Discovery of the most metal-poor damped Lyman-alpha system

TL;DR

This paper reports the discovery of the most metal-poor damped Lyman-alpha system, analyzing its elemental abundances to infer properties of early Population III stars and implications for early galaxy formation.

Contribution

It presents the first detailed abundance analysis of a highly metal-poor DLA, constraining the progenitor star mass and its nucleosynthesis yields, and discusses its significance for early universe evolution.

Findings

Abundances: [C/H] = -3.43, [O/H] = -3.05, [Si/H] = -3.21

Consistent with yields of a 20.5 M_sun metal-free star

Crossed the 'transition discriminant' threshold for low-mass star formation

Abstract

We report the discovery and analysis of the most metal-poor damped Lyman-alpha (DLA) system currently known, based on observations made with the Keck HIRES spectrograph. The metal paucity of this system has only permitted the determination of three element abundances: [C/H] = -3.43 +/- 0.06, [O/H] = -3.05 +/- 0.05, and [Si/H] = -3.21 +/- 0.05, as well as an upper limit on the abundance of iron: [Fe/H] < -2.81. This DLA is among the most carbon-poor environment currently known with detectable metals. By comparing the abundance pattern of this DLA to detailed models of metal-free nucleosynthesis, we find that the chemistry of the gas is consistent with the yields of a 20.5 M_sun metal-free star that ended its life as a core-collapse supernova; the abundances we measure are inconsistent with the yields of pair-instability supernovae. Such a tight constraint on the mass of the progenitor Population III star is afforded by the well-determined C/O ratio, which we show depends almost monotonically on the progenitor mass when the kinetic energy of the supernova explosion is E_exp > 1.5x10^51 erg. We find that the DLA presented here has just crossed the critical 'transition discriminant' threshold, rendering the DLA gas now suitable for low mass star formation. We also discuss the chemistry of this system in the context of recent models that suggest some of the most metal-poor DLAs are the precursors of the 'first galaxies', and are the antecedents of the ultra-faint dwarf galaxies.

Tables

Table 1. Table 1: Column densities for each component of the DLA at z abs = 3.07759 subscript 𝑧 abs 3.07759 z_{\rm abs}=3.07759 towards J0903 + + 2628

X	log $ϵ$(X)${}_{}{}^a{}_{\odot }{}^{}$	Component 1	Component 2	Component 3	Totalb	[X/H]	[X/O]
H i	$12.0$	…	…	…	$20.32\pm 0.05$	…	…
C ii	$8.43$	$13.05\pm 0.03$	$12.99\pm 0.04$	$12.72\pm 0.08$	$13.32\pm 0.03$	$-3.43\pm 0.06$	$-0.38\pm 0.03$
C iv	$8.43$	…	…	…	$\le {12.56}^c$	…	…
O i	$8.69$	$13.74\pm 0.02$	$13.57\pm 0.03$	…	$13.96\pm 0.02$	$-3.05\pm 0.05$	…
Si ii	$7.51$	$12.40\pm 0.01$	$12.23\pm 0.03$	$12.54\pm 0.02$	$12.62\pm 0.01$	$-3.21\pm 0.05$	$-0.16\pm 0.02$
Si iv	$7.51$	…	…	…	$\le {12.10}^c$	…	…
Fe ii	$7.47$	$\le 12.76$	…	…	…	$\le -{2.81}^d$	$\le +{0.23}^e$

Table 2. Table 2: Compilation of DLA C/O and O/H measurements

QSO	$z_{\mathrm{abs}}$	[C/O]	[O/H]	refa
J0035$-$0918	2.34010	$0.08\pm 0.16$	$-2.44\pm 0.07$	1,2
HS0105$+$1619	2.53651	$0.110\pm 0.045$	$-1.776\pm 0.021$	3
J0140$-$0839	3.69660	$-0.30\pm 0.08$	$-2.75\pm 0.15$	4
J0311$-$1722	3.73400	$-0.42\pm 0.11$	$-2.29\pm 0.10$	5
J0903$+$2628	3.07759	$-0.38\pm 0.03$	$-3.05\pm 0.05$	6
Q0913$+$072	2.61829	$-0.36\pm 0.012$	$-2.416\pm 0.011$	3, 7
J0953$-$0504	4.20287	$-0.50\pm 0.03$	$-2.55\pm 0.10$	2
J1001$+$0343	3.07841	$-0.41\pm 0.03$	$-2.65\pm 0.05$	5
J1016$+$4040	2.81633	$-0.21\pm 0.05$	$-2.46\pm 0.11$	7
Q1111$+$1332	2.27094	$-0.18\pm 0.11$	$-1.92\pm 0.08$	1
Q1202$+$3235	4.97700	$-0.33\pm 0.11$	$-2.02\pm 0.13$	8
J1337$+$3153	3.16768	$-0.19\pm 0.11$	$-2.67\pm 0.18$	9
J1358$+$6522	3.06730	$-0.27\pm 0.06$	$-2.335\pm 0.02$	3, 10
J1558$+$4053	2.55332	$-0.06\pm 0.07$	$-2.45\pm 0.06$	7
Q2206$-$199	2.07624	$-0.38\pm 0.04$	$-2.07\pm 0.05$	7
J2155$+$1358	4.21244	$-0.29\pm 0.08$	$-1.80\pm 0.11$	11

Full text

††pagerange: Discovery of the most metal-poor damped Lyman-$\alpha$ system††thanks: Based on observations collected at the W.M. Keck Observatory which is operated as a scientific partnership among the California Institute of Technology, the University of California and the National Aeronautics and Space Administration. The Observatory was made possible by the generous financial support of the W.M. Keck Foundation. –Discovery of the most metal-poor damped Lyman-$\alpha$ system††thanks: Based on observations collected at the W.M. Keck Observatory which is operated as a scientific partnership among the California Institute of Technology, the University of California and the National Aeronautics and Space Administration. The Observatory was made possible by the generous financial support of the W.M. Keck Foundation. ††pubyear: 2015

1 Introduction

Almost every astrophysical environment is contaminated by the nucleosynthesis of stars. To date, only a small handful of pristine environments have been identified, including a mostly ionized cloud of gas at redshift $z\simeq 3$ which may be associated with cold flows (Fumagalli, O’Meara, & Prochaska, 2011), and a mostly neutral cloud of gas at redshift $z\simeq 7$ attributed to either a neutral protogalaxy or the intergalactic medium (Simcoe et al. 2012; but see Bosman & Becker 2015 for an alternative interpretation). If pockets of absolutely pristine gas still exist at redshift $z\sim 3$, there may also be some environments that were enriched exclusively by the first generation of stars (also called Population III, or Pop III, stars).

The nature of Pop III stars, in particular their mass spectrum, is still a matter of debate. Numerical simulations that follow the collapse of primordial material from cosmological initial conditions originally suggested that Pop III stars were predominantly very massive, with typical masses in excess of $100~{}{\rm M}_{\odot}$ (Bromm, Coppi, & Larson, 1999; Nakamura & Umemura, 2001; Abel, Bryan, & Norman, 2002). Modern calculations suggest a somewhat lower typical mass of the first stars, and indicate that a small multiple of Pop III stars are formed in a given minihalo (Clark, Glover, & Klessen, 2008; Turk, Abel, & O’Shea, 2009; Stacy, Greif, & Bromm, 2010; Stacy & Bromm, 2013; Hirano et al., 2014; Stacy, Bromm, & Lee, 2016). Although the form of the mass function has not yet been pinned down, a general conclusion borne out of the above cosmological simulations is that the initial mass function of the first stars is top heavy, with most of the total stellar mass concentrated in stars with masses $M\gtrsim 10~{}{\rm M}_{\odot}$.

In principle, the mass spectrum of the first stars can be inferred observationally, by identifying the chemical fingerprint of the elements that were made during the life of a Pop III star. Finding this chemical signature is a challenging prospect, since a region must be identified that is solely enriched by the first stars. Dedicated surveys to find putative second generation stars in the Milky Way have identified several excellent candidates (see Frebel & Norris 2015 for a review). These searches have also uncovered a striking diversity of chemical abundance patterns, including many metal-poor stars enhanced with light elements relative to heavy elements (collectively known as carbon-enhanced metal-poor [CEMP] stars; for an overview, see Beers & Christlieb 2005), a star with apparently no iron (Keller et al., 2014), and an almost pristine star with roughly solar-scaled chemical abundances (Caffau et al., 2011). The abundance patterns of these stars are successfully reproduced by models of Pop III stellar nucleosynthesis (Heger & Woosley, 2010; Cooke & Madau, 2014; Ishigaki et al., 2014; Marassi et al., 2014; Tominaga, Iwamoto, & Nomoto, 2014), however, there is still some flexibility in the models due to the uncertain explosion mechanism of core-collapse supernovae (see e.g. Janka 2012).

A similar quest has been undertaken to identify the chemical signature of the first stars among the most metal-poor damped Ly$\alpha$ (DLA) systems (Erni et al., 2006; Pettini et al., 2008; Penprase et al., 2010; Cooke et al., 2011a, b; Cooke, Pettini, & Murphy, 2012). DLAs are clouds of mostly neutral gas that are observed in absorption against a bright background source, typically a quasar (for a general review of DLAs, see Wolfe, Gawiser, & Prochaska 2005). DLAs are typically observed at redshift $z\simeq 3$ ($\sim 2$ Gyr after the Big Bang) where many of the rest frame ultraviolet absorption lines of interest are conveniently redshifted into the optical spectral range. The absorption lines associated with the most metal-poor DLAs are typically very weak, owing to the low metal abundance; in such systems, only the most abundant elements can be reliably measured with current telescope facilities (Cooke et al., 2013).

The most metal-poor DLAs are thought to be the antecedents of the lowest mass galaxies (Salvadori & Ferrara, 2012; Webster, Bland-Hawthorn, & Sutherland, 2015; Cooke, Pettini, & Jorgenson, 2015), as well as cold gas streams being accreted onto galaxies (Yuan & Cen, 2016). Such an affiliation marks these DLAs as a promising environment to measure – at high redshift – the chemical abundance pattern of the earliest generation of stars.

In this paper, we present the discovery of the most metal-poor DLA currently known, and discuss the chemistry of this near-pristine environment. In Section 2, we describe the details of our observations and absorption line profile fitting. We discuss the chemistry of this newly discovered DLA in Section 3, and summarise the main conclusions of our work in Section 4.

2 Observations and Analysis

Among the DLAs discovered by the Sloan Digital Sky Survey (Noterdaeme et al., 2009), we first identified the extremely metal-poor DLA towards the $z_{\rm em}=3.22$ quasar J0903+2628 (R.A. = $09^{\rm h}03^{\rm m}33.\!\!^{\rm s}55$, decl. = $+26^{\circ}28^{\prime}36.\!\!^{\prime\prime}3$; $m_{\rm r}=19.0$) based on the strong, damped Ly$\alpha$ absorption feature at redshift $z_{\rm abs}=3.076$, combined with the non-detection of the strongest associated absorption lines of C, O and Si (see Pettini et al., 2008, for a description of the technique).

We observed J0903+2628 with the Keck High Resolution Spectrometer (HIRES; Vogt et al. 1994) on 2016 March 1, 2, 30 for a total exposure time of $9\times 3600$ s, in seeing conditions (full width at half maximum, ${\rm FWHM}\simeq 0.8^{\prime\prime}$) that were well-matched to the chosen slit width ($0.861^{\prime\prime}$; decker C1). The nominal instrument resolution of our setup is $R\simeq 49,000$, assumed to be a Gaussian profile with a full width at half maximum of $v_{\rm fwhm}=6.1$ km s*-1*. We used the blue cross-disperser to cover the wavelength range 3700–6530 Å, with small gaps near 4600 Å and 5600 Å, corresponding to the gaps between the detector mosaic. All frames were binned $2\times 2$ during readout.

The data were processed with the makee data reduction pipeline111makee is available for download from:

http://www.astro.caltech.edu/$\sim$tb/ipac_staff/tab/makee/index.html. This pipeline first subtracts the detector bias level. The locations of the echelle orders are traced using an exposure of a quartz lamp through a pinhole decker (HIRES decker D5). The pixel-to-pixel variations and the blaze function were removed by dividing the science frames by an exposure of a quartz lamp through the science slit (decker C1). A one-dimensional spectrum was optimally extracted from each reduced frame, and the data were finally calibrated to a vacuum and heliocentric wavelength scale. The individual spectra were resampled and combined using the uves_popler code222uves_popler is maintained by Michael T. Murphy, and is available from the following url:

http://astronomy.swin.edu.au/$\sim$mmurphy/UVES_popler. Deviant pixels and ghosts were identified by visual inspection, and masked prior to combination.

The absorption lines were analyzed using the Absorption LIne Software (alis) package333alis is available for download from:

https://github.com/rcooke-ast/ALIS, which uses the atomic data tabulated by Morton (2003). Both the quasar continuum emission and the DLA absorption lines were fit simultaneously. The quasar continuum was modeled with a low order Legendre polynomial locally around each absorption line, while each DLA absorption component was modeled with a Voigt profile. The Ly$\alpha$ profile of the DLA is presented in Figure 1 (black histogram) overlaid with the best fitting Voigt profile (red curve), corresponding to a neutral hydrogen column density of log $N$(H i)/cm${}^{-2}=20.32\pm 0.05$.

We detect metal absorption lines of C ii, O i and Si ii, reproduced in Figure 2 together with the best fitting model profiles. The absorption profiles can be described by three components, located in two distinct velocity intervals. The primary absorption component is located at $z_{\rm abs}=3.077588\pm 0.000002$, and exhibits a turbulent Doppler parameter of $b_{1}=9.4\pm 0.2$ km s*-1*. The two additional ‘satellite’ components are located at a velocity of $\Delta v_{2}\simeq-68.9$ km s*-1* and $\Delta v_{3}\simeq-93.1$ km s*-1* relative to the primary component, with Doppler parameters $b_{2}=10.8\pm 0.5$ km s*-1* and $b_{3}=15.9\pm 0.7$ km s*-1*, respectively. In addition to the turbulent broadening quoted above, we assume that the line profiles of all absorption components are thermally broadened by gas at a kinetic temperature of $T_{\rm kin}=10^{4}$ K (see Cooke, Pettini, & Jorgenson, 2015)444We note that the choice of $T_{\rm kin}=10^{4}$ K does not impact the derived column densities, since the thermal Doppler parameter is much narrower than the turbulent Doppler parameter and, in any case, all of the fitted absorption lines are on the linear part of the curve of growth where the derived column density is independent of the Doppler parameter.. Some of the DLA absorption lines shown in Figure 2 are mildly blended with absorption unrelated to the DLA. We simultaneously fit the DLA absorption and the unrelated absorption, and display the model profile of the unrelated blends in Figure 2, represented by the light purple line.

The column densities of each component of the detected ions are listed in Table 1, and are mostly driven by C ii $\lambda 1334$, O i $\lambda 1302$, and Si ii $\lambda 1260$. We also list a $3\sigma$ upper limit on the Fe ii column density, which is based on the non-detection of Fe ii $\lambda 1260$, integrated over the velocity interval of Component 1. The errors on the column density measurements are computed using the diagonal terms of the covariance matrix. We do not detect any absorption originating from the higher ionization absorption lines of C iv and Si iv. In Table 1 we quote $3\sigma$ upper limits on the column density of these ions. Other high ionization absorption lines of interest (such as C iii $\lambda 977$ and Si iii $\lambda 1206$) are blended with the Ly$\alpha$ forest.

We note that Component 3 is only detected in Si ii $\lambda 1260$ and is marginally detected in both C ii lines; this component is not observed in O i. Thus, this component may be an unfortunate blend of unrelated features, not physically associated with the DLA.. However, if this component is associated with the DLA, it probably arises in a region of ionized gas, since no absorption is detected in O i. For these reasons, we consider only the column density from absorption components 1 and 2 when assessing the relative element composition of this DLA. This ensures that we only consider the mostly neutral gas, where the observed ions are all the dominant ionization stage, thus obviating the need to perform ionization corrections.

3 Results

3.1 Metal-poor chemistry

The DLA reported here is the most metal-poor system currently known, and has an oxygen abundance ([O/H] $=-3.05$) a factor of 2 lower than the next most metal-poor DLA (reported recently by Cooke et al. 2016; [O/H] $=-2.85$). Similarly, this system exhibits the lowest C and Si abundance of any known DLA. Other extremely metal-poor quasar absorption line systems are known with lower H i column density (known as Lyman limit systems, LLSs), including two systems with no detectable metals (${\rm[Z/H]}\lesssim-4.0$; Fumagalli, O’Meara, & Prochaska 2011) and two systems with comparable (lower) metal abundances to the DLA reported here (Crighton, O’Meara, & Murphy, 2016; Lehner et al., 2016). Given that the abundance measurements of LLSs require an ionization correction, DLAs tend to afford somewhat higher precision abundances. This, in turn, allows for a more informative test of nucleosynthesis models.

The relative element abundances555In what follows, we assume that the metals and hydrogen of this DLA are uniformly mixed. This assumption receives some support from observations of three very metal-poor DLAs (Q0913$+$072, J1419$+$0829, J1558$-$0031) which exhibit two absorption components that share a similar metallicity and abundance pattern (see Table 2 of Cooke, Pettini, & Jorgenson 2015; see also, Prochaska 2003). of this DLA are [C/O] $=-0.38\pm 0.03$ and [Si/O] $=-0.16\pm 0.02$. These abundances are somewhat lower than the typical values seen in other very metal-poor DLAs (i.e. DLAs with [Fe/H] $\leq-2.0$), which have [$\langle$C/O$\rangle$] $=-0.28\pm 0.12$ and [$\langle$Si/O$\rangle$] $=-0.08\pm 0.10$ (Cooke et al., 2011b), where the quoted uncertainty represents the $1\sigma$ spread in the very metal-poor DLA population. We also find that the [Si/O] abundance is identical between Components 1 and 2, whereas the [C/O] abundance of these components differs by 0.11 dex (a $1.8\sigma$ difference). This difference is still well within the scatter of all DLA measurements (see Section 3.3).

We first compare the abundance pattern of this DLA to nucleosynthesis models of very massive metal-free stars ($M\simeq 140-260~{}{\rm M}_{\odot}$) that ended their life as pair-instability supernovae (Heger & Woosley, 2002). In these models, stars with a mass in this range produce element yields of $-0.66\lesssim{\rm[C/O]}\lesssim-0.60$ and $+0.20\lesssim{\rm[Si/O]}\lesssim+0.80$, which are inconsistent with the values we measure. It thus seems unlikely that a pair-instability supernova enriched this DLA.

We also compare the abundance pattern of this DLA to nucleosynthesis models of massive metal-free stars that ended their lives as type-II core-collapse supernovae (Heger & Woosley, 2010). This set of models consists of 120 simulated stars covering a mass range $M=10-100~{}{\rm M_{\odot}}$, with a mass resolution of $\Delta M\gtrsim 0.1~{}{\rm M_{\odot}}$. Since the explosion mechanism of a type-II supernova is still uncertain, these models adopt a parameterised ‘mixing and fallback’ scheme. The mixing between stellar layers during the explosion is achieved by ‘smoothing’ the star over the mass (i.e. radial) coordinate 4 times with a boxcar filter of a specified width666Heger & Woosley (2010) adopt this parameterisation as it provides a good fit to the hard X-ray and optical light curves of SN 1987A.; a grid of 14 different widths are considered by Heger & Woosley (2010), where the mixing width is defined as a fraction of the He core size. Then, the explosion is simulated as a moving piston that deposits momentum at a specified mass coordinate. In what follows, we consider the standard case recommended by Heger & Woosley (2010), which places the piston near the base of the oxygen burning shell, where the entropy per baryon $S\simeq 4\,k_{\rm B}$, where $k_{\rm B}$ is the Boltzmann constant. The explosion is parameterised by the kinetic energy of the material that escapes the binding energy of the star, $E_{\rm exp}$ (hereafter referred to as the explosion energy). These models consider a grid of 10 values of the explosion energy for each star, in the range $(0.3-10)\times 10^{51}~{}{\rm erg}$. Thus, this model suite comprises a total of 16 800 combinations of the three parameters: Stellar mass, explosion energy, and mixing width. The ranges of the model parameters are observationally motivated (see the discussion by Heger & Woosley 2010), and encompass all realistic model values.

To find the range of parameters that are an acceptable solution to the abundance pattern of the DLA reported here, we have linearly interpolated this three-dimensional space, and conducted a Markov chain Monte Carlo analysis to search for the most likely set of parameters, using the emcee software (Foreman-Mackey et al., 2013). In Figure 3, we plot the one- and two-dimensional projections of the samples to identify the covariance between model parameters. (dark and light shades in the 2D projections represent the 68 and 95 per cent confidence contours, respectively). We find that the progenitor mass is well-determined, centered on a value $M=20.5~{}{\rm M_{\odot}}$, while the supernova explosion energy tends towards the upper end of the range considered, with a favoured value $E_{\rm exp}\sim 6$–$8\times 10^{51}$ erg. The mixing parameter is unconstrained.777Since we only have two measured relative abundances to fit three model parameters, we are unable to fully constrain the best-fitting model parameters. The mixing parameter cannot be determined with the currently available data.

It may seem surprising that the progenitor mass of the model Population III star is so well-determined relative to the other parameters of the model. This strong bound is largely due to the well-determined [C/O] value. The most abundant isotopes of C and O (i.e. 12C and 16O) are primarily formed by helium burning, with some 16O resulting from neon burning (see e.g. Woosley & Weaver, 1995). These models suggest that the [C/O] abundance is largely insensitive to the details of the explosion when $E_{\rm exp}\gtrsim 1.5\times 10^{51}~{}{\rm erg}$, and is almost uniquely dependent on the mass of the progenitor star (see top panel of Figure 4). When the explosion energy is lower than this value, a higher fraction of 16O (relative to 12C) falls back onto the compact remnant, thereby causing the C/O ratio to increase (see e.g. Figure 4 of Heger & Woosley, 2010). This is matched by a decrease in the Si/O ratio, since more Si falls back relative to O.

The monotonicity of the relationship between the [C/O] abundance and the progenitor mass (as well as the invariance of [C/O] with $E_{\rm exp}$) breaks down above $\sim 35~{}M_{\odot}$, where higher C/O values are recovered by the models. We note that model stars with a mass above $\sim 35~{}M_{\odot}$ are ruled out by the measured [Si/O] abundance of the DLA reported here (see middle panel of Figure 4).

Given enough data, we speculate that the strong monotonic dependence of the C/O ratio on progenitor mass might allow a robust measure of the Pop III initial mass function. The main uncertainty governing the calculated [C/O] ratio is the 12C($\alpha,\gamma)^{16}$O reaction rate that is used as input into the stellar models; the value of this rate adopted by Heger & Woosley (2010) is consistent with the latest empirical determination reported by An et al. (2016). However, we caution that other effects such as rotation (Hirschi, 2007; Ekström et al., 2008; Joggerst, Woosley, & Heger, 2009), mass loss via rotation/winds (Meynet & Maeder, 2002), and the three-dimensional modelling of metal-free stars (Joggerst, Almgren, & Woosley, 2010) may alter the dependence of C/O on the progenitor mass.

The somewhat weaker bound on $E_{\rm exp}$ shown in Figure 3 is derived from the measured [Si/O] ratio (see middle panel of Figure 4). We note that a tighter bound on the explosion energy could be afforded by the [Fe/O] ratio (bottom panel of Figure 4). Unfortunately, the currently measured upper limit on this ratio ([Fe/O] $\leq+0.23$), is unable to provide an informative bound on $E_{\rm exp}$. Since the [Fe/O] abundance is also highly sensitive to the chosen mixing parameter, a measurement of the [Ni/Fe] abundance is needed to break the degeneracy (see Cooke et al. 2013). These measurements might become possible with the next generation of 30+ m telescope facilities (see Section 3.2).

Finally, we have only two relative element abundances with which to constrain the nucleosynthesis models, and we are unable at this stage to rule out the possibility that this system may be contaminated by other forms of nucleosynthesis, including Population II core-collapse supernovae. At a redshift of $z_{\rm abs}=3.07759$, the Universe has aged by $\simeq 1.5$ Gyr after the epoch of the first stars ($z\simeq 10-15$; e.g. Maio et al. 2010), and there is enough time for a second generation of stars to have already operated in this DLA. It is nevertheless encouraging that the chemistry of this one DLA is in good agreement with the chemistry of absorption line systems at $z_{\rm abs}\simeq 6$ (Becker et al., 2012), which are captured only a few hundred Myr after the putative epoch of the first stars. It is also worth noting that the low metallicity of this DLA is consistent with the metallicity regime expected for gas that is enriched solely by metal-free Population III stars (Smith & Sigurdsson, 2007; Bromm & Yoshida, 2011; Wise et al., 2012; Cooke & Madau, 2014; Webster, Bland-Hawthorn, & Sutherland, 2015; Ritter et al., 2016).

3.2 The antecedents of metal-poor stars and the ultra-faint dwarf galaxies

According to the Stellar Abundances for Galactic Archaeology (SAGA) database888The SAGA database is maintained by Takuma Suda, Yutaka Katsuta, and Shimako Yamada, and is available at the following address:

http://sagadatabase.jp/wiki/doku.php (Suda et al., 2008), the only star currently known with a lower [C/H] abundance than the DLA reported here is the Leo star (Caffau et al., 2011). The remaining stars in the SAGA database are either red giant branch stars999As a star ascends the red giant branch (RGB), carbon that has been processed through the CN cycle is mixed to the stellar surface, reducing the surface abundance of C (e.g. Gratton et al., 2000). As a result, the [C/H] abundance that an upper RGB star was born with can only be recovered by applying a correction based on stellar evolution models (Placco et al., 2014). After correcting for this effect, Placco et al. (2014) find $\sim 10$ RGB stars that exhibit a corrected [C/H] value that is lower than the DLA we report here. or display a higher abundance of [C/H]. The star with the closest abundances to the DLA we report here is SDSS J0259$+$0057 (Aoki et al., 2013), which has a carbon abundance [C/H] $=-3.33\pm 0.22$, and is not carbon-enhanced ([C/Fe] $=-0.02\pm 0.22$).

If the DLA we consider here has a chemical abundance pattern that is similar to the Milky Way halo star SDSS J0259$+$0057, we estimate that the Fe abundance of the DLA would be ${\rm[Fe/H]}\simeq-3.4$. The Fe abundance of the DLA can also be estimated by subtracting the typical Si/Fe abundance of very metal-poor DLAs ([Si/Fe] $=+0.32\pm 0.09$; Cooke et al. 2011b) from the observed [Si/H] abundance, which would give ${\rm[Fe/H]}\simeq-3.5$. We note, however, that the [Fe/H] abundance of this DLA may be considerably lower than these estimates; there exist some Milky Way halo stars with slightly higher [C/H] that exhibit a strong overabundance of [C/Fe], collectively known as CEMP stars (see e.g. Beers & Christlieb, 2005). Measuring the detailed chemical abundance pattern of this DLA (including the abundances of N, Mg, Al, and Fe) will become feasible with the next generation of 30+ m telescope facilities. Assuming that the abundances of these elements are in the same proportions as a typical very metal-poor DLA (see Table 13 of Cooke et al. 2011b), this goal could be achieved with a spectrum of ${\rm S/N}~{}\simeq~{}150$.101010The next detectable elements include S and Ni, which would require a ${\rm S/N}~{}\simeq~{}600$ and $2000$, respectively, to detect an individual absorption line at $5\sigma$ confidence. Stacking the available Ni ii lines (see Cooke et al., 2013) could reduce this requirement to a ${\rm S/N}~{}\simeq~{}1000$.

As stated above, there is only one star currently known that is more carbon-poor than the DLA reported here. We now speculate how many DLAs might need to be observed before a DLA with an abundance as low as the Leo star may be found. The SAGA database lists 7 stars with a carbon abundance [C/H] $\leq-3.0$, including the Leo star which has an abundance [C/H] $\leq-3.8$. There are currently four DLAs with a carbon abundance [C/H] $\leq-3.0$ (see Section 3.3 and Table 2); thus, doubling the number of DLAs in this extremely metal-poor regime may be sufficient to uncover a DLA of similar metallicity to the Leo star.

Given the metal paucity of the DLA reported here, it is conceivable that just a single (Population III) star is responsible for the metal enrichment. Indeed, hydrodynamic models that trace the internal feedback111111Including both radiative feedback from massive stars and the subsequent supernova feedback. within idealized dwarf galaxies, do entertain this idea (Webster, Bland-Hawthorn, & Sutherland, 2015; Bland-Hawthorn, Sutherland, & Webster, 2015). In this scenario, the DLA reported here would be considered an immediate progenitor of a “first galaxy” that is poised to form its first generation of low mass stars (see also, Section 3.3). This scenario receives some support from abundance measurements of stars in the lowest luminosity and most metal-poor galaxies in the Local Group – the so-called ultra-faint dwarf galaxies (UFDs; Kirby et al. 2008; Norris et al. 2010; Simon et al. 2011; Lai et al. 2011; Vargas et al. 2013; Brown et al. 2014); the presumed [Fe/H] abundance of the DLA reported here ([Fe/H] $\simeq-3.5$) is comparable to the most Fe-poor stars in the UFDs.121212We note that a direct comparison between the [Si/H] abundance of this DLA and the stars in UFDs is not possible, since [Si/H] has not been measured for the most Fe-poor stars in UFDs (Vargas et al., 2013).

3.3 Metallicity evolution of the C/O ratio

We now extend this discussion to the metallicity evolution of the C/O ratio. In Figure 5, we present the latest complement of C/O and O/H measurements available in the literature. These include abundance measurements of stars (Bensby & Feltzing, 2006; Fabbian et al., 2009; Nissen et al., 2014) in the Milky Way thin and thick disc (solid and open black circles respectively) and in the Milky Way halo (black squares). To this we add C/O measurements of local H ii regions (Dufour, Shields, & Talbot, 1982; Garnett et al., 1995; Kurt et al., 1999; Esteban et al., 2002, 2009, 2014; García-Rojas & Esteban, 2007; López-Sánchez et al., 2007; Berg et al., 2016), using either optical recombination lines or ultraviolet collisionally excited lines (blue diamonds and squares respectively).131313We apply a correction of $+0.24$ dex to the [O/H] abundance measured using collisionally excited lines (Esteban et al., 2014; Steidel et al., 2016). Finally, we overplot measurements of C/O and O/H in the lowest metallicity DLAs (red triangles) and in the new DLA reported here (green triangle). For convenience, in Table 2 we provide all of the DLA measurements used in this work, scaled to our adopted solar abundance scale (Asplund et al., 2009, see also the second column of Table 1).

All of the observations displayed in Figure 5 are generally in good mutual agreement, regardless of the different environments probed and the different techniques used. There are, however, a few DLAs and H ii regions that exhibit somewhat enhanced C/O values relative to the locus defined by Milky Way stars. Similar enhancements have also been seen in the [C/$\alpha$] ratios measured in some LLSs (Lehner et al. 2016; see also, Lehner et al. 2013). The origin of such enhancements is unclear at present, but could be related to the production of C by intermediate mass stars (Sharma et al., 2016).

As discussed previously in the literature (Henry, Edmunds, & Köppen, 2000; Carigi, 2000; Akerman et al., 2004; Cescutti et al., 2009; Romano et al., 2010), the increase of [C/O] when [O/H] $\gtrsim-1.0$ is thought to be due to the metallicity dependent winds of massive stars combined with the delayed release of C from low and intermediate mass stars. The upturn of [C/O] when [O/H] $\lesssim-1.5$, on the other hand, is still open to debate. Akerman et al. (2004, see also, ) propose that the increased C/O values seen at low O/H is the result of an increased carbon yield from metal-free Population III stars. This scenario has previously been used to explain the lowest metallicity population of CEMP stars (Umeda & Nomoto, 2003; Ryan et al., 2005; Cooke & Madau, 2014).

In addition to the C/O upturn at low O/H, we also draw attention to the scatter in the DLA [C/O] measurements. With the improved statistics of the data collected in Table 2, we now see that the DLA values of [C/O] at a given [O/H] differ by more than their errors, suggesting an intrinsic dispersion in these elements relative abundances that was not immediately apparent in our earlier studies of carbon and oxygen in very metal-poor DLAs (Pettini et al., 2008; Cooke et al., 2011b). The total range in [C/O] exhibited by DLAs with [O/H] $\lesssim-2.0$ spans a factor of $\sim 4$, from [C/O] $\simeq-0.5$ to $\simeq+0.1$. There are several possibilities that could explain such a scatter. As shown in the top panel of Figure 4 (see also, Section 3.1), the [C/O] ratio is sensitive to the mass of the stars that enriched the DLAs. If the first stars formed in isolation or in small multiples, as in currently favoured scenarios (Clark, Glover, & Klessen, 2008; Turk, Abel, & O’Shea, 2009; Stacy, Greif, & Bromm, 2010; Stacy & Bromm, 2013; Hirano et al., 2014; Stacy, Bromm, & Lee, 2016), the intrinsic C/O scatter could be explained if these DLAs were enriched by a single or a small multiple of Pop III stars. The predicted mass range of the stars responsible for the enrichment of these metal-poor DLAs is $10\lesssim M/{\rm M_{\odot}}\lesssim 25$, with a bias towards higher mass stars. An alternative possibility, suggested recently by Sharma et al. (2016), is that the C/O scatter is due to enrichment by a combination of massive and intermediate mass (Population II) stars; the high C/O values are the result of preferential enrichment by intermediate mass stars while the low C/O values are primarily due to massive stars. The observed C/O scatter is the result of poor mixing between these two channels of carbon production.

We point out that massive Population II stars alone are less likely to produce the observed C/O scatter than Population III stars, even though the relationship between C/O and progenitor star mass is nearly identical for Population III stars and stars with progenitor metallicity $Z=10^{-4}~{}Z_{\odot}$ (see the black curve in the top panel of Figure 4). This is because the stellar initial mass function of Population II stars is more fully sampled, unlike the sparsely sampled initial mass function of Population III stars, as discussed above. Likewise, we suggest that the scatter of C/O measurements in the most metal-poor DLAs might provide a measure of the stellar initial mass function and multiplicity of the first stars. Measuring the intrinsic scatter of C/O in almost pristine environments should be considered a key goal of future observations.

In Figure 5, we also overlay the ‘transition discriminant’ (Frebel, Johnson, & Bromm, 2007, see also Bromm & Loeb 2003), shown by the solid grey shaded region, which marks the transition from a predominantly high mass generation of (Population III) stars to a low mass generation of (Population II) stars. This criterion is based on the fine structure cooling lines of [C ii] and [O i]. Once a critical abundance of C and O is reached in a cloud of cold neutral gas (i.e. to the right of the grey band shown in Figure 5), the gas is able to fragment to low mass scales and produce a long-lived generation of stars. In principle, DLAs are not restricted to the right of this region; a DLA that overlaps this region would be an ideal environment to empirically study the transition from Population III to Population II star formation. The DLA reported here lies very close to this region,141414The DLA reported here exhibits a value of the transition discriminant, ${D_{\rm trans}=-3.19\pm 0.04}$, which lies just above the critical value calculated by Frebel, Johnson, & Bromm (2007), $D_{\rm trans,crit}=-3.5\pm 0.2$. and appears to have just crossed the threshold for low mass star formation.

Finally, given that: (1) The gas in DLAs is mostly neutral (i.e. conducive to star formation); (2) this particular DLA has just crossed into a chemical regime where a generation of long-lived stars is expected to form; and (3) the presumed [Fe/H] abundance of the DLA reported here is consistent with the most Fe-poor stars seen in UFDs, we suggest that this DLA could be representative of one of the antecedents of the UFD galaxy population, as discussed previously in Section 3.2.

4 Summary and Conclusions

We report the discovery of the most metal-poor damped Lyman-$\alpha$ system currently known, located at a redshift $z_{\rm abs}=3.07759$, based on observations of the quasar J0903$+$2628 taken with the Keck HIRES spectrograph. This DLA was identified as part of our ongoing programme to search for the nucleosynthetic imprints of the first stars in neutral gas at high redshift. We draw the following main conclusions:

(i) The extremely low metallicity of this DLA has allowed only the most abundant chemical elements to be detected. Based on Voigt profile fitting, we deduce the following abundances: [C/H] $=-3.43\pm 0.06$, [O/H] $=-3.05\pm 0.05$, and [Si/H] $=-3.21\pm 0.05$, and an upper limit on the iron abundance of [Fe/H] $\leq-2.81$. This DLA is a factor of $\sim 2$ more oxygen-poor than the next most metal-poor DLA known (Cooke et al. 2016; [O/H] $=-2.85$).

(ii) This DLA exhibits an abundance of carbon that is lower than any metal-poor star currently known, other than the Leo star (Caffau et al., 2011). The low metallicity is consistent with the view that this DLA may have been enriched solely by the products of a first generation of stars.

(iii) We have compared the relative chemical abundances of this DLA to nucleosynthesis calculations of massive metal-free stars that ended their lives as core-collapse supernovae. We show that the C/O yield of massive metal-free stars is a strong (almost monotonic) function of the progenitor mass, and is virtually independent of the other parameters in the nucleosynthesis modelling when the explosion energy is $E_{\rm exp}\gtrsim 1.5\times 10^{51}~{}{\rm erg}$. Using these models, we estimate that the DLA was enriched by a star of mass $M\simeq 20.5~{}{\rm M}_{\odot}$. We also rule out the possibility that the enrichment was due to a star that ended its life as a pair-instability supernova.

(iv) The carbon and oxygen abundances of this DLA yield a measure of the transition discriminant, $D_{\rm trans}=-3.19\pm 0.04$, which is just over the critical threshold for low mass star formation $D_{\rm trans,crit}=-3.5\pm 0.2$ (Frebel, Johnson, & Bromm, 2007). Given the large column density of neutral gas hosted by the DLA, which may be conducive to star formation, we propose that this environment might represent an immediate precursor to the formation of a ‘first galaxy’ (i.e. a dark matter halo that has not yet formed a generation of long-lived, low mass stars).

(v) We compile all available literature determinations of the C/O abundance in very metal-poor DLAs. The scatter of these measurements is larger than the measurement uncertainties, indicating that there is an intrinsic dispersion in the population. Given the sensitivity of the C/O ratio to progenitor mass, we propose that the scatter can be explained if a single or small multiple of metal-free stars is responsible for the enrichment of extremely metal-poor DLAs. We also propose that the distribution of C/O measurements of extremely metal-poor DLAs can be used to determine both the stellar initial mass function and the multiplicity of the first stars.

(vi) Recent numerical models have also suggested that the most metal-poor DLAs may have been enriched by just a single massive star, and could have similar properties to the progenitors of the Milky Way ultra-faint dwarf galaxies (Salvadori & Ferrara, 2012; Webster, Bland-Hawthorn, & Sutherland, 2015). We note that the presumed [Fe/H] of the DLA reported here is consistent with the most Fe-poor stars found in the UFD galaxies. A key goal of future work will be to pin down the iron abundance of the DLA reported here and of other extremely metal-poor DLAs.

Our work highlights the importance of identifying and measuring the detailed chemical abundance patterns of the most metal-poor DLAs. The high precision and accuracy that can be achieved for the relative chemical abundances ($\lesssim 0.05$ dex), allow for a strong and informative test of nucleosynthesis models of metal-free stars. In particular, the C/O ratio offers a sensitive conversion to the progenitor mass of the metal-free stars that may have enriched the most metal-poor DLAs. With future telescope facilities, it may also be possible to measure the [Ni/Fe] ratio to comparable precision, thereby allowing a strong bound on the supernova explosion energy (Cooke et al., 2013). Finally, we estimate that by merely doubling the number of DLAs with a carbon abundance [C/H] $\leq-3.0$ (from 4 to 8 DLAs), it is statistically possible to find a DLA that is as metal-poor as the Leo star. Furthermore, the future prospect of finding extremely metal-poor (and possibly, pristine) DLAs at high redshift is very promising; the steep slope of the quasar luminosity function combined with the collecting area of the 30+ m telescopes, will ensure access to $\sim 100$ times more quasars than those that are within reach of current facilities (see the discussion by Cooke et al. 2016).

Acknowledgements

We are grateful to the staff astronomers at Keck Observatory for their assistance with the observations. We thank the anonymous referee for their prompt review, and for offering several helpful suggestions that improved the presentation of this paper. During this work, R. J. C. was supported by a Royal Society University Research Fellowship, and 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. CCS has been supported by grant AST-1313472 from the U.S. NSF. This research was also supported by a NASA Keck PI Data Award, administered by the NASA Exoplanet Science Institute. Data presented herein were obtained at the W. M. Keck Observatory from telescope time partially allocated to the National Aeronautics and Space Administration through the agency’s scientific partnership with the California Institute of Technology and the University of California. Our work made use of the matplotlib (Hunter, 2007), emcee (Foreman-Mackey et al., 2013), and corner (Foreman-Mackey et al., 2016) python packages, which we gratefully acknowledge. The Observatory was made possible by the generous financial support of the W. M. Keck Foundation. We thank the Hawaiian people for the opportunity to observe from Mauna Kea; without their hospitality, this work would not have been possible.