lines. We overcome this uncertainty by assuming that the instrumental broadening function is closely approximated by a Gaussian, and allow the FWHM of the Gaussian profile to be a free parameter in our modeling procedure. Therefore, our uncertainty in the instrumental profile is folded into the uncertainty in both the cloud model and chemical abundances. Our profile analysis also folds in the uncertainty in the placement of the quasar continuum level, by fitting a low order polynomial to the quasar continuum during the absorption line fitting procedure.

DLAs are generally considered to host a CNM and a WNM in pressure equilibrium (e.g., Wolfire et al. 1995). We have performed a suite of Monte Carlo simulations to test the sensitivity of the above technique to the presence of a two-phase medium. We consider two test cases:

• 1.  
  A system with only metal-absorption lines (J0035−0918), and
• 2.  
  A system with metal, D i, and H i absorption lines (J1419+0829).Using the best-fitting absorption line profiles, we generated a new model where 50 per cent of the gas along the line-of-sight resides in a CNM (100 K), and the remaining gas is in a WNM with the measured kinetic temperature. One hundred Monte Carlo realizations were generated by perturbing these model profiles by the error spectrum of the data. We first perform a fit to each realization that includes both a CNM and a WNM. On the basis of these simulations we conclude that we are unable to recover the properties of the CNM; our technique to measure the kinetic temperature is therefore insensitive to the presence of a CNM. We then performed a second fit to each realization but this time we only included a WNM component in the fit. With this test, we were able to recover the properties of the WNM, without any significant bias. Therefore, if the DLAs that we probe in our sample contain a CNM, the physical properties that we derive are largely unaffected, although we are unable to detect the presence of a CNM. In future studies, it may be possible to identify a CNM in metal-poor DLAs by detecting absorption from molecular hydrogen or C i. Absorption lines from these species are not present in our sample.

The majority of our sample is described in our previous work (Pettini et al. 2008; Cooke et al. 2011b, 2014). In this Appendix, we present our analysis of a newly discovered metal-poor DLA at redshift z_abs ≃ 2.271 toward J1111+1332 and provide an improved reanalysis of the metal-poor DLA reported by Cooke et al. (2011a) at z_abs ≃ 2.340 toward J0035−0918. In both of these systems, the data cover absorption lines with a range of oscillator strengths for several elements between C and Ni. The difference in atomic mass for these elements, combined with the broad range of oscillator strengths for many transitions, has allowed us to decouple the thermal broadening of the line profile (which depends on the atomic mass) from that of turbulent broadening (assumed to be the same for all elements), using metal absorption lines alone.

B.1. J0035−0918, z_abs = 2.340097

B.2. J1111+1332, z_abs = 2.270940

The second system reported here is the z_abs ≃ 2.271 DLA toward the m_r ≃ 17.1 quasar J1111+1332. We first recognized this DLA as an excellent candidate VMP DLA from the relatively high S/N SDSS discovery spectrum, which exhibits strong H i Lyα absorption line together with the apparent absence of associated metal absorption lines. We performed follow-up observations of J1111+1332 with HIRES (Vogt et al. 1994) on 2011 March 24 and 25 (Program ID: A152Hb). We employed the 1.148 arcsec wide C5 decker (with a nominal resolution of v_FWHM = 8.3 km  s⁻¹) and the red cross-disperser. We recorded 6 × 2700 s exposures, covering a wavelength range 4645–8155 with small gaps near 5100 Å and 6700 Å. We used 2 × 2 on-chip binning. The data were reduced with makee,¹⁴ following the standard reduction steps of bias subtraction and flat fielding. The orders were defined and traced using a quartz lamp exposure with a pinhole decker (D5). After the orders were extracted, the data were converted to a vacuum heliocentric wavelength frame, with reference to an exposure of a ThAr hollow-cathode lamp. The final spectrum was combined with UVES_popler.¹⁵ The combined S/N of the data near 6000 Å is ∼60.

Cooke et al. (2013) included the Fe-peak element ratios in this DLA in their study. Here, we report a detailed profile analysis which includes all of the available metal species. In total, we detect the absorption lines for 10 elements and place an interesting upper limit on the Zn abundance. The absorption profiles are very well fit by a single absorption component centered at redshift z_abs = 2.270940 ± 0.000001 with turbulent Doppler parameter b_turb = 2.2 ± 0.1 km  s⁻¹ and gas kinetic temperature T_gas = 13, 000 ± 2000 K. We reproduce the data and our best-fitting model for this system in Figure 8 where the top panel shows the damped Lyα line profile, and best-fitting model for $\log N({\rm H\,\scriptsize {I}})/{\rm cm}^{-2}=20.39\pm 0.04$. The remaining panels are a selection of metal absorption lines. The y-axis scale has been reduced in the six panels showing the weakest absorption lines. Column densities for all detected elements and ion stages in this system are collected in Table 6.

Zoom In Zoom Out Reset image size

Table 6. Ion Column Densities of the DLA in J1111+1332 at z_abs = 2.270940

Ion	Transitions Used	log N(X)/cm−2	[X/H]	[X/Fe]
H i	1215	20.39 ± 0.04	...	...
C ii	1334	14.72 ± 0.09	−2.10 ± 0.10	+0.17 ± 0.09
N i	1134.4, 1134.9, 1199.5, 1200.2, 1200.7	13.53 ± 0.02	−2.69 ± 0.05	−0.42 ± 0.02
O i	1302	15.16 ± 0.07	−1.92 ± 0.08	+0.35 ± 0.07
Al ii	1670	12.37 ± 0.02	−2.46 ± 0.05	−0.19 ± 0.02
Al iii	1854, 1862	11.90 ± 0.03	...	...
Si ii	1190, 1193, 1260, 1304, 1526, 1808	13.95 ± 0.02	−1.95 ± 0.05	+0.32 ± 0.02
Si iii	1206	13.36 ± 0.05	...	...
Si iv	1393, 1402	13.09 ± 0.02	...	...
S ii	1259	13.69 ± 0.09	−1.84 ± 0.10	+0.43 ± 0.09
Cr ii	2062, 2066	12.00 ± 0.09	−2.03 ± 0.10	+0.24 ± 0.09
Fe ii	1121, 1125, 1143, 1144, 1608, 1611, 2260, 2374, 2382	13.59 ± 0.01	−2.27 ± 0.04	...
Ni ii	1317, 1370, 1454, 1751	12.15 ± 0.06	−2.45 ± 0.07	−0.18 ± 0.06
Zn ii	2026	⩽11.08a	⩽ − 1.94	⩽ + 0.33

Note. ^a3σ upper limit.

Download table as:  ASCIITypeset image