THE MOST METAL-POOR STARS. II. CHEMICAL ABUNDANCES OF 190 METAL-POOR STARS INCLUDING 10 NEW STARS WITH [Fe/H] ⩽ −3.5^*,^†,^‡

In Norris et al. (2013a; Paper I) we describe high-resolution spectroscopic observations of 38 program stars (obtained using the Keck, Magellan, and VLT telescopes), including the discovery and sample selection, equivalent-width measurements, radial velocities, and line list. Our sample comprises 34 stars original to the present work, together with four for which published abundances already exist, or which are the subject of analyses currently underway. In Paper I, we also describe the temperature scale, which consists of spectrophotometry and Balmer-line analysis. We refer the reader to these works for the details of the observational data upon which the present analysis is based.

With effective temperatures, T_eff, and equivalent widths in hand, our analysis proceeded in the following manner. Surface gravities,¹⁰ log g, were taken from the Y² isochrones (Demarque et al. 2004), assuming an age of 10 Gyr and [α/Fe] = +0.3. We note that changing the age from 10 Gyr to 13 Gyr (or 7 Gyr) would only introduce a systematic difference in log g of ⩽0.1 dex. We also note that these isochrones only extend down to [Fe/H] = −3.5; therefore, the surface gravity we obtain for more metal-poor stars involves a linear extrapolation, from [Fe/H] = −3.5 down to [Fe/H] = −4.1, for the most metal-poor stars in our program sample. For our four most metal-poor stars, we note that the average difference between the surface gravity inferred using [Fe/H] = −3.5 (the boundary value of the Y² isochrones) and the extrapolated surface gravity using the actual [Fe/H] is 0.06 dex (for the giant/subgiant case) and 0.01 dex (for the dwarf case). Initial estimates of the metallicity came from the medium-resolution follow-up spectroscopy.

Model atmospheres were taken from the α-enhanced, [α/Fe] = +0.4, NEWODF grid of ATLAS9 models by Castelli & Kurucz (2003). These one-dimensional (1D), plane-parallel, local thermodynamic equilibrium (LTE) models were computed using a microturbulent velocity of 2 km s⁻¹ and no convective overshooting. Interpolation within the grid was necessary to produce models with the required combination of T_eff, log g, and [M/H]. The interpolation software, kindly provided by Dr. Carlos Allende Prieto, has been used extensively (e.g., Reddy et al. 2003; Allende Prieto et al. 2004), and involves linear interpolation in three dimensions (T_eff, log g, and [M/H]) to produce the required model.

The final tool in our analysis kit was the LTE stellar line-analysis program MOOG (Sneden 1973). The particular version of MOOG that we used includes a proper treatment of continuum scattering (see Sobeck et al. 2011 for further details). We refer the reader to Cayrel et al. (2004) and Sobeck et al. (2011) for a discussion regarding the importance of Raleigh scattering (Griffin et al. 1982) at blue wavelengths in metal-poor stars.

Having computed the abundance for each line, the microturbulent velocity was determined, in the usual way, by forcing the abundances from Fe i lines to have no trend with the reduced equivalent width, log (W_λ/λ). The metallicity, [Fe/H], was inferred exclusively from Fe i lines. While we are mindful that such lines are more susceptible to non-LTE effects than Fe ii lines (e.g., Asplund 2005), we were unable to measure any Fe ii lines for a number of program stars. Higher-quality spectra are necessary to measure additional Fe ii lines in our sample. In Section 6, we shall compare iron abundances derived from neutral and ionized species for those stars for which data are available.

With an updated estimate of the metallicity, [Fe/H]_star, we then compared this value with the metallicity assumed when generating the model atmosphere, [M/H]_model. If the difference exceeded 0.1 dex, we computed an updated model atmosphere with [M/H]_new = [Fe/H]_star. Based on the new metallicity, the surface gravity was revised and the star was re-analyzed. When this was required, we note that the abundance from Fe i lines and the metallicity of the model atmosphere converged within one or two iterations. That is, the inferred abundance, [Fe/H], is only weakly dependent on the metallicity of the model, [M/H], provided the initial guess is close to the final value. During the analysis process, we removed Fe i lines for which the abundance differed from (1) the median abundance by more than 0.5 dex or (2) the median abundance by more than 3σ. (Lines yielding abundances higher or lower than the median value by more than 0.5 dex or 3σ were rejected.) This criterion resulted in the rejection of a handful of lines for a given star. The largest number of rejected lines for a given star was 6, leaving 33 accepted lines, while the largest fraction of rejected lines was 3, leaving 15 accepted lines.

In the course of our analysis, another consideration was whether or not a given Fe line might be blended with CH molecular lines. Therefore, we repeated the entire analysis using a subset of lines which spectrum synthesis suggests are not blended with CH (see Norris et al. 1997, 2010a for further details). For the microturbulent velocity and metallicity, the results from the two approaches are very similar. Once we had measured the [C/Fe] abundance ratio, we adopted the results using the CH-free line list if the program star was a carbon-enhanced metal-poor (CEMP) object (we applied the Aoki et al. 2007 CEMP definition). Details on the C measurements and CEMP definition are provided in Sections 2.2 and 7.1. In Table 1, we present the stellar parameters for the program stars. The evolutionary status, T_eff versus log g, for the program stars is shown in Figure 1.

Zoom In Zoom Out Reset image size

We note here that for the nine stars for which we conducted separate analyses using a dwarf gravity and a subgiant gravity, the average difference in iron abundance between the dwarf and subgiant analyses is 〈[Fe/H]_dwarf − [Fe/H]_subgiant〉 = 0.02 ± 0.01 dex (σ = 0.03); the relative abundance is 〈[X/Fe]_dwarf − [X/Fe]_subgiant〉 = 0.05 ± 0.02 dex (σ = 0.16). Given these modest differences, for a given star we average the abundance ratios [Fe/H] and [X/Fe] for the dwarf and subgiant cases. Unless noted otherwise, we use these average values for a given star throughout the remainder of the paper. We report the values for the individual analyses in the relevant tables.

The abundances of atomic lines were computed using our measured equivalent widths and log gf values from Paper I, final model atmospheres, and MOOG. Lines of Sc ii, Mn i, Co i, and Ba ii are affected by hyperfine splitting. In our analysis, we treated these lines appropriately, using the data from Kurucz & Bell (1995). Additionally, Ba ii lines are affected by isotopic splitting. Our Ba abundances were computed assuming the McWilliam (1998) r-process isotopic composition and hyperfine splitting. For a restricted number of elements in some stars, we determined upper limits to the chemical abundance, based on equivalent-width limits presented in Paper I.

Given the low metallicities of our sample and the S/N of our spectra, we are not well positioned to measure the abundance of oxygen. That said, we have determined the abundance (or its upper limit) of this element for six C-rich (i.e., CEMP) stars in our sample, which we shall discuss further in Paper IV (Norris et al. 2013b). For HE 2139−5432, the oxygen abundance was derived from analysis of the 7771.94 Å, 7774.17 Å, and 7775.39 Å lines. The measured equivalent widths for these lines are 20.1 mÅ, 18.7 mÅ, and 12.2 mÅ and the adopted log gf values are 0.324, 0.174, and −0.046, respectively. Thus, we obtained an LTE abundance of A(O) = 7.82 (σ = 0.05) for HE 2139−5432, which corresponds to [O/Fe] = +3.15 (we adopt the Asplund et al. 2009 solar abundances). For HE 0146–1548 and HE 1506−0113, the O limits ([O/Fe] < +1.63 and +2.32, respectively) were determined using an equivalent width of <10mÅ for the 6300.30 Å line, adopting log gf = −9.820. Finally, for 53327-2044-515, HE 0057−5959, and HE 1201−1512, the O limits ([O/Fe] < +2.81, +2.77, and +2.64, respectively) were determined using an equivalent width of 10 mÅ for the 7771.94 Å line (where for 53327-2044-515 and HE 1201−1512 the abundance value for each star is the mean of the low- and high-gravity solutions).

For C and N, abundances (or upper limits) were determined from analysis of the (0–0) and (1–1) bands of the A−X electronic transitions of the CH molecule (4290 Å to 4330 Å) and the NH molecule (3350 Å to 3370 Å). We compared synthetic spectra, generated using MOOG, with the observed spectra and adjusted the input abundance until the two spectra matched. The macroturbulent broadening was determined using a Gaussian representing the combined effects of the instrumental profile, atmospheric turbulence, and stellar rotation. The width of this Gaussian was estimated during the course of the spectrum synthesis fitting, and the C and N abundances are thus (slightly) sensitive to the adopted broadening. Following the analysis described in Norris et al. (2010b), the CH line list was that compiled by B. Plez et al. (2009, private communication). We used a dissociation energy of 3.465 eV. The abundances of C and O are coupled through the CO molecule. In the absence of an O abundance measurement, we arbitrarily adopted a halo-like value of [O/Fe] = +0.4, and note that for our program stars, the derived C abundance is insensitive to the adopted O abundance (i.e., for a handful of stars, we adopted [O/Fe] = 0.0 and [O/Fe] = +1.5, and the derived C abundance was unchanged). The NH line list was the same as in Johnson et al. (2007), in which the Kurucz-gf values were reduced by a factor of two, and the dissociation energy was 3.450 eV. Given the low S/N at these wavelengths, we smoothed the observed spectra with a 5 pixel boxcar function to increase the S/N (at the expense of spectral resolution). The N abundance was adjusted until the synthetic spectra matched the observed spectra. In Figures 2–4, we show examples of the spectrum synthesis.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

As with the analysis of Fe lines, we repeated the element abundance analysis using the subset of lines believed to be unaffected by CH blends, combined with the stellar parameters obtained using the set of “CH clean” Fe lines. Depending on whether the star was found to be a CEMP object, we adopted the element abundances associated with the relevant element abundance analysis. In particular, we note that our line list has only one Si line, 3905.52 Å, and that line is excluded in the “CH clean” set of lines. Therefore, there are CEMP stars with Si equivalent-width measurements in Paper I without Si abundance measurements. For these CEMP stars, we tried to measure Si abundances (or limits) from spectrum synthesis of the 4102.94 Å Si line. We present abundance ratios in Table 2 (C and N) and Table 3 (Na to Ba). The adopted solar abundances for all elements were from Asplund et al. (2009).

Table 3. Chemical Abundances (Na–Ba) for the Program Stars

Star	A(X)	Nlines	s.e.loga	Total Errorb	[X/Fe]
(1)	(2)	(3)	(4)	(5)	(6)
Na
52972-1213-507	4.86	2	0.32	0.33	1.60
53327-2044-515c	2.35	1	⋅⋅⋅	⋅⋅⋅	0.11
53327-2044-515d	2.32	1	⋅⋅⋅	⋅⋅⋅	0.17
53436-1996-093	2.58	1	⋅⋅⋅	⋅⋅⋅	−0.13
54142-2667-094	⋅⋅⋅	⋅⋅⋅	⋅⋅⋅	⋅⋅⋅	⋅⋅⋅

Notes. ^aStandard error of the mean. ^bTotal error is the quadratic sum of the updated random error and the systematic error (see Section 2.3 for details). ^cDwarf gravity is assumed (see Section 2.1 for details). ^dSubgiant gravity is assumed (see Section 2.1 for details). ^eAbundances, or limits, were determined from spectrum synthesis of the 4102.94 Å Si line. ^fAbundances limits were determined from the 4077.71 Å Sr line. ^gAbundances limits were determined from the 4554.03 Å Ba line.

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image

Our abundance measurements are subject to uncertainties in the model parameters. We estimated these uncertainties to be T_eff ± 100 K, log g ± 0.3 dex, ξ_t ± 0.3 km s⁻¹, and [M/H] ± 0.3 dex. To determine the abundance errors, we repeated our analysis varying our parameters, one at a time, assuming that the errors are symmetric for positive and negative changes. We present these abundance uncertainty estimates in Table 4, in which the final column is the accumulated error when the four uncertainties are added quadratically. Strictly speaking, quadratic addition is appropriate if the errors are fully independent. Additionally, our approach neglects covariances, and we refer the interested reader to McWilliam et al. (1995), Johnson (2002), and Barklem et al. (2005) for a more detailed discussion. Note that the contribution from the uncertainties in [M/H] to the total error budget is small, in general. Therefore, our condition to re-compute a model atmosphere only when |[M/H]_model − [Fe/H]_star| > 0.1 dex does not adversely affect our results.

To obtain the total error estimates given in Table 3, we follow Norris et al. (2010b). We replace the random error in Table 3 (s.e._log) by max(s.e._log, 0.20/$\sqrt{N_{\rm lines}}$) where the second term is what would be expected for a set of N_lines with a dispersion of 0.20 dex (a conservative estimate for the abundance dispersion of Fe i lines). The total error is obtained by quadratically adding this updated random error with the systematic error in Table 4. Finally, we note that this 1D LTE analysis is subject to abundance uncertainties from three-dimensional (3D) and non-LTE effects (Asplund 2005).

For a subset of the program stars, we computed abundances using the MARCS (Gustafsson et al. 2008) grid of model atmospheres. Three stars were chosen to sample a reasonable range of stellar parameters: two giants and one main-sequence turnoff star. In Table 5 we show the abundance differences for A(X), in the sense Castelli & Kurucz (2003) − MARCS. For these representative objects, the maximum abundance difference was ΔA(X) = 0.05 dex, and the minimum abundance difference was ΔA(X) = −0.02 dex. When considering the ratio [X/Fe], we note that the maximum abundance difference was Δ[X/Fe] = 0.02 dex, and the minimum abundance difference was Δ[X/Fe] = −0.01 dex. We regard these abundance differences, ΔA(X) and Δ[X/Fe], to be small, especially when compared to the abundance uncertainties and errors in Tables 3 and 4. Therefore, we do not expect the choice of model atmosphere grid, Castelli & Kurucz (2003) or MARCS (Gustafsson et al. 2008), to significantly alter our abundance results or subsequent interpretation. However, we note that a more complete chemical abundance analysis would require, among other things, tailored models with appropriate CNO abundances.

We adopted the Aoki et al. (2007) definition for CEMP stars, which accounts for nucleosynthesis and mixing in evolved giants, namely, (1) [C/Fe] ⩾ +0.70, for log (L/L_☉) ⩽ 2.3, and (2) [C/Fe] ⩾ +3.0 − log (L/L_☉), for log (L/L_☉) > 2.3. The original definition proposed by Beers & Christlieb (2005) is [C/Fe] ⩾ +1.0. In Figure 13, we plot [C/Fe] versus log (L/L_☉), showing both CEMP definitions. We refer the reader to Paper III for more discussion of the CEMP fraction (and the metallicity distribution function) at lowest metallicity. (Throughout the present paper, we use CEMP and C-rich interchangeably.) None of our program stars are C-normal objects. This is likely due to selection biases and that we could only obtain [C/Fe] limits for many program stars.

Zoom In Zoom Out Reset image size

In Figure 14, we plot abundance ratios [X/Fe] versus [Fe/H] for the 38 program stars. The CEMP stars are marked in red in each panel. The green line in this figure represents the predictions from the Galactic chemical enrichment models of Kobayashi et al. (2006), which will also be discussed in Sections 7.5 and 7.7. For all elements, the abundance dispersion exceeds the measurement uncertainty. In particular, Na, Sr, and Ba exhibit very large dispersions, as do C and N. Before we consider the complete sample (program stars + literature stars), we note that the abundances of Mg and Si appear to be lower than the canonical halo value of [α/Fe] = +0.4. Furthermore, for a given element, the outliers are often, but not always, CEMP objects. However, for many of our stars we could only obtain upper limits for the C abundance. It would be interesting to re-assess whether the abundance outliers are always CEMP objects by either (1) restricting the sample to those stars with C measurements and/or (2) obtaining higher-quality spectra to convert upper limits for C into detections.

Zoom In Zoom Out Reset image size

For each element from C to Ba, we plot the abundance ratio [X/Fe] versus [Fe/H] for the combined sample, i.e., program stars and literature stars (Figures 15–30). In all figures, the left panels show only the dwarf stars (log g > 3.0), and the right panels show only the giant stars (log g < 3.0). In Figures 17–30, we determine the linear fit to the data in the following manner. First, we exclude all CEMP stars from the fit. Second, we determine the linear fit and measure the dispersion about the mean trend. Third, we eliminate 2σ outliers from the fit. Fourth, we re-determine the linear fit and show in the plots (1) the slope of the fit and its associated uncertainty, (2) the dispersion about the fit, and (3) the mean abundance, [X/Fe], and the standard deviation. The reason for excluding C-rich objects and 2σ outliers was that we were seeking to (1) identify a “normal” population of metal-poor stars and (2) characterize the mean trend between [X/Fe] and [Fe/H] for this “normal” population. We emphasize that while CEMP objects and 2σ outliers were not included in determining the best fits to the data, these objects are included in all plots.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Our motivation for attempting to define a normal population comes from the First Stars analyses (Cayrel et al. 2004; Spite et al. 2005; François et al. 2007; Bonifacio et al. 2009). As mentioned, these studies (along with Arnone et al. 2005) found extremely small scatter, which is likely due in part to the high-quality data and analysis, and to the fact that their samples included only two CEMP giants (CS 22949−037 and CS 22892−052) and one CEMP dwarf (CS 29527−015). Thus, they are “biased against carbon-rich objects and cannot be used to constrain the full dispersion of carbon abundances at the lowest metallicities” (Cayrel et al. 2004, p. 10). As identified in the literature, and confirmed in the present series, CEMP stars often have anomalous abundance ratios for elements other than C. However, it has also become evident that some C-normal stars show peculiar abundances for other elements (e.g., Cohen et al. 2008). Therefore, when searching for a “normal” population of metal-poor stars, we eliminate CEMP objects as well as 2σ outliers from the fit, but retain them in the plots.

In these figures, we also present contour plots illustrating the density in the [X/Fe] versus [Fe/H] plane. We consider (1) C-normal dwarfs, (2) CEMP dwarfs, (3) C-normal giants, and (4) CEMP giants. For each sample, we represent a given data point with a two-dimensional Gaussian, for which the FWHM in the [Fe/H] and [X/Fe] directions corresponds to our estimates of the typical measurement uncertainties for [Fe/H] and [X/Fe], respectively. The height of each Gaussian is set to 1.0. The Gaussians are then summed and a contour plot is generated. In the event that a given panel has N data points with a single value of [Fe/H] and [X/Fe], the contour would have a maximum height of N. One advantage of using such plots is that the density of data points can be more readily seen, which may be useful given our sample size.

Although we determine linear fits to the data throughout this paper, we are not suggesting that this is the appropriate function to use. Rather, we consider this a first pass to begin to understand how the abundance ratios, [X/Fe], evolve with metallicity, [Fe/H]. Such an approach has the advantage of enabling comparisons between different objects (dwarfs versus giants), different elements, and different studies. Without knowing what is the appropriate function to use, another option might be to follow Norris et al. (2001), and use LOESS regression lines (Cleveland 1979).

The reason for separating the dwarfs from giants is that we want to compare, as best we can, stars with similar stellar parameters to minimize systematic errors (e.g., Cayrel et al. 2004; Meléndez et al. 2008; Bonifacio et al. 2009; Alves-Brito et al. 2010; Nissen & Schuster 2010).¹³ Additionally, we distinguish dwarfs from giants because, for some elements (notably C and N), we anticipate abundance differences due to stellar evolution. Thus, by considering dwarfs and giants separately, we hope to minimize such effects and thereby more clearly identify a “normal” population of stars, and quantify the abundance trends [X/Fe] versus [Fe/H] for the “normal” dwarf sample and the “normal” giant sample.

A measure of (1) whether a “normal” population indeed exists, and (2) whether our selection criteria are able to identify such a population, is to compare the dispersion about the linear fit to the representative measurement uncertainty (the average “total error” for the program stars). (Such a comparison is only meaningful if, as we assume, the dependence of [X/Fe] versus [Fe/H] is linear. We reiterate that while we use linear functions, we are not suggesting that they are correct. Instead, this assumption represents a first step to understanding trends between [X/Fe] and [Fe/H].) In Table 8, we compare the observed dispersion about the linear fit to the typical, i.e., average, measurement uncertainty (the lower panel of Figure 31 shows the comparison for a subset of elements in giant stars). In many cases (e.g., Al, Ca, Ti, Mn, Co, Ni), the dispersion in [X/Fe] about the linear fit is in good agreement with the representative measurement uncertainty. There are examples in which the representative uncertainty exceeds the dispersion (e.g., Sc, Cr), which may indicate that the uncertainties are overestimated. For other cases (e.g., Sr, Ba), the dispersion about the linear trend far exceeds the measurement uncertainty, suggesting that the uncertainties are underestimated, there is a large abundance dispersion, and/or the “normal” population, if present, was not successfully identified.

Zoom In Zoom Out Reset image size

Bonifacio et al. (2009) conducted a detailed analysis of metal-poor dwarf stars, and then compared their results with abundances from the Cayrel et al. (2004) metal-poor giants. For Ca, Ni, Sr, and Ba, they found very good agreement between the abundances from dwarfs and giants. However, for C, Sc, Ti, Cr, Mn, and Co, the abundances from dwarfs were roughly 0.2 dex higher than those for giants; for Mg and Si, the abundances from dwarfs were approximately 0.2 dex lower than for giants. Carbon was the only element in which the abundance difference between dwarfs and giants could be attributed to an astrophysical cause, namely mixing and nucleosynthesis in giants (Iben 1964). For some elements, the abundance discrepancies could be the result of neglecting non-LTE and/or 3D effects in the analysis. Bonifacio et al. (2009) advocated using the abundances from giants for comparisons with chemical evolution models, because their 3D corrections were typically smaller than for dwarfs. On the other hand, Asplund (2005) found that non-LTE effects tend to be larger for giants. Also, Collet et al. (2011) argued that the CO⁵BOLD 3D models used by Bonifacio et al. (2009) suffer from systematic errors in the high atmospheric layers, and thus underestimate the 3D effects.

For the complete sample, program stars + literature stars, we now use Figures 17–30 to compare abundances between giants and dwarfs. (For C and N, we defer to Spite et al. 2005 and Bonifacio et al. 2009, who have more accurate measurements for homogeneous samples of dwarfs and giants.) We consider the slope in [X/Fe] versus [Fe/H] and the mean abundance, for each element, having excluded CEMP stars and 2σ outliers. For Na, Al, Sc, Ti ii, Cr, Mn, Ni, and Sr, we note that the slope differs between dwarfs and giants at the 2σ level or higher. The remaining elements, Mg, Si, Ca, Ti i, Co, and Ba, exhibit slopes in [X/Fe] versus [Fe/H] that agree between dwarfs and giants. For Na, Si, Ti i, Cr, Co, and Ba, the mean abundance between dwarfs and giants differs by more than 0.20 dex. For Na and Si, the mean abundance for giants exceeds the mean abundance for dwarfs, while for Ti i, Cr, Co, and Ba, the mean abundance for dwarfs is higher than for giants. The sign of the differences for Si, Ti, and Co is the same as that found by Bonifacio et al. (2009). Consideration of the standard error of the mean would indicate that, for all elements except Al, Ca, Ni, Sr, and Ba, the differences in the average abundances between the dwarf and giant samples are significant at the 3σ level or higher. There is no obvious astrophysical cause for these abundance differences between dwarfs and giants; thus, we would attribute these abundance differences to non-LTE and/or 3D effects. We remind the reader that we have employed the same set of lines for giants and dwarfs, although the giants and dwarfs may use different subsets of lines for a given element.

Another way to view the abundance differences between dwarfs and giants is to plot the abundance ratios [X/Fe] versus T_eff (see Figures 32–34). Many elements exhibit clear trends, in particular, Na, Si, Ti i, Cr, Co, Sr, and Ba are significant at the 3σ level or higher. For Na, Si, and Sr, the trend is negative (decreasing [X/Fe] with increasing T_eff), while for the other elements, the trend is positive (increasing [X/Fe] with increasing T_eff). Lai et al. (2008) found similar results for Si, Ti, and Cr, and speculated that they could be due to blends, deficiencies in the model atmospheres and/or inadequacies in spectral-line analysis techniques (e.g., non-LTE, 3D). They also caution that due care must be taken when comparing abundances of these elements with chemical evolution models or supernova yields. On looking at the data in Barklem et al. (2005), considering only stars with [Fe/H] ⩽ −2.5, we find that many elements (e.g., Ca, Co, Cr, Sc, and Ti) exhibit a very significant trend (3σ or higher) between [X/Fe] versus T_eff. Therefore, we echo the Lai et al. (2008) concerns, and note the importance of restricting the range in T_eff when possible (e.g., comparing dwarfs with dwarfs or giants with giants) in order to minimize systematic uncertainties, as done successfully by Arnone et al. (2005).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Finally, we note that radiative levitation and gravitational settling (sometimes called atomic diffusion) is suspected to play a role in altering the photospheric abundances between dwarf and giant stars at low metallicity (e.g., Richard et al. 2002a, 2002b). At present, the observational tests have been limited to the moderately metal-poor globular clusters NGC 6397 (Korn et al. 2007; Lind et al. 2008, 2009b) and NGC 6752 (Korn 2010). These analyses support the view that radiative levitation and gravitational settling can play an important role in producing abundance differences between dwarfs and giants. On looking at Figure 11 of Richard et al. (2002a), the model with [Fe/H] = −3.31 predicts that [Na/Fe] will be ∼0.3 dex higher at T_eff = 4500 K compared with T_eff = 6200 K; this is in fair agreement with our observations. However, for Si and Cr, the same models predict abundance differences between warmer dwarfs and cooler giants that are in the opposite sense to our findings. Ultimately, understanding the abundance differences between dwarfs and giants will require a combination of improved modeling in terms of non-LTE, 3D, and/or radiative levitation and gravitational settling.

For Na, Lind et al. (2011) computed non-LTE abundance corrections for a number of lines, including the resonance lines used in our analysis. Their non-LTE corrections covered a large range in stellar parameters (T_eff, log g, [Fe/H]), as well as a large range in [Na/Fe]. In Figure 35, we apply the Lind et al. (2011) non-LTE corrections to the C-normal sample presented in Figure 17, for which Na abundances were all determined from the resonance lines. To reiterate, we exclude CEMP objects in this plot. To obtain the non-LTE corrections, we used linear interpolation for stars having stellar parameters (T_eff, log g, [Fe/H]) within the grid. For stars beyond the grid (those with log g < 1.0 or [Fe/H] ⩽ −5.0), we applied the non-LTE correction at the nearest boundary of the grid; thus, the corrections for those stars are uncertain. (Had we excluded the stars that lie beyond the grid, our results would be unchanged.)

Zoom In Zoom Out Reset image size

In Figure 35, the left columns show LTE abundances. The top panels show dwarf stars and the middle panels show giants. In these panels, we again determine the linear fit to the data (excluding 2σ outliers), and show the slopes and uncertainties, dispersions about the linear fit, and the mean abundance and dispersion. For the upper left and middle left panels, the numbers are the same as in Figure 17, as they should be. Furthermore, we note that Na is an element in which dwarfs and giants exhibit significant differences in their mean abundance, [Na/Fe], and for the slope, [Na/Fe] versus [Fe/H].

In the right panels of Figure 35, we apply the Lind et al. (2011) non-LTE corrections,¹⁴ and determine the linear fit. (For clarity, we stress that only the Na abundances have been corrected, not the Fe abundances.) These corrections generally result in lower Na abundances, and therefore lower [Na/Fe] ratios, than in the LTE case. For the giant sample, the slope changes from 0.30 dex/dex to −0.05 dex/dex, while for the dwarf sample, it changes from −0.14 dex/dex to −0.22 dex/dex. Therefore, application of the non-LTE Na abundance corrections results in an improved agreement for the slope of [Na/Fe] versus [Fe/H] between dwarfs and giants. The slopes only differ at the 1.4σ level, although the mean abundance difference is 0.21 dex. Similar results are obtained when using the Andrievsky et al. (2007) Na non-LTE corrections, although we note that their corrections assume an equivalent width corresponding to an LTE abundance of [Na/Fe] = 0. The Lind et al. (2011) non-LTE corrections cover a large range in [Na/Fe] at a given T_eff/log g/[Fe/H], and we expect (and find) the magnitude of the non-LTE correction to be a function of LTE Na abundance. For this sample, the non-LTE corrections are larger for giants than for dwarfs.

In the bottom panels of Figure 35, we plot the [Na/Fe] abundances (LTE and non-LTE) versus T_eff. Application of the Lind et al. (2011) Na non-LTE corrections results in a significantly shallower slope between [Na/Fe] versus T_eff. This plot serves to highlight the importance of taking into account non-LTE effects, when possible, and as a useful exercise in assessing the importance of non-LTE corrections to the (1) mean abundance, (2) slope of [X/Fe] versus [Fe/H], and (3) trends between [X/Fe] versus T_eff. We await with great interest more detailed grids of non-LTE abundance corrections for additional elements as well as grids of 3D abundance corrections, although we recognize the magnitude of such efforts currently underway (e.g., Andrievsky et al. 2008, 2011; Collet et al. 2011; Bergemann & Gehren 2008; Bergemann et al. 2010; Bergemann 2011; Bergemann et al. 2012; Dobrovolskas et al. 2012; Lind et al. 2009a, 2011, 2012; Spite et al. 2012).

As discussed by Cayrel et al. (2004), one possibility to aid the interpretation of the abundances in metal-poor stars is to use Mg as the reference element rather than Fe. An advantage of such an approach is that the nucleosynthetic production of Mg during hydrostatic burning is well understood, whereas the synthesis of Fe is more complicated and not unique. On the other hand, a disadvantage of using Mg over Fe is that there are fewer lines from which the abundance is measured, thus the measurements are less accurate. Nevertheless, Cayrel et al. (2004) took this alternate approach and noted that, in the regime [Mg/H] ⩽ −3.0, there was a suggestion that all abundance ratios, [X/Mg], were flat. The plateau value of [X/Mg] at lowest [Mg/H] may therefore reflect yields from the first generation of supernovae.

In Figures 36–38, we plot [X/Mg] versus [Mg/H] for dwarfs and giants; for comparison with the work of Cayrel et al. (2004), in the range [Mg/Fe] ⩽ −3.0, we fit the data excluding CEMP stars and 2σ outliers. In these figures we again include the predictions from Kobayashi et al. (2006). We reach similar conclusions to those of Cayrel et al. (2004). Namely, the dispersion about the linear fit is generally greater when plotting [X/Mg] versus [Mg/H] than for [X/Fe] versus [Fe/H] (although there are a few cases in which the opposite is true, e.g., Al, Si, and Ti ii). This is presumably due to the Mg measurements being more uncertain than Fe due to the smaller number of lines. (Note that we are only fitting stars with [Mg/H] ⩽ −3.0 rather than the full abundance range.) For most elements, the linear fit to the data with [Mg/H] ⩽ −3.0 shows zero slope (at the 2σ level), with notable exceptions including Al (dwarfs), Ti i (giants), and Co (dwarfs and giants). Considering stars in the range [Mg/H] ⩽ −2.0, rather than [Mg/H] ⩽ −3.0, we note that the linear fit to the data is not consistent with zero slope for a larger number of elements. In general, the behavior of [X/Mg] versus [Mg/H] exhibits a similar behavior at all metallicities. It is difficult to assess whether the Kobayashi et al. (2006) predictions are a better match to the [X/Mg] or the [X/Fe] plots. Ultimately, using Mg as the reference element does not seem to offer any major advantages over Fe, at least in this analysis.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Furthermore, we note that Chieffi et al. (2000) have suggested that Si, S, Ar, and Ca (or combinations of these elements) may be better reference elements and tracers of Galactic chemical evolution than Fe (or O). Chieffi et al. (2000) argue that the yields of these elements are not strongly dependent on the location of the mass cuts employed in the supernova explosion calculation or with the pre-explosive chemical composition. Of these elements, Ca has the most number of lines measured in our program stars, although the average number of lines is smaller than for Mg, and the average total error for Ca exceeds that of Mg.

In the upper panel of Figure 31, we compare the slope from the linear fit of [X/Fe] versus [Fe/H] between our sample (program + literature stars) and the Cayrel et al. (2004) study. That is, we are comparing the coefficient a from the relation, [X/Fe] = a× [Fe/H] + b, describing the best fit to our data and the best fit to their data. Since their sample consists exclusively of giants, we use the slope as determined from our giant sample in the comparison. To ensure that this was a proper comparison of gradients, we used our software to determine the slope (and uncertainty) to the Cayrel et al. (2004) data, rather than relying upon the values in their Table 9. As with our sample, we exclude the two CEMP stars and 2σ outliers when determining the linear fit to the Cayrel et al. (2004) data. Figure 31 shows that, for Mg, Ca, Sc, Cr, and Ni, the gradients measured in this study and Cayrel et al. (2004) are in very good agreement. For Si, Mn, and Co, the slopes differ by more than 2σ. For Si, the sign of the gradient differs between the two studies.

In the lower panel of this figure, we also compare the scatter about the linear fit between our giant sample and that of Cayrel et al. (2004). For their data, we again used our software to determine the scatter about the linear fit after eliminating 2σ outliers and the two CEMP stars. (We used the relation given in Taylor (1997) for the fractional uncertainty in the dispersion, 1/$\sqrt{2(N-1)}$.) Some elements (Mg, Ca, Mn, Co, and Ni) exhibit very good agreement. The elements that differ by more than 2σ are Si, Sc, and Cr. It is not surprising that, for all elements, the dispersions about the linear fit to the Cayrel et al. (2004) data are always equal to or smaller than for our sample of program and literature stars. This could be due, in part, to the fact that the Cayrel et al. (2004) sample is very homogeneous, and their spectra are of very high quality. While our sample includes their stars, we have a more heterogeneous sample, albeit one that was analyzed in a homogeneous manner.

In the lower panel of Figure 31, we also plot a representative measurement uncertainty for each element, the average “total error” for the program stars. For the Cayrel et al. (2004) data, we also plot their estimate of the expected scatter for each element, based on their measurement uncertainties. These are shown as upward facing arrows, and we would expect the observed dispersions to lie on or above these values (for species having an astrophysically significant dispersion). For the Cayrel et al. (2004) data, the observed dispersions are in good agreement with the expected scatter. For our data, some elements (Mg, Mn, Co) exhibit an observed dispersion in good agreement with the expected scatter. However, for Si, the observed dispersion exceeds the expected value; for Sc, the observed dispersion is considerably smaller than the expected value, which may indicate that our measurement uncertainties are overestimated for this element. Nevertheless, it is reassuring that in several cases (Mg, Ca, Co, and Ni) our dispersions about the linear fit are comparable with the values of Cayrel et al. (2004), suggesting that if indeed a “normal” population exists, our selection criteria were effective in identifying this population, even though our sample of program and literature stars is quite heterogeneous.

In Figure 14 and Figures 36–38, we overplot the Kobayashi et al. (2006) predictions of Galactic chemical enrichment. Their chemical evolution model includes the following assumptions: (1) one zone centered on the solar neighborhood, (2) no instantaneous recycling approximation, (3) contributions from hypernovae with large explosion energy (E₅₁ > 10), Type II supernovae, and Type Ia supernovae, (4) no contributions from low- and intermediate-mass stars, and (5) infall of primordial gas. Figure 14 shows the evolution of [X/Fe] against [Fe/H]. In general, the predictions provide a fair fit to the data in terms of the slope (or lack thereof). For many elements, the mean predictions differ from the mean observations by ∼0.5 dex or more.

As reported in other investigations (e.g., Kobayashi et al. 2006), the predictions for Ti and Sc are underabundant relative to the LTE measurements. In our study, Sc measurements are exclusively from Sc ii lines. We find that the abundance ratios [Ti ii/Fe] and [Sc/Fe] (and [Ti ii/H] versus [Sc/H]) exhibit a highly significant, ∼10σ, correlation (see Figure 39). In this figure, we show all stars (upper panel), dwarfs (middle panel), and giants (lower panel). For each sample, the scatter about the mean trend is only ∼0.10 dex, which is substantially lower than the average total error for either [Ti ii/Fe] (0.14 dex) or [Sc/Fe] (0.20 dex). We also find a correlation between [Ti i/Fe] and [Sc/Fe], but with a lower significance (∼2σ) and shallower slope (0.19 dex/dex to 0.38 dex/dex). Although Sc and Ti are produced via different processes, the correlation we find suggests that the two elements might be produced in similar conditions. Umeda & Nomoto (2005) suggest that the yield of Sc in metal-poor supernovae can be greatly increased in low-density (i.e., high-entropy) regions. Kobayashi et al. (2011a) suggest that the ν-process in core-collapse supernovae may produce Sc, although to our knowledge there have been few, if any, studies of the yields of Sc from the ν-process. The strong correlation between Sc and Ti found here might suggest that the ν-process does not provide a complete explanation of the production of Sc at lowest metallicities.

Zoom In Zoom Out Reset image size

Figures 36–38 show the evolution of [X/Mg] against [Mg/H]. Again, the predictions provide a fair fit to the slope and the mean abundance. In this case, the elements which are poorly fit include Na, Sc, and Ti.

Presumably, the Kobayashi et al. (2006) model, or other chemical evolution models, could be fine tuned to provide a better fit to this set of observations. We note in particular that the inclusion of yields from intermediate-mass stars cannot account for the underproduction of Na at low metallicity (Kobayashi et al. 2011b). As discussed earlier, we caution that non-LTE and 3D effects should be taken into consideration. Chemo-dynamical models of Galactic formation and evolution (e.g., Kobayashi & Nakasato 2011), and/or inhomogeneous chemical enrichment models (e.g., Karlsson & Gustafsson 2005) will enable more comprehensive comparisons with the available observations, including the predicted dispersion in abundance ratios as a function of stellar population characteristics such as metallicity, kinematics, age, and spatial distribution.

In Figures 40–44, we plot for each program star the abundance pattern [X/Fe] versus element. (In these figures, we include, as double entries, results of both the dwarf and subgiant analyses of the nine stars for which there was disagreement between spectrophotometric and Balmer-line analyses of the evolutionary status.) The solid line in each panel represents the abundance ratio [X/Fe] that a “normal” dwarf (log g > 3.0) or giant (log g < 3.0) star would have at the metallicity of the program star. The “normal” star abundance was taken from the linear trends described above, and plotted in Figures 17–30. For elements that deviate from the solid line by more than 0.50 dex, we regard these abundances to be peculiar, and mark them in red. (Note that some peculiar abundances lie above the solid line, while others lie below.) For CEMP objects, we mark their C abundance in red. For N and O, we regard ratios [X/Fe] ⩾ +1.0 to be unusual, and also mark them in red. Inspection of Figures 40–44 then readily highlights whether a given star has elements that may be regarded as peculiar. Any star with several such elements can be considered as a chemically peculiar star.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Similarly, we examined the 152 star literature sample in order to identify the chemically peculiar objects. In Figures 45–47, we plot the abundance pattern [X/Fe] versus element for the subset of program stars with [Fe/H] ⩽ −2.55 in which there are at least two elements that are unusual (i.e., elements that are at least 0.5 dex above, or below, the [X/Fe] ratio of a “normal” dwarf or giant at the same [Fe/H]). For the two most Fe-poor stars in these figures, the solid lines showing the [X/Fe] ratios of a normal star are for [Fe/H] = −4.2.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In Section 7.2, we noted that in the [X/Fe] versus [Fe/H] plane, abundance outliers were often, but not always, CEMP objects. We now comment briefly on the C-normal population (i.e., only those stars with [C/Fe] measurements, or limits, that enable us to confirm that they are not CEMP stars). We find that of the 109 C-normal stars, some 32 are chemically peculiar (i.e., these 32 stars have at least one element, from Na to Ba, for which the [X/Fe] ratio is at least 0.5 dex above, or below, that of a normal star at the same metallicity). These 32 objects have a mean metallicity of [Fe/H] = −3.02. If we exclude the neutron-capture elements Sr and Ba (these elements exhibit a very large dispersion such that many stars will have [X/Fe] ratios at least 0.5 dex from the “normal” star abundance), there are 23 C-normal stars that are chemically peculiar, and these stars have a mean metallicity of [Fe/H] = −2.91.

We note that for the majority of these objects, only one element in a given star may be regarded as unusual. If we consider only those C-normal stars that are chemically peculiar for two or more elements, there are seven such objects with a mean metallicity of [Fe/H] = −3.31. When excluding Sr and Ba, there are only four C-normal stars, with an average metallicity of [Fe/H] = −3.14, that are chemically peculiar for two or more elements—HE 1207−3108 (This Study), HE 0024−2523 (Carretta et al. 2002; Cohen et al. 2002), HE 1424−0241 (Cohen et al. 2008), and CS 22873−055 (Cayrel et al. 2004). The main point of this analysis is to note that indeed there are C-normal objects that are chemically peculiar, although the fraction is small, four of 109 objects (4% ± 2%). We now discuss some interesting examples of chemically unusual stars.

7.8.1. Stars with Enhanced C, N, O, Na, Mg, and/or Al

7.8.2. A Star with Enhanced Si

HE 2136−6030 is a C-normal star with an unusually high Si abundance, [Si/Fe] = +1.20. However, for all other elements measured, the abundance ratios are in good agreement with a “normal” star at the same metallicity. This object has T_eff = 6409 K and [Fe/H] = −2.88. Figure 20 shows that for dwarf stars, the [Si/Fe] ratio is almost constant as [Fe/H] evolves from −4.0 to −2.5. While Figure 32 shows a strong trend between [Si/Fe] and T_eff, stars with T_eff similar to that of HE 2136−6030 all have [Si/Fe] ≲ +0.5 dex. In Figure 48, we plot the spectra of HE 2136−6030, along with two Si-normal stars with similar stellar parameters. The Si line is substantially stronger in HE 2136−6030, relative to the two comparison stars. Examination of our spectra shows that CH blending of the 3905.52 Å Si line is unlikely, and that the subordinate 4102.94 Å Si line is likely present. Spectrum synthesis of the 4102.94 Å Si line returns a [Si/Fe] ratio in good agreement with the abundance from the 3905.52 Å line. We are thus confident that the Si abundance is particularly high in this star.

Zoom In Zoom Out Reset image size

At the other extreme, Cohen et al. (2007) found a star with an unusually low Si abundance, [Si/Fe] = −1.01. This star, HE 1424−0241, also has low abundances of [Ca/Fe] and [Ti/Fe] but a normal [Mg/Fe] ratio. These two stars, HE 2136−6030 and HE 1424−0241, reveal that the [Si/Fe] ratio in metal-poor stars can vary by a factor of 100.

7.8.3. “Fe-enhanced” Metal-poor Stars

HE 1207−3108 (T_eff/log g/[Fe/H] = 5294/2.85/−2.70) is notable for having unusually low abundance ratios [X/Fe] for Mg, Al, Ca, Sc, Ti, and Sr. (With the exception of Ti, and possibly Sr, none of these elements exhibit significant trends with T_eff, thus it is highly unlikely that the peculiar abundance pattern can be attributed to any T_eff-dependent trends.) The effect is shown in Figure 49, where the filled circles represent [X/Fe] for HE 1207−3108 as a function of atomic species. Indeed, one sees here that for all elements other than Na, the abundance ratios [X/Fe] lie below that of a “normal” star at the same metallicity. (We note here that the two Na lines yield abundances that differ by 0.48 dex. Without knowing which of the lines to reject, we retain both lines and the large error reflects the discordant measurements.) Cayrel et al. (2004) identified another star, CS 22169−035, as being deficient in Mg, Si, Ca, Ti, Co, Ni, and Zn with respect to Fe. They suggested that (p. 1135) “the abundance anomalies are most simply characterized as an enhancement of Fe;” the same description may be applied to HE 1207−3108. Inspection of Figure 42 suggests that if the Fe abundance were lowered, by say 0.6 dex, HE 1207−3108 would have normal abundance ratios [X/Fe] for Mg and all heavier elements, but [Na/Fe] might be regarded as being unusually high.

Zoom In Zoom Out Reset image size

In Figure 49, we also present the data for CS 22169−035. Here, we we arbitrarily select Sc as the element to which we normalize the abundances; this requires a shift of −0.47 dex for CS 22169−035. The solid line shows the abundance ratio that a “normal” giant would have at the metallicity of HE 1207−3108. (An equivalent line for CS 22169−035 would be essentially identical, given that both stars are giants with comparable metallicity, [Fe/H] = −2.70 and −2.95 for HE 1207−3108 and CS 22169−035, respectively.) HE 1207−3108 and CS 22169−035 appear to have similar and unusual abundance patterns, i.e., they are chemically distinct from the bulk of the halo stars at the same metallicity. We note that their chemical abundance patterns are not identical, namely, there are differences for Cr, Co, Ni, and Sr. That said, their abundance ratios for Mg, Al, Ca, Ti, and Mn are indistinguishable and unusually low. We suggest that HE 1207−3108 and CS 22169−035 may belong to a relatively new and growing class of “Fe-enhanced metal-poor” stars.

A small number of field halo stars, both dwarfs and giants, are known to have low [α/Fe] ratios relative to field stars at the same metallicity (e.g., Carney et al. 1997; Nissen & Schuster 1997; Fulbright 2002; Stephens & Boesgaard 2002; Ivans et al. 2003; Nissen & Schuster 2010). These stars cover the metallicity range −2 ≲ [Fe/H] ≲ −0.8. One suggestion is that these stars likely formed from regions in which the interstellar gas was unusually enriched in the products of Type Ia supernovae, relative to Type II supernovae, and as such they may be regarded as having an unusually high Fe-content relative to their α-content. One explanation is that these stars may have been accreted from dwarf galaxies, many of which are known to contain stars that are chemically distinct from the majority of field halo stars (e.g., Tolstoy et al. 2009). While it may be tempting to assign the low [α/Fe] stars and HE 1207−3108 and CS 22169−035 into a single group, there are differences in [Fe/H] and other [X/Fe] ratios. Nevertheless, this intriguing possibility exists.

Venn et al. (2012) have identified a giant star in the Carina dwarf galaxy with a similar chemical pattern to that of HE 1207−3108 and CS 22169−035. Their star, Car-612, with [Fe/H] = −1.3, is considerably more metal-rich than HE 1207−3108 and CS 22169−035, with [Fe/H] = −2.70 and −2.95, respectively. Venn et al. (2012) suggest that this star formed from gas with an unusually high ratio of Type Ia supernovae to Type II supernovae products. They estimate the magnitude of the “excess of Fe” to be 0.7 dex, a value comparable to that of HE 1207−3108 and CS 22169−035. Had we overplotted their data for this star in Figure 49, again normalizing to the Sc abundances, there would have been striking similarities in the some abundance ratios (Mg, Ti, and Ni), as well as notable differences for other elements (Ca, Cr, and Mn). As before, we do not overplot these data, since we do not wish to conduct inhomogeneous comparisons. Nevertheless, it is tempting to associate the three stars (HE 1207−3108, CS 22169−035, and Car-612) as belonging to the same class of object. We tentatively speculate that any differences in [X/Fe] ratios between Car-612, HE 1207−3108, and CS 22169−035 may reflect differences in the currently unknown formation processes for such stars. (During publication, we were informed of another “Fe-enhanced metal-poor star” discovered by Cohen & Huang 2010 in the Ursa Minor dwarf spheroidal galaxy.)

7.8.4. A CEMP-s Star with Enhanced [Ba/Sr]

HE 0207−1423 ([Fe/H] = −2.95) is a CEMP-s star with the unusually large Ba/Sr ratio of [Ba/Sr] = +1.40. (CEMP-s stars are a subclass of CEMP objects that exhibit overabundances of the s-process elements, defined by [Ba/Fe] > +1.0 and [Ba/Eu] > +0.5; Beers & Christlieb 2005.) That is to say, the nucleosynthetic site(s) that produced the large Ba abundance, [Ba/Fe] = +1.73, in this object, yielded only [Sr/Fe] = +0.33, considerably smaller than one might have expected. Inspection of Figure 50 presents the dependence of [Ba/Sr] on [Fe/H] (which we shall discuss in the following section), and shows that there are other (literature) stars that exhibit unusually large [Ba/Sr] values (or low [Sr/Ba]). These stars are all CEMP stars with [Fe/H] ⩾ −3.0.

Zoom In Zoom Out Reset image size

Although our spectra are of moderate-to-high quality, we were only able to measure the abundance of two neutron-capture elements, Sr and Ba. The absence of other neutron-capture elements was not for lack of effort: in the course of our analysis, we tried to measure the equivalent widths of numerous lines of La ii, Ce ii, Nd ii, and Eu ii. That said, our measurements of Sr and Ba enable us to comment on three issues.

First, previous studies have found a very large scatter in Sr and Ba abundances at low metallicity. In Figures 29 and 30, there is a large dispersion about the linear fit to the [Sr/Fe] versus [Fe/H] and [Ba/Fe] versus [Fe/H] trends, for both dwarfs and giants, even after excluding CEMP stars and 2σ outliers. In all four cases, the dispersion (ranging from 0.30 dex to 0.60 dex) exceeds the typical measurement uncertainties (ranging from 0.15 dex to 0.22 dex). In Figure 50, we consider [Ba/Sr] versus [Fe/H]. For dwarfs, the dispersion about the linear fit to the data (with the usual exclusions) is 0.28 dex. This is comparable to the dispersions for [Sr/Fe] (0.36 dex, Figure 29) and [Ba/Fe] (0.30 dex, Figure 30). For giants, the dispersion about the linear fit to [Ba/Sr] versus [Fe/H] is 0.56 dex. Again, the value is comparable to the dispersions for [Sr/Fe] (0.60 dex, Figure 29) and [Ba/Fe] (0.43 dex, Figure 30). Therefore, our data set suggests that the dispersion about the linear fit to [Ba/Sr] versus [Fe/H] is not smaller than the individual dispersions about the linear fit to [Sr/Fe] or [Ba/Fe] versus [Fe/H], as might have been expected if a correlation existed between the abundances of the two elements.

Second, there is evidence that an additional process (or processes) dominates the production of the lighter neutron-capture elements. François et al. (2007) plotted [(Sr, Y, Zr)/Ba] versus [Ba/H] (see their Figure 15), and noted that (1) for [Ba/H] ⩾ −2.5, all ratios are close to the solar value, (2) for −4.5 ⩽ [Ba/H] ⩽ −2.5 [(Sr, Y, Zr)/Ba] increases as [Ba/H] decreases, and (3) for [Ba/H] ⩽ −4.5, all ratios drop to solar. They concluded that such observations are inconsistent with a single r-process. In Figure 51 (upper panel), we consider the ratio [Sr/Ba] versus [Ba/H]. Our data are qualitatively consistent with those of François et al. (2007) and their results, had we overplotted them, would have provided a very good match to ours. However, such an inhomogeneous comparison would be contrary to the spirit of the homogeneous analysis presented herein. Considering only our data below [Ba/H] = −2.5, the dispersion in [Sr/Ba] appears to increase with decreasing [Ba/H]. Indeed, there is a hint of two populations below [Ba/H] = −4.0, one with high [Sr/Ba] and the other with solar, or sub-solar, [Sr/Ba], although the statistics are still relatively poor.

Zoom In Zoom Out Reset image size

The lower panel of Figure 51 shows [Ba/H] versus [Sr/H]. Within the limited data, there is a hint of two (overlapping) populations, with the boundary at [Sr/H] ≃ −3.2. Excluding 2σ outliers (and in this case retaining CEMP stars), we find that the two populations have significantly different slopes, but very similar dispersions about the linear fit. These data may indicate that at lowest Sr and Ba abundances, the nucleosynthetic process(es) that create Sr and Ba produces a different ratio of Sr/Ba than the nucleosynthetic process(es) that operate when the Sr and Ba abundances are higher. The scaled-solar r-process distribution would be represented by a line of gradient 1.00 in the [Sr/H] versus [Ba/H] plane. For the two populations that we identify, neither slope matches the main r-process line, which may suggest that there are two (or more) nucleosynthetic sites producing Sr and Ba at low metallicity that are distinct from the main r-process. Ignoring the sole point with [Sr/H] ⩾ −1.5, one might still argue that the data with highest [Sr/H] (admittedly mainly CEMP objects) do not appear to lie on a linear extrapolation of the line fitting the data with lower [Sr/H]. Again, we note in passing that the François et al. (2007) data confirm the trends seen in our data. Had we included their measurements in the linear fits to the data described above, we would have obtained essentially identical slopes, uncertainties, and dispersions about the linear fits. However, once again we do not overplot these data to avoid inhomogeneous comparisons. Travaglio et al. (2004) noted the need for a primary process to produce Sr, Y, and Zr at low metallicity that was different from the s- and the r-process. They referred to this as a lighter element primary process (LEPP). At face value, our results are consistent with the suggestion there may be two components to the Travaglio et al. (2004) LEPP. Roederer et al. (2010) suggest that high-entropy neutrino winds from core-collapse supernovae can explain the diversity of neutron-capture element abundances found at low metallicity. It would be interesting to examine whether our Sr and Ba measurements can be explained by this scenario.

Third, the relation between the abundances of C and the neutron capture elements Ba and Sr may shed light on the process(es) that created the CEMP-s and CEMP-no classes of stars. As already noted, Beers & Christlieb (2005) defined CEMP-s stars as those with [C/Fe] ⩾ +1.0, [Ba/Fe] > +1.0, and [Ba/Eu] > +0.5, and CEMP-no stars as those with [C/Fe] ⩾ +1.0 and [Ba/Fe] < 0. The CEMP-s stars are the majority population, ∼80% of all CEMP objects (Aoki 2010), and radial-velocity studies have revealed that the observed binary frequency in this subclass is consistent with a binary fraction of 100% (Lucatello et al. 2005). Mass transfer between an asymptotic giant branch (AGB) primary onto the currently CEMP-s star is believed to be responsible for the C and Ba enhancements (see, e.g., Izzard et al. 2009 for a detailed examination of the viability of this scenario), a process that also produces the more metal-rich classes of CH and Ba stars. Additionally, the metallicity distributions differ, with CEMP-s being generally more metal rich, [Fe/H] > −3.0, and CEMP-no being generally more metal poor, [Fe/H] < −3.0 (e.g., Cohen et al. 2006; Aoki et al. 2007; Aoki 2010). Cohen et al. (2006) and Masseron et al. (2010) have suggested that the CEMP-no stars are likely the extremely metal-poor counterparts of the CEMP-s stars.

Figure 52 shows [Sr/Fe], [Ba/Fe], and [Ba/Sr] versus [C/H]. In this figure, the CEMP stars with the highest [Sr/Fe] and [Ba/Fe] ratios tend to have the highest [C/H] values. While more data are urgently needed, we speculate that the stars which produce CEMP objects with [C/H] ⩾ −1.0 synthesize large amounts of Sr and Ba. The diverse set of abundance ratios seen in the observations calls for production sites capable of diverse yields. Yields of C and s-process elements from AGB stars are strongly dependent on mass and metallicity, such that they may be good candidates for explaining part, or perhaps most, of the large range of C and s-process element abundances (e.g., Karakas & Lattanzio 2007; Karakas 2010; Cristallo et al. 2011; Lugaro et al. 2012). Regardless, the limited data indicate that stars with [C/H] > −1.0 have Sr and Ba enhancements and [Ba/Sr] > 0. See Papers III and IV in this series for additional discussions of CEMP stars.

Zoom In Zoom Out Reset image size

In Table 9, we provide the coordinates, stellar parameters, and abundance ratios for all of the program stars and literature stars presented in this work. Below is a description of the columns in the table.

(1) Star; (2) RA2000; (3) DEC2000; (4) effective temperature (T_eff); (5) surface gravity (log g); (6) stellar metallicity [Fe/H]_derived; (7) CEMP (0 = no, 1 = yes); (8) [C/Fe]; (9) [N/Fe]; (10) [Na/Fe]; (11) [Mg/Fe]; (12) [Al/Fe]; (13) [Si/Fe]; (14) [Ca/Fe]; (15) [Sc/Fe]; (16) [Ti i/Fe]; (17) [Ti ii/Fe]; (18) [Cr/Fe]; (19) [Mn/Fe]; (20) [Fe ii/H]; (21) [Co/Fe]; (22) [Ni/Fe]; (23) [Sr/Fe]; (24) [Ba/Fe]; (25) Source.

The C and N abundances for literature stars were taken directly from the literature sources. For the abundances of Sr and Ba, in the cases in which the literature sources did not provide an equivalent width (i.e., they determined abundances using spectrum synthesis), we include (when available) the abundance measurements or limits. These literature values are flagged appropriately in the table.