NUCLEOSYNTHESIS AND THE INHOMOGENEOUS CHEMICAL EVOLUTION OF THE CARINA DWARF GALAXY^*

The new MARCS spherical models have been adopted in this analysis (Gustafsson et al. 2003, 2008; further expanded by B. Plez 2008, private communications). The grid covers the range of parameters (temperature, luminosity, microturbulence, and metallicity) of RGB stars. The models used in this analysis adopt the Galactic abundance pattern, i.e., [α/Fe] = +0.4 for [Fe/H] ⩽−1.0, which is the metallicity range of the stars in Carina. We adopt the standard version of the MOOG¹⁸ (Sneden 1973) spectral synthesis code to determine the individual line abundances. Corrections were applied when necessary for continuum scattering effects that can affect the results for lines below 5000 Å, as described in Section 3.8.2. The photometric gravities were retained throughout the analysis (see Section 3.8.3).

The effective temperatures were refined from the photometric values through examination of the Fe i lines. Only lines redward 5000 Å were considered to avoid uncertainties in the standard MOOG treatment of scattering in the continuous opacity (see Section 3.8.2). Microturbulence values were found by minimizing the slope in the Fe i line abundances with EW. Similarly, the excitation temperatures were found by minimizing the slope in the same Fe i line abundances with excitation potential (χ, in eV). While the M68 stars had lower excitation than color temperatures (by an average of 67 K), the Carina stars had higher excitation temperatures (by an average of 200 K). The final metallicity (for which we adopt the [Fe i/H] values) is also found at the end of these minimization iterations.

No significant relationship was found in the Fe i abundances versus wavelength (>5000 Å) for either the globular cluster or the Carina targets; this indicates that the sky subtraction and overall data reduction methods were successful. Eliminating lines with χ < 1.4 eV, which can be particularly sensitive to departures from local thermodynamic equilibrium (LTE) effects, also had no significant effect on the temperature or microturbulence values. Samples of these relationships are shown in Figure 4 for two stars, one observed with the FLAMES/UVES spectrograph and the other observed with the MIKE spectrograph. Adjusting the metallicities from the initial CaT values had no significant effect on the excitation temperatures (ΔT < 10 K). Adopting the spectroscopic temperatures had only a small effect on the photometric gravities (⩽ + 0.08 in the Carina stars and ⩽ − 0.02 for the M68 stars).

Zoom In Zoom Out Reset image size

Four stars in Carina have been observed and analyzed independently by Koch et al. (2008a). Their analysis with ATLAS9 models atmospheres results in excitation temperatures that are in excellent agreement with ours, with differences ranging from Δ T = +116 to −70 K, and with an average offset of +32 K. The differences in microturbulence are more significant, with Koch et al.'s values being ∼0.50 km s⁻¹ lower. This is linked to differences in our gravity determinations discussed in Section 3.8.3.

Spherical MARCS models represent a significant improvement in the modeling of stellar atmospheres, e.g., Heiter & Eriksson (2006) recommend the use of spherical model atmospheres in abundance analyses for stars with log g <2 and 4000 K ⩽ T_eff ⩽ 6500 K, which encompasses the range in stellar parameters of our target stars. Tests performed by Tafelmeyer et al. (2010) showed systematic offsets between the spherical and plane-parallel models are below 0.15 dex (this includes iron lines and other elements, e.g., [C/Fe]).

We performed similar tests, comparing the metallicities we determined from spherical MARCS models to those from plane-parallel MARCS (pp-MARCS) and Kurucz models. The pp-MARCS models had the effect of reducing the excitation temperature results by ∼100 K, but the Kurucz models had a different effect, increasing the microturbulence values by ∼0.1 km s⁻¹. In both cases, all other parameters were unchanged. These offsets, when applied to the abundance analysis, changed the Fe i abundances by only ∼0.05 dex (+0.05 with pp-MARCS models and −0.05 with Kurucz models). This is very similar to the offsets found by Tafelmeyer et al. (2010) and Heiter & Eriksson (2006) for iron.

The standard version of MOOG treats continuum scattering (σ_ν) as if it were absorption (κ_ν) in the source function, i.e., S_ν = B_ν (the Planck function), an approximation that is only valid at long wavelengths in RGB stars. At shorter wavelengths (⩽5000 Å), Cayrel et al. (2004) have shown that the scattering term must be taken into account such that the source function becomes S_ν = (κ_ν B_ν + σ_ν J_ν)/(κ_ν + σ_ν). A new version of MOOG (MOOG-SCAT; Sobeck et al. 2010) has been used to test our results and calculate corrections to our abundances line by line. As shown in Figure 5, the scattering corrections are negligible for red lines, but can approach 0.4 dex in the blue. Lines with the largest corrections are the resonance lines of Mn i. Note also that the corrections are sensitive to the atmospheric parameters, in particular metallicity; the three stars in Carina that are shown were choosen because they have the largest corrections and they are the most metal poor.

Zoom In Zoom Out Reset image size

We compared the elemental abundances between the programs MOOG and CALRAI (Spite 1967; Cayrel et al. 1991; V. Hill et al. 2012, in preparation), and also the scattering corrections calculated with MOOG-SCAT to those determined with the program Turbospectrum (Alvarez & Plez 1998); the standard abundance results and scattering corrections were nearly identical for all lines between these codes; in particular, the scattering corrections were in agreement to within ∼0.01 dex. Only the resonance lines of Mn i showed slightly larger differences in the scattering corrections (∼0.04 dex); since resonance lines form over more atmospheric layers, we expect that these lines are more sensitive to the details in the models and the line formation calculations between the codes. Thus, we consider the consistency in the abundance results from the line formation codes MOOG and CALRAI, and the scattering corrections between MOOG-SCAT and Turbospectrum to be in excellent agreement.

As a final note, the scattering corrections had no effect on our spectroscopic temperature or microturbulence determinations, nor the metallicities for [Fe i/H], since only Fe i lines with λ > 5000 Å were used in those steps.

The analysis of ten stars in Carina by Koch et al. (2008a) and five more stars by Shetrone et al. (2003) used the Fe ii lines to determine spectroscopic gravities from ionization equilibrium (where [Fe i/H] = [Fe ii/H]). For four stars in common with Koch et al., our gravities are smaller than theirs by log g ∼0.5 dex. Since turbulent velocity and ionization equilibrium are correlated in RGBs, then the higher microturbulent values found by Koch et al. (see Section 3.5.2) are correlated with their higher spectroscopic gravities.

There are sufficient Fe ii lines in many of the stars in our analysis to examine the ionization equilibrium of iron and determine spectroscopic gravities, however we do not use this method in this analysis. For consistency with the FLAMES/GIRAFFE analysis, where a shorter wavelength interval meant fewer Fe ii lines, then we adopt photometric gravities using the same methods as in Lemasle et al. (2012). On the other hand, we note that the Fe i and Fe ii abundances are in good agreement for our Carina targets, with a mean [Fe ii/Fe i] = +0.02, and range of −0.24 to +0.20, which is ∼1σ(Fe i). The stars in M68 have higher offsets, with a mean [Fe ii/Fe i] = +0.35, which may reflect uncertainties in its distance or reddening. When the abundance of Fe ii is found to be larger than Fe i in red giants, it can be due to the overionization of Fe i by the radiation field, which is neglected under the assumption of LTE. Corrections for this effect can be as large as 0.3 dex (Mashonkina et al. 2010), which is similar to the offset in the M68 stars. Thus, we consider our Fe i and Fe ii results to be in good agreement for all of our targets when physical gravities are adopted.

Note, Fe ii lines at all wavelengths were examined in this analysis, i.e., they were not restricted to λ < 5000 Å like Fe i.

Carbon forms during the helium burning phases, whether the hydrostatic helium core burning phases in massive stars, or helium shell burning phases in asymptotic giant branch (AGB) stars (Woosley & Weaver 1995). The chemical evolution of carbon is further complicated by CNO cycling, where carbon is reduced when exposed to hot protons and the CN cycle runs to equilibrium values. This evolution in carbon can be seen in RGB stars in globular clusters, where the [C/H] abundance and ¹²C/¹³C ratio are both reduced as stars ascend the RGB due to internal (self) mixing (Gratton et al. 2004).

Carbon abundances in this analysis were determined from spectrum syntheses of portions of the A²Π - X²Δ CH G band near λ4320 in the Magellan spectra. The best fit was taken as the carbon abundance from the CH line list by Brown et al. (1987) and Carbon et al. (1982), with acceptable fits as the abundance uncertainty. A C¹²/C¹³ ratio of six was adopted (the typical value for bright RGB stars that have undergone the first dredge-up); this ratio made only a small difference such that increasing the value to 50 caused Δlog(C/H) = 0.06. Similarly, reducing the oxygen abundance had only a small effect (since C can be locked into the CO molecule); we adopted [O/Fe] = +0.4, but reducing this by 3× lowered the carbon abundance by only ⩽0.05 dex. The spectrum of M68-6023 around the CH 4320 Å feature is shown in Figure 7, with spectrum syntheses for three carbon abundances.

Zoom In Zoom Out Reset image size

The carbon abundances for the M68 stars are in good agreement with typical red giants in globular clusters above the RGB bump (e.g., Smith et al. 2005; Gratton et al. 2000), where [C/Fe] ∼−0.4 at the RGB bump and slowly decreases to ∼ − 1.0 dex at the RGB tip due to mixing of CNO-cycled gas. This trend can be seen in Figure 8 (lower panel, data from Gratton et al. 2000). The carbon abundances compiled by Frebel (2010) suggests a much larger scatter in carbon in metal-poor stars than seen in clusters (Figure 8, upper panel), however Frebel's compilation includes field stars of all evolutionary stages, including carbon-enhanced metal-poor (CEMP) stars. CEMPs have [C/Fe] >1.0 and are thought to be both stars that have undergone mass transfer in a binary system with an AGB companion (CEMP-s, e.g., Beers & Christlieb 2005) and stars with high primordial carbon (e.g., from high-mass rotating stars; Meynet et al. 2006). Aoki et al. (2007) suggested that mixing on the RGB and extra mixing at the tip of the RGB will lower the surface carbon abundance, such that the definition of a CEMP star should depend on luminosity. This luminosity dependent range is shown in Figure 8 (lower panel), and clearly shows that stars in M68 and the Carina dSph are not CEMP stars by any definition.

Zoom In Zoom Out Reset image size

The mean carbon abundance in M68 is 〈[C/Fe]〉 = −0.7  ±  0.1, in agreement with typical upper RGB stars. Upper limits on the carbon abundances in the metal-poor Carina stars are even lower ([C/Fe] ⩽−1.0); this value is typical of upper RGB stars in globular clusters, but on the low side of the values found for the bright, metal-poor field stars in the Galaxy, as seen in Figure 8.

The α-elements (O, Mg, Si, Ca, and Ti) are built from multiple captures of He nuclei (α particle) during various stages in the evolution of massive stars (>8 M_☉, e.g., carbon burning, neon burning, and complete and incomplete Si burning) and dispersed during SN II events. Thus, the [α/Fe] ratio in a star is a way to trace the relative contributions from SN II to SN Ia products in the interstellar medium (ISM) when it formed (the nucleosynthesis of iron is discussed in the next section). Although Ti is not a true α-element, the dominant isotope ⁴⁸Ti forms through explosive Si burning and the α-rich freezeout,²⁰ thus it behaves like an α-element (Woosley & Weaver 1995).

Our [α/Fe] results are plotted in Figures 9 and 10. Koch et al. (2008a) determined α-element abundances for four stars in common with ours, and a solid line connects those results in the figures. In general, the α-element abundances in Carina tend to be lower than the Galactic distributions at all intermediate metallicities, i.e., −2.2 < [Fe/H] <−1.3. In the most metal-poor stars at [Fe/H] = −2.9, most of the α-elements are in good agreement with the Galactic distributions. Two stars stand out in their [α/Fe] distributions, Car-612 and Car-5070, which have very low ratios of [Mg, Ca, and Ti/Fe] for their metallicities. These two stars will be discussed further in the following sections.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Oxygen. Oxygen abundances in this analysis are solely from the [O i] 6300 line. Oxygen could only be measured in the higher metallicity stars in the sample where the line is stronger. Excellent agreement was found for one star in common with Koch et al. (2008a; Car-484, where they found [O/Fe] = +0.39 and we find = +0.48).

Silicon. Silicon abundances are only available for the higher metallicity stars. Line measurements of the Si i lines at 5684 Å and 6155 Å are similar to those in Koch et al. (2008a), however our oscillator strength for one line is lower by 0.4 dex. Only upper limits were available from the Magellan spectra taken of the metal-poor stars.

Magnesium. Mg i abundances were calculated from the λ5528 line in all of the Carina stars, combined with λ5172 and λ5183 when <200 mÅ, and λ5711 in the FLAMES/UVES spectra (it was too weak to measure in the Magellan spectra). The abundances between these lines are in good agreement. We note that the Carina Mg abundances are in good agreement with those from Koch et al. (2008a) and Shetrone et al. (2003); see Figure 10.

Calcium. Many lines of Ca i were available in all of the program stars, over a wide range of wavelengths, and of appropriate line strengths. Results are in good agreement with Koch et al. (2008a); see Figure 10.

Titanium. Titanium abundances were determined from a range of Ti i and Ti ii lines across the entire spectral region. The abundances between the two species are in good agreement, typically within 1σ of each other, despite the possibility that titanium can be overionized by the radiation field resulting in lower Ti i abundances when LTE is assumed. Letarte et al. (2010) suggested that better agreement is found when [Ti i/Fe i] is compared to [Ti ii/Fe ii] in their sample of Fornax RGB stars, due to similarities in the NLTE effects and possibly due to temperature scale variations. Examination of our M68 comparison stars are in agreement with their finding, such that the agreement improves when Ti ii is compared with Fe ii; however this had no effect on our Carina abundances. Thus, in this paper, the abundances from both species are averaged together. We note that our Carina Ti abundances are in good agreement with those from Koch et al. (2008a) and Shetrone et al. (2003); see Figure 10.

Sodium is produced primarily during the carbon burning stages, but as metallicity increases a sufficient amount of neutron-rich material also allows sodium to be produced through the Ne–Na cycle during H burning (Woosley & Weaver 1995). Thus, the chemical evolution of Na initially follows the α-elements, but then deviates from this (rising) when AGB stars contribute. Furthermore, Herwig (2004) have shown that metal-poor AGB stars from 2 to 5 M_☉ will overproduce sodium by factors of 10–100. This Na overproduction has multiple sites dependent on the AGB mass.

In the FLAMES/UVES spectra, sodium abundances were determined from a single Na i line at 5688 Å; this was because the resonance Na D lines at 5895 and 5889 Å were too strong (∼300 mÅ), and the line at 5682 Å (used by Koch et al. 2008a) was too weak. Only the most metal-poor stars in Carina have Na D lines that are weak enough to be used in this analysis, but since those stars were observed with Magellan/MIKE spectra, their lower S/N spectra required spectrum syntheses for the line abundance determinations. The Fe i 5615 and 5634 lines were examined to check the broadening parameters (Gaussian broadening with FWHM = 0.25 Å was adopted, which is the effective resolution of the Magellan MIKE spectra). When the Na D lines could not be fit simultaneously, then an average of the best-fit abundances for each line was adopted, and the range used to estimate the uncertainty.

Andrievsky et al. (2007) have examined the NLTE corrections for the Na D resonance lines in metal-poor RGB stars. Using their Table 2, combined with their Figure 6 showing the effect of the correction on the line EWs, then we estimated an NLTE correction for our abundances. For the M68 stars, the higher gravities and strong line strengths imply corrections of ∼ − 0.5 dex. For the three metal-poor Carina stars (Car-1087, Car-5070, and Car-7002), the atmospheric parameters and line strengths suggest corrections of ∼ − 0.6 dex. These corrections are included in Figure 11. Of course, no corrections are applied to the other Carina objects since we did not use the Na D lines for those.

Zoom In Zoom Out Reset image size

For Car-612, we only determined an upper limit, EW < 30 mÅ for Na i λ5688, which corresponds to a very low upper limit of [Na/Fe] = −0.8. This result is comparable to that from Koch et al. (2008a), who measured an EW (=24 mÅ) for the same weak line and determined [Na/Fe] = −0.5. Our lower upper limit is due to small differences in the atmospheric parameters and adopted solar Na abundance.

In the early universe, the iron-peak elements (Sc to Zn: 23 ⩽ Z ⩽ 30) are exclusively synthesized during Type II supernovae explosions by explosive oxygen and neon burning, and complete and incomplete explosive Si burning. Abundances of these elements show a strong odd–even effect (odd nuclei have lower abundances than the even nuclei). Their yields depend on the mass of the progenitors (e.g., Woosley & Weaver 1995), the mass cut adopted (mass expelled relative to the mass that falls back onto the remnant, e.g., Nakamura et al. 1999), and SNe II explosive energies (e.g., Umeda & Nomoto 2005; Heger & Woosley 2002). Only at later times (∼1 Gyr, e.g., Maoz et al. 2010), when lower mass stars reach the end of their life time, do SNe Ia become a significant, possibly the dominant, contributor to the total iron-group inventory. The onset of SNe Ia in the chemical evolution of the Galaxy is observed as a knee in the [α/Fe] versus [Fe/H] near [Fe/H] ∼−1.0 (e.g., McWilliam et al. 1995). At lower metallicities, SNe Ia [X/Fe] yields could vary (e.g., Kobayashi & Nomoto 2009). We have measured several of the Fe-peak elements in the Carina stars to compare with the Galactic trends.

Scandium. Several (4–13) lines of Sc ii were measured and HFS corrections applied. The distribution in the Sc abundances at intermediate metallicities is larger than most of the other iron-peak elements (see Figures 12 and 13), although this includes Car-612, which has several chemical peculiarities and stars analyzed by Shetrone et al. (2003). Other than those objects, the Sc abundances are similar to the Galactic distribution.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Vanadium. For most of the intermediate-metallicity stars, the V abundance is well determined from ∼10 V i lines and HFS corrections were applied. However, in the metal-poor stars and Car-612, it is determined from only 1–2 lines.

Manganese. The Mn i abundances were determined from several lines (4–7) in the intermediate-metallicity stars, and these line abundances showed good agreement with one another. Generally, [Mn/Fe] in Carina follows the Galactic distribution above [Fe/H] = −2. In the metal-poor Carina stars and M68 stars, a different set of spectral lines was used (because of the wavelength coverage and resolution of the Magellan/MIKE spectra). The blue resonance lines at λ4030, 4033, and 4034, as well as several additional subordinate lines were measured, with only λ4823 in common between the MIKE/Magellan and FLAMES/UVES data sets. The Mn results from the resonance lines were lower than from λ4823 and the other subordinate lines in the M68 standard stars, most likely due to non-LTE effects. Bergemann & Gehren (2008) have calculated non-LTE corrections for Mn i lines and do find large corrections for the blue resonance lines that are metallicity dependent. They do not give corrections for red giants, but for red main-sequence stars the corrections can be as large as +0.5 dex at [Fe/H] = −3. These corrections are similar to the offsets that we find between resonance and subordinate lines, therefore we apply a correction of Δlog(Mn) = +0.5 dex to the Mn i resonance line abundances—this correction should be checked with proper modeling of these metal-poor red giant atmospheres. Correcting the resonance line abundances improves the mean [Mn/Fe] results, but the mean values in Car-5070 and Car-1087 remain lower than for similar stars in the Galactic halo.

Chromium. Cr abundances were determined from 4–7 lines of Cr i, which showed good agreement from line to line. A recent calculation of the NLTE effects on the Cr i lines by Bergemann & Cescutti (2010) showed that the corrections are small (⩽0.1 dex), but that Cr i/Cr ii ionization equilibrium and the solar [Cr/Fe] ratio is regained for metal poor stars, rather than the downward trend seen in Figure 13. Our Carina data generally follow the Galactic trend, with the exception of very low abundances in Car-5070, Car-1087, Car-705, and Car-612.

Cobalt. Co was determined from only 1–2 lines in this analysis. The [Co/Fe] ratios lie slightly below the Galactic abundances, although they are not well constrained from so few lines. The only exception is the very low upper limit to [Co/Fe] for Car-612 (see Figure 13). Bergemann et al. (2010) calculated NLTE corrections for Co i and Co ii; they suggest the NLTE corrections depend on metallicity and can become as large as +0.6 to +0.8 dex at [Fe/H] ∼−3.0. This would not affect the very low Co upper limit found for Car-612.

Nickel. Several Ni i lines (13–17) were measured in the intermediate-metallicity stars over a range of wavelengths, showing that [Ni/Fe] is in agreement with the Galactic distribution (Figure 13). Only 1–3 Ni i lines were available in the most metal-poor stars though, thus the uncertainties are large because σ(Fe i) was adopted for the error estimates. Only the Ni result for Car-612 stands out, but this result is robust, determined from 14 Ni i lines with a small error in the mean. Combined with the low Na upper limit, then Car-612 fits the Na–Ni relationship observed for some α-poor stars in the Galaxy by Nissen & Schuster (1997, 2010).

While contiguous on the periodic table, the main nucleosynthetic sites of copper and zinc are difficult to ascertain. In massive stars, Cu and Zn form during complete Si burning, the α-rich freezeout, and even the weak s-process (e.g., Timmes et al. 1995). The contributions to the production of Cu and Zn in AGB stars are uncertain, though recent calculations suggest that small amounts of Cu and even smaller amounts of Zn can be produced in more massive (5 M_☉) AGB stars (Karakas et al. 2008; Karakas 2010). Precise estimates of the AGB yields can also depend on uncertain parameters such as the mass-loss law and number of dredge-up episodes (Travaglio et al. 2004). The majority of these elements are thought to form in SN Ia events though, where the yields of Cu and Zn are sensitive to the neutron excess and thus metallicity (Matteucci et al. 1993; Mishenina et al. 2002; Travaglio et al. 2005; Kobayashi & Nomoto 2009).

Copper. The Cu i line at 5105 Å was observed in some of our target stars. Upper limits were calculated adopting an EW of 30 mÅ for the FLAMES/UVES spectra and 50 mÅ for the most metal-poor Carina stars. HFS corrections were applied. Generally, copper follows the Galactic trend, and though we have very few actual measurements, the upper limit for Car-612 is very low and provides a very strong constraint.

Zinc. The Zn i line at λ4810 was observed in most of our sample, or used to determine the upper limits to the Zn abundance (adopting <40 mÅ); see Figure 14. The zinc abundances at intermediate metallicities were slightly lower than the Galactic distribution, especially Car-612.

Zoom In Zoom Out Reset image size

Neutron-capture elements (Sr to U, 38 ⩽ Z ⩽ 92) originate in the rapid neutron-capture process (r-process) that occurs during explosive nucleosynthesis in SNe II. The r-process has a main source in 8–10 M_☉ stars that form elements with Z > 50, and a weak source in more massive progenitors (>20 M_☉) that contribute primarily to the lighter neutron-capture elements (e.g., Sr, Y, Zr; Travaglio et al. 2004). A more detailed examination of the r-process shows that a more accurate representation is in terms of a high entropy wind model (Farouqi et al. 2009; Roederer et al. 2010); nucleosynthesis in low entropy winds proceeds primarily through a chared-particle (α-) process, whereas a neutron-capture component (the classical weak and main r-processes) occur in the high entropy winds. Thermal pulsing in AGB stars also contributes to these elements through the slow neutron-capture process (s-process). With intermediate masses (2–4 M_☉), AGB stars have longer lifetimes than the sites of the r-process, and therefore in simple chemical evolution models AGB stars do not produce any of the heavy elements in the most metal-poor stars, but dominate the formation of some of these elements at later times and higher metallicities. In the Galaxy, this is seen as the rise in the s-process that begins near [Fe/H] = −2.5 (McWilliam 1998; Burris et al. 2000; François et al. 2007); this is in contrast to the knee in [α/Fe] due to contributions from SNe Ia near [Fe/H] =−1.0. In the Sun, the r-process contributes 11%, 15%, 25%, 28%, 53%, and 97% of Sr, Ba, La, Y, Nd, and Eu, respectively (Burris et al. 2000). Thus, Eu is critically important as a nearly pure r-process indicator.

Yttrium. [Y/Fe] in the intermediate-metallicity stars was determined from Y ii 4883, 5087, and 5200, which gave very consistent abundances from line to line and star to star. The Y ii 4900 line could not be used because it is severely blended in all of the spectra. The [Y/Fe] ratios are similar to those found by Shetrone et al. (2003), i.e., slightly below the Galactic values at intermediate metallicities; see Figure 15. One exception is Car-612, which has an extremely low [Y/Fe] result, and another is Car-5070, which has a low upper limit value. The low [Y/Fe] in Car-612 can be seen directly in comparison with Car-705—Figure 16 shows that the Fe i line strengths are similar but the Y ii 5087 line is much weaker in Car-612. Upper limits were determined in the most metal-poor stars using 40 mÅ as the EW upper limit for the Y ii λ4883 line.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Barium. Five lines of Ba ii were analyzed (at 4554, 4934, 5853, 6141, and 6496 Å), which yielded consistent abundances from line to line and star to star, although they were not all observed in one star. Spectrum syntheses were used to confirm the abundances when the S/N was low. HFS corrections for three lines (5853, 6141, and 6496 Å) are negligible (⩽0.02). NLTE corrections are also negligible (⩽0.03, Short & Hauschildt 2006), with the exception of the λ4554 resonance line, however we have chosen to not correct that line since the estimated correction is not large (⩽0.15 dex, Short & Hauschildt 2006; Andrievsky et al. 2009) and the LTE abundance from that line is consistent with the other Ba ii line results. Most stars in Carina have the same [Ba/Fe] distribution as stars in the Galaxy, with the exception of three stars, Car-612, Car-5070, and Car-705; see Figure 15.

Lanthanum. La was determined from three lines of La ii at λ6320, 6390, 6774 (additional lines at λ4333, 5301, 5303 could not be used due to the S/N of the spectra). Negligible HFS (⩽0.02) were calculated for two La lines (λ6320 and 6390). [La/Fe] follows the Galactic distribution for the few stars where we could measure it (see Figure 15).

Strontium. [Sr/Fe] was determined from two Sr ii lines at very blue wavelengths (4077 and 4215 Å) reached only by our Magellan/MIKE spectra. The S/N is poor in that region, thus we adopted the results from spectrum syntheses for those lines in the three Carina stellar spectra; see Figure 17. Nevertheless, the interpretation of the synthetic results remains difficult; the λ4077 line is the stronger of the Sr ii resonance lines, and yet the detection of the λ4215 line is more clear in Car-7002, and possibly Car-1087. The best estimates for the Sr abundances in these two stars are listed in Table 13. An Sr upper limit for Car-5070 could not be determined due to noise near both the λ4215 and 4077 lines. NLTE corrections were considered, but not applied to our results. Short & Hauschildt (2006) estimate abundance corrections of −0.07 dex, whereas Andrievsky et al. (2011) estimate corrections of ∼0.0 and +0.1 dex for λ4077 and 4215, respectively. We chose to not apply either correction since these estimates are small and in opposite directions.

Zoom In Zoom Out Reset image size

Zirconium. Upper limits for Zr were determined from the Zr ii λ4208 line in the Magellan spectra. These upper limits are quite high and do not add the discussion on the heavy element abundances.

Neodymium. Nd was determined from two Nd ii lines (λ5319 and 5249) for all stars with FLAMES/UVES spectra, and these lines were used to calculate the upper limits from the Magellan spectra. Nd is consistent with the Galactic distribution, though it is very low in Car-612 (Figure 18).

Zoom In Zoom Out Reset image size

Europium. The europium abundance in the Sun is nearly entirely due to the r-process, therefore Eu is identified as an important indicator in any stellar analysis to establish the ratio of r-process to s-process contributions of the neutron-capture elements. In this analysis, three Eu ii lines were examined (λ4129, 4205, and 6645). Unfortunately, all three lines were not observed in any one star; the Magellan spectra have too low resolution to detect the weak 6645 Å line, and the FLAMES/UVES data do not cover the bluer wavelengths. Nevertheless, the Eu results are consistent with the Galactic distribution (see Figure 18). HFS and isotopic splitting corrections are negligible (<0.05 dex).

To assess the relative amounts of s-process to r-process contributions among the neutron-capture elements, then abundance ratios relative to Eu are examined. In Figure 19, [Y/Eu] and [Ba/Eu] of stars in Carina and the Galaxy are shown and compared to the solar r-process fractions from Burris et al. (2000; also see Arlandini et al. 1999). In both [Y/Eu] and [Ba/Eu] (as well as [La/Eu] and [Nd/Eu], not shown), the gradual rise in these ratios can be seen with increasing metallicity. McWilliam (1998) showed that the rise in the s-process starts at [Fe/H] = −2.5 in the Galaxy. In Carina, this rise appears to begin at a higher metallicity, [Fe/H] ⩾−2.0. This is most likely due to the metallicity dependence of the AGB yields, as modeled by Travaglio et al. (2004; also Pignatari et al. 2010), where fewer iron seed nuclei in a high neutron density wind can collect more neutrons, thus underproducing the first s-process peak elements,²¹ and overproducing the second and/or third s-process peak elements. The slightly lower [Y/Eu] and slightly higher [Ba/Eu] ratios in Carina are consistent with metal-poor AGB stars contributions.

Zoom In Zoom Out Reset image size

As a final note, the high Ba abundance in M68 is somewhat surprising. This was also found by Lee et al. (2005). Since [Ba/Eu] is high, but [Eu/Fe] is normal, this suggests M68 has had an unusual chemical evolution, possibly having been enriched (or self-enriched) by metal-poor AGB stars.

In Figure 21, we show only the metal-poor tail in the Carina abundances and compare the [Mg/Fe] and [Ca/Fe] ratios to those of metal-poor stars in five UFD galaxies and the Draco dSph. Most of the UFD stars have high values of [Mg/Fe] and [Ca/Fe], like the Galactic halo, whereas Carina and Draco show a range extending to very low values.

Zoom In Zoom Out Reset image size

Some stars in all of these systems show very high [Mg/Ca] values (>0.5), e.g., two stars in the Hercules UFD (Koch et al. 2008b), one in Böotes I (Feltzing et al. 2009), one in Draco (Fulbright et al. 2004; Shetrone et al. 2001), and one in Carina (this paper). In the UFDs, it has been proposed that high [Mg/Ca] may be due to the chemical enrichment of its interstellar gas by as few as one SN II explosion from a massive progenitor (e.g., a >35 M_☉ star). In this scenario, a unique chemical signature can be imprinted onto the gas that is used to form stars, while the rest of the gas is expelled to quench further star formation (e.g., Koch et al. 2008b; Frebel et al. 2010b; Simon et al. 2010). This scenario is aided by peculiar neutron-capture ratios, e.g., in the Hercules stars, the Ba ii lines are not detected (Koch et al. 2008b). Similarly, the Ba ii (and Sr ii) lines in the high [Mg/Ca] star in Draco (Draco-119) are also not detected (Fulbright et al. 2004 suggested that this star was enriched by an SN II in the mass range of 20–25 M_☉ and is lacking in contributions from higher mass progenitors). However, Ba ii (and Sr ii) lines are detected in the high [Mg/Ca] star in Carina. Those lines are weak, leading to very low abundances of [Ba/Fe] and [Sr/Fe], but if the source of these elements is the same as in the other dSphs, then Carina appears to have been able to retain a bit more of those early (high-mass progenitor) SN II enrichments.

One question worth considering is whether the large dispersion in [Mg/Fe] is simply due to measurement errors. For example, there are only 1–2 Mg i lines available in each star, whereas there are 3–21 lines of Ca i (depending on the metallicity) in this analysis. Similarly, due to the limited wavelength coverage of the FLAMES/GIRAFFE spectra, there are ⩽13 Ca i lines measured by Lemasle et al. (2012) and only one Mg i line at λ5528. A larger number of spectral lines reduces the random measurement errors in an elemental abundance (by $\sqrt{N}$). This is reflected in the smaller average error in [Ca/Fe] than [Mg/Fe] in Table 10 of Lemasle et al. (2012). In this paper, even if the measurement error from fitting a line is small, as in a high S/N spectrum, the analysis of a single line means that the adopted error is the standard deviation in the Fe i line abundance (σ(Fe); see Section 4.2). These (conservative) errors are about as large as the [Mg/Ca] dispersion, but smaller than the [Mg/Fe] dispersion, therefore the large dispersion in [Mg/Fe] does not appear to be due to measurement errors alone.

Another consideration is that variations in the [Mg/Fe] and [Mg/Ca] ratios have been found in a number of metal-poor stars in the UFD galaxies, and attributed to incomplete mixing and poor sampling of the full mass function, as shown in Figure 21. Only the largest outliers are examined in terms of pollution by a single SN II, and similarly large outliers are found in our [Mg/Ca] and [Mg/Fe] data as well. The consistency between these analyses suggests that the signatures in the largest outliers are not due only to measurement errors.

An offset in the mean [Mg/Fe] abundance in the IA population relative to the old population can clearly be seen in Figure 22. A Kolmogorov–Smirnov (K-S) test shows that the distribution of the [Mg/Fe] values for the old population is broader and with lower mean values, i.e., 〈[Mg/Fe]〉_old = 0.05 ± 0.32 and 〈[Mg/Fe]〉_IA = 0.22 ± 0.15, both with moderately high probabilities of having "normal" distributions once the outlier in the IA data set (at [Mg/Fe] = −0.95) is removed. This was also seen by Lemasle et al. (2012) for both [Mg/Fe] and [Ca/Fe], although the offset in [Ca/Fe] is not as large, which we confirm with a K-S test. We also note that these offsets in the [Mg/Fe] and [Ca/Fe] ratios with age are still present when only the high-resolution data are examined (this paper; Koch et al. 2008a; Shetrone et al. 2003), although they are not as clear. The same signature is also hinted at in the [Ti/Fe] ratios (from the high-resolution data), but there is insufficient data in the IA population for a meaningful K-S test. Since the large dispersions in the [α/Fe] ratios discussed above (larger than seen in most dSph galaxies) are an indication of inhomogeneous mixing, the simplest explanation for the offset in [α/Fe] between the old and IA population is that the second epoch of star formation occurred in α-enriched gas. The small range in the [α/Fe] and [Fe/H] ratios suggests this gas was well mixed.

Clearly the old and IA stellar populations in Carina show different chemical signatures, and it is interesting to investigate how the two populations might be connected. While we have discussed the α-elements, possibly the most valuable abundance ratios for studying the chemical evolution of a dwarf galaxy are of the neutron-capture elements.

In Figure 23, it is possible to see that AGB stars have contributed to both age groups in Carina, but not until [Fe/H] >−2. This can be seen by the [Ba/Eu] ratios, where the lower dashed line represents the r-process contributions (in the Sun), such that stars with this ratio have been enriched by SNe II products only. Above [Fe/H] =−2, the [Ba/Eu] ratios slowly increase from the low r-process value through contributions to Ba from the s-process (the rise in the s-process). This is seen in the Galaxy, Fornax, and Sculptor as well, although the rise begins at higher metallicities in those systems. From the [Ba/Fe] ratios, it is also possible to see how the AGB contributions begin to dominate the nucleosynthesis of Ba, e.g., the dispersion in [Ba/Fe] at low metallicities in the Galaxy is interpreted as the stochastic sampling of the (small amounts of) Ba from SNe II products early on, but this scatter lessens when the dominant contributions from AGB stars come at later times. The flatness of the relationship in [Ba/Fe] at higher metallicities in the Galaxy implies that the timescale and yields of Ba from the AGB stars and Fe from SNe Ia are similar, and the small scatter implies that the gas is well mixed. In Carina, [Ba/Fe] in the IA population may have a large scatter. If the large scatter is real, then this would suggest that the AGB contributions were not well mixed in the ISM at any age, which would be an interesting contrast to the uniformity in the [α/Fe] ratios in the IA population. However, several of these data points are from Lemasle et al. (2012) where the measurements of a single Ba ii line were very difficult due to the low S/N in that spectral region. The data from our high-resolution analysis is far more uniform above [Fe/H] = −2, as shown in Figure 15.

Below [Fe/H] = −2 it is difficult to ascertain whether there are any stars with AGB contributions. There is one star (Car-5070) with [Ba/Eu] below the r-process ratio, implying that this star shows no signs of AGB products; however, Carina was poorly mixed at early times, and metal-poor AGB stars could produce very little Ba. The high [Ba/Y] ratios in stars over [Fe/H] = −2 show that most of the s-process elements came from metal-poor AGB stars, where the first s-process peak (Y) was bypassed in favor of the second s-process peak (Ba; as described further in Section 4.4.7)—at even lower metallicities, both can be bypassed for the third s-process peak. Therefore, we cannot ascertain the precise metallicity at which AGB stars began to contribute in Carina, but certainly there are contributions from metal-poor AGB stars in the stars with [Fe/H] >−2, and a hint of the rise in the s-process between −2.0 < [Fe/H] <−1.6.

Further evidence for the rise in the s-process can be seen in the evolution in the [Na/Fe]; see Figure 24. Cayrel et al. (2004) and Andrievsky et al. (2007) showed that the [Na/Mg] and [Na/Fe] ratios are flat and low in metal-poor stars in the Galaxy, suggesting that Na is produced with the α-elements in SNe II at a (low) fixed ratio. The Na abundance rises when AGB stars begin to contribute (see Section 4.4.3). We see a very similar trend in the [Na/Fe] ratios in Carina; in fact, the rise in [Na/Fe] is very similar to the rise in the s-process elements. One significant difference though is that the initial [Na/Fe] value appears to be much lower in Carina than in the Galaxy. The final [Na/Fe] ratios may be similar in Carina as in the Galaxy (and Sculptor), and we see no offset in the Na abundances between the old and IA populations. A dispersion in Na is not clear in this small sample.

It is interesting that the evolution of [Mn/Fe] and [Cr/Fe] are also similar to Na in Carina. The abundances of these elements are much lower than in the Galaxy at low metallicities, but then rise above [Fe/H] = −2.0 to abundances that are similar to stars in the Galaxy with the same metallicity. There is also no difference in these elements between the old and IA populations. Like Na, once AGB stars and SNe Ia begin to produce iron-group elements, those contributions will dominate over the initial (low) SNe II yields. The timescale for AGB contributions is thought to be equal to or shorter than that of SNe Ia (∼1 Gyr, Maoz et al. 2010), thus it is likely that both AGB stars and SNe Ia begin to contribute to the ISM in Carina near [Fe/H] =−2.0. The SNe Ia contributions appear to be well mixed in the ISM, and there is no offset between the old and IA stars.

The earliest stages of chemical evolution in Carina can be examined from the element ratios of the stars below [Fe/H] = −2. As discussed above, these stars appear to have been enriched only in SNe II products. While there is a large range in the [α/Fe] and [Mg/Ca] ratios (when the whole data set is examined, including previously published results), the ratios of Mn, Cr, and Na all start out very low but then rise to the Galactic values at intermediate metallicities.

In addition, the [Sr/Fe] values are among the lowest of all stars analyzed in the Galaxy, other dSphs, and the UFDs (see Figure 25). The [Sr/Fe] values for Car-1087 and Car-7002 are similar to the lowest values found for metal-poor stars in Draco, Boötes I, and Com Ber. Tafelmeyer et al. (2010) suggested that dwarf galaxies may have a lower floor in the [Sr/Fe] ratio (∼ − 1.2) when their mass is equal to or lower than Draco. The low [Sr/Fe] ratios in Carina lead to very low [Sr/Ba] ratios as well. [Sr/Ba] ratios are quite interesting because they tend to be high in metal-poor stars in the Galactic halo, which has been interpreted as evidence for an extra nucleosynthetic source in the formation of these elements in massive stars (Travaglio et al. 2004; Ishimaru et al. 2004; Qian & Wasserburg 2007; Farouqi et al. 2009; Pignatari et al. 2010).

Zoom In Zoom Out Reset image size

The absence of an [Sr/Ba] enhancement at low metallicities in Carina suggests a lack of the excesses seen in the Galactic stars. It is not clear from which stellar mass range these excesses arise: Travaglio et al. (2004) suggest the main r-process occurs in 8–10 M_☉ stars, and that the excess must arise from the more massive SNe II, whereas Farouqi et al. (2009) suggest the excess occurs in SN II with lower entropy winds where a neutron-rich, α-freezeout can occur. The specific mass or energy range in Farouqi's models is not clear. If we follow Travaglio's suggestions, then the excess Sr may form in more massive, or higher energy SNe II (i.e., hypernovae), and therefore those SNe II seem to be missing in Carina—either those stars did not form or their ejecta was driven out of Carina.²³

Mn and Cr also form in hypernovae as the decay products of complete and incomplete Si burning (Umeda & Nomoto 2002; Nomoto et al. 2001), thus a lack of hypernovae might also explain the very low initial values of these elements. A neutron-rich, α-freezeout in hypernovae would also contribute to Na production, and thus a lack of hypernovae (i.e., if this is the source of the Sr excess, as in Farouqi et al.'s 2009 models) could explain the very low [Na/Fe] ratios in the most metal-poor stars. Evolution models (e.g., Salvadori et al. 2008; Brooks et al. 2009, 2007; Ferrara & Tolstoy 2000) suggest that very low mass galaxies can lose gas after the first star-forming epoch through SN II driven winds (possibly reaccreting cold gas for later star formation events). Koch et al. (2006) also suggested that the loss of metal-rich winds would help to explain the metallicity distribution function of Carina, which suffers from the well-known G dwarf problem. Detailed models of the star formation and chemical evolution of Carina by Lanfranchi & Matteucci (2004) invoked two major epochs of outflows through winds. Their best model included a wind efficiency that was 7× the star formation rate and associated with the initial and IA star formation episodes. Thus, the detailed chemical composition of the most metal-poor stars in Carina suggest a lack of hypernovae contributions, possibly implying that these stars did not form, but more likely indicating that their gas was removed from Carina by SN-driven winds.

One alternative to this scenario is that the metal-poor stars are displaying the imprint of the hypernovae, rather than the lack of hypernovae, which would provide a very strong constraint on the nucleosynthetic models. An excellent test of this alternative would be the [Zn/Fe] abundances, which are predicted to be enriched by hypernovae because of an increase in the mass ratio between the complete and incomplete Si-burning regions (Nomoto et al. 2001). We are only able to determine upper limits to the [Zn/Fe] ratios in three stars in Carina below [Fe/H] =−2 from our moderate S/N Magellan/MIKE spectra; however, these limits are tantalizingly close to providing an interesting constraint (see Figure 14). In particular, Car-5070 has a low [Zn/Fe] upper limit and suggests that Car-5070 lacks hypernova enrichments.

Carina and Draco (and possibly Boötes I) appear to be quite different in their early chemical evolution from the other dSphs and UFDs. These two (three) galaxies may be at the critical mass where SN-driven winds remove the gas from the most massive or energetic SNe II progenitors, but the products of the remaining SNe II are retained, and contribute to the (inhomogeneous) chemical evolution of the host. This is unlike the dSphs, which appear to retain the gas from the earliest epochs and undergo a smooth chemical evolution that is not too different from stars in the Galactic halo (other than contributions occurring at lower metallicities, e.g., the AGB yields are from metal-poor stars and the SNe Ia contribute at lower metallicities). Both of these galactic systems are unlike the UFDs, where a single massive SN II may remove all of the gas, quenching the star formation event, and imprinting their unique chemical signatures on the few stars that will complete the star formation process—in this case, any abundance variations within an UFD are due to stochastic sampling of that hypernova and not chemical evolution.