The Cosmic Evolution of the Metallicity Distribution of Ionized Gas Traced by Lyman Limit Systems
We present the first results from our KODIAQ Z survey aimed to determine the metallicity distribution and physical properties of the partial and full Lyman limit systems (pLLSs and LLSs; ), which are probed of the interface regions between the intergalactic medium (IGM) and galaxies. We study 31 HI-selected pLLSs and LLSs at observed with Keck/HIRES in absorption against background QSOs. We compare the column densities of metal-ions to HI and use photoionization models to assess the metallicity. The metallicity distribution of the pLLSs/LLSs at is consistent with a unimodal distribution peaking at . The metallicity distribution of these absorbers therefore evolves markedly with since at it is bimodal with peaks at and . There is a substantial fraction (25–41%) of pLLSs/LLSs with metallicities well below those of damped Ly absorbers (DLAs) at any studied from to –4, implying reservoirs of metal-poor cool, dense gas in the IGM/galaxy interface at all . However, the gas probed by pLLSs and LLSs is rarely pristine, with a fraction 3–18% for pLLSs/LLSs with . We find C/ enhancement in several pLLSs and LLSs in the metallicity range , where C/ is 2â5 times larger than observed in Galactic metal-poor stars or high redshift DLAs at similar metallicities. This is likely caused by preferential ejection of carbon from metal-poor galaxies into their surroundings.
Modern theory and simulations agree that the star formation of galaxies and the properties of their circumgalactic medium (CGM, defined here as the gas between the inner regions of galaxies and the diffuse intergalactic medium, IGM) should be intimately connected. This is especially true for the dense flows through the CGM: feedback from star formation is understood to drive outflows that carry mass and metals away from galaxies, while infall from the IGM is thought to bring in fresh gas to fuel on-going star formation. In fact, each of these is a necessary component for our current understanding of galaxy evolution. Without significant feedback, most baryons would cool into the centers of halos to form prodigious quantities of stars (e.g., White & Rees, 1978; Kereš et al., 2009), but with feedback, the baryon content of stars and cold gas in galaxies can be matched (% of their cosmic baryons; e.g., Fukugita et al., 1998; Conroy & Wechsler, 2009) by driving matter into the CGM and beyond. Similarly, without continued infall of IGM material, star-forming galaxies would consume their interstellar gas in 1 Gyr (e.g., Genzel et al., 2010; Prochaska et al., 2005). The absence of star formation in some galaxies may be explained by the strangulation of IGM infall, wherein the hot ambient coronal matter in high-mass galaxies is sufficient to heat the infalling gas to temperatures that make it unavailable for immediate star formation (Dekel & Birnboim 2006; Kereš & Hernquist 2009).
These exchanges of matter, both in and out, through the CGM thus play critical roles in the evolution of galaxies. The competition between these large-scale inflows and outflows and its behavior with galactic mass is thought to shape such disparate properties of galaxies as the galactic mass-metallicity relation, the galaxy color bimodality, the maintenance of star formation in galaxies over billions of years, and the (stellar) baryonic mass fraction of galaxies (e.g., Kereš et al., 2005; Dekel & Birnboim, 2006; Faucher-Giguère et al., 2011). It has, however, been difficult to verify these predictions. There is good reason to believe feedback-driven outflows are important carriers of mass and metals through the CGM since ubiquitous outflows are observed toward galaxy centers (e.g., Pettini et al., 2001; Shapley et al., 2003; Steidel et al., 2004, 2010; Weiner et al., 2009; Rubin et al., 2014). The COS-Halos and COS-Dwarfs surveys have demonstrated that the CGM is a massive reservoir of galactic metals, with galaxies having ejected at least as much metal mass as they have retained (Tumlinson et al. 2011; Werk et al. 2014; Peeples et al. 2014; Bordoloi et al. 2014, and see also, e.g., Stocke et al. 2013; Liang & Chen 2014; Lehner et al. 2015 for other works). Similarly, characterizing the infall of matter requires that the accreting gas is first found. It is not often seen in absorption against the galaxies themselves (e.g., Martin et al., 2012; Rubin et al., 2012) and has been difficult to observe directly in the CGM.
To study the relationship between galaxy and CGM properties requires the development of methods for identifying gas infall, outflows, or other phenomena. Our team has approached this problem by using absorption lines toward background QSOs, searching for CGM gas with an HI selection technique and determining the gas metallicity as a “tracer” of the origin(s) of the gas (Ribaudo et al., 2011; Fumagalli et al., 2011b, a; Lehner et al., 2013). The selection based only on its HI column density avoids biases that can be present with metal-line selection (e.g., via MgII absorption). We target absorbers with a detectable break at the Lyman limit and/or with the Lyman series so that the HI column density is in the interval . These are known as the partial Lyman limit systems (pLLS, defined in this work as ) and LLSs (defined in this work as ). The reasons for targeting these absorbers are twofold. First, in cosmological simulations, the LLSs have been shown to be good tracers of cold flows at –3 (e.g., Fumagalli et al., 2011b, 2014; Faucher-Giguère et al., 2011, 2015; van de Voort et al., 2012). Second, Empirically, at , the pLLSs and LLSs have been associated with the dense CGM (Lanzetta et al., 1995; Penton et al., 2002; Bowen et al., 2002; Chen et al., 2005), and in particular for each specific pLLS and LLS with some galaxy information, they have been found well within the virial radius of galaxies (typically at impact parameter kpc, (Lehner et al. 2013, hereafter L13). Higher redshift studies can only observe the most luminous galaxies, but notably the Keck Baryonic Structure Survey (KBSS) shows that at –3 there is a strong incidence of absorbers with with galaxies at transverse physical distance kpc and velocity separation between the absorber and galaxy redshifts , but not for the lower absorbers (Rudie et al., 2012). The same survey also found that nearly half of the absorbers with are found in the CGM of (massive) galaxies, which also implies that some of the absorbers (especially the pLLSs) may probe more diffuse gas or the CGM of less massive galaxies at high . In any case, at all , by definition of their HI column densities, the pLLSs/LLSs are at the interface between the IGM probed by Ly forest (LYAF) absorbers with and virialized structures traced by super-LLSs (SLLS; ) and damped Ly absorbers (DLAs; ).
Recently, we have shown that the dense CGM of galaxies traced by pLLSs and LLSs has a bimodal metallicity distribution function (MDF) with two well-separated peaks at and and with about equal proportions in each branch (L13). We have now doubled the initial sample of pLLSs and LLSs at and found the same MDF (Wotta et al. 2016, hereafter W16). However, as shown in W16, the bimodal nature of the MDF is dominated by the pLLS population and may start to transition to a unimodal distribution in the LLS regime. As argued in these papers, the metal-rich branch must trace expelled matter: galactic winds, recycled outflows, and tidally-stripped gas, i.e., it traces gas that has been in a galaxy previously in view of the relatively large metal enrichment of the gas. On the other hand, the metallicities of pLLSs and LLSs in the metal-poor branch are extremely low for the universe, lower than the metallicities of dwarf galaxies accreting onto central massive galaxies (e.g., Skillman et al., 1989; Tremonti et al., 2004; Nicholls et al., 2014; Jimmy et al., 2015) and much lower than the lowest metallicities observed for the typical DLAs at similar redshift (L13; W16). These metal-poor LLSs appear to have all the properties of those expected for infalling matter, including the temperature, ionization structure, kinematic properties, and metallicity (Fumagalli et al., 2011b; van de Voort et al., 2012; Shen et al., 2013).
Having identified low-metallicity gas in the halos of galaxies at low redshift, we now want to determine how the metallicity of the pLLSs and LLSs evolves with and at using the same selection criteria and method to derive the metallicity. This program directly builds on our Keck Observatory Database of Ionized Absorbers towards Quasars (KODIAQ) survey (Lehner et al., 2014; O’Meara et al., 2015), which has used the NASA Keck Observatory Archive (KOA) to characterize the properties of the highly ionized gas associated with pLLSs and LLSs. With our new KODIAQ Z program, we will expand this effort to now determine the MDF and physical properties of the pLLSs and LLSs at in an unprecedently large sample.
In this paper, we present the results from a pilot study from a subset of the KODIAQ Z sample with the goal to assemble a sample of pLLSs and LLSs at with a similar size as in L13 at . The total sample consists of 32 HI selected pLLSs and LLSs (19 pLLSs and 13 LLSs); the statistical sample for the metallicity distribution analysis is 31 (18 pLLSs and 13 LLSs; two pLLSs having similar metallicity and are only separated by 50 in the redshift rest-frame of the absorbers). We emphasize that our study contrasts from the recent HD-LLS survey at (Prochaska et al. 2015; Fumagalli et al. 2016b, hereafter FOP16) or from the survey of low-metallicity LLSs at (Cooper et al., 2015; Glidden et al., 2016). The HD-LLS survey targets HI-selected LLSs and SLLSs with at –3.0, but only 9 LLSs have , while all the others have . Similarly the Cooper et al. study also targeted a sample of 17 high LLS (typically ), but selected them on the absence of metal absorption in Sloan Digital Sky Survey (SDSS) spectra, i.e., they targeted a priori low-metallicity LLSs. These programs are therefore complementary to ours and we will use their results for comparison with our samples.
Our paper is organized as follows. In §2 we describe the new and archival pLLS and LLS samples. In §3, we describe the different steps to estimate the metallicities of the absorbers with additional technical details (including the description of each absorber) provided in the Appendix for interested readers. Our main results are presented in §§4 and 5 where we discuss the metallicity distribution of the pLLSs and LLSs at and the evolution of their properties. In §6 we discuss some of the implications of our new observational results. Finally, in §7 we summarize our main results.
2 Data, sample selection and definition
With this pilot study, we assemble a sample of pLLSs and LLSs at similar in size and coverage to the original sample of pLLSs and LLSs in L13. Our final sample for this study consists of 25 new HI-selected absorbers with and 7 from the literature with . We note that some of the high absorbers in the new sample were part of the LLS survey by Steidel (1990), but, in the present work, all the HI and metal column densities were estimated using high resolution Keck spectra; the Steidel’s study used much lower (35–80 ) resolution observations, which led to metallicities being typically crudely estimated.
For the literature sample, we searched for HI-selected absorbers with , where we carefully excluded any absorbers that were selected for D/H or using metal diagnostics to preselect them. Two pLLSs are drawn from Crighton et al. (2013, 2015). The rest of the sample comes from our KODIAQ survey used to search for OVI absorption in HI-selected LLSs with five LLSs () (Lehner et al., 2014).
Many of the other pLLSs/LLSs found in the KODIAQ database could not be used to study OVI owing to the contamination of the Ly forest near the OVI doublet transitions, but are useful for studying the metallicity distribution of these absorbers. In this sample, we selected pLLSs and LLSs for which we could derive reasonably well (specifically with a 1 error less than 0.3 dex, see §3.2) and estimate column densities (or column density limits) for SiII, SiIII, and SiIV (at least two of these ions are required to be uncontaminated), which are key ions to derive the metallicity of the pLLSs and LLSs at –3 (see §3.1).
All the new data presented here are from our KODIAQ database as part of our new KODIAQ Z survey (Lehner et al., 2014; O’Meara et al., 2015). In short, these data were acquired with the HIgh Resolution Echelle Spectrometer (HIRES) (Vogt et al., 1994) on the Keck I telescope on MaunaKea. These data were obtained by different PIs from different institutions with Keck access, and hundreds of spectra of QSOs at (most being at –) were collected. As part of our previous NASA KODIAQ program, we have uniformly reduced, coadded, and normalized the Keck HIRES QSO spectra (for a full information regarding the data processing, see O’Meara et al. 2015). A significant fraction of the reduced KODIAQ data is now publicly available from the KOA (O’Meara et al., 2015).111Available online at http://koa.ipac.caltech.edu/Datasets/.
Before proceeding to our main analysis, we emphasize two aspects of our sample of the pLLSs and LLSs. First, there is no proximate pLLS or LLS in our sample, i.e., all the absorbers in our sample have velocity separations from the redshift QSOs well above 3000 . Second, as we emphasize further below, we derive the column densities of HI and the metal lines in the main absorption associated with the pLLSs or LLSs, so the integration of the velocity profiles are over about 40 to 130 . This contrasts from the HD-LLS survey (Prochaska et al., 2015), where they consider that a LLS is all of the optically thick gas within a velocity interval of 500 from the redshift of the LLS. Owing to that we use higher resolution spectra in our survey and that the values are typically below cm, we can consider reliably smaller velocity intervals. However, we note there is one case in our sample where a pLLS has evidence for two pLLSs ( toward J144453+291905), but the signal-to-noise (S/N) level is not good enough to accurately model them separately. There is also one case where two pLLSs are separated only by 50 ( and 2.43359 toward J170100+641209) and where we find a similar metallicity for each absorber; in that case we only kept one for our analysis of the metallicity distribution (there is also one similar case in the Crighton et al. 2015, but in this case we adopted their results based on the total column density since there was little variation in the metallicity across the velocity profile). Finally, for two cases, a pLLS is associated with a SLLS, i.e., there is a velocity separation less than 300 between the pLLS and SLLS (one in our new sample – toward J012156+144823, see Appendix, and one in Crighton et al. 2013). It is unclear at this stage if this could bias in any ways the sample, but since there are only two such cases presently, any effect would be marginal (in the case of the Crighton et al. 2013 sample, the metallicity of pLLS is factor 50 than the SLLS, and hence the two absorbers do not have the same origin). In the future, with larger samples, we will be able to investigate more systematically pLLSs in the redshift vicinity of SLLSs or DLAs.
3 Estimation of the metallicity
The most robust approach to measure the metallicity of the pLLSs and LLSs would be to use the OI/HI ratio given that charge exchange reactions with hydrogen ensure that the ionizations of HI and OI are strongly coupled. However, for absorbers with , OI is rarely detected, and the limit that can be placed on is generally not sensitive enough. Hence to determine the metallicity of the pLLSs and LLSs, we have to compare the column densities of metal ions with HI. Since the pLLSs and LLSs are not pre-dominantly neutral like DLAs, but nearly completely ionized, we need to constrain the ionization of this gas to be able to derive its metallicity (e.g., Prochaska 1999; Lehner et al. 2009, 2013; Prochaska et al. 2015; FOP16; and see below for more details). LLSs and pLLSs are often multiphase, with absorption seen in different ionization stages, and the low to intermediate ions (e.g., SiII, SiIII, SiIV, CII, CIII, and sometimes CIV) and high ions (OVI) often show distinct kinematics (e.g., Lehner et al. 2009, 2013; Fox et al. 2013; Crighton et al. 2013; FOP16). This is illustrated in Figs. 1 and 2, where we show two examples of pLLSs at from our new sample with and 16.63, respectively. In the left panel of these figures, the HI transitions used to determine the HI column density are shown; the right panel shows some of the metal ions used to determine the metallicity. Other examples of high- LLS absorption profiles can be found, for example, in Lehner et al. (2014), Prochaska et al. (2015), and Crighton et al. (2013, 2015) as well as in the Appendix for the metal lines. For the ionizing radiation field and for pLLSs with typical metallicities at –3 (about 0.1% solar or , see below and FOP16), even strong transitions like CII 1334 and SiII 1260 are often not detected, so we have to use SiIII and SiIV to determine the metallicity. However, as in our study at low redshift (Lehner et al., 2013), we typically do not use high ions (specifically OVI at –3) because the distinct kinematics of these ions (see Fig. 2 and Lehner et al. 2014) imply that the bulk of the highest ions (i.e., OVI) are not produced by the same mechanism that ionizes the lower ions in the pLLSs/LLSs or at the same density.
In order to estimate the metallicity, we therefore need accurate column densities of HI and metal ions. We describe in §3.1 and §3.2 how we estimate the column densities of the metal ions and HI. To correct for the large ionization when comparing HI to metal ions (e.g., SiII, SiIII, SiIV, CII, CIII, CIV) to determine the metallicity, we use Cloudy (Ferland et al., 2013) models; a full description of this method and its limitations are presented in §3.3.
3.1 Metals and their column densities
The main ions and transitions used in our study are SiII 1190, 1193, 1260, 1304, 1526, SiIII 1206, SiIV 1393, 1402, CII 1036, 1334, CIII 977, and CIV 1548, 1550. In some cases, we can also use OI 1039, 1302, AlII 1670, FeII 1608, FeIII 1122. We also consider OVI 1031, 1037 and NV 1238, 1242 in order to assess whether CIV is likely to arise in the same gas-phase as the low ions. In the Appendix, we show for each pLLS or LLS the normalized profiles of the metal ions or atoms and discuss the specific ions used to determine the metallicity. We emphasize that understanding the physical conditions of all the gas-phases is beyond the scope of this paper. However, to determine the metallicity requires one to determine the column densities of the metal ions that are tracing the ionized gas associated with the HI of the pLLS or LLS. Following L13, the preferred species to constrain the ionization parameter (see below) are those for which the velocity structures of their profiles best follow the HI velocity profiles and that are produced mostly by a single phase ionization model.
To estimate the column density of the metal ions, we use the apparent optical depth (AOD) method described by Savage & Sembach (1991). The absorption profiles are converted into apparent column densities per unit velocity, cm (), where and are the modeled continuum and observed fluxes as a function of velocity, respectively, is the oscillator strength of the transition and is the wavelength in Å (the atomic parameters are from Morton 2003). Although the KODIAQ spectra are normalized (O’Meara et al., 2015), we still model the continuum with a Legendre polynomial within –2000 of the absorption feature of interest since the original continuum model may have sometimes over/under fitted some regions of the spectrum.222In this paper, we use high S/N data, so the continuum errors are typically at the 5% level or less depending on the redshift and if the feature of interest is deep in the LYAF or not. The velocity ranges used to model the continuum depend on the number of absorbing features and the overall complexity of the continuum in this region of the spectrum. To determine the total column densities, we integrate the profiles over the velocities that correspond to the main absorption of the HI of the pLLS or LLS. In the Appendix, we discuss for each pLLS/LLS the velocity structure of the metals and HI and show the integration range used to estimate (see the listed values in Table 1, which can vary somewhat between different ions); typically the integration range is over in the rest-frame of the absorber. There can be several velocity components within that velocity range, but we do not consider higher-velocity components that correspond to typically weaker HI absorbers clustered around the pLLSs or LLSs since the metallicity can be substantially different in these higher velocity components relative to the pLLSs or LLSs (e.g., Prochter et al., 2010; Crighton et al., 2013).
For doublets or ions with several available atomic transitions (e.g., CIV, SiIV, SiII), the levels of contamination or saturation can be assessed directly by comparing the values. In that case if there is no evidence of contamination, the absorption is typically resolved, i.e., there is no hidden saturation in the absorption profiles. For ions or atoms with only a single transition available, we require similar velocity structures between different species in the velocity intervals used for integrating to rule out contamination from unrelated absorbers. If the absorption reaches zero flux, the absorption is saturated, and we can only estimate a lower limit on the column density using the AOD method. If the peak optical depth is or similar to that of absorption lines observed with two or more transitions where there is no evidence of saturation, we infer that the absorption is not saturated. For strong absorption (–2), however, we allow in the photoionization modeling for the possibility that the line is saturated if needed by the models (i.e., we treat the column densities as possible lower limits).
In many cases, absorption from an ion or atom is not detected. If there is no contamination, we can estimate 2 upper limits on the equivalent widths, simply defined as the 1 error times 2. The 1 error is determined by integrating the spectrum over a similar velocity interval to that of a detected ion or over when no metals are detected in the absorber based on the typical smallest velocity intervals in other pLLSs/LLSs with detection of metals. The 2 upper limit on the column density is then derived assuming the absorption line lies on the linear part of the curve of growth. In Table 1, we summarize our apparent column density estimates of the metals as well the velocity interval used to integrate the profiles. For species with more than one transition, we list the results for each transition and in the row with no wavelength information the adopted weighted average column densities and velocities (see notes in this table for more information). Note that the errors are errors and include statistical and continuum placement errors following the methodology described in Sembach & Savage (1992). These errors do not, however, include errors arising from the original continuum fits to coadd the data (see O’Meara et al. 2015 and footnote 2).
3.2 HI column density
The estimation of for each LLS () was made using a procedure similar to that described in Lehner et al. (2014). We use the graphical package x_fitlls333As part of the xidl distribution package available at http://www.ucolick.org/xavier/IDL/xidl_doc.html that allows us to create Voigt profiles to model the data. We iteratively varied the redshift, -value, and of each system until a good fit was obtained. In many cases, the absorption in a LLS is complicated, requiring multiple absorption lines to produce a good fit. For the LLSs presented here, we consider all absorption that produces significant absorption (normalized flux at line center ) through at least Lyman-5 (i.e., all components with ) that might affect our total estimate. In most cases, such absorption impacts the total estimate at a level well below our error estimate on the , but in some cases multiple components of similar strength in are seen and cannot be ignored in the final estimate. Since we are fitting the absorption of the LLSs by eye (as opposed to using a reduced- approach, see below), we adopt very conservative errors, with a minimum error on the for any LLS of fitted using this methodology. We finally note that we must appeal to further constraints to accurately determine for the strong LLSs, as the higher order Lyman series lines remain saturated for many more transitions than the pLLS or weak LLSs (see below). We have, however, two important constraints. First, the onset of weak damping features in the Ly line can be used to constrain the from above, as if the is too large, excess absorption appears on either side of the line-center. Second, the break in flux level below the Lyman limit can be used to determine if there is enough S/N in the data and no nearby absorption from other strong systems.
For the pLLSs () and one LLS, the primary tool used to constrain are the higher order Lyman series transitions (see Figs. 1, 2, 3). Two authors (O’Meara, Lehner) undertook the analysis of the pLLSs where the continuum placement near each HI transition and profile fits to the pLLSs were independently assessed.444The only exception is the pLLS at toward J212912-153841 where the S/N is too low to use the high order Lyman series transitions. In that case, we use the combined information of the Lyman series transitions and the flux decrement at the Lyman limit. O’Meara used the same method described above for the LLSs, but instead fitted high order Lyman series transitions. For example, at the resolution of our HIRES data, a pLLS absorber with and becomes unsaturated (the normalized flux at the line-center being ) at Lyman-9. This and higher order Lyman series transitions can then be used to accurately determine the combination of , , and (or in the redshift rest-frame of the absorber) that best fits the observed absorption (see Fig. 3). Lehner fitted the HI profiles with a minimum reduced- method using a modified version of the code described by Fitzpatrick & Spitzer (1997). The best-fit values describing the pLLSs were determined by comparing the model profiles convolved with an instrumental Gaussian LSF with the data. The three parameters , , and for each component, i (typically ), are input as initial guesses and were subsequently varied to minimize . Since the Lyman series transitions are often blended with the Ly and Ly forest absorbers, the fitting was an iterative process to select transitions that were not blended or with minimum blending. In the case of small blends, we iteratively masked the blended regions. Figs. 1 and 2 show 2 pLLSs with various levels of contamination, while Fig. 3 shows a rare pLLS where 10 Lyman series transitions have little contamination. Despite some contamination, the use of different HI transitions with small oscillator strengths allows us to determine accurately . For each pLLS, the independently derived values were in excellent agreement. We adopted and errors from the Voigt profile fitting with the minimum reduced .
Our results are summarized in Table 2 and in Fig. 4 where we show the HI column density distribution for the entire sample of pLLSs and LLSs. There are 32 HI-selected absorbers listed in Table 2, 19 pLLSs () and 13 LLSs (). However, two pLLSs are at essentially the same redshift (separated by about 50 ) and have similar metallicities; we therefore treat these pLLSs as one, so that our total sample for the rest of the paper is 31. This is similar in size to the L13 sample of pLLSs and LLSs at (28 absorbers in total, 24 pLLSs and 4 LLSs). Our newer sample at has now doubled with 44 pLLSs and 11 LLSs (W16). Our sample is also complementary to the HD-LLS survey, which, by definition of their sample, targets only LLSs with all but 9 LLSs at –3.3 having (Prochaska et al. 2015; FOP16).
3.3 Photoionization modeling and metallicity determination
With the column densities of HI and metals determined, we can estimate the metallicity of each pLLS or LLS. This requires large ionization corrections since the fraction of H that is ionized always exceeds 90% and is often close to 100% (i.e., ). To determine the metallicity we follow closely L13, modeling the ionization using Cloudy (version c13.02; Ferland et al., 2013) and assuming the gas is a uniform slab geometry photoionized by the Haardt-Madau background radiation field from quasars and galaxies (HM05, as implemented within Cloudy – see also Haardt & Madau 1996, 2012; by adopting HM05 we also reduce any systematics in the comparison with the low redshift pLLSs/LLSs studied by L13 and W16). For each absorber, we vary the ionization parameter, which is by definition the ratio of H ionizing photon density to total hydrogen number density (), and the metallicity (we use the usual notation for the metallicity , where X is a given element) to search for models that are consistent with the constraints set by the column densities determined from the observations.
We assume solar relative heavy element abundances from Asplund et al. (2009), i.e., we do not include a priori the effects of dust or nucleosynthesis on the relative abundances. We note that for the main elements (C, Si, see below) that we use to model the photoionization and for the densities that the pLLSs and LLSs typically probe, the dust depletion levels of C and Si are expected to be small. In the Milky Way, the depletions observed in the so-called “warm-disk” and “cool-halo” clouds for Si and C are dex (e.g., Savage & Sembach, 1996; Welty et al., 1999; Jenkins, 2009). At the studied redshift intervals in our survey, even smaller depletion levels of Si are typically observed in the denser environments probed by DLAs and SLLSs (e.g., Ledoux et al., 2002; Prochaska et al., 2003b; Rafelski et al., 2012; Quiret et al., 2016); e.g., Rafelski et al. (2012) found on average for gas metallicities . Furthermore, FOP16 has shown that the strong LLSs reside typically in dust-poor enviromnents. We nevertheless consider these possibilities a posteriori (especially for carbon that can have a different nucleosynthesis history than elements as silicon or oxygen for example). This can be done a posteriori because the dust depletion or nucleosynthesis effects should affect all the ionization levels of a given element by the same factor. A posteriori, we find that typically dust depletion does not need to be invoked to explain the relative abundances of the pLLSs and LLSs in our sample, a finding consistent with the results from FOP16.
The metallicity for each pLLS or LLS is determined using elements (usually Si), but the ionization model is constrained using the suite of Si and C ions (SiII, SiIII, SiIV, CII, CIII, CIV), and sometimes other atoms or ions (e.g., OI, AlII, etc.). In the Appendix, we provide the set of ions that determines and for each LLS or pLLS. In Table 2, we list the derived metallicities while in Table 6 of the Appendix, we provide for each pLLS and LLS the Cloudy output parameters from our models (total column density of H – , , , , ionized fraction – /, temperature – , and the linear scale of the absorber – ).
The errors on the metallicity and (listed in Table 2 and Appendix) reflect the range of values allowed by the uncertainties on the observed column densities. They do not include errors from the limitation of the models used to estimate the ionization corrections, which are about 0.3–0.5 dex on the metallicity (see L13; W16). As discussed in L13, uncertainties in the assumed radiation field largely do not affect the shape— of the metallicity distribution. W16 explore the effect of changing the ionizing background from HM05 to HM12 (Haardt & Madau, 2012) for the pLLSs and LLSs at and found that on average it would increase the metallicity of the pLLSs and LLSs by about dex, well within the 0.3–0.5 dex uncertainty quoted above. This is, however, a systematic effect, i.e., both low and high metallicity absorbers are affected the same way, and hence the overall shape of the metallicity distribution would be very similar. FOP16 also provide a thorough analysis of a large sample of LLSs where they use several ionization models and Bayesian techniques to derive the physical properties and metallicities of the LLSs. They find as well that the metallicity estimates are typically not very sensitive to the assumptions behind the ionization corrections.
4 Metallicity of the pLLSs and LLSs at
4.1 Metallicity distribution of the pLLSs and LLSs
Figure 5 shows the metallicity distribution function (MDF) for the 31 HI-selected pLLSs and LLSs in our sample at summarized in Table 2. Visually, the MDF is unimodal (see below). The MDF extends from dex () to dex (), but most of the values are dispersed around dex. Using the Kaplain-Meier (KM) product limit estimator from the survival analysis (Feigelson & Nelson 1985; Isobe et al. 1986) to account for the upper limits in the sample, we estimate for the pLLSs and LLSs that (where the quoted error is the KM error on the mean value). Treating the 5 upper limits as values, the median and standard deviation are and 0.83 dex, respectively (under that assumption the mean of the MDF would be dex).
There is no evidence of a strong dip in the distribution as observed at low redshift (\al@lehner13,wotta16; \al@lehner13,wotta16), and there is a prominent peak near the mean. A Dip test (Hartigan & Hartigan, 1985) shows that the significance level with which a unimodal distribution can be rejected is only 26%.555See Muratov & Gnedin 2010 for the description of the Dip test code. Treating censored data as actual values, a Kolmogorov-Smirnov (KS) test finds the metallicity distribution is not inconsistent with a normal distribution with -value where the normal distribution has a mean and With future larger KODIAQ Z samples, we will be able to determine more robustly the shape of the MDF of both the pLLSs and LLSs. With the current sample, the MDF of the pLLSs+LLSs at can therefore be described by a unimodal distribution (possibly as a Gaussian distribution) with a large spread in both high and low metallicities.
4.2 Variation of the metallicity with
In Fig. 6, we show the distribution of the metallicity against at , which allows us to separate the pLLSs and LLSs (and other absorbers) and to visualize the unbinned measurements. There is a large spread in the data for both the pLLS and LLS samples. In Table 3, we list the mean, median, standard deviation, and fraction of very metal poor (VMP) absorbers with (value corresponding to below the mean metallicity of the DLAs). The LLSs and pLLSs have similar dispersions in their metallicity distributions, but from the KM method, we estimate that the mean metallicity of the LLSs is a factor 5 smaller (0.7 dex) than that of the pLLSs, vs. (although they overlap within less than KM error). There is also less evidence of VMP pLLSs than LLSs (6% vs. 43%). A Gehan’s generalized Wilcoxon test and log-rank tests (which take into account that there are censored data – upper limits – in both the pLLS and LLS samples, see Feigelson & Nelson 1985) indicate a marginal statistical difference between the MDFs of the pLLSs (18 data points including 2 upper limits) and LLSs (13 data points including 4 upper limits) at significance levels % and %, respectively. Yhe samples of LLSs and pLLSs are still small and there is a large overall dispersion in the metallicity distribution of both the pLLSs and LLSs; hence we consider any difference between the pLLS and LLS samples as tentative and marginal.
In Fig. 6, we also show the metallicity for lower and higher absorbers. For the LYAF, we show the mean and standard deviation from Simcoe et al. (2004) who determined in the spectra of 7 QSOs the metallicity using OVI and CIV for absorbers with (most between , which is highlighted by the asymmetric error on the horizontal axis) at . We also note the pixel optical depth method leads to similar results at (Ellison et al., 2000; Schaye et al., 2003; Aguirre et al., 2004). In the LYAF sample, about 60–70% of the LYAF absorbers are enriched to (observable) levels of , while the remaining have even lower abundances. The LLSs and SLLSs shown with grey squares and associated vertical error bars are from the HD-LLS survey and represent the medians and the 25th/75th percentiles of the composite posterior metallicity PDFs (FOP16; the horizontal error bars show the range of and are centered on the average values). For completeness and reference, we also show in this figure (in light-yellow squares) the SLLS metallicities recently compiled from the literature as well as a few new metallicity estimates by Quiret et al. (2016). For that sample, we only consider metallicities that were derived using an -element (i.e., OI, SiII, MgII) and within the redshift interval . We have also attempted to remove from that sample any proximate SLLSs or absorbers that may be possibly biased (e.g., a D/H target). In that sample, the 5 estimated metallicities with OI are all for SLLSs with and resulted in metallicities within the range . Note that for several of these metallicites (including those derived with singly ionized species) no ionization correction was realized, which may play in part a role in some of the observed elevated values (), especially since 5 of these have comparatively low values with . Owing to the clean selections of the LLSs and SLLSs and the uniform analysis of the HD-LLS survey (both similar to the KODIAQ Z survey), we favor HD-LLS survey for comparison with our sample. For the DLAs, we use the measurements and compilation from Rafelski et al. (2012).666We note that Quiret et al. (2016) also compile all the existing DLA metallicities from the literature. Unfortunately, for our purposes, this compilation lacks key information regarding any selection biases (e.g., D/H targets, DLAs pre-selected owing to the absence of metal absorption in SDSS spectra, etc.). In Table 3, we summarize the mean, median, and dispersion for each of these classes of absorbers. We also estimated the fraction of VMP DLAs with (see Table 3), which by definition of this threshold value ( below the mean metallicity of the DLAs) is small. For the HD-LLS survey, owing to the method used to determine the metallicity, we list in Table 3 the probability of finding absorbers lower than .
Considering the entire range of plotted in Fig. 6 () at , several immediate conclusions can be drawn: 1) there is a gradual decrease in the mean (or median) metallicity from the DLAs to the LYAF (with possibly the exception of the pLLSs, but see above); 2) the dispersion around the mean for the LYAF, pLLSs, LLSs, and SLLSs is large (about 0.8 dex on average), but for the DLAs the dispersion is a factor 2 smaller (0.5 dex); 3) there is a substantial fraction of LYAF, pLLSs, LLSs, and SLLSs that has metallicities below while of the DLAs have such low metallicities; 4) only for the LYAF, pLLSs, and LLSs, there is evidence of metallicity below (see Fig. 6): for the pLLSs and LLSs, the fraction with is in 2.5–17.7% (68% confidence interval), while of the LYAF absorbers have (Simcoe et al., 2004; Simcoe, 2011).
5 Redshift evolution of the pLLSs and LLSs
Our selection of the pLLSs and LLSs at and follows the same criteria: first, they are HI-selected to have HI column densities between ; second, the HI column density can be estimated reasonably accurately (within 0.3 dex and often better than 0.1 dex); and third, there is enough information from the metal lines to derive sensitively the metallicities. Therefore we can directly compare the high and low redshift samples to study the evolution of the metallicity for these systems. However, the overdensities of the structures change as function of . At the critical density of the universe is about a factor 8 lower than at . Using, e.g., the empirical relationship for the overdensity derived by Davé et al. (1999) for absorbers with , , the change in is similarly a factor 8 between the mean redshifts of the W16 () and this study (). This implies that absorbers at some given at high and low redshifts are not necessarily physically analogous (see also Davé et al., 1999). For the LYAF absorbers, SLLSs, and DLAs, the redshift evolution of the density does not change the fact that LYAF absorbers trace very diffuse gas () and SLLSs/DLAs trace virialized structures () at both high and low . On the other hand, for the LLSs and especially the pLLSs, while at they probe gas well within the CGM of galaxies, at , can be , and hence pLLSs could probe more diffuse ionized gas at . KBSS shows that only half of the absorbers with are found in the CGM of (massive) galaxies at ; the other half may probe more diffuse gas or the CGM of dwarf galaxies (Rudie et al., 2012). Hence while high LLSs and pLLSs are by definition at the interface between the denser and more diffuse gas, they may not trace necessarily the same dense CGM of galaxies as their counterparts at . We keep this caveat in mind as we now review the evolution of the properties of the pLLSs and LLSs with .
5.1 Evolution of the physical properties with
While the main goals of our study are to determine the shape of the metallicity distribution of the pLLSs/LLSs at high and how it evolves with , we can also highlight similarities and differences in other properties (densities, , etc.) of the pLLSs and LLSs at low and high . In Table 4, we summarize the mean, median, standard deviation, and minimum, maximum values of and several physical parameters derived from the Cloudy models for the pLLS/LLS samples at (from L13) and (this paper as well as the results from Crighton et al. 2013, 2015; Lehner et al. 2014). Note that here we have treated upper or lower limits as actual values, but this has limited effect on the statistics and comparison.777We have removed for this analysis the two absorbers where we set by hand owing to too little constraints from the observations; including these would, however, not have changed the results. For example, we find for the sample of pLLSs and LLSs at using the KM estimator instead of assuming that the lower limits are actual values. As demonstrated by FOP16, we emphasize that while the metallicities derived from the Cloudy simulations are quite reliable, there is a degeneracy between ionization parameter and intensity of the radiation field, which hinders robust estimates of the densities and sizes of the absorbers. Hence the hydrogen density () and linear scale () are not as robustly derived as the metallicities or the total H column density ().
Unsurprisingly, the statistics for at low and high are not too dissimilar owing to a similar initial selection of the pLLSs and LLSs (see Table 4). A two-sided KS test on the low and high samples gives a maximum deviation between the cumulative distributions and a -value , implying no significant difference between the samples at low and high . On the other hand, the ionization parameter derived from the Cloudy simulations evolves significantly with . In Fig. 7, we show the histogram distribution of and distribution of against for the pLLSs and LLSs in our sample at (see Appendix) and the L13 sample at . There is some evidence that strong LLSs with have smaller -values at any studied , but the sample of these strong LLSs is still small. For absorbers with , there is no obvious trend between and at any . Most of the pLLSs/LLSs at have (consistent with the early compilation made for the LLSs by Fumagalli et al. 2011a and from the HD-LLS analysis, see FOP16) while at , most have . A two-sided KS test on the samples at low and high gives and , implying a significant difference in the distributions at low and high . The mean and median of are a factor 10 times larger at than at . The higher -values at high redshift explain why highly ionized species (SiIV, CIV) can be modeled by photoionization, while a single-phase photoionization model typically fails to produce the same highly-ionized species (especially CIV) at for the pLLSs and LLSs (L13 and see also Fox et al. 2013).
In Fig. 8, we show the hydrogen density, hydrogen column density, and physical scale as a function of the HI column density for the pLLSs and LLSs at from our sample and at from L13 (note that we ignore the very few lower/upper limits in this figure). For the densities, while there are few more high values at for weak pLLSs, overall at high and low redshifts overlaps and have the same mean with a dispersion of about dex. These densities are very similar to the densities estimated by FOP16 for stronger LLSs. A two-sided KS test on the samples at low and high gives and , implying indeed no significant difference in the distributions at low and high .
For the total H column densities, their typically values are higher at high redshift than at low redshift over the entire range probed by the pLLSs and LLSs. On average, is a factor 10 times larger at high than low . A similar trend is also observed for where large-scale structures (– kpc) for the pLLSs and LLSs are not rare at (a result also found by FOP16 and Cooper et al. 2015 at higher and for the LLSs at the boundary with the SLLSs). In the pLLS regime, while there is a large fraction of low- pLLSs with kpc, there is also an overlap between high- and low- pLLSs with kpc. A two-sided KS test on the and samples at low and high gives and , respectively, implying in both cases significant differences in the distributions of these quantities at low and high .
Finally, the last entry of Table 4 shows that the temperature of the gas probed by the pLLSs and LLSs is higher at high , but with a similar large dispersion at both low and high . FOP16 found that the probability distribution function of the gas temperature peaks strongly at a similar value for the photoionized gas than the mean of our high redshift sample. A two-sided KS test on the temperatures samples at low and high gives and , implying a significant difference in the distributions at low and high .
Hence this strongly suggests based on simple overdensity arguments and the Cloudy results that the pLLSs and LLSs have different physical parameters at high and low (except for the densities), implying that the pLLSs and LLSs at do not evolve directly into their low analogs. Using the empirical relationship from Davé et al. (1999), the pLLSs and LLSs at should evolve into strong LYAF absorbers () and pLLSs at , respectively.
5.2 Evolution of the metallicity with
The cosmic evolution of the DLAs (e.g., Prochaska et al., 2003a; Rafelski et al., 2012; Battisti et al., 2012; Jorgenson et al., 2013) and SLLSs (e.g., Som et al. 2013, 2015; FOP16; Quiret et al. 2016) have been well studied for several years. In Fig. 9, we show the metallicity evolution of the pLLSs and LLSs as a function of redshift (and look-back time) where the low and high absorbers were selected and analyzed using the same methodology. At all the peak-to-peak scatter in the metallicities of the pLLSs and LLSs is large (over 2 dex spread in ). Owing to this large scatter, there is an overlap in the MDFs of the pLLSs and LLSs at low and high , but the MDF is also changing drastically with : at , the MDF is unimodal, peaking at with a long tail to higher metallicities, while at low , the MDF is bimodal, peaking at and with about the same number of absorbers in each branch of the distribution (see also \al@lehner13,wotta16; \al@lehner13,wotta16). At low , only one system has a metallicity well below , although there are several upper limits near this lower bound metallicity. The quasi-absence of very low metallicity gas at can be attributable in part to the lower sensitivity of the UV data (typically, S/N – for HST/COS observations compared to – for data obtained with Keck HIRES, see L13 and O’Meara et al. 2015), but it is also possible that low metallicity gas with is rare at low .
As noted above, pLLSs and LLSs at low are probably not always their direct high redshift analogs. Based on the overdensity argument, LLSs at could evolve into the low pLLSs. Using the results from this work (see Fig. 6 and §4.2) and FOP16, the MDF of the LLSs at is consistent with a unimodal distribution, significantly different from the bimodal MDF of the pLLSs at (W16). Therefore, even considering the redshift evolution of the cosmic structures, there is a significant evolution of the MDF of the LLSs with .
The change in the MDF of the pLLSs and LLSs between and is also quite significant and distinct from DLA and SLLS evolution. The MDF of the pLLSs and LLSs is not simply shifting to higher metallicity as observed for the SLLSs and DLAs, but the shape of the MDF is evolving significantly to lower . In Fig. 9, we also show the redshift evolution of DLA metallicities from the Rafelski et al. (2012) survey for comparison. As noted by Rafelski et al. (2012) and others, there is an overall increase of the metallicity with decreasing , but the shape of the MDF for the DLAs does not evolve with ; it is unimodal with similar scatter about the mean at all redshifts. This scatter in metallicities is also smaller than that observed for the pLLSs and LLSs. The “lower envelope” of the metallicity of the DLAs (mean metallicity of the DLAs minus ) changes from at to dex at . Below these metallicities at the respective redshifts, there is a large number of pLLSs or LLSs, implying that a large fraction of the pLLSs and LLSs follows a different metal enrichment than the DLAs. At all , however, there is also a large overlap in the metallicities of the DLAs and the more metal-enriched pLLSs and LLSs; these higher-metallicity pLLSs and LLSs may follow a similar metal enrichment evolution similar to that of the DLAs.
5.3 Relative Abundances of C/
So far we have only presented the results for the absolute abundances of the gas. Although we have limited information on the relative abundances, at both high and low redshifts (see L13), we have some constraints on the C/ ratio. This ratio is a good indicator of the nucleosynthesis history since in low density, diffuse gas, carbon and the elements used in these works are not expected to be strongly depleted into dust grains (see §3.3), and hence this ratio provides additional information regarding the origin of the gas. For the pLLSs and LLSs, this ratio was principally derived from the photoionization models (see §3.3). In these models, C/Si was set a priori to a solar value, but was allowed to vary in order to determine the best , -values that fit the data. Although, this ratio is derived using photoionization models and subtle changes in the radiation field could change its value, we feel it is robustly derived for the following reasons. Firstly, W16 show that while modifying the radiation field from HM05 to HM12 can change in a systematic manner by about dex, it does not affect as much the C/ ratio. Secondly and independently from any ionization assumption, we can directly estimate C/ from the observations using the column density ratios at and at (CIV and SiIV are not considered at lower redshift because these are typically produced in a different gas-phase, see L13). We summarize these results in Table 5. There is only a small fraction of the sample where we have simultaneously column densities for all these ions, but it is striking that for all but one, the direct and modeling methods provide consistent results (the only discrepancy toward J131215423900 could be possibly arising owing to some contamination in the CIII 977 absorption). As a reminder for the pLLSs and LLSs, at high redshift, the -element is mostly Si, but at low redshift it can also be O, Mg, and/or S depending on the system (see L13).
In Fig. 10, we show [C/] vs. [/H] for the pLLSs and LLSs from both the high- and low-redshift samples from this and L13 surveys (note that the most metal poor LLS in this figure is from Crighton et al. 2015). We note that in the regions of overlapping metallicities, there is no obvious difference between the low and high redshift samples, and we therefore treat them together in the remainder of this section. For comparison, we also show the results for high redshift DLAs and SLLSs and Milky Way (MW) stars. For the DLAs and SLLSs, we use the results from Pettini et al. (2008), Penprase et al. (2010), and Cooke et al. (2011) (and references therein and see also Becker et al. 2012 for measurements). For the MW thin and thick stars, we use the results from Bensby & Feltzing (2006), and for the MW halo stars, Fabbian et al. (2010) and Akerman et al. (2004). For the stars, is O, while for the DLAs and SLLSs, is O or Si (changing O to Si or vice-versa for the DLAs would have little effect on the distribution of these data). As noted by Pettini et al. (2008), Penprase et al. (2010), and Cooke et al. (2011), the metal-poor SLLSs/DLAs follow well the overall behavior of [C/] with [/H] having a similar dispersion as observed in the MW metal-poor stars and confirm the overall increase of [C/] seen in metal-poor stars (Akerman et al., 2004; Spite et al., 2005). Where DLAs and stars overlap (), the overall agreement in the distribution of C/ suggests a universal origin for the production of C relative to -elements (Cooke et al., 2011).
The overall trend observed in Fig. 10 in the stellar and SLLS/DLA samples can be separated in roughly two regions. Region 1: At , [C/] decreases with increasing metallicity from super-solar values to about dex. Region 2: at , [C/] increases with increasing metallicity from about dex to super-solar values. The behavior in region 2 has been well known for some time and is thought to occur as a result of the delayed release of carbon from low- and intermediate-mass stars combined with a strong metallicity dependence of the yields of carbon by massive stars with mass-loss (e.g., Akerman et al., 2004; Fabbian et al., 2010). The increase of [C/] to lower metallicity at was somewhat surprising at first, but has now been confirmed independently in both stellar atmospheres and SLLSs/DLAs. One possible interpretation for the high values of C/ at low metallicity could be the leftovers from the enhanced production of C (relative to -elements, and in particular O) in Population III (Pop III) stars. As shown by Frebel et al. (2007) and Bromm & Loeb (2003), the gas progenitor of Pop III stars must have had high C abundance to efficiently cool the gas in order to actually form stars and to drive the transition from Pop III to Pop II stars (see also Cooke et al. 2011 for more discussion). We show in Fig. 10 that condition (hatched orange region) defined as the “transition discriminant” criterion. No Pop II stars should be found in that zone, but any gas in this region will likely have been polluted by Pop III stars (twoo LLSs are found in that“forbidden” zone, see Fig. 10 and below).
Considering now the pLLSs/LLSs, about half the sample of the pLLSs and LLSs follows a similar distribution to that observed for the DLAs and stars over the entire range of metallicity, i.e., . For these, their chemical enrichment (at least of C and -elements) appears to be similar to that of the MW stars and the bulk of the SLLSs/DLAs. However, the other half — mostly clustered at and — does not follow the trend observed in MW stars or DLAs as first pointed out by L13. These gas clouds are carbon-enhanced by a factor –5 (– dex) compared to stars or most DLAs with similar . This effect is not artificially caused by the ionization modeling since near solar [C/] over are confirmed directly by the observations (see Table 5), and hence the carbon-enhancement observed at is real.
Finally, we highlight the lowest metallicity LLS in our sample with and at observed toward J095852+120245 that lies in the Pop III/Pop II transition (orange-zone in Fig. 10). The properties of this LLS are reminiscent of another one at with and described by Crighton et al. (2016) (shown with green data point in Fig. 10). This implies that there are now two LLSs at with expected [C/] and that are consistent with gas polluted from Pop III stars.
Our present study explores the properties (in particular the metallicity) of the pLLSs and LLSs at , a redshift epoch corresponding to the ascending part of the cosmic star formation rate (SFR) density, near its peak (e.g., Madau et al., 2014). Our previous studies (\al@lehner13,wotta16; \al@lehner13,wotta16) have explored the metallicity of the pLLSs and LLSs with similar at where the cosmic SFR density has significantly decreased. According to cosmological simulations, the exchanges of matter in and out through the CGM play critical roles in the evolution of galaxies and in the evolution of the cosmic star-formation (e.g., Kereš et al., 2005; Dekel & Birnboim, 2006; Faucher-Giguère et al., 2011). We therefore expect that some of the properties of the pLLSs and LLSs should be intimately coupled to those of star formation in galaxies. This should also be reflected in changes of the properties of the IGM/galaxy interface region as a function of . As we lay out below, there are clear differences but also similarities between the low and high CGM probed by pLLSs and LLSs.
Before going further we emphasize that at both high and low redshift studies the samples were HI-selected absorbers with in order to avoid introducing any bias in the metallicity of the gas probed by these absorbers. We also use the same technique to derive the metallicity of the absorbers, so any changes in the MDF of the pLLSs and LLSs as a function of should be genuine, not some effect from comparing different samples or metallicities derived using different techniques. However, owing to the redshift evolution of the universe, pLLSs and LLSs at high are not the direct analogs of the low redshift pLLSs and LLSs (see §5.1).
We also note that at low we make a direct association between the CGM and absorbers with since all the pLLSs and LLSs with galaxy information have been found so far well within the virial radius of relatively bright galaxies ( to , see, e.g., L13; Lehner et al. 2009; Cooksey et al. 2008). At high , galaxy information is still scant.
Observations with the Multi Unit Spectroscopic Explorer (MUSE) found no bright, star forming galaxy in the vicinity of the most metal-poor LLS in our sample (Fumagalli et al., 2016a). This LLS could probe an IGM structure888The path length of 2 Mpc and density cm derived using our Cloudy model for this absorbe rare consistent with an IGM origin. However, we note this absorber is unique among our sample. or the CGM of a faint galaxy with a SFR M yr. Furthermore, we note that the KODIAQ OVI survey of HI-selected absorbers with shows a large fraction of the pLLSs and LLSs at high has strong and broad OVI absorption associated with these absorbers, which contrasts remarkably from the OVI properties in the IGM (typically much narrower and weaker). The strength and breadth of the OVI make these absorbers likely probes of the CGM of some very actively star-forming galaxies (Lehner et al. 2014 and see §6.5). In any case and at all , the pLLSs and LLSs are at the interface between the very diffuse IGM probed by LYAF absorbers and virialized structures probed by SLLSs and DLAs, and it is in this context that we discuss our results below.
6.1 Evolution of the MDF of pLLSs and LLSs with
In the ascending part of the cosmic SFR density at , we find that the MDF of the pLLSs/LLSs is heavily weighted to low metallicities, unimodally distributed around . At , well past the peak SFR density, the overall MDF has shifted to higher metallicity. For the pLLSs at , the MDF is bimodal with about the same weight in each of the metallicity branches that peak at and , i.e., the low-metallicity branch has on average a metallicity 20 times lower than those in the high-metallicity branch (\al@wotta16,lehner13; \al@wotta16,lehner13). These results for the low-redshift universe show that there are clearly two main populations of gaseous flows through the CGM at . The metal-enriched CGM gas has properties consistent with those expected for matter being ejected by large-scale galaxy outflows, for matter being tidally-stripped from satellite galaxies, or for material tracing the remnants of earlier outflows that are recycled by galaxies. The other half has an extremely low metallicity for the universe. For all the cases so far, these metal-poor pLLSs and LLSs have been found well within the virial radius of some galaxy and have column densities, temperatures, and metal-enrichment levels about consistent with cold accretion gas as observed in cosmological simulations at –3 and (see L13 and simulations by, e.g.,Fumagalli et al. 2011b; van de Voort et al. 2012; Shen et al. 2013; Hafen et al. 2016; and see also §6.6).
On average the metallicity of the gas also increases with increasing at and (see Fig. 6 and \al@lehner13,wotta16; \al@lehner13,wotta16). As noted by W16, the difference in the MDFs of the pLLSs/LLSs compared to the SLLSs and DLAs implies there is a fundamental change in the physical origins with . DLAs are likely probing gas that have been enriched recently at a given , while the bulk of the LYAF probes typically the diffuse IGM with little metal content. The pLLSs and LLSs appear to probe both types of gas, recent metal-enrichment as well as very ancient metal-enrichment. The SLLSs predominantly probe recent enrichment, but a non-negligible fraction may also be more pristine IGM-like metallicity (see Table 3).
Naively, if the interpretation that low-metallicity pLLSs and LLSs are mostly probing infalling gaseous streams or clouds, then the gas at the interface between galaxies and diffuse IGM at high would be infall-dominated at . However, at these redshifts, the median metallicity of the pLLSs and LLSs is , and hence a large proportion of the pLLSs and LLSs has metallicity overlapping with those of the DLAs (Table 3 and see Figs. 6 and 9). At , only the high metallicity branch overlaps with the DLA MDF (W16); the mean metallicity of the DLAs at is , very similar to that of the pLLSs/LLSs in the high metallicity branch. The mean metallicities of the DLAs and pLLSs/LLSs at are, however, much closer than at low redshift (a factor 4 compared to a factor 20).
In view of the overlap of metallicities between pLLSs/LLSs and DLAs at high , a better approach to separate at all potential metal-poor cold accretion candidates from other processes is to consider the fraction of VMP pLLSs/LLSs that we define as absorbers with metallicities below the mean metallicity of the DLAs in any given redshift interval. At , that threshold is ; at , it is ; and at , it is . At , the proportion of VMP pLLSs/LLSs is 25–41% in our sample (see Table 3). Similar numbers in the same redshift interval are found for the HD-LLS survey (31% for the LLSs, 21% for the SLLSs, see Table 3). At , W16 derive 28–44% of the pLLSs are VMP. Using the recent sample at of very strong LLSs from Glidden et al. (2016) (, except for 2 systems), we calculate that the fraction of VMP strong LLSs is 18–34% (sample size is 31 as we exclude the two SLLSs, which is similar to the present KODIAQ Z sample). Since many of these absorbers overlap with the SLLS regime, if we include only systems with from the Glidden et al. sample, then the fraction of VMP strong LLSs would be 30–51% (sample size 20).999At with a smaller sample probing extremely strong LLSs () and an indirect method, Cooper et al. (2015) also found 28%–40% of the LLS population could trace VMP gas. All these intervals are at the 68% confidence level.
While in the future we will improve the confidence intervals and refine these fractions over smaller redshift bins, it is striking that the proportion of VMP pLLSs and LLSs do not evolve much with redshift (although we emphasize the values sampled in the interval are quite higher than in our sample). The average metallicities of the VMP pLLSs/LLSs increase with increasing redshift, but their fractions remain about the same over 12 billion years.101010We also note that the total hydrogen column densities or scale-lengths of the VMP pLLSs and LLSs evolve in the same way as for the more metal rich pLLSs and LLSs, i.e., on average is 10 times larger at than at and there is no obvious difference between the VMP pLLSs/LLSs and the rest of the sample. These VMP pLLSs and LLSs have metallicities that are consistent with the IGM metallicities in each redshift interval (although at , the metallicity of the IGM is unknown as a result of the limited sensitivity of the space-based UV observations). Hence these VMP pLLSs/LLSs appear to be the reservoirs of metal-poor gas in the interface between galaxies and the IGM, which appear to remain constant over cosmic time and which may feed galaxies with metal-poor gas to continue to form stars over billions of years. These VMP pLLSs/LLSs are also good candidates for cold flow accretions as seen in cosmological simulations (see §6.6).
6.2 The fraction of pristine gas at
We found two pLLSs and LLSs with no metals (see Appendix) that might be reminiscent of the pristine LLSs that were discovered at and 3.1, down to a limit and (Fumagalli et al., 2011a). Unfortunately, SiIII is contaminated for each of these cases, and hence we cannot place a stringent constraint on their metallicities. For example, the conservative limit on the LLSs at toward J025905+001121 is (and ); if instead we adopt the mean derived in our sample, then (see Appendix), a limit similar to those found by Fumagalli et al. (2011a).
To better understand the level of mixing of metals in the gas probed by pLLSs and LLSs in the early universe, we will need a much larger sample to reliably determine the frequency of pristine gas at in the interface regions between galaxies and the LYAF. With our sample, we determine that the fraction of pLLSs/LLSs with is – (2/31) at (68% confidence interval), consistent with the FOP16 results for stronger LLSs. This fraction includes the lowest metallicity absorbers in our sample that have metals detected. If we push to metallicities down to to exclude any pLLS or LLS with some metals detected, that fraction becomes (68% confidence interval), implying that pristine pLLSs/LLSs at are rare.
As noted by Crighton et al. (2016) (see also Cooke et al., 2011), the extremely metal-poor LLSs ( at ) with detected metal absorption may provide a new path to study the Pop III/Pop II metal-enrichment transition. The use of both the low metallicity and C ratio indeed provides a strong method to find metal-pollution at the transition from Pop III to Pop II star formation. In our sample of 31 pLLS/LLSs, we have found one such absorber (corresponding to a proportion of –) with and [C/, both consistent with a Pop III origin.
6.3 Super metal-rich gas at
On the other end of the metallicity spectrum, we have also discovered a supersolar pLLS () at toward J172409+531405. This absorber is extraordinary on several levels. It has a metallicity of at a redshift . This is the only pLLS with a detection of OI, which is remarkable for such a low absorber. The physical-scale ( pc), density ( cm), and temperature ( K) are all extremely atypical for any pLLSs at any . The non-detection of FeII implies a /Fe enhancement (or possibly some dust depletion of Fe relative to Si). This pLLS is detected in several ions and transitions, so its properties are well-constrained. It is a multiphase absorber since the CIV and singly-ionized species have very different velocity profiles (see Appendix)
This is clearly an outlier in our sample (1/31 or – at the 68% confidence interval). Its properties (in particular its high metallicity and multiphase nature) suggests that it directly probes an active outflow from a proto-galaxy at . As our KODIAQ Z survey will grow, we will more robustly determine the frequency and properties of both metal rich and pristine pLLSs and LLSs at .
6.4 C/ in pLLSs and LLSs over cosmic time
The combined sample of pLLSs and LLSs at and shows that the scatter in C/ with metallicity is very large at any and C/ does not follow the trend observed in stars or DLAs (see Fig. 10 and Table 5). Stated in another way, about half the sample of pLLSs and LLSs has an enhanced C/ ratio in the metallicity range compared to Galactic halo stars and DLAs, while the other half follows more closely C/ patterns seen in Galactic metal-poor stars or DLAs. The enhanced C/ ratio in the metallicity range implies that this gas must have been polluted by preferential ejection of C from low metallicity galaxies. A recent study in fact shows that at least some local metal-poor dwarf galaxies have also enhanced C/ over similar metallicities (Berg et al., 2016). While their C/ ratios are not as high as observed for the pLLSs and LLSs and their sample is small (12 galaxies), the absence of clear trend between [C/ and /H] is similar to that observed in pLLSs and LLSs.
On the other hand, in the IGM (probed by the LYAF) at –, using the pixel optical depth analysis of CIV, OVI, and SiIV, low C/ ratios were derived: and (Schaye et al., 2003; Aguirre et al., 2004, 2008). As discussed in Aguirre et al. (2004), they only use the CIV/SiIV and OVI/SiIV ratio to determine C/Si and O/Si, respectively, which is dependent on the assumed ionizing background (and if collisional ionizing processes take place). While such low values are found for some of pLLSs and LLSs (see Fig. 10), our results imply a very large scatter in C/ that does not depend on the redshift or the metallicity. It would seem likely that this should also happen in the IGM.
6.5 OVi associated with pLLSs and LLSs
Although we focus throughout on the metallicity of the cool gas of the pLLSs and LLSs, some of the surveys described above have also revealed that OVI absorption with overlapping velocities with HI is found at any (Lehner et al., 2013, 2014; Fox et al., 2013). When OVI is detected, these pLLSs and LLSs have typically multiple gas-phases as evidenced by the presence of low ions (e.g., CII, SiII, SiIII) and OVI (or other high ions) that have often very different kinematics and cannot be explained by a single photoionization model (e.g., Lehner et al. 2009, 2013; Crighton et al. 2013; Fox et al. 2013). At , among the 23 pLLSs/LLSs with OVI coverage, only 6 have no OVI absorption, and hence the detection rate of OVI absorption associated with pLLSs/LLSs is about 70% and even higher (75–91%) if only sensitive limits on are considered (Fox et al., 2013). At , a similar number is found with the KODIAQ survey (Lehner et al., 2014). While there is a high frequency of OVI absorption associated with pLLSs/LLSs at both high and low , the similarities in the highly ionized gas properties between the high and low pLLS/LLS sample end there.
The KODIAQ survey shows that for HI-selected absorbers at –3.5 with , the OVI absorption has typically total column densities and full-widths (Lehner et al. 2014; Burns 2014; Lehner 2017; N. Lehner, J.C. Howk, J. O’Meara al. 2016, in prep., and see also Fig. 2 and Appendix). More than half of the KODIAQ sample has and . The breadth and strength of the OVI absorption in strong HI absorbers at –3.5 are quite similar to those observed in starburst galaxies at low redshift (see, e.g., Grimes et al., 2009; Tripp et al., 2011; Muzahid et al., 2015) and remarkably different from those of the OVI absorption in the IGM at similar redshifts (typically and , see Simcoe et al. 2002; Muzahid et al. 2012). This strongly suggests that the bulk of the strong and broad OVI associated with pLLSs and LLSs traces large-scale outflows from high-redshift star-forming galaxies. In contrast, at , OVI absorption in the pLLS sample has typically and (Fox et al., 2013). There is overlap between the low and high surveys, but broad and strong OVI absorption associated with LLSs and pLLSs at is the exception, not the norm. Only two strong HI absorbers with broad ( ) and strong OVI absorption at have been reported so far, both associated with a massive, large-scale outflow from a massive star-forming galaxy (Tripp et al., 2011; Fox et al., 2013; Muzahid et al., 2015). Therefore randomly HI-selected pLLSs and LLSs at and show a dramatic change not only in the MDF of their cool gas but also in the properties of the associated highly ionized gas.
It is likely that the difference in frequency of strong and broad OVI between the low and high pLLS/LLS surveys reflects the fact that low- galaxies are much more quiescent than their high-redshift counterparts. The weaker OVI absorbers associated with pLLSs/LLSs at both low and high have, however, likely a wider range of origins; according to simulations these may include outflows, inflows, ambient CGM (e.g., Shen et al., 2013; Ford et al., 2014).
6.6 pLLSs and LLSs in cosmological simulations
With the first study that extends into the pLLS and low column density LLS regime with at high , we provide new stringent empirical results to test cosmological hydrodynamical simulations. In particular, we demonstrate there is a strong evolution of the metallicity of the pLLSs/LLSs with , but also a remarkably constant fraction of VMP pLLSs/LLSs over cosmic time. For a large proportion of the pLLSs/LLSs at and , C/ also does not follow the typical trend observed in metal-poor Galactic stars or high redshift DLAs (see Fig. 10 and Table 5). As shown by Bird et al. (2014), the simultaneous knowledge of the DLA MDF and column density function can provide strong constraints on the feedback model in cosmological simulations. The same applies for the pLLSs and LLSs for which the evolution of the MDF with starts to be constrained (and more refinement and improvement will come in the near future) and their column density function is also constrained over cosmic time (e.g., Lehner et al., 2007; O’Meara et al., 2007; Prochaska et al., 2010; Ribaudo et al., 2011; Fumagalli et al., 2013).
Simulations have already shown that pLLSs/LLSs may be used to trace cold flows (Faucher-Giguère et al., 2011, 2015; Fumagalli et al., 2011b, 2014; van de Voort & Schaye, 2012; van de Voort et al., 2012; Hafen et al., 2016). Simulated pLLSs and LLSs at –3 and appear, however, to have too many metals (see also discussion in FOP16). Only in simulations with very mild stellar feedback (Fumagalli et al., 2011a), there is some agreement between the observed and simulated metallicity distributions; in this simulation, cold streams are traced mostly by LLSs within 1 or 2 virial radii of galaxies where the gas has only been enriched to with similar scatter to that observed at high or low . However, while mild feedback produces better agreement with the observed MDF at –3, the disagreement with the baryon fraction in stars worsens (Fumagalli et al., 2011a). The zoom-in Eris2 simulations by Shen et al. (2013) include much stronger galactic outflows (but possibly more realistic at these redshifts, see Lehner et al. 2014) and show that cold flows are metal-poor, but with a median value dex, much larger than observed. van de Voort & Schaye (2012) similarly show that cold mode accretion is generally metal-poor with for any halo mass at , and only for does the metallicity of the cold mode accretion go below dex. The FIRE zoom-in simulations at have also recently studied the physical nature of the pLLSs and LLSs (Hafen et al., 2016). These simulations confirm the general interpretation of the bimodal metallicity distribution observed at : very low metallicity LLSs are predominantly associated with inflows at , but higher metallicity LLSs trace gas with roughly equal probability of having recycled outflows (inflows) or outflows. However, the simulated metallicity distribution is not bimodal and has a metallicity plateau between about and dex at . Furthermore, while very low metallicity pLLSs and LLSs are prevalent in the observations, they are not in the FIRE simulations, showing again that the gas is typically too metal rich in simulations.
Nevertheless despite some disagreements between the simulations and the observations, there is a consensus in the simulations that a large fraction of the metal-poor LLSs and pLLSs should probe cold flow accretions onto galaxies. Future simulations with the goals of studying absorbers such as the pLLSs and LLSs (such as in Hafen et al. 2016) that include advanced radiative transfer techniques (crucial for correctly predicting the pLLS/LLS properties) and varying feedback prescriptions will help guiding the interpretation of these observational results, and in turn these observational results should help refining the sub-grid simulation physics and feedback prescriptions.
We have undertaken a study of the properties of the gas probed by pLLSs and LLSs at and the evolution of their properties over cosmic time. Here we present the first results from our KODIAQ Z survey with which we have assembled the first sizable sample of HI-selected pLLSs and LLSs at with (most with ) for which we have determined the metallicity for each absorber. This sample of 31 absorbers therefore probes gas at the transition in between the LYAF () and stronger LLSs (). It provides a direct comparison sample with the sample of L13 and W16 and complements other samples of typically stronger LLSs at similar and higher redshifts (FOP16; Cooper et al. 2015; Glidden et al. 2016).
To derive the metallicity we have used Cloudy simulations assuming a single gas-phase model following the methodology of our early work at low redshift (L13). In particular we have used the same ionizing background (HM05) to avoid introducing additional systematics in our comparison between low and high redshift absorbers. As in L13, we only model the absorption seen in the metals that is associated with the pLLS or LLS HI absorption, i.e., the metallicity is determined by comparing estimated column densities of metal ions and HI in the strongest HI component (not over the entire velocity profile where metal-line absorption may be observed). Our main results are as follows.
Typically the following ions SiII SiIII, SiIV, CII, CIII, CIV associated with the pLLSs or LLSs at are satisfactorily modeled with ionization models with (with a dispersion of 0.6 dex), which imply temperatures (1– K. Based on these Cloudy models, about half of the sample has physical scale kpc and the other half kpc (see Table 4).
We empirically establish that the metallicity distribution of the pLLSs and LLSs at is unimodal peaking at (error on the mean from the survival analysis) with a standard deviation of dex. The mean and distribution are quite similar to those derived for the stronger LLSs () from the HD-LLS survey over the same redshifts. On the other hand, the mean metallicities of the SLLSs () and DLAs () at are higher, and dex, respectively (the dispersion of the metallicities for the DLAs is also also factor 2 smaller). For the LYAF (), the mean metallicity is significantly smaller at similar redshifts, (with a similar dispersion). The mean metallicity of the gas at therefore increases with increasing (with a possible exception for the pLLSs, although a larger sample will be needed to robustly determine this).
There is a substantial fraction (–) of VMP pLLSs and LLSs with metallicities below the mean metallicity of the DLAs (i.e., at ). These VMP pLLSs and LLSs are good candidates of metal-poor cold gas feeding galaxies as seen in cosmological simulations.
At , we determine that the fraction of pLLSs and LLSs with , i.e., at the Pop III remnant level, is – at (68% confidence interval). The lowest metallicity LLS in our sample with a metallicity of has some metals detected with , consistent with a Pop III enrichment. There is no strong evidence ( at the 68% confidence interval) in this sample of pristine pLLS or LLS (i.e., with no metal absorption) with .
About half the sample of the pLLSs and LLSs at and has C/ ratios similar to those derived for MW stars and SLLSs/DLAs with similar metallicities over the entire probed metallicity interval (). The other half has enhanced C/ ratios (near solar values) in the metallicity range , implying that this gas must have been polluted by preferential ejection of C from low metallicity galaxies.
The comparison of the pLLSs and LLSs at and that were selected using the same selection criteria and analyzed using the same procedures shows that some of their properties have not evolved strongly with . The absence of trend between C/ and the metallicity for the pLLSs and LLSs is observed at both high and low . At overlapping metallicities, similar scatter and range of values are observed in C/ at high and low . We show that the fraction of VMP pLLSs/LLSs is 20–47% (68% confidence interval) over the redshift interval to , i.e., over the last 12 billion years the fraction of VMP pLLSs and LLSs appears to remain relatively constant. The hydrogen densities of the pLLSs and LLSs are also similar at both low and high .
On the other hand, several properties of the pLLSs and LLSs have evolved strongly with . The MDF of the pLLSs and LLSs evolves markedly with , changing from a unimodal distribution at that peaks at to a bimodal distribution at with peaks at and . In contrast, the MDF of the DLAs over the same redshift intervals stays unimodal with only an increase of the mean metallicity with decreasing . The ionization parameters, linear scales, and total hydrogen column densities are a factor larger on average at than at .
These first results from the KODIAQ Z survey already put some strong empirical constraints on the dense ionized gas probed by absorbers with and their evolution over 12 billion years of cosmic time, before and after the peak of cosmic star formation. However, our sample is still too small to robustly determine if the pLLS and LLS populations at probe similar or widely different physical structures. At , by doubling the initial sample of pLLSs and LLSs in L13, W16 have demonstrated that the MDF of the pLLSs is bimodal, but likely transitions to a unimodal distribution in the LLS regime. Our ongoing KODIAQ Z survey at and COS Legacy survey at will yield much larger samples of pLLSs, LLSs, as well absorbers with at both high and low , which will provide new stringent constraints on the properties of the diffuse and dense ionized gas at .
Support for this research was partially made by NASA through the Astrophysics Data Analysis Program (ADAP) grant NNX16AF52G. MF acknowledges support by the Science and Technology Facilities Council through grant ST/L00075X/1. Part of this manuscript was written at the winter 2016 retreat hosted by IMPS of UC Santa Cruz Department of Astronomy. We thank the Esalen Institute for its great setting and wonderful hospitality during that retreat. All the data presented in this work were obtained from the Keck Observatory Database of Ionized Absorbers toward QSOs (KODIAQ), which was funded through NASA ADAP grant NNX10AE84G. This research has made use of the Keck Observatory Archive (KOA), which is operated by the W. M. Keck Observatory and the NASA Exoplanet Science Institute (NExScI), under contract with the National Aeronautics and Space Administration.The authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of Mauna Kea has always had within the indigenous Hawaiian community.
- Aguirre et al. (2008) Aguirre, A., Dow-Hygelund, C., Schaye, J., & Theuns, T. 2008, ApJ, 689, 851
- Aguirre et al. (2004) Aguirre, A., Schaye, J., Kim, T.-S., et al. 2004, ApJ, 602, 38
- Akerman et al. (2004) Akerman, C. J., Carigi, L., Nissen, P. E., Pettini, M., & Asplund, M. 2004, A&A, 414, 931
- Asplund et al. (2009) Asplund, M., Grevesse, N., Sauval, A. J., & Scott, P. 2009, ARA&A, 47, 481
- Battisti et al. (2012) Battisti, A. J., Meiring, J. D., Tripp, T. M., et al. 2012, ApJ, 744, 93
- Becker et al. (2012) Becker, G. D., Sargent, W. L. W., Rauch, M., & Carswell, R. F. 2012, ApJ, 744, 91
- Bensby & Feltzing (2006) Bensby, T., & Feltzing, S. 2006, MNRAS, 367, 1181
- Berg et al. (2016) Berg, D. A., Skillman, E. D., Henry, R. B. C., Erb, D. K., & Carigi, L. 2016, ApJ, 827, 126
- Bird et al. (2014) Bird, S., Vogelsberger, M., Haehnelt, M., et al. 2014, MNRAS, 445, 2313
- Bordoloi et al. (2014) Bordoloi, R., Tumlinson, J., Werk, J. K., et al. 2014, ApJ, 796, 136
- Bowen et al. (2002) Bowen, D. V., Pettini, M., & Blades, J. C. 2002, ApJ, 580, 169
- Bromm & Loeb (2003) Bromm, V., & Loeb, A. 2003, Nature, 425, 812
- Burns (2014) Burns, V. 2014, A High Resolution Study of Circumgalactic O VI Absorbers at , University of Notre Dame Senior Honors Thesis, , , undergraduate Thesis
- Chen et al. (2005) Chen, H.-W., Prochaska, J. X., Weiner, B. J., Mulchaey, J. S., & Williger, G. M. 2005, ApJ, 629, L25
- Conroy & Wechsler (2009) Conroy, C., & Wechsler, R. H. 2009, ApJ, 696, 620
- Cooke et al. (2011) Cooke, R., Pettini, M., Steidel, C. C., Rudie, G. C., & Nissen, P. E. 2011, MNRAS, 417, 1534
- Cooksey et al. (2008) Cooksey, K. L., Prochaska, J. X., Chen, H.-W., Mulchaey, J. S., & Weiner, B. J. 2008, ApJ, 676, 262
- Cooper et al. (2015) Cooper, T. J., Simcoe, R. A., Cooksey, K. L., O’Meara, J. M., & Torrey, P. 2015, ApJ, 812, 58
- Crighton et al. (2013) Crighton, N. H. M., Hennawi, J. F., & Prochaska, J. X. 2013, ApJ, 776, L18
- Crighton et al. (2015) Crighton, N. H. M., Hennawi, J. F., Simcoe, R. A., et al. 2015, MNRAS, 446, 18
- Crighton et al. (2016) Crighton, N. H. M., O’Meara, J. M., & Murphy, M. T. 2016, MNRAS, 457, L44
- Davé et al. (1999) Davé, R., Hernquist, L., Katz, N., & Weinberg, D. H. 1999, ApJ, 511, 521
- Dekel & Birnboim (2006) Dekel, A., & Birnboim, Y. 2006, MNRAS, 368, 2
- Ellison et al. (2000) Ellison, S. L., Songaila, A., Schaye, J., & Pettini, M. 2000, AJ, 120, 1175
- Fabbian et al. (2010) Fabbian, D., Khomenko, E., Moreno-Insertis, F., & Nordlund, Å. 2010, ApJ, 724, 1536
- Faucher-Giguère et al. (2015) Faucher-Giguère, C.-A., Hopkins, P. F., Kereš, D., et al. 2015, MNRAS, 449, 987
- Faucher-Giguère et al. (2011) Faucher-Giguère, C.-A., Kereš, D., & Ma, C.-P. 2011, MNRAS, 417, 2982
- Feigelson & Nelson (1985) Feigelson, E. D., & Nelson, P. I. 1985, ApJ, 293, 192
- Ferland et al. (2013) Ferland, G. J., Porter, R. L., van Hoof, P. A. M., et al. 2013, Rev. Mexicana Astron. Astrofis., 49, 137
- Fitzpatrick & Spitzer (1997) Fitzpatrick, E. L., & Spitzer, Jr., L. 1997, ApJ, 475, 623
- Ford et al. (2014) Ford, A. B., Davé, R., Oppenheimer, B. D., et al. 2014, MNRAS, 444, 1260
- Fox et al. (2013) Fox, A. J., Lehner, N., Tumlinson, J., et al. 2013, ApJ, 778, 187
- Frebel et al. (2007) Frebel, A., Johnson, J. L., & Bromm, V. 2007, MNRAS, 380, L40
- Fukugita et al. (1998) Fukugita, M., Hogan, C. J., & Peebles, P. J. E. 1998, ApJ, 503, 518
- Fumagalli et al. (2016a) Fumagalli, M., Cantalupo, S., Dekel, A., et al. 2016a, MNRAS, 462, 1978
- Fumagalli et al. (2014) Fumagalli, M., Hennawi, J. F., Prochaska, J. X., et al. 2014, ApJ, 780, 74
- Fumagalli et al. (2011a) Fumagalli, M., O’Meara, J. M., & Prochaska, J. X. 2011a, Science, 334, 1245
- Fumagalli et al. (2016b) —. 2016b, MNRAS, 455, 4100
- Fumagalli et al. (2013) Fumagalli, M., O’Meara, J. M., Prochaska, J. X., & Worseck, G. 2013, ApJ, 775, 78
- Fumagalli et al. (2011b) Fumagalli, M., Prochaska, J. X., Kasen, D., et al. 2011b, MNRAS, 418, 1796
- Genzel et al. (2010) Genzel, R., Tacconi, L. J., Gracia-Carpio, J., et al. 2010, MNRAS, 407, 2091
- Glidden et al. (2016) Glidden, A., Cooper, T. J., Cooksey, K. L., Simcoe, R. A., & O’Meara, J. M. 2016, ArXiv e-prints, arXiv:1604.02144, submitted to the ApJ
- Grimes et al. (2009) Grimes, J. P., Heckman, T., Aloisi, A., et al. 2009, ApJS, 181, 272
- Haardt & Madau (1996) Haardt, F., & Madau, P. 1996, ApJ, 461, 20
- Haardt & Madau (2012) —. 2012, ApJ, 746, 125
- Hafen et al. (2016) Hafen, Z., Faucher-Giguere, C.-A., Angles-Alcazar, D., et al. 2016, ApJ, arXiv:1608.05712, submitted
- Hartigan & Hartigan (1985) Hartigan, J. A., & Hartigan, P. M. 1985, The Annals of Statistics, 13, 70
- Isobe et al. (1986) Isobe, T., Feigelson, E. D., & Nelson, P. I. 1986, ApJ, 306, 490
- Jenkins (2009) Jenkins, E. B. 2009, ApJ, 700, 1299
- Jimmy et al. (2015) Jimmy, Tran, K.-V., Saintonge, A., et al. 2015, ApJ, 812, 98
- Jorgenson et al. (2013) Jorgenson, R. A., Murphy, M. T., & Thompson, R. 2013, MNRAS, 435, 482
- Kereš & Hernquist (2009) Kereš, D., & Hernquist, L. 2009, ApJ, 700, L1
- Kereš et al. (2009) Kereš, D., Katz, N., Fardal, M., Davé, R., & Weinberg, D. H. 2009, MNRAS, 395, 160
- Kereš et al. (2005) Kereš, D., Katz, N., Weinberg, D. H., & Davé, R. 2005, MNRAS, 363, 2
- Lanzetta et al. (1995) Lanzetta, K. M., Bowen, D. V., Tytler, D., & Webb, J. K. 1995, ApJ, 442, 538
- Ledoux et al. (2002) Ledoux, C., Bergeron, J., & Petitjean, P. 2002, A&A, 385, 802
- Lehner (2017) Lehner, N. 2017, Gas Accretion via Lyman Limit Systems, ed. A. Fox & R. Davé, Astrophysics and Space Science Library Series (Springer)
- Lehner et al. (2015) Lehner, N., Howk, J. C., & Wakker, B. P. 2015, ApJ, 804, 79
- Lehner et al. (2014) Lehner, N., O’Meara, J. M., Fox, A. J., et al. 2014, ApJ, 788, 119
- Lehner et al. (2009) Lehner, N., Prochaska, J. X., Kobulnicky, H. A., et al. 2009, ApJ, 694, 734
- Lehner et al. (2007) Lehner, N., Savage, B. D., Richter, P., et al. 2007, ApJ, 658, 680
- Lehner et al. (2013) Lehner, N., Howk, J. C., Tripp, T. M., et al. 2013, ApJ, 770, 138
- Liang & Chen (2014) Liang, C. J., & Chen, H.-W. 2014, MNRAS, 445, 2061
- Madau et al. (2014) Madau, P., Shen, S., & Governato, F. 2014, ApJ, 789, L17
- Martin et al. (2012) Martin, C. L., Shapley, A. E., Coil, A. L., et al. 2012, ApJ, 760, 127
- Morton (2003) Morton, D. C. 2003, ApJS, 149, 205
- Muratov & Gnedin (2010) Muratov, A. L., & Gnedin, O. Y. 2010, ApJ, 718, 1266
- Muzahid et al. (2015) Muzahid, S., Kacprzak, G. G., Churchill, C. W., et al. 2015, ApJ, 811, 132
- Muzahid et al. (2012) Muzahid, S., Srianand, R., Bergeron, J., & Petitjean, P. 2012, MNRAS, 421, 446
- Nicholls et al. (2014) Nicholls, D. C., Dopita, M. A., Sutherland, R. S., et al. 2014, ApJ, 786, 155
- O’Meara et al. (2007) O’Meara, J. M., Prochaska, J. X., Burles, S., et al. 2007, ApJ, 656, 666
- O’Meara et al. (2015) O’Meara, J. M., Lehner, N., Howk, J. C., et al. 2015, AJ, 150, 111
- Peeples et al. (2014) Peeples, M. S., Werk, J. K., Tumlinson, J., et al. 2014, ApJ, 786, 54
- Penprase et al. (2010) Penprase, B. E., Prochaska, J. X., Sargent, W. L. W., Toro-Martinez, I., & Beeler, D. J. 2010, ApJ, 721, 1
- Penton et al. (2002) Penton, S. V., Stocke, J. T., & Shull, J. M. 2002, ApJ, 565, 720
- Pettini et al. (2001) Pettini, M., Shapley, A. E., Steidel, C. C., et al. 2001, ApJ, 554, 981
- Pettini et al. (2008) Pettini, M., Zych, B. J., Steidel, C. C., & Chaffee, F. H. 2008, MNRAS, 385, 2011
- Prochaska (1999) Prochaska, J. X. 1999, ApJ, 511, L71
- Prochaska et al. (2003a) Prochaska, J. X., Gawiser, E., Wolfe, A. M., Castro, S., & Djorgovski, S. G. 2003a, ApJ, 595, L9
- Prochaska et al. (2003b) Prochaska, J. X., Gawiser, E., Wolfe, A. M., Cooke, J., & Gelino, D. 2003b, ApJS, 147, 227
- Prochaska et al. (2005) Prochaska, J. X., Herbert-Fort, S., & Wolfe, A. M. 2005, ApJ, 635, 123
- Prochaska et al. (2015) Prochaska, J. X., O’Meara, J. M., Fumagalli, M., Bernstein, R. A., & Burles, S. M. 2015, ApJS, 221, 2
- Prochaska et al. (2010) Prochaska, J. X., O’Meara, J. M., & Worseck, G. 2010, ApJ, 718, 392
- Prochter et al. (2010) Prochter, G. E., Prochaska, J. X., O’Meara, J. M., Burles, S., & Bernstein, R. A. 2010, ApJ, 708, 1221
- Quiret et al. (2016) Quiret, S., Péroux, C., Zafar, T., et al. 2016, MNRAS, 458, 4074
- Rafelski et al. (2012) Rafelski, M., Wolfe, A. M., Prochaska, J. X., Neeleman, M., & Mendez, A. J. 2012, ApJ, 755, 89
- Ribaudo et al. (2011) Ribaudo, J., Lehner, N., Howk, J. C., et al. 2011, ApJ, 743, 207
- Rubin et al. (2012) Rubin, K. H. R., Prochaska, J. X., Koo, D. C., & Phillips, A. C. 2012, ApJ, 747, L26
- Rubin et al. (2014) Rubin, K. H. R., Prochaska, J. X., Koo, D. C., et al. 2014, ApJ, 794, 156
- Rudie et al. (2012) Rudie, G. C., Steidel, C. C., & Pettini, M. 2012, ApJ, 757, L30
- Savage & Sembach (1991) Savage, B. D., & Sembach, K. R. 1991, ApJ, 379, 245
- Savage & Sembach (1996) —. 1996, ARA&A, 34, 279
- Schaye et al. (2003) Schaye, J., Aguirre, A., Kim, T.-S., et al. 2003, ApJ, 596, 768
- Sembach & Savage (1992) Sembach, K. R., & Savage, B. D. 1992, ApJS, 83, 147
- Shapley et al. (2003) Shapley, A. E., Steidel, C. C., Pettini, M., & Adelberger, K. L. 2003, ApJ, 588, 65
- Shen et al. (2013) Shen, S., Madau, P., Guedes, J., et al. 2013, ApJ, 765, 89
- Simcoe (2011) Simcoe, R. A. 2011, ApJ, 738, 159
- Simcoe et al. (2002) Simcoe, R. A., Sargent, W. L. W., & Rauch, M. 2002, ApJ, 578, 737
- Simcoe et al. (2004) —. 2004, ApJ, 606, 92
- Skillman et al. (1989) Skillman, E. D., Kennicutt, R. C., & Hodge, P. W. 1989, ApJ, 347, 875
- Som et al. (2013) Som, D., Kulkarni, V. P., Meiring, J., et al. 2013, MNRAS, 435, 1469
- Som et al. (2015) —. 2015, ApJ, 806, 25
- Spite et al. (2005) Spite, M., Cayrel, R., Plez, B., et al. 2005, A&A, 430, 655
- Steidel (1990) Steidel, C. C. 1990, ApJS, 74, 37
- Steidel et al. (2010) Steidel, C. C., Erb, D. K., Shapley, A. E., et al. 2010, ApJ, 717, 289
- Steidel et al. (2004) Steidel, C. C., Shapley, A. E., Pettini, M., et al. 2004, ApJ, 604, 534
- Stocke et al. (2013) Stocke, J. T., Keeney, B. A., Danforth, C. W., et al. 2013, ApJ, 763, 148
- Tremonti et al. (2004) Tremonti, C. A., Heckman, T. M., Kauffmann, G., et al. 2004, ApJ, 613, 898
- Tripp et al. (2011) Tripp, T. M., Meiring, J. D., Prochaska, J. X., et al. 2011, Science, 334, 952
- Tumlinson et al. (2011) Tumlinson, J., Thom, C., Werk, J. K., et al. 2011, Science, 334, 948
- van de Voort & Schaye (2012) van de Voort, F., & Schaye, J. 2012, MNRAS, 423, 2991
- van de Voort et al. (2012) van de Voort, F., Schaye, J., Altay, G., & Theuns, T. 2012, MNRAS, 421, 2809
- Vogt et al. (1994) Vogt, S. S., Allen, S. L., Bigelow, B. C., et al. 1994, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, Vol. 2198, Instrumentation in Astronomy VIII, ed. D. L. Crawford & E. R. Craine, 362
- Weiner et al. (2009) Weiner, B. J., Coil, A. L., Prochaska, J. X., et al. 2009, ApJ, 692, 187
- Welty et al. (1999) Welty, D. E., Hobbs, L. M., Lauroesch, J. T., et al. 1999, ApJS, 124, 465
- Werk et al. (2014) Werk, J. K., Prochaska, J. X., Tumlinson, J., et al. 2014, ApJ, 792, 8
- White & Rees (1978) White, S. D. M., & Rees, M. J. 1978, MNRAS, 183, 341
- Wotta et al. (2016) Wotta, C. B., Lehner, N., Howk, J. C., O’Meara, J. M., & Prochaska, J. X. 2016, ApJ, arXiv:1608.02584, in press
|J143316+313126 – –|
|J030341–002321 – –|
|J014516–094517A – –|
|J172409+531405 – –|