\omega Centauri Abundances

Chemical Abundances for 855 Giants in the Globular Cluster Omega Centauri (NGC 5139)

Christian I. Johnson11affiliation: Department of Astronomy, Indiana University, Swain West 319, 727 East Third Street, Bloomington, IN 47405–7105, USA; cijohnson@astro.ucla.edu; catyp@astro.indiana.edu 22affiliation: Department of Physics and Astronomy, University of California, Los Angeles, 430 Portola Plaza, Box 951547, Los Angeles, CA 90095–1547, USA 33affiliation: Visiting astronomer, Cerro Tololo Inter–American Observatory, National Optical Astronomy Observatory, which are operated by the Association of Universities for Research in Astronomy, under contract with the National Science Foundation. and Catherine A. Pilachowski11affiliation: Department of Astronomy, Indiana University, Swain West 319, 727 East Third Street, Bloomington, IN 47405–7105, USA; cijohnson@astro.ucla.edu; catyp@astro.indiana.edu 33affiliation: Visiting astronomer, Cerro Tololo Inter–American Observatory, National Optical Astronomy Observatory, which are operated by the Association of Universities for Research in Astronomy, under contract with the National Science Foundation.
Abstract

We present elemental abundances for 855 red giant branch (RGB) stars in the globular cluster Omega Centauri (\omega Cen) from spectra obtained with the Blanco 4m telescope and Hydra multifiber spectrograph. The sample includes nearly all RGB stars brighter than V=13.5, and span’s \omega Cen’s full metallicity range. The heavy \alpha elements (Si, Ca, and Ti) are generally enhanced by \sim+0.3 dex, and exhibit a metallicity dependent morphology that may be attributed to mass and metallicity dependent Type II supernova (SN) yields. The heavy \alpha and Fe–peak abundances suggest minimal contributions from Type Ia SNe. The light elements (O, Na, and Al) exhibit >0.5 dex abundance dispersions at all metallicities, and a majority of stars with [Fe/H]>–1.6 have [O/Fe], [Na/Fe], and [Al/Fe] abundances similar to those in monometallic globular clusters, as well as O–Na, O–Al anticorrelations and the Na–Al correlation in all but the most metal–rich stars. A combination of pollution from intermediate mass asymptotic giant branch (AGB) stars and in situ mixing may explain the light element abundance patterns. A large fraction (27\%) of \omega Cen stars are O–poor ([O/Fe]<0) and are preferentially located within 5–10\arcmin of the cluster center. The O–poor giants are spatially similar, located in the same metallicity range, and are present in nearly equal proportions to blue main sequence stars. This suggests the O–poor giants and blue main sequence stars may share a common origin. [La/Fe] increases sharply at [Fe/H]\gtrsim–1.6, and the [La/Eu] ratios indicate the increase is due to almost pure s–process production.

stars: abundances, globular clusters: general, globular clusters: individual (\omega Centauri, NGC 5139), stars: Population II

1 INTRODUCTION

For many years, globular clusters were regarded as prototypical simple stellar populations. However, recent observations have revealed that several of the most massive known globular clusters contain multiple main sequence, subgiant, and/or red giant branch (RGB) populations (Piotto et al. 2007; Marino et al. 2008; Milone et al. 2008; Anderson et al. 2009; Moretti et al. 2009; Piotto 2009; Milone et al. 2010). These data, combined with the well–known and pervasive light element abundance correlations and anticorrelations that appear to be unique to the globular cluster environment, suggest that many, if not all, globular clusters undergo at least some degree of self–enrichment (e.g., Carretta et al. 2009a). While nearly all of these clusters exhibit small (\lesssim0.1 dex) star–to–star metallicity variations (e.g., see review by Gratton et al. 2004), Omega Centauri (\omega Cen) has long been known to exhibit both a complex color–magnitude diagram and a metallicity spread of more than a factor of ten.

Early color–magnitude diagrams of \omega Cen indicated that it hosts an unusually broad RGB (e.g., Woolley 1966; Cannon & Stobie 1973). Subsequent photometric surveys have discovered that this trend continues into both the main sequence and subgiant branch regions as well (Anderson et al. 1997; Lee et al. 1999; Hilker & Richtler 2000; Hughes & Wallerstein et al. 2000; Pancino et al. 2000; van Leeuwen et al. 2000; Bedin et al. 2004; Ferraro et al. 2004; Rey et al. 2004; Sollima et al. 2005; Castellani et al. 2007; Sollima et al. 2007; Villanova et al. 2007; Bellini et al. 2009a; Calamida et al. 2009). Additionally, detailed photometric and spectroscopic analyses have shown that at least 4–5 discrete populations are present in the cluster (Norris et al. 1996; Lee et al. 1999; Hilker & Richtler 2000; Pancino et al. 2000; Bedin et al. 2004; Rey et al. 2004; Sollima et al. 2005; Castellani et al. 2007; Villanova et al. 2007; Johnson et al. 2008; Bellini et al. 2009a,b; Calamida et al. 2009; Johnson et al. 2009; Marino et al. 2010). These individual populations span a metallicity range from [Fe/H]111We make use of the standard spectroscopic notations where [A/B]\equivlog(N{}_{\rm A}/N{}_{\rm B}){}_{\rm star}– log(N{}_{\rm A}/N{}_{\rm B}){}_{\sun} and log \epsilon(A)\equivlog(N{}_{\rm A}/N{}_{\rm H})+12.0 for elements A and B.\approx–2.2 to –0.5. However, few stars are found with [Fe/H]<–2, and more than half of \omega Cen’s stars reside in a population peaked near [Fe/H]\approx–1.7 (Norris & Da Costa 1995; Suntzeff & Kraft 1996; Hilker & Richtler 2000; Smith et al. 2000; Cunha et al. 2002; Sollima et al. 2005; Kayser et al. 2006; Stanford et al. 2006; Villanova et al. 2007; Johnson et al. 2008; Calamida et al. 2009; Johnson et al. 2009; Marino et al. 2010). The rest of the stars reside in the intermediate metallicity populations, and a minority (\lesssim5\%) of stars are found to lie along the “anomalous”, metal–rich sequence (Lee et al. 1999; Pancino et al. 2000; Ferraro et al. 2004; Sollima et al. 2005; Villanova et al. 2007).

The large metallicity spread in \omega Cen is commonly believed to be due to significant self–enrichment induced by multiple star formation episodes (e.g., Ikuta & Arimoto 2000; Tsujimoto & Shigeyama 2003; Marcolini et al. 2007; Romano et al. 2007, 2010). Despite being the most massive known cluster in the Galaxy, with an estimated mass of \sim2–7\times10{}^{\rm 6} M{}_{\sun} (Mandushev et al. 1991; Richer et al. 1991; Meylan et al. 1995; van de Ven et al. 2006), Gnedin et al. (2002) showed that \omega Cen does not currently possess an abnormally deep gravitational potential well. Additionally, the cluster’s Galactic orbit indicates that it should pass through the disk at least every 1–2\times10{}^{\rm 8} years (Dinescu et al. 1999). This makes it hard to believe that \omega Cen could have experienced the 2–4 Gyr period of star formation that seems required to fit observations of the main sequence turnoff (e.g., Stanford et al. 2006). However, the cluster’s retrograde motion through the Galaxy (Dinescu et al. 1999) suggests that it may be a captured system and therefore could have been more massive in the past. In fact, the most popular scenario is that \omega Cen, and perhaps several other globular clusters containing multiple populations, are the remnant cores of tidally disrupted dwarf galaxies (e.g., Dinescu et al. 1999; Majewski et al. 2000; Smith et al. 2000; Gnedin et al. 2002; McWilliam & Smecker–Hane 2005; Bekki & Norris 2006). This is now favored over an accretion or merger scenario because the individual stellar populations within \omega Cen all exhibit the same proper motion, rotation, and average radial velocity (e.g., Pancino et al. 2007; Bellini et al. 2009a).

Although the observed evolutionary sequences have now been mostly matched to the different populations derived from spectroscopy, one of the remaining puzzles is how these populations relate to \omega Cen’s bifurcated main sequence. The discovery that \omega Cen’s main sequence splits into a red and blue sequence over a span of at least two magnitudes (e.g., Anderson 1997; Bedin et al. 2004) is difficult to explain because the blue main sequence is more metal–rich than the red main sequence (Piotto et al. 2005). A possible explanation for this is that the blue main sequence stars are selectively enhanced in helium at Y\sim0.38 (e.g., Norris 2004; Piotto et al. 2005). While the source of the proposed helium enrichment is not clear, the leading candidate appears to be intermediate mass (\sim3–8 M{}_{\sun}) asymptotic giant branch (AGB) stars with perhaps some contribution from massive, rapidly rotating main sequence stars (e.g., Renzini 2008; Romano et al. 2010). Interestingly, the blue main sequence stars appear to be preferentially located near the cluster core (Sollima et al. 2007; Bellini et al. 2009b), which is an indication that He–rich material may have collected there at some point in the cluster’s evolution. A similar radial segregation near the core has been found for stars with [Fe/H]\gtrsim–1.2, but the more metal–poor stars appear to be rather uniformly distributed across the cluster (Suntzeff & Kraft 1996; Norris et al. 1997; Hilker & Richtler 2000; Pancino 2000; Rey 2004; Johnson et al. 2008; Bellini et al. 2009b; Johnson et al. 2009). It is worth noting that helium enrichment may also play a role in determining the chemical composition of stars in monometallic globular clusters (e.g., Bragaglia et al. 2010).

\omega Cen shows clear signs of extended star formation and chemical self–enrichment, and the large abundance dispersion is not limited to just the Fe–peak elements. Instead, the [X/H] ratios for all elements analyzed so far are found to vary by at least a factor of ten as well (e.g., Cohen 1981; Paltoglou & Norris 1989; Norris & Da Costa 1995; Smith et al. 2000; Cunha et al. 2002; Johnson et al. 2009; Villanova et al. 2009; Cunha et al. 2010; Stanford et al. 2010). Despite the current interpretation that \omega Cen may be the surviving core of a disrupted dwarf galaxy, the [X/Fe] abundance ratios for the light elements (O, Na, Mg, and Al) and heavy \alpha elements (Si, Ca, and Ti) more closely resemble the patterns found in individual globular clusters. These patterns include the O–Na, O–Al, and Mg–Al anticorrelations concurrent with the Na–Al correlation and consistently supersolar [\alpha/Fe] ratios (e.g., Norris & Da Costa 1995; Smith et al. 2000; Johnson et al. 2009). This suggests that both Type II supernovae (SNe) and the products of proton–capture nucleosynthesis have played a significant role in shaping \omega Cen’s chemical enrichment. However, the abundance patterns of neutron–capture elements in \omega Cen stars with [Fe/H]\gtrsim–1.5 indicate that the slow neutron–capture process (s–process) was also a dominant production mechanism. This strongly contradicts the trends found in other globular clusters, and is instead more similar to observations of dwarf galaxies (e.g., see reviews by Venn et al. 2004; Geisler et al. 2007). While dwarf galaxies also contain many s–process enhanced stars, the low [X/Fe] ratios for the light and especially \alpha elements suggests a significant contribution from Type Ia SNe. In contrast, the enhanced [\alpha/Fe] ratios, high [Na/Fe] and [Al/Fe] abundances, and low [Cu/Fe] ratios seem to indicate that Type Ia SNe have played only a minimal role in \omega Cen. However, Type Ia SNe may have contributed in the most metal–rich stars, as is evidenced by a potential downturn in [\alpha/Fe] and rise in [Cu/Fe] at [Fe/H]>–1 (Pancino et al. 2002; Origlia et al. 2003; but see also Cunha et al. 2002).

In this paper we have obtained a nearly complete sample that includes 855 RGB stars and covers \omega Cen’s full metallicity range down to V=13.5. We present new chemical abundance measurements of several light odd–Z, \alpha, Fe–peak, and neutron–capture elements, and compare these results with abundance trends found in the Galactic disk, halo, bulge, globular cluster, and nearby dwarf galaxy populations. We also compare the abundance patterns found for the different \omega Cen populations in an effort to understand the cluster’s formation and chemical enrichment history.

2 OBSERVATIONS AND REDUCTIONS

The observations for this project were taken at the Cerro Tololo Inter–American Observatory (CTIO) using the Blanco 4m telescope equipped with the Hydra multifiber positioner and bench spectrograph. We obtained all of the spectra in two separate runs spanning 24–28 March 2008 and 6–8 March 2009. We employed two different wavelength setups encompassing \sim6135–6365 and \sim6500–6800 Å with wavelength centers near 6250 and 6670 Å, respectively. This required the use of two separate order blocking filters for each echelle spectrograph setup. The red setup centered near 6670 Å used the echelle filter #6 (E6757); however, neither the echelle filter #7 nor #8 provides sufficient transmissivity over the bluer region spanning 6135–6365 Å. Since a primary goal of this project is to obtain both oxygen and sodium abundances from the 6300 Å [O I] line and 6154/6160 Å Na I lines, we purchased a new, single–piece echelle filter (E6257222This filter is on long–term loan at CTIO and available for public use.) that provides >75\% transmissivity from \sim6135–6365 Å and allowed for the simultaneous observation of both the oxygen and sodium lines. For both setups, the “large” 300\micron (2\arcsec) fibers combined with the 400 mm Bench Schmidt Camera and 316 line mm{}^{\rm-1} echelle grating to yield a resolving power of R(\lambda/\Delta\lambda)\approx18,000 (0.35 Å FWHM). A summary of the Hydra observations is provided in Table 1.

Photometry, coordinates, and membership probabilities for all stars were taken from the proper motion study by van Leeuwen et al. (2000). We targeted stars with V\leq13.5 and 0.70\leqB–V\leq1.85 while excluding those with membership probabilities below 70\%. Field stars located along \omega Cen’s line–of–sight are easily removed due to the cluster’s comparatively large radial velocity and small velocity dispersion (\langleV{}_{\rm R}\rangle\sim232 km s{}^{\rm-1}; \sigma\sim10 km s{}^{\rm-1}; e.g., Reijns et al. 2006; Sollima et al. 2009). The magnitude and color restrictions provide a balance between maximizing the signal–to–noise ratio (S/N) of observations and limiting the number of required Hydra configurations. At V=13.5, one can obtain a S/N\approx100 after three hours of integration. This luminosity cutoff also allows for the observation of all giant branches in \omega Cen, and is at least 1 mag. below the RGB tip of the most metal–rich stellar population (see Figure 1).

In order to limit the number of repeat observations, stars were given a low priority in the Hydra assignment code following their inclusion into a Hydra configuration, and stars were completely removed from the fiber assignment process if incorporated into two Hydra configurations. The total number of fibers assigned to objects ranged from 50–110, and the co–added S/N ratio for almost all stars extended from about 100 to more than 350. The full sample obtained for this project is shown in Figure 1 along with the non–repeat stars from our previous papers on the cluster.

The complexity of \omega Cen’s color–magnitude diagram requires a large sample of stars to fully interpret its chemical history. Therefore, we have obtained a nearly 100\% complete sample of RGB members with V\leq13.0 and achieved more than 75\% completion for V\leq13.5. Since \omega Cen exhibits a moderate radial metallicity gradient (Norris et al. 1996; Suntzeff & Kraft 1996; Norris et al. 1997; Hilker & Richtler 2000; Pancino et al. 2000; Rey et al. 2004; Johnson et al. 2008; Bellini et al. 2009b; Johnson et al. 2009), we targeted stars spanning a wide range of cluster radii. Figure 2 shows the observed completion fraction in terms of V magnitude, B–V color, and distance from the cluster center, and Figure 3 illustrates the spatial location of our sample relative to the cluster center. For V\leq13.0, B–V>1.1, and 10\arcmin<D<24\arcmin, the completion fraction exceeds 0.90. However, the completion fraction for the inner 10\arcmin of the cluster ranges from 0.52–0.90. The decrease is due to both stellar crowding near the cluster core and physical limitations on fiber placement. Despite the large sample size, a modest evolutionary selection effect is present because the most metal–rich stars have both lower V magnitudes and tend to be located closer to the cluster center. Therefore, we have only observed stars along the most metal–rich giant branch that are within \sim1 mag. of the RGB tip.

Data reduction was handled using the necessary tasks provided in standard IRAF333IRAF is distributed by the National Optical Astronomy Observatories, which are operated by the Association of Universities for Research in Astronomy, Inc., under cooperative agreement with the National Science Foundation. packages. We used ccdproc to trim the overscan region and apply the bias level correction. However, the majority of the data reduction process was carried out with the dohydra package, which was used to trace the fiber locations on the detector, remove scattered light, apply the flat–field correction, identify lines in the ThAr comparison spectrum, apply the wavelength calibration, remove cosmic rays, subtract the background sky spectra, and extract the one–dimensional spectra. The reduction processes were identical for both the 6250 and 6750 Å data with the exception of the wavelength calibration. A problem with the calibration lamp during the 6750 Å observations meant that we had to use a high S/N, daylight solar spectrum for wavelength calibration instead of the ThAr comparison source.

Following completion of the dohydra task, the data were continuum fit and normalized before being corrected for telluric contamination. We obtained high S/N spectra of multiple bright, rapidly rotating B–stars spanning air masses ranging from 1.05 to 1.75. These spectra were used as the templates for removing the telluric features in the 6270–6350 Å window. Fortunately, the cluster’s radial velocity corresponds to a wavelength shift of roughly +4.8 Å. This moves the 6300 Å [O I] stellar absorption line away from the telluric emission feature at 6300 Å, and places it cleanly between the 6302 and 6306 Å telluric absorption doublets. After applying the telluric correction, the spectra were then co–added to remove any remaining cosmic rays and increase the S/N.

3 Analysis

3.1 Model Stellar Atmospheres

Effective temperatures (T{}_{\rm eff}) were determined via the empirical V–K color–temperature relation from Alonso et al. (1999, 2001; their equations 8 & 9), which is based on the infrared flux method (Blackwell & Shallis 1977). The V magnitudes were taken from van Leeuwen et al. (2000) and the K magnitudes were taken from the Two Micron All Sky Survey (2MASS; Skrutskie et al. 2006) database444The 2MASS catalog can be accessed online at: http://irsa.ipac.caltech.edu/applications/Gator/.. All photometry was corrected for interstellar reddening and extinction using the recommended values of E(B–V)=0.12 (Harris 1996) and E(V–K)/E(B–V)=2.70 (McCall 2004). While there is some evidence for minor differential reddening near the cluster’s core (Cannon & Strobie 1973; Calamida et al. 2005; van Loon et al. 2007; McDonald et al. 2009), the well–defined evolutionary sequences observed in the photometry by Villanova et al. (2007) seem to suggest that differential reddening is not a major issue. Therefore, we have applied a uniform reddening correction that is independent of a star’s location in the cluster. Although our data set did not contain enough Fe I lines of varying excitation potential to strongly constrain T{}_{\rm eff} via excitation equilibrium, we did not find any strong, systematic trends in plots of Fe I abundance versus excitation potential. It is likely that our photometric temperatures are accurate to within the roughly 25–50 K uncertainty range given by the Alonso et al. (1999) empirical fits.

Surface gravity (log g) estimates were obtained using the photometric temperatures and absolute bolometric magnitudes (M{}_{\rm bol}). The bolometric corrections were taken from Alonso et al. (1999; their equations 17 and 18) and applied to the absolute visual magnitudes (M{}_{\rm V}), which assumed a distance modulus of (m–M){}_{\rm V}=13.7 (van de Ven et al. 2006). Final surface gravity values were calculated with the standard relation,

 log(g_{*})=0.40(M_{bol.}-M_{bol.\sun})+log(g_{\sun})+4[log(T/T_{\sun})]+log(M/% M_{\sun}), (1)

and assumed stellar mass of 0.80 M{}_{\rm\sun}. However, the likely age spread of \sim2–4 Gyr (e.g., Stanford et al. 2006) among the stars in different populations means that a mass spread among RGB stars undoubtedly exists as well. This is further complicated by the inferred existence of a helium–rich population (Bedin et al. 2004; Norris 2004; Piotto et al. 2005) in which stars will evolve more rapidly. When one also includes “contamination” of first ascent RGB stars with AGB stars, which could account for as much as 20–40\% of the RGB above the horizontal branch (e.g., Norris et al. 1996), it is not unreasonable to assume \omega Cen giants will have a mass range spanning \sim0.60–0.80 M{}_{\rm\sun}. Fortunately, the surface gravity estimates scale with log(M) and are thus relatively insensitive to small changes in the assumed stellar mass. We estimate that the uncertainty introduced into our surface gravity values due to the inherent mass range on \omega Cen’s RGB does not exceed \Deltalog g=0.15. Comparison between the abundances of elements in different ionization states seems to substantiate this with \langle[FeI/H]-[FeII/H]\rangle=–0.09 (\sigma=0.10) and \langle[ScI/Fe]-[ScII/Fe]\rangle=–0.18 (\sigma=0.21). We provide a more detailed analysis regarding how surface gravity uncertainties affect abundance ratio determinations in \lx@sectionsign3.3.

In addition to effective temperature and surface gravity, metallicity and microturbulence (v{}_{\rm t}) information are required to generate a suitable 1–D model atmosphere. For an initial metallicity estimate, we used the empirical [Ca/H] calibration provided by van Leeuwen et al. (2000; their equation 15) with the assumptions that stars with [Fe/H]<–1 have [Ca/Fe]=+0.30 and those with [Fe/H]>–1 decline to [Ca/Fe]=0 at solar metallicity. This assumption is verified in our new [Ca/Fe] data (see \lx@sectionsign4.6). An initial microturbulence value was determined from the empirical v{}_{\rm t}–T{}_{\rm eff} relation given in Johnson et al. (2008; their equation 2). The initial T{}_{\rm eff}, log g, [Fe/H], and v{}_{\rm t} values were used to generate the necessary model atmospheres via interpolation within the available ATLAS9555The model atmosphere grids can be downloaded from http://cfaku5.cfa.harvard.edu/grids.html. grid. The final determination of all microturbulence values followed the prescription outlined by Magain (1984) in which the microturbulence was adjusted until trends of Fe I abundance versus line strength were removed. The overall model atmosphere metallicity was then adjusted to match the derived [Fe/H] abundance for each star. This value was also used to further refine the calculated effective temperature, which has a slight metallicity dependence. A full listing of star identifiers, photometry, model atmosphere parameters, and S/N ratios is provided in Table 2.

3.2 Derivation of Abundances

3.2.1 Equivalent Width Analysis

Chemical abundances for Na, Si, Ca, Sc, Ti, Fe, and Ni were determined through standard equivalent width (EW) analyses using the abfind driver in the LTE line analysis code MOOG666The MOOG code can be downloaded at: http://www.as.utexas.edu/ chris/moog.html. (Sneden 1973). Individual EWs were measured by fitting single or multiple Gaussian profiles to isolated and blended stellar absorption lines using the interactive EW fitting code developed for Johnson et al. (2008). The high resolution, high S/N solar and Arcturus atlases777These are available online from the NOAO Data Archives at: http://www.noao.edu/archives.html. (Hinkle et al. 2000) were used to aid in line identification and continuum placement. The Arcturus atlas was also used as a reference for selecting suitable spectral lines. However, the atomic log gf values were determined by an inverse solar analysis in which the EWs measured in the Sun were forced to match the photospheric abundances given in Anders & Grevesse (1989)888The solar log \epsilon(Fe) abundance was assumed to be 7.52 (Sneden et al. 1991a).. When comparing our derived log gf values to those given in the NIST999The NIST Atomic Line Database can be accessed at: http://www.nist.gov/physlab/data/asd.cfm. (Ralchenko et al. 2008), Thevenin (1990), and VALD101010The VALD linelist can be accessed at: http://www.astro.uu.se/ vald/php/vald.php. (Kupka et al. 2000) compilations, we find very good agreement such that \langlelog gf{}_{\rm ours}–log gf{}_{\rm lit.}\rangle=–0.02 (\sigma=0.08).

While most abundances were determined through a straight–forward EW analysis, the odd–Z Fe–peak and neutron–capture elements have line profiles that may be affected by hyperfine structure. For the purposes of this study, this includes the elements Sc, La, and Eu. Prochaska & McWilliam (2000) give hyperfine log gf values for the 6245 Å Sc II line, but unfortunately their work does not include the 6309 Å Sc II nor the 6210 and 6305 Sc I lines used here. Similarly, the Zhang et al. (2008) analysis of solar Sc abundances only includes the 6245 Å line as well. However, the error introduced by ignoring hyperfine structure increases as a function of EW, and the small Sc EWs in our sample (\langleEW\rangle=42 mÅ, \sigma=26 mÅ) lead us to believe a standard EW abundance analysis is a reasonable approach for this element.

For abundance determinations of the neutron–capture elements La and Eu, we have employed the hyperfine structure linelists available in Lawler et al. (2001a) for the 6262 Å La II line and Lawler et al. (2001b) for the 6645 Å Eu II line. However, the La abundances were determined by spectrum synthesis and are described in \lx@sectionsign3.2.2. Eu abundance determinations are more complicated than those for most elements because the line profiles are both affected by hyperfine splitting and Eu has two stable, naturally occurring isotopes ({}^{\rm 151}Eu and {}^{\rm 153}Eu) with solar system isotopic fractions of 47.8\% and 52.2\%, respectively (Lawler et al. 2001b). The Eu EWs were measured using the same interactive fitting code mentioned previously, and the EWs were combined with the Eu linelist and isotope fractions as inputs into the MOOG blends driver to obtain the final abundances.

All EWs measured for this project and the atomic linelists are provided in Tables 3a–3b for Fe and Tables 4a–4b for all other elements. Similarly, the chemical abundance ratios for all elements are given in Table 5, and the number of lines measured for each element per star along with the \sigma/\surd(N) values are available in Tables 6a–6b. The log gf values listed for La and Eu in Table 4b represent the total gf values instead of an individual hyperfine component. The interested reader can find the full linelists for these elements in the references given above.

3.2.2 Spectrum Synthesis Analysis

The abundances of O, Al, and La were determined by spectrum synthesis rather than the EW fitting method described in \lx@sectionsign3.2.1. The primary motivation for using synthesis instead of an EW analysis for these elements is that the available lines suffer from varying degrees of contamination with either nearby metal lines or molecular CN. The 6300.31 Å [O I] line is blended with the nearby 6300.70 Sc II feature, and is also moderately sensitive to the C+N abundance. Similarly, the 6696/6698 Å Al I lines are both moderately blended with nearby Fe I and CN features, and the 6262 Å La II line is lightly blended with both CN and a Co I line at 6262.81 Å.

Spectrum synthesis modeling was carried out using the synth driver in MOOG. The atomic linelist was generated primarily from the Kurucz online database111111The online database can be found at: http://kurucz.harvard.edu/LINELISTS/GF100/ with updated log gf values provided by C. Sneden (2008, private communication). The atomic linelist was merged with a molecular CN linelist that was created through a combination of the Kurucz molecular linelist121212The molecular linelist can be found at: http://kurucz.harvard.edu/LINELISTS/LINESMOL/ and one provided by B. Plez (2007, private communication; see also Hill et al. 2002). Individual log gf values for lines of interest were verified through spectrum synthesis of both the solar and Arcturus atlases. As was mentioned in \lx@sectionsign3.2.1, the La II hyperfine structure linelist was taken from Lawler et al. (2001a).

Since most stars in our sample do not have published [C/Fe], [N/Fe], and/or {}^{\rm 12}C/{}^{\rm 13}C ratios, we set [C/Fe]=–0.5, {}^{\rm 12}C/{}^{\rm 13}C=4, and treated the N abundance as a free parameter to fit the available CN features. Previous work on evolved RGB stars in \omega Cen (e.g., Norris & Da Costa 1995; Smith et al. 2002; Origlia et al. 2003; Stanford et al. 2010) has shown that our set values for [C/Fe] and {}^{\rm 12}C/{}^{\rm 13}C are a reasonable approximation given that all of the stars in our sample will have already undergone first dredge–up and are above the RGB luminosity bump. With these assumptions, values of +0.80\lesssim[N/Fe]\lesssim+1.50 tended to provide the best fits to the CN lines.

Figure 4 shows sample spectra of four moderately metal–poor ([Fe/H]\approx–1.45) program stars along with synthetic spectrum fits to the O, La, and Al regions. The bottom panels of Figure 4 indicate the uncertainty introduced when the abundances of CN and other nearby, blended metals are altered by \pm0.50 dex. In warmer stars and those that are moderately metal–poor, the CN contamination does not provide a significant change in the derived abundance. However, cooler and more metal–rich stars have O, La, and Al abundances that can deviate by at least 0.10–0.20 dex compared to an analysis that does not properly account for molecular blends. For the Al lines, the nearby Fe lines are generally not much of an issue in cool giants because the Fe transitions have excitation potentials \gtrsim5 eV. The O and La lines are also not significantly affected by blends from neighboring Fe–peak element features unless the [Fe–peak/Fe] abundance exceeds roughly +0.3 dex. However, the O and Sc lines are blended strongly enough at this resolution to warrant spectrum synthesis regardless of the [Sc/Fe] abundance.

3.2.3 Comparison to Other Studies

As described in \lx@sectionsign1, the chemical composition of \omega Cen has been extensively studied using a variety of abundance indicators. However, there are only four high resolution spectroscopic studies for which we have more than five stars in common: Norris & Da Costa (1995; 35 stars), Smith et al. (2000; 7 stars), Johnson et al. (2008; 171 stars), and Johnson et al. (2009; 59 stars). Figure 5 illustrates the differences between our adopted model atmosphere parameters and those found in the literature. The average differences in T{}_{\rm eff}, log g, [Fe/H], and v{}_{\rm t}, in the sense present minus literature values, are 0 K (\sigma=61 K), –0.02 cgs (\sigma=0.09), –0.03 dex (\sigma=0.17 dex), and 0.02 km s{}^{\rm-1} (\sigma=0.24 km s{}^{\rm-1}), respectively. We conclude from these results that there are no strong systematic offsets among the studies with regard to the adopted model atmosphere parameters. This conclusion is in agreement with the [X/Fe] abundances comparisons shown in Figure 6. The average differences in the chemical abundances between this study and those in the literature tend to be <0.10 dex (\sigma\lesssim0.20 dex).

The paucity of Al and Eu comparisons shown in Figure 6 is due to two effects: (1) we only obtained about \sim40\% as many spectra in the spectral region that contains the Al and Eu lines and (2) we purposely chose to observe stars in the Al/Eu region for which Al and/or Eu abundances were not already available in the literature. Further examination of Figure 6 indicates that La is the only element showing a systematic abundance offset. We tend to find systematically lower [La/Fe] ratios, especially at [Fe/H]\gtrsim–1.7, because of our inclusion of hyperfine structure for the 6262 Å La II line. Norris & Da Costa (1995), Smith et al. (2000), and Johnson et al. (2009) base all or part of their La abundances on the 6774 Å La II line, which suffers from hyperfine broadening. However, there are no hyperfine linelists available in the literature for this transition. Since our present data, combined with that from Johnson et al. (2009), include both the 6262 and 6774 Å lines, we have derived an empirical hyperfine structure correction factor for the 6774 Å line that is described in Appendix A.

In addition to the studies mentioned above, we also have five stars in common (ROA 211, 300, 371, WFI 618854, and WFI 222068) with the Pancino et al. (2002) work that measured [Fe/H], [Si/Fe], [Ca/Fe], and [Cu/Fe] in six relatively metal–rich ([Fe/H]\geq–1.2) \omega Cen giants. However, despite sharing small differences in our derived T{}_{\rm eff}, log g, [Fe/H], and v{}_{\rm t} values, we find noticeably different [\alpha/Fe] abundances for two of the most metal–rich stars (ROA 300 and WFI 222068). This is important because the Pancino et al. (2002) result is one of the primary studies suggesting that Type Ia SNe may have significantly affected \omega Cen’s chemical enrichment. Origlia et al. (2003) also find a decrease in [\alpha/Fe] at [Fe/H]>–1, and we note similar discrepancies in our derived abundances for the most metal–rich stars. However, their abundances are based on low resolution, infrared spectra and may be subject to systematic offsets with our data.

For a direct comparison with the Pancino et al. (2002) data, in star ROA 300 we find Si and Ca offsets of \Delta[Si/Fe]=+1.14 and \Delta[Ca/Fe]=+0.17. Similarly, WFI 222068 exhibits differences of \Delta[Si/Fe]=+0.60 and \Delta[Ca/Fe]=+0.42. To investigate this discrepancy, we ran spectrum syntheses for the 6155 Å Si I line and 6156, 6161, and 6162 Å Ca I lines (see Figure 7). The results shown in Figure 7 indicate that the Pancino et al. (2002) [Si/Fe] and [Ca/Fe] abundances are too low to match the observed spectra using our linelist and model atmospheres. Instead, we find better agreement by using the upper limits on the error bars given by Pancino et al. (2002; their table 3), which results in increasing their [Si/Fe] and [Ca/Fe] abundances by \sim+0.2 dex.

Further inspection of Figure 7 shows that our EW–based [Si/Fe] abundances may have overestimated the true [Si/Fe] abundances for these two stars by \sim+0.3 dex. We did not find any clear reason for this discrepancy because the EW–based abundances for calcium and all other elements were in agreement with the spectrum synthesis fits, but it is possible that an unaccounted for (probably CN) blend is present near the silicon line in these very cool (T{}_{\rm eff}<4000 K), relatively metal–rich ([Fe/H]\sim–0.7) giants. It is worth noting that the [Si/Fe] and [Ca/Fe] abundance values are in much better agreement for two of the warmer, more metal–poor stars where the differences between EW– and synthesis–based abundances are negligible. In the remaining star (ROA 371), the difference between our derived [Ca/Fe] abundance and that from Pancino et al. (2002) is mostly negligible, but the [Si/Fe] abundance offset is noticeably larger at \Delta[Si/Fe]=+0.48. However, this star was also analyzed by both Paltoglou & Norris (1989) and Norris & Da Costa (1995), and we find in agreement with those two studies that ROA 371 is Si–rich with [Si/Fe]\gtrsim+0.5. It seems likely that most, if not all, of the discrepancy between our derived abundance values and Pancino et al. (2002) are the result of differences in adopted log gf values, model atmospheres, and line choice.

3.3 Abundance Sensitivity to Model Atmosphere Parameters

Table 7 shows the sensitivity of our derived log \epsilon(X) abundances to changes in the adopted model atmosphere parameters. The tests were conducted at T{}_{\rm eff}=4200 K and T{}_{\rm eff}=4600 K, values typical of stars in our sample, and metallicities ranging from [Fe/H]=–2.0 to –0.50. The analyses for each test star were run by adjusting T{}_{\rm eff}\pm100 K, log g\pm0.30 cgs, [Fe/H]\pm0.30 dex, and v{}_{\rm t}\pm0.30 km s{}^{\rm-1} individually while holding the other parameters constant.

We find that the chemical abundances derived from subordinate ionization state transitions (e.g., most neutral metals) are most sensitive to changes in T{}_{\rm eff}. However, abundances derived from dominant ionization state transitions (e.g., neutral oxygen; singly ionized transition metals and heavy elements) are more sensitive to uncertainties in surface gravity and metallicity because of their stronger dependence on electron pressure and H{}^{\rm-} opacity. For stars with [Fe/H]<–1, microturbulence was found to have a negligible effect on abundances derived from all transitions except Fe I and Ca I. Abundances derived from Fe I and Ca I lines were more sensitive to microturbulence uncertainties because of their typically larger EWs than other lines at a given metallicity. In stars with [Fe/H]>–1, most abundances were affected at the 0.05–0.10 dex level due to the increased line strengths. Similarly, the abundances of most elements in warmer stars were less sensitive to changes in microturbulence because of the generally weaker line strengths. It seems likely that our derived log \epsilon(X) abundance uncertainties do not exceed \sim0.20 dex based on our choices of model atmosphere parameters. Additionally, the [X/Fe] abundance ratios for most elements are expected to exhibit an even weaker dependence on model atmosphere parameter uncertainties because of their similar behavior to Fe I.

In addition to the parameters shown in Table 7, we also tested the abundance uncertainties based on changes to CN and He. Since CN lines are strongest in the O–poor stars, it is possible that standard EW and spectrum synthesis analyses may not give the same abundances for lines significantly blended with CN. However, we find that none of the lines chosen for this study that were analyzed via a standard EW approach were significantly affected by continuum suppression or blending from CN. The robust agreement between the synthesis and EW–based analyses for elements other than O, Al, and La is demonstrated in Figure 4, where the abundances of all other elements studied here were preset to those values obtained from a standard EW analysis.

Since the current interpretation of \omega Cen’s blue main sequence is that stars belonging to that population are He–rich (Y\sim0.38), we investigated the effects helium enrichment might have on our analyses. To test this, we ran both EW and spectrum synthesis analyses using He–normal (Y=0.27) and He–rich (Y=0.35) ATLAS9 models131313The He–rich models can be downloaded at http://wwwuser.oat.ts.astro.it/castelli/grids.html.. We find that the He–rich model does not result in a significantly different abundance (\Deltalog \epsilon(X)<0.1 dex), and the effects on our derived [X/Fe]141414Also note that the decrease in N(H) for He–rich stars will not affect abundances reported as [X/Fe] ratios because [X/H] and [Fe/H] both increase by the same amount. ratios are further mitigated for the low ionization potential metals. These results are in agreement with helium enrichment predictions by Boehm–Vitense (1979), and are consistent with similar tests on \omega Cen stars in Piotto et al. (2005), Johnson et al. (2009), and Cunha et al. (2010). Furthermore, Girardi et al. (2007) conclude that increasing the He abundance to the extreme values predicted in some \omega Cen stars should not significantly alter either the bolometric correction or V–K color–temperature relation. We therefore believe that our adopted atmospheric parameters are reliable even for He–rich giants.

4 RESULTS

4.1 Iron and the Metallicity Distribution Function

As discussed in \lx@sectionsign1, \omega Cen’s large metallicity spread has been previously verified in many photometric and spectroscopic analyses. However, the results presented here are based on direct measurements from high resolution, high S/N spectra in a nearly complete sample of \omega Cen giants with V\leq13.5. These new data cover the cluster’s full metallicity regime, and are also nearly complete out to \sim50\% of the tidal radius. The data presented here, along with that from Johnson et al. (2008; 2009), yield spectroscopic [Fe/H] measurements for 867 giants.

In Figure 8, we plot our derived metallicity distribution function and compare with the results of two other large spectroscopic surveys that spanned the upper RGB and SGB (Norris et al. 1996; Suntzeff & Kraft 1996). The general trend among all studies is that a dominant, metal–poor stellar population exists at [Fe/H]\approx–1.7 along with a higher metallicity tail terminating around [Fe/H]\approx–0.5. Our data confirm this result, and also support previous observations that found multiple peaks in the metallicity distribution function but a paucity of stars with [Fe/H]<–2. The full range of iron abundances in our sample extends from [Fe/H]=–2.26 to –0.32, and in Figure 8 we find five peaks in the metallicity distribution function located at [Fe/H]\approx–1.75, –1.50, –1.15, –1.05, and –0.75. These peaks correspond to the RGB–MP, RGB–MInt, and RGB–a populations identified by Pancino et al. (2000) and Sollima et al. (2005), and also generally agree with Strömgren photometry estimates (Hilker & Richtler 2000; Hughes & Wallerstein 2000; Calamida et al. 2009). It is difficult to accurately deblend the two populations near [Fe/H]=–1.15 and –1.05 because the separation is comparable to the line–to–line dispersion of Fe abundance measurements in individual stars. Instead, we will combine these two populations during further analyses. Taking this into account, the (now four) stellar populations make up roughly 61\%, 27\%, 10\%, and 2\% of our sample, respectively. For brevity, we will follow a similar naming scheme used by Sollima et al. (2005) when referring to the different metallicity populations: RGB–MP ([Fe/H]\leq–1.6), RGB–Int1 (–1.6<[Fe/H]\leq–1.3), RGB–Int2+3 (–1.3<[Fe/H]\leq–0.9), and RGB–a ([Fe/H]>–0.9).

The most metal–poor stars ([Fe/H]\leq–2) make up about 2\% (17/867) of the full sample and only about 3\% (17/541) of the RGB–MP stellar population. However, the RGB–a stars are slightly underrepresented because of our V magnitude cutoff. To test for any selection effects, we rebinned the data to only include stars within \sim1 mag of each giant branch’s RGB tip, which is the approximate magnitude range over which we sampled the RGB–a. We did not find any significant differences in the relative population mix, and different magnitude cutoffs only raised the RGB–a population fraction to \sim5\%. These estimates are consistent with those derived from number counts in photometric analyses (Pancino et al. 2000; Sollima et al. 2005; Villanova et al. 2007; Calamida et al. 2009). It should also be noted that AGB contamination may affect the number counts of each population differently. Lee et al. (2005a) found that if the intermediate metallicity and most metal–rich stars are in fact He–rich then these stars will populate the “extreme” horizontal branch. Furthermore, D’Cruz et al. (2000) estimate that as much as 30\% of the cluster’s horizontal branch population may reside on the “extreme” horizontal branch, and it is likely that these stars evolve directly to white dwarfs rather than first ascending the AGB (e.g., Sweigart et al. 1974). Since it is difficult to differentiate between RGB and AGB stars in \omega Cen’s color–magnitude diagram, it is possible that the number counts for the two most metal–poor populations contain a disproportionate number of AGB stars compared to the more metal–rich populations. However, our estimated population fractions are consistent with those found along the main sequence and subgiant branch (e.g., Villanova et al. 2007) where AGB contamination is not an issue.

In addition to the existence of multiple, discrete stellar populations in \omega Cen, there is some evidence that the metal–rich stars are more centrally located than the more metal–poor populations (Norris et al. 1996; Suntzeff & Kraft 1996; Pancino et al. 2000; Hilker & Richtler 2000; Pancino et al. 2003; Rey et al. 2004; Sollima et al. 2005; Johnson et al. 2008; Bellini et al. 2009b; Johnson et al. 2009). In Figure 9, we plot our derived abundances as a function of projected distance from the cluster center. A two–sided Kolmogorov–Smirnov (K–S) test (Press et al. 1992) confirms that the metal–rich stars ([Fe/H]>–1.3) are more centrally located than the metal–poor stars at the 96\% level151515We adopt the notion that the null hypothesis (i.e., that the two distributions are the same) can be rejected if the p value is “small” (<0.05).. Additionally, all of the stars with [Fe/H]\geq–0.9 are located within 13\arcmin of the cluster center, with most of those residing inside 10\arcmin.

Further inspection of Figure 9 reveals another interesting radial distribution trend; all stars with [Fe/H]\leq–2 are located within 12\arcmin of the cluster core, and 88\% (15/17) of these stars reside inside 5\arcmin. A two–sided K–S test comparing the radial distribution of stars with –2.0<[Fe/H]\leq–1.60 versus those with [Fe/H]\leq–2 indicates that the two distributions are drawn from different parent populations at the 99\% level. Additionally, the star–to–star metallicity dispersion decreases with increasing distance from the cluster center, but this is mostly driven by the metallicity gradient and paucity of stars with [Fe/H]\geq–1.3 outside \sim15\arcmin from the cluster center. If one only considers stars with [Fe/H]<–1.3, the standard deviation in [Fe/H] between 0–10\arcmin and 10–20\arcmin differs by less than 0.02 dex. This indicates that the two most metal–poor stellar populations are well mixed inside the cluster.

4.2 Oxygen

The chemical evolution of oxygen in \omega Cen has previously been analyzed via high resolution spectroscopy in several studies containing sample sizes ranging from \sim5–40 RGB stars (e.g., Cohen 1981; Paltoglou & Norris 1989; Brown & Wallerstein 1993; Norris & Da Costa 1995; Zucker et al. 1996; Smith et al. 2000), and more recently in a sample of \sim200 RGB stars (Marino et al. 2010). The main results from past studies indicate that: (1) \omega Cen giants exhibit large star–to–star dispersions in [O/Fe] abundance, (2) many of the intermediate metallicity stars have [O/Fe]<0, (3) the majority of metal–poor stars are O–rich with [O/Fe]\sim+0.3, and (4) oxygen is anticorrelated with both sodium and aluminum. The results presented here add 848 new [O/Fe] abundance measurements.

Figure 9 includes a plot of our derived [O/Fe] abundances as a function of projected distance from the cluster center. Compared to the other elements in Figure 9, oxygen appears to exhibit a unique radial distribution. Stars with [O/Fe]\leq0, and especially those with [O/Fe]<–0.4, are more centrally concentrated than the bulk of stars with [O/Fe]>0. In our sample, 62\% (145/233) of stars with [O/Fe]\leq0 are located inside 5\arcmin from the core and 91\% (213/233) are inside 10\arcmin. This is compared to just 42\% (261/615) and 77\% (472/615) for the stars with [O/Fe]>0, respectively. A two–sided K–S test reveals that the O–poor stars exhibit a different spatial distribution than the O–rich stars at the 99\% level. This result may have important implications regarding the origin of the blue main–sequence, and will be discussed further in \lx@sectionsign5.2.2.

Figure 10 shows the chemical evolution of [O/Fe] plotted as a function of [Fe/H]. This plot reveals that the O–poor stars, in addition to being preferentially located near the cluster core, are also well separated from the O–rich stars over a large metallicity range. Figure 11 shows the [O/Fe] data binned in 0.10 dex increments and separated into the population subclasses defined in \lx@sectionsign4.1. The resultant histograms support the existence of two subpopulations, one O–rich ([O/Fe]>0) and the other O–poor ([O/Fe]<0), residing inside the RGB–Int1 and RGB–Int2+3 populations. Interestingly, neither the RGB–MP nor the RGB–a populations appear to exhibit this bimodal behavior. Instead, the RGB–MP stars are predominantly O–rich with a median [O/Fe]=+0.32, and the RGB–a stars are moderately O–poor with a median [O/Fe]=–0.15. In the RGB–MP population, the percentages of O–poor and O–rich stars are 13\% (71/535) and 87\% (464/535), respectively. The two intermediate metallicity populations show quite different distributions, with the percentages being 46\% (100/218) to 54\% (118/218) in the RGB–Int1 group and 64\% (47/74) to 36\% (27/74) in the RGB–Int2+3 group. The relative distribution in the RGB–a stars is 71\% (15/21) O–poor to 29\% (6/21) O–rich, respectively.

Examining the bulk properties of the [O/Fe] abundances reveals that, in all but the most metal–poor and metal–rich stars, a significant star–to–star dispersion is present with \Delta[O/Fe]>2 over a large metallicity range. The full range of [O/Fe] abundances found in our sample spans from [O/Fe]=–1.30 to +0.80. The stars with [Fe/H]\leq–2 are overwhelmingly O–rich with 94\% (15/16) having [O/Fe]>0 and \langle[O/Fe]\rangle=+0.38, and the single O–poor star is only moderately depleted at [O/Fe]=–0.13. The [O/Fe] abundance “ceiling” decreases for stars with [Fe/H]\gtrsim–1.3, dropping from [O/Fe]\approx+0.6 at [Fe/H]=–1.3 to [O/Fe]\approx+0.0 at [Fe/H]=–0.3. Additionally, the super O–poor stars ([O/Fe]\leq–0.4) are only found in the range –1.9\lesssim[Fe/H]\lesssim–1.0. When considering all stars in our sample, the relative percentages of O–rich ([O/Fe]>0), O–poor (–0.4<[O/Fe]\leq0.0), and super O–poor ([O/Fe]\leq–0.4) stars are 73\% (615/848), 14\% (118/848), and 13\% (115/848), respectively. We also find that in stars with [Fe/H]\lesssim–1, [O/Fe] is anticorrelated with both [Na/Fe] and [Al/Fe]. The implications of these anticorrelations, along with the possible significance of the super O–poor stars, will be discussed further in \lx@sectionsign5.

4.3 Sodium

Previous sodium abundance measurements support the idea that \omega Cen experienced a significantly different chemical evolutionary path than any other stellar system (Cohen 1981; Paltoglou & Norris 1989; Brown & Wallerstein 1993; Norris & Da Costa 1995; Zucker et al. 1996; Smith et al. 2000; Johnson et al. 2009; Villanova et al. 2009; Marino et al. 2010). The results from these studies have shown that: (1) [Na/Fe] appears to increase as a function of increasing [Fe/H], (2) \Delta[Na/Fe]>1 for most values of [Fe/H] in the cluster, (3) no strong [Na/Fe] abundance gradient is observed, and (4) [Na/Fe] is correlated with [Al/Fe] and anticorrelated with [O/Fe]. Our new results, combined with those from Johnson et al. (2009), give [Na/Fe] abundances for 848 cluster giants. Although it is likely that our derived sodium abundances suffer from moderate non–LTE (NLTE) effects, abundances derived from the 6154/6160 Å doublet used here are expected to have NLTE offsets <0.2 dex for giants in our metallicity regime (e.g., Gratton et al. 1999; Mashonkina et al. 2000; Gehren et al. 2004). Since no standard NLTE corrections are available in the literature, the abundances reported in Table 5 and shown in the figures do not include NLTE corrections.

Inspection of Figure 11 indicates that [Na/Fe] exhibits a similar bimodal abundance pattern as shown by [O/Fe]. That is, the most metal–poor and metal–rich stellar populations show a single primary peak in the [Na/Fe] distribution function, and the two intermediate metallicity populations may be best described as having two peaks in the [Na/Fe] distribution function. However, unlike the case with oxygen, we do not find an obvious centrally concentrated population that correlates with any [Na/Fe] abundance range. We do find that the most Na–rich stars in our sample ([Na/Fe]\geq+0.6) are all found inside 13\arcmin from the cluster center, but this observation is unlikely to be significant because (1) the two most metal–rich stellar populations contain 69\% (44/64) of the most Na–rich stars (see Figures 1011) and (2) these stellar populations are already known to be centrally concentrated. This is in contrast to the O–poor radial trend that is found in stars with –1.6<[Fe/H]\leq–1.3, which do not exhibit a preferred radial location. However, Figure 9 shows that a weak, declining [Na/Fe] gradient may exist such that the median [Na/Fe] values for 0–5\arcmin, 5–10\arcmin, 10–15\arcmin, and 15–20\arcmin are +0.22, +0.14, +0.08, and –0.03, respectively.

Figure 10 highlights the chemical evolution of [Na/Fe] as a function of [Fe/H]. We find that a large star–to–star dispersion is present at all metallicities, and that the full range extends from [Na/Fe]=–1.02 to +1.36. Since we have many stars of the same temperature, surface gravity, and metallicity, the line strength differences confirm that the observed abundance spread is a real effect and not due to possible underlying NLTE effects. In addition to displaying a significant star–to–star dispersion, the sodium abundances also exhibit a strong metallicity dependence such that the median [Na/Fe] value increases with increasing [Fe/H]. The median [Na/Fe] value rises from +0.08 in the RGB–MP population to +0.78 in the RGB–a population. As mentioned above, the two most metal–rich populations contain the most Na–rich stars in the cluster. Despite the complex nature of sodium’s evolution in \omega Cen, the O–Na anticorrelation and Na–Al correlation are present in all but the most metal–rich stars.

4.4 Aluminum

Except for iron and calcium, aluminum has been the most highly studied element in \omega Cen. Previous high resolution spectroscopic work has targeted more than 200 RGB stars (Cohen 1981; Brown & Wallerstein 1993; Norris & Da Costa 1995; Zucker et al. 1996; Smith et al. 2000; Johnson et al. 2008; Johnson et al. 2009) and shown: (1) \Delta[Al/Fe]>0.5 at all metallicities and exceeds more than a factor of ten in the most metal–poor stars, (2) the range of observed [Al/Fe] abundances decreases at [Fe/H]>–1.3, (3) there is a paucity of stars with [Al/Fe]<+0.3 at intermediate and high metallicities, and (4) a Na–Al correlation and O–Al anticorrelation are present in most, if not all, cluster stars. In this paper we present 133 new [Al/Fe] abundance measurements, and when combined with the data from Johnson et al. (2008; 2009) provide [Al/Fe] values for 332 \omega Cen giants. As with sodium (see \lx@sectionsign4.3), we have not applied any NLTE corrections to our derived aluminum abundances. However, all aluminum abundances determined here utilized the non–resonance 6696/6698 Å lines, which are not expected to have large NLTE offsets in the temperature, gravity, and metallicity range of stars in our sample (e.g., Gehren et al. 2004; Andrievsky et al. 2008).

Unlike oxygen, and to a lesser extent sodium, aluminum does not show any obvious correlation between [Al/Fe] abundance and radial location. However, aluminum does show the same bimodal abundance distribution for the RGB–Int1 and RGB–Int2+3 populations (see Figure 11). By dividing the samples at [Al/Fe]=+0.6, we find that the percentage of “Al–enhanced” ([Al/Fe]\geq+0.6) stars in the RGB–Int1 population is 50\% (50/100) compared to 50\% (50/100) as well for the “Al–normal” stars ([Al/Fe]<+0.6). Similarly, the RGB–Int2+3 stars are distributed as 69\% (31/45) enhanced and 31\% (14/45) normal, respectively. Interestingly, the [Al/Fe] distribution also shows complex substructure in the RGB–MP population, which is not observed in the [O/Fe] and [Na/Fe] data. In this population, 41\% (70/172) of the stars are Al–enhanced and 59\% (102/172) are Al–normal. Furthermore, this is the only \omega Cen population that contains a significant number of stars over the full [Al/Fe] range. For the RGB–a population, a single peak is observed at [Al/Fe]=+0.5 in the [Al/Fe] distribution function.

The full range of [Al/Fe] abundances observed here spans from –0.34 to +1.37, but only 4\% (12/332) of the stars have [Al/Fe]<0. Similarly, we find that \Delta[Al/Fe]\sim1.5 dex for [Fe/H]<–1.3. However, the star–to–star dispersion decreases noticeably at higher metallicities. Inspection of Figure 10 shows that [Al/Fe] exhibits an interesting trend as a function of [Fe/H]. The maximum value reached for stars with [Fe/H]\lesssim–1.3 remains steady near [Al/Fe]\approx+1.3, but above [Fe/H]\sim–1.3 the maximum abundance decreases to only [Al/Fe]\approx+0.6 in the RGB–a stars. Furthermore, the number of stars with [Al/Fe]<+0.3 strongly decreases at [Fe/H]>–1.3. In the RGB–MP and RGB-Int1 populations, stars with [Al/Fe]<+0.3 constitute 25\% (69/272) of the distribution, but this decreases to only 7\% (1/15) of the RGB–a population.

4.5 Silicon

Previous analyses (Cohen 1981; Paltoglou & Norris 1989; Brown & Wallerstein 1993; Norris & Da Costa 1995; Smith et al. 2000; Pancino et al. 2002; Villanova et al. 2009) have used the heavy \alpha element (Si, Ca, and Ti) abundances to assess the dominance of Type II versus Type Ia supernovae in \omega Cen and other clusters. In terms of silicon abundances, it has been shown that: (1) silicon is enhanced with [Si/Fe]>+0.3 in nearly all cluster stars, (2) the star–to–star dispersion in [Si/Fe] is significantly smaller than for the lighter \alpha and odd–Z elements, and (3) the most metal–rich stars may have appreciably lower [Si/Fe] abundances compared to the more metal–poor populations. From this study, we add 821 new [Si/Fe] measurements over \omega Cen’s full metallicity range.

While we find that the lighter \alpha element oxygen shows a distinctly unique distribution versus distance from the cluster center, [Si/Fe] does not show the same trend. Figure 9 suggests that a weak [Si/Fe] gradient may be present such that the stars inside 5\arcmin have a higher average silicon abundance than those outside 5\arcmin. We find that stars inside 5\arcmin have \langle[Si/Fe]\rangle=+0.37, which is noticeably higher than the \langle[Si/Fe]\rangle=+0.29 for those at r>5\arcmin. This result does not change even if we limit examination to stars only between 0–5\arcmin and 5–10\arcmin. Except near the cluster core, the average [Si/Fe]\approx+0.3 at all radii. It should be noted that Villanova et al. (2009) find \langle[Si/Fe]\rangle=+0.5 in the outer 20–30\arcmin of \omega Cen, which is larger by about 0.2 dex than we find in the same region. However, we do not presently have sufficient data to assess whether the average [Si/Fe] ratio increases at larger radii or if this merely reflects a systematic offset.

The full range of [Si/Fe] abundances in our data span from –0.30 to +1.15, but the average over all stars is [Si/Fe]=+0.33 (\sigma=0.17). While we do find a few Si–poor stars ([Si/Fe]<0), these stars comprise only 2\% (16/821) of the total sample. Similarly, the very Si–rich stars ([Si/Fe]>+0.6) only represent 6\% (52/821) of the total sample. Figure 10 reveals that [Si/Fe] may have a more complex morphology as a function of [Fe/H] than previously thought. The average [Si/Fe] ratio decreases from \langle[Si/Fe]\rangle=+0.46 (\sigma=0.19) in stars with [Fe/H]\leq–2 to \langle[Si/Fe]\rangle=+0.29 (\sigma=0.16) in the stars that comprise the majority of the RGB–MP population (–2.0<[Fe/H]\leq–1.6). In the subsequent populations, the average [Si/Fe] abundance monotonically increases with [Fe/H] to \langle[Si/Fe]\rangle=+0.45 (\sigma=0.23) in the RGB–a population. This is in agreement with Norris & Da Costa (1995) and Smith et al. (2000), but contrasts with the claims by Pancino et al. (2002) and Origlia et al. (2003) that stars with [Fe/H]>–1 have lower [\alpha/Fe] abundances (see \lx@sectionsign3.2.3 for a brief discussion).

4.6 Calcium

In addition to iron, calcium abundances have been analyzed in great detail for \omega Cen stars. Previous analyses have used calcium as a proxy metallicity indicator (Freeman & Rodgers 1975; Cohen 1981; Norris et al. 1996; Suntzeff & Kraft 1996; Rey et al. 2004; Sollima et al. 2005; Stanford et al. 2006; Lee et al. 2009) and as an \alpha element tracer (Paltoglou & Norris 1989; Norris & Da Costa 1995; Smith et al. 2000; Pancino et al. 2002; Kayser et al. 2006; Villanova et al. 2007; Johnson et al. 2009; Villanova et al. 2009). These studies have shown: (1) there is a large spread of at least 1 dex in [Ca/H] with multiple peaks in the distribution function (i.e., confirms the different populations found when using [Fe/H] as a metallicity tracer), (2) nearly all stars have enhanced [Ca/Fe]\approx+0.3 at all metallicities, (3) the star–to–star dispersion is significantly smaller than for the lighter elements, and (4) there may be a downturn in [Ca/Fe] at [Fe/H]>–1. Combining our new data with that of Johnson et al. (2009), we add 857 [Ca/Fe] abundance measurements.

Unlike silicon, which provides some evidence for a weak radial abundance gradient, [Ca/Fe] does not vary ostensibly between the inner and outer regions of the cluster. When considering all stars in our sample, the majority are Ca–rich with \langle[Ca/Fe]\rangle=+0.29 (\sigma=0.12). However, the full range of observed [Ca/Fe] abundances is smaller than for [Si/Fe], with [Ca/Fe] varying between –0.13 and +0.65. Figure 10 shows that [Ca/Fe] displays a similar morphology to [Si/Fe] when plotted as a function of [Fe/H]. That is, stars with [Fe/H]\leq–2 tend to be more Ca–rich with \langle[Ca/Fe]\rangle=+0.37 (\sigma=0.16) compared to the majority of stars in the RGB–MP population with \langle[Ca/Fe]\rangle=+0.26 (\sigma=0.11). Similarly, the average [Ca/Fe] abundance rises for the RGB–Int1 and RGB–Int2+3 populations to \langle[Ca/Fe]\rangle=+0.34 (\sigma=0.11; see also Figure 12). However, unlike the case for [Si/Fe], the average [Ca/Fe] abundance decreases for [Fe/H]\gtrsim–1, and the RGB–a stars have \langle[Ca/Fe]\rangle=+0.26 (\sigma=0.12).

Further inspection of Figure 10 reveals that the distribution of [Ca/Fe] among the RGB–Int2+3 stars may be bimodal. Figure 12 also suggests that the RGB–Int2+3 stars may exhibit a bimodal distribution, and shows that the other populations appear to exhibit a mostly unimodal [Ca/Fe] distribution. Interestingly, the two RGB–Int2+3 subsets occur in nearly equal proportions with the stars peaked near [Ca/Fe]=+0.45 constituting 47\% (36/76) of the subsample and the stars peaked near [Ca/Fe]=+0.25 making up 53\% (40/76) of the subsample. However, a two–sided K–S test does not rule out that the [Ca/Fe] distributions for the RGB–Int1 and RGB–Int2+3 are different at more than the 95\% level. While we caution the reader that the apparent bimodality may be a product of small number statistics, it would be interesting to investigate this possible trend further with additional calcium abundance indicators (e.g., HK index).

4.7 Scandium

Scandium is typically used as a tracer of Fe–peak element production in stellar populations, and Galactic halo and globular cluster stars with [Fe/H]>–2.5 tend to exhibit solar–scaled [Sc/Fe] abundances. Although scandium has been analyzed in only a handful of studies for \omega Cen stars (Cohen 1981; Paltoglou & Norris 1989; Norris & Da Costa 1995; Zucker et al. 1996; Smith et al. 2000; Johnson et al. 2009), the results typically show that: (1) the observed star–to–star scatter in [Sc/Fe] is significantly smaller than for lighter elements and (2) \langle[Sc/Fe]\rangle\approx0 at all metallicities. Combined with the results from Johnson et al. (2009), we are able to add 821 [Sc/Fe] abundance measurements.

As can be seen in Figure 9, we do not find any evidence for a radial [Sc/Fe] abundance gradient. Similarly, Figure 10 indicates that the [Sc/Fe] ratio is approximately constant over the full metallicity regime. However, a weak metallicity dependence may be present such that the average [Sc/Fe] abundance decreases from \langle[Sc/Fe]\rangle=+0.08 (\sigma=0.13) in the RGB–MP population to \langle[Sc/Fe]\rangle=–0.07 (\sigma=0.19) in the RGB–a stars. The full range of observed [Sc/Fe] abundances spans from –0.49 to +0.44, but most stars exhibit a solar–scaled [Sc/Fe] ratio. When considering the entire sample, we find \langle[Sc/Fe]\rangle=+0.05 (\sigma=0.15).

4.8 Titanium

Titanium is generally considered either the heaviest \alpha element or one of the lightest Fe–peak elements. Previous titanium abundance measurements for \omega Cen stars (Cohen 1981; Paltoglou & Norris 1989; Brown & Wallerstein 1993; Norris & Da Costa 1995; Smith et al. 2000; Villanova et al. 2007; Johnson et al. 2009; Villanova et al. 2009) have shown: (1) the star–to–star dispersion in [Ti/Fe] is comparable to that found in [Si/Fe] and [Ca/Fe], (2) the titanium abundance is generally enhanced at [Ti/Fe]\sim+0.3, and (3) there may be evidence for an increase in [Ti/Fe] with increasing [Fe/H]. Our new results, combined with Johnson et al. (2009), provide 826 [Ti/Fe] measurements.

Inspection of Figure 9 confirms that we do not find any correlation between our determined [Ti/Fe] abundance and a star’s radial location. In a similar fashion to the behavior of silicon and calcium, Figure 10 shows that titanium also exhibits a metallicity dependent morphology. The average [Ti/Fe] ratio is roughly constant across the RGB–MP population’s full metallicity range ([Fe/H]\leq–1.6) at \langle[Ti/Fe]\rangle=+0.13 (\sigma=0.12), which is \sim0.2 dex lower than the [Si/Fe] and [Ca/Fe] ratios in those same stars. However, the average [Ti/Fe] abundance rises monotonically to \langle[Ti/Fe]\rangle=+0.34 (\sigma=0.25) in the RGB–a population (see also Figure 12). The full range of abundances in our sample spans from [Ti/Fe]=–0.42 to +0.85, but most stars are at least moderately Ti–enhanced with \langle[Ti/Fe]\rangle=+0.18 (\sigma=0.16).

4.9 Nickel

Aside from iron, nickel is the only other “true” Fe–peak element analyzed here. The chemical evolution of nickel in a stellar population often tracks very closely to iron, and \omega Cen appears to follow that trend (Cohen 1981; Paltoglou& Norris 1989; Norris & Da Costa 1995; Smith et al. 2000; Johnson et al. 2009; Villanova et al. 2009). Previous studies agree that: (1) the derived [Ni/Fe] abundances show the smallest intrinsic dispersion of any element and (2) the average [Ni/Fe] abundance is nearly solar at all metallicities and locations in the cluster. We add to these results 806 new [Ni/Fe] abundance determinations.

Figure 9 shows that, like the other transition metals, [Ni/Fe] abundances do not exhibit any signs of a radial gradient. Similarly, Figure 10 indicates that the distribution of [Ni/Fe] is essentially constant as a function of [Fe/H] with a small intrinsic scatter, but there may be a slight decrease in [Ni/Fe] at [Fe/H]\gtrsim–1.3. The full spread of [Ni/Fe] values found in our sample ranges from –0.48 to +0.69, and the cluster as a whole gives \langle[Ni/Fe]\rangle=–0.03 (\sigma=0.12).

4.10 Lanthanum

The heavy element lanthanum is often used as a tracer of the slow neutron–capture process (s–process), and its evolution has proved to be particularly interesting in \omega Cen. Previous analyses (Cohen 1981; Paltoglou & Norris 1989; Norris & Da Costa 1995; Smith et al. 2000; Johnson et al. 2009; Marino et al. 2010) have examined the [La/Fe] ratios in \sim100 RGB stars and found: (1) the most metal–poor stars tend to have [La/Fe] abundances consistent with those found in monometallic globular clusters, (2) a large increase in [La/Fe] is seen between [Fe/H]\approx–1.7 and –1.4, (3) the intermediate metallicity stars are almost exclusively La–rich, and (4) the average [La/Fe] ratio remains super–solar in the most metal–rich stars. When combined with the data from Johnson et al. 2009, we add to these past results 810 new [La/Fe] abundances.

As can be seen in Figure 9, we find no evidence supporting the existence of a radial [La/Fe] gradient, and the star–to–star dispersion remains approximately constant across all radii sampled here. On the other hand, our data shown in Figure 10 support previous claims that [La/Fe] abundances exhibit an unusual morphology when plotted as a function of [Fe/H]. There is a strong increase in [La/Fe] for stars with [Fe/H]\gtrsim–1.7, and a large intrinsic scatter of \Delta[La/Fe]\geq1 is present an nearly all metallicities. Furthermore, the average [La/Fe] abundance monotonically increases from +0.05 in the RGB–MP population to +0.49 in the RGB–Int2+3 population (see also Figure 12). However, the RGB–a stars have \langle[La/Fe]\rangle=+0.43, which suggests either a leveling off or slight decline in [La/Fe] at [Fe/H]\gtrsim–1.

The full range of [La/Fe] abundances observed here spans from –0.78 to +1.17, and it is worth noting that the proper accounting of hyperfine structure in the [La/Fe] derivations has decreased the maximum abundance values found in Johnson et al. (2009) from [La/Fe]\sim+2 to [La/Fe]\sim+1.2. These lower abundance ratios suggest that a large fraction of binary transfer systems may not be required to account for the significant lanthanum enhancements. However, we still find that only 29\% (232/810) of the stars in our sample have [La/Fe]<0, and 94\% (217/232) of those stars reside in the RGB–MP population. Interestingly, the stars with [Fe/H]\leq–2 tend to exhibit rather high [La/Fe] abundances. These stars have \langle[La/Fe]\rangle=+0.19, which is distinctly larger than the \langle[La/Fe]\rangle=+0.05 found for the full sample of RGB–MP stars. Unfortunately, a two–sided K–S test indicates that the data are insufficient to reject the null hypothesis with more than 94\% confidence.

4.11 Europium

In an analogous fashion to lanthanum, the heavy element europium is often used as an indicator of the rapid neutron–capture process (r–process). However, europium has been analyzed in far fewer stars than lanthanum (Norris & Da Costa 1995; Zucker et al. 1996; Smith et al. 2000; Johnson et al. 2009). The primary results from these studies are: (1) [Eu/Fe] tends to be somewhat underabundant relative to monometallic globular clusters of similar metallicity, (2) a significant intrinsic scatter is observed, but it is smaller than that found in [La/Fe], and (3) [Eu/Fe] remains relatively constant as a function of [Fe/H]. Combined with the data from Johnson et al. (2009), we provide [Eu/Fe] abundances for 194 stars.

Given the significantly smaller sample for europium compared to the other elements analyzed here, it is difficult to assess whether any true radial trends exist. Figure 9 provides weak evidence that the average [Eu/Fe] abundance may increase away from the cluster center. The available data support this by showing an increase from \langle[Eu/Fe]\rangle=+0.12 for stars between 0–5\arcmin to \langle[Eu/Fe]\rangle=+0.23 for stars between 5–10\arcmin from the core. Unfortunately, the sample size becomes too small outside \sim10\arcmin to conclude whether this trend continues.

Figure 10 reveals that [Eu/Fe] exhibits a significantly different behavior than [La/Fe] when plotted as a function of [Fe/H]. The full range is somewhat smaller with [Eu/Fe] spanning –0.46 to +0.83, and the average [Eu/Fe] abundance appears to decrease in the metallicity range where [La/Fe] shows its greatest increase. While the intermediate metallicity populations generally contain the lowest [Eu/Fe] abundances, the average [Eu/Fe] ratios differ by only \sim+0.1 dex among the different stellar populations.

5 DISCUSSION

The results of our analyses support previous observations that \omega Cen hosts multiple stellar populations exhibiting a complex history of chemical enrichment. To briefly summarize, we have confirmed five peaks in the metallicity distribution function located at [Fe/H]\approx–1.75, –1.50, –1.15, –1.05, and –0.75; however, for discussion purposes the [Fe/H]=–1.15 and –1.05 populations are treated as a single group. The RGB–MP, RGB–Int1, RGB–Int2+3, and RGB–a populations constitute 61\%, 27\%, 10\%, and 2\% of stars in our sample, respectively. We also find large intrinsic abundance dispersions for O, Na, and Al, and, except for perhaps in the most metal–rich stars, these elements exhibit the well–known abundance correlations and anticorrelations found in “normal” globular clusters. Additionally, the O–poor ([O/Fe]\leq0) stars are located almost exclusively within \sim5–10\arcmin of the cluster center, but the O–rich ([O/Fe]\sim+0.3) stars are rather evenly distributed at all cluster radii. The heavier \alpha elements Si, Ca, and Ti exhibit smaller star–to–star dispersions than the lighter elements and are generally enhanced by about a factor of two. The average [\alpha/Fe] ratio tends to increase with metallicity up to [Fe/H]\approx–1, and above this metallicity the average [Ca/Fe] ratio begins to decline while the average [Si/Fe] and [Ti/Fe] abundances remain roughly constant. The two Fe–peak elements scandium and nickel exhibit little star–to–star dispersion and their [X/Fe] ratios are nearly constant as a function of metallicity. We find a strong increase in the [La/Fe] abundances when comparing stars in the RGB–MP and RGB–Int1 populations, but the average [La/Fe] ratios for stars in the RGB–Int2+3 and RGB–a populations remain roughly the same. In contrast, [Eu/Fe] does not vary strongly with metallicity and is only modestly enhanced. We now aim to interpret what these results reveal about \omega Cen’s complex evolutionary history.

5.1 Supernova Nucleosynthesis: Evidence from Heavy \alpha and Fe–peak Elements

The standard theory of Galactic chemical evolution suggests that massive stars (\gtrsim10 M{}_{\sun}) produce the majority of elements up to the Fe–peak during various hydrostatic and/or explosive burning stages, and return the newly synthesized material to the interstellar medium (ISM) primarily through Type II SN explosions (e.g., Arnett & Thielemann 1985; Thielemann & Arnett 1985; Woosley & Weaver 1995; Nomoto et al. 2006). Theoretical yields indicate that stellar populations where Type II SNe have played the dominant role in polluting the ISM should produce future generations of stars with [\alpha/Fe] ratios that are about 0.3–0.5 dex larger than the solar–scaled value, and exhibit abundance ratios in the range –0.5\lesssim[X/Fe]\lesssim+0.3 for other elements lighter than about zinc. The massive stars provide chemical enrichment on time scales of \sim2\times10{}^{\rm 7} years or less, and are believed to be the dominant production sources of most elements in the Galactic halo and disk up to [Fe/H]\approx–1 (e.g., Timmes et al. 1995; Samland 1998). In contrast, Type Ia SNe primarily produce Fe–peak elements, and can contribute to a stellar population’s ISM about 5\times10{}^{\rm 8} to 3\times10{}^{\rm 9} years after the onset of star formation (e.g., Yoshii et al. 1996; Nomoto et al. 1997). Significant contributions from Type Ia SNe are believed to drive the observed decrease in the Galactic [\alpha/Fe] abundance trend at [Fe/H]>–1.

Figure 13 shows our measured [X/Fe] ratios as a function of [Fe/H], and overplots the expected abundance trends if (1) Type II SNe are responsible for all of \omega Cen’s chemical enrichment and (2) Type Ia ejecta are mixed with Type II ejecta in a 75/25\% ratio. For consistency, we show only the supernova yields from Nomoto et al. (1997; Type Ia) and Nomoto et al. (2006; Type II), but the theoretical yields from other groups (e.g., Woosley & Weaver 1995) follow approximately the same trends. We find that the \alpha and Fe–peak abundance distributions are generally well described by pollution from Type II SNe. However, Figures 10 and 13 indicate that the behavior of [Si/Fe], [Ca/Fe], and [Ti/Fe] as a function of increasing [Fe/H] is more complex than for [Sc/Fe] and [Ni/Fe]. For all three \alpha elements, the average [\alpha/Fe] abundance noticeably increases between the most metal–poor and intermediate metallicity populations. Additionally, the stars with [Fe/H]<–2 tend to exhibit larger [Si,Ca/Fe] ratios than the rest of the RGB–MP stars, but the [Ti/Fe] abundances are mostly uniform across the full RGB–MP metallicity range.

Some of this behavior may be at least qualitatively explained by examining the mass and/or metallicity dependent yields of massive stars. In Figure 14, we plot the predicted production factors from Woosley & Weaver (1995) for various elements as a function of progenitor mass. The increase in the average [Si/Fe] and [Ca/Fe] abundances for \omega Cen stars at [Fe/H]>–1.6 may be explained by the metallicity dependence of the Si and Ca yields, especially for stars more massive than about 18–20 M{}_{\sun}. As can be seen in Figure 14, the most massive stars are predicted to produce higher yields as the metallicity increases from [Fe/H]=–2 to –1, but the difference between the Si and Ca yields are expected to remain roughly constant. This means that as long as \omega Cen was able to retain and mix the ejecta of \gtrsim18 M{}_{\sun} stars, we should expect (1) that the average [Si/Fe] and [Ca/Fe] abundances should increase with [Fe/H] and (2) that both Si and Ca should exhibit the same general morphology until at least [Fe/H]\approx–1. Both of these predictions are seen in Figures 10 and 13. However, the similar increase found for [Ti/Fe] may not be due to Type II SNe. The theoretical yields do not predict a significant increase in [Ti/Fe] as a function of either progenitor mass or metallicity, and the situation does not improve if >25 M{}_{\sun} stars are included (e.g., McWilliam 1997). Instead, it seems likely that titanium has additional production sources. We should note that this all follows the assumption that the observed abundances trace {}^{\rm 48}Ti, in addition to {}^{\rm 28}Si and {}^{\rm 40}Ca, but an increase in the production of other stable isotopes could alter this scenario.

Mass dependent yields may also be responsible for explaining the discrepancy in [Si,Ca/Fe] between the stars with [Fe/H]<–2 and the rest of the RGB–MP population. As noted in \lx@sectionsign4, the average [Si/Fe] and [Ca/Fe] abundances are 0.17 and 0.11 dex larger for the [Fe/H]<–2 stars. This trend can be reconciled if the most metal–poor stars in the cluster, which represent only 3\% of the RGB–MP population, preferentially formed from the ejecta of \gtrsim20 M{}_{\sun} stars. However, this would require very rapid enrichment of the early \omega Cen environment because >20 M{}_{\sun} stars live \lesssim10{}^{\rm 7} years (e.g., Schaller et al. 1992). Note that this scenario is compatible with the observation that the [Fe/H]<–2 stars have the same mean [Ti/Fe] abundance as the rest of the RGB–MP population because, as mentioned above, the titanium yields from >20 M{}_{\sun} stars are comparable to those of lower mass stars. Additionally, if a monotonic relationship between [Fe/H] and formation time exists for at least the RGB–MP stars, then the mass dependent yields may also explain the apparent decrease in [Si,Ca/Fe] as [Fe/H] increases from \sim–2 to –1.6, as well as, the steeper decline for [Si/Fe] compared to [Ca/Fe]. As indicated by Figure 14, the decline in Si yield is a stronger function of progenitor mass between 18–25 M{}_{\sun} than for Ca. Therefore, forming stars from gas polluted by progressively less massive SNe should qualitatively reproduce the observed trend. The sudden increase in [Si,Ca/Fe] in the RGB–Int1 population would then make sense if a new round of star formation began with >20 M{}_{\sun} stars contributing once again.

5.1.1 Are Type Ia SNe Required?

Since previous analyses have estimated that the age spread among the various \omega Cen populations is \sim2–4 Gyr (e.g., Stanford et al. 2006), it would seem reasonable to assume that Type Ia SNe could have contributed to the cluster’s chemical enrichment. However, the consistently elevated [\alpha/Fe] ratios observed for nearly all stars in the cluster suggests that Type Ia enrichment has been limited. Pancino et al. (2002) and Origlia et al. (2003) found in a small sample of \omega Cen giants that the RGB–a stars had noticeably lower [\alpha/Fe] and higher [Cu/Fe] abundances than the lower metallicity stars, and attributed these trends to the onset of Type Ia SNe at [Fe/H]>–1. On the other hand, Cunha et al. (2002) analyzed [Cu/Fe] abundances in a larger sample spanning [Fe/H]\sim–2 to –0.8, and did not find evidence for an increase in [Cu/Fe]. Similarly, Norris & Da Costa (1995) and Smith et al. (2000) did not find evidence for a decrease in [\alpha/Fe] or an increase in [Cu/Fe].

While the primary production source of Cu is uncertain (e.g., Sneden et al. 1991b; Matteucci et al. 1993), it is clear that ambiguity remains regarding the significance of Type Ia SNe to \omega Cen’s chemical evolution. Our data are generally inconsistent with the rather extreme 75\% Type Ia to 25\% Type II mixture plotted in Figure 13, especially at [Fe/H]<–1. Although we find a slight decrease in [Ca/Fe] at [Fe/H]>–0.7, at least part of this decrease may be explained by a reduction in calcium yields from more metal–rich Type II SNe (e.g., see Figures 1314). Interestingly, the [Si/Fe] and [Ti/Fe] ratios do not exhibit similar decreases at [Fe/H]>–0.7. However, the larger measurement error for silicon compared to calcium may be masking any subtle trends, and although titanium is often enhanced like other \alpha elements in globular cluster stars its dominant isotope {}^{\rm 48}Ti is not an \alpha isotope. Additionally, analyzing different mixtures of Type II versus Ia ejecta requires inherent assumptions about the massive star IMF and the source of Type Ia SNe, which in Figure 13 is the “standard” white dwarf deflagration model. The model values shown in Figure 13 could easily be changed by using different assumptions and adjusting the aforementioned parameters.

A more empirical approach is to compare the evolution of \alpha and Fe–peak elements with other stellar populations exhibiting different levels of Type Ia enrichment. In Figures 1517 we plot our derived abundances for \omega Cen stars as a function of [Fe/H], and compare with data from the literature tracing the chemical evolution of other globular clusters, the Galactic thin/thick disk, halo, bulge, and nearby dwarf galaxies (see Table 8 for literature references). Focusing on the heavy \alpha and Fe–peak elements at the metal–rich end of the distribution shows that, at least for stars with [Fe/H]<–0.7, \omega Cen generally follows a morphology similar to that found in monometallic globular clusters, the Galactic halo, and the Galactic bulge. In contrast, the most metal–rich \omega Cen stars ([Fe/H]>–0.7) exhibit [Ca/Fe] ratios that are more similar to those found in Galactic thick disk stars (e.g., see Brewer & Carney 2006). Additionally, the most metal–rich \omega Cen stars tend to exhibit [Ca/Fe] ratios that are, on average, at least 0.1–0.2 dex lower than those found in the more metal–poor stars. This may indicate that the level of Type Ia enrichment in the most metal–rich \omega Cen stars and the thick disk were comparable. However, at [Fe/H]>–0.7 the [Ni/Fe] ratios are noticeably low in the \omega Cen stars, and as mentioned previously the [Si/Fe] and [Ti/Fe] data do not exhibit similar abundance decreases in concert with [Ca/Fe]. Although \omega Cen is widely believed to be the remnant core of a dwarf spheroidal galaxy, the heavy \alpha elements are enhanced in \omega Cen stars by a factor of 2–3 compared with other dwarf galaxies, at least for [Fe/H]\gtrsim–1.5.

In addition to the heavy \alpha and Fe–peak elements, the lighter elements O, Na, and Al are also inconsistent with significant contributions from Type Ia SNe. Figure 13 shows that nearly all of the stars with [Fe/H]>–1 have [Na/Fe] and [Al/Fe] abundances that are well above even the levels predicted by Type II SNe, but [O/Fe] is abnormally low. The abundance patterns expected from Type Ia production should lead to an overall decrease in the average abundance of all three elements as [Fe/H] increases. However, these elements can be altered by either in situ mixing or pollution from other sources, and therefore may not be reliable indicators of a star’s original composition. While the heavy \alpha element data, in particular [Ca/Fe], provide some evidence for Type Ia SN contributions in the most metal–rich stars, the light element data are in better agreement with a Type II SN pollution model that includes an additional proton–capture production mechanism. The apparent suppression of Type Ia SNe in \omega Cen remains an open problem, but it may be at least partially tied to the cluster’s several Gyr relaxation time scale (e.g., van de Ven et al. 2006) and low (\sim3–4\%) binary frequency (Mayor et al. 1996).

5.2 Proton–Capture Processing: Light Element Variations

The light elements oxygen through aluminum provide sensitive diagnostics for determining the chemical enrichment history of stellar populations. These elements are primarily produced in the hydrostatic helium, carbon, and/or neon burning stages of massive (\gtrsim10 M{}_{\rm\sun}) stars (e.g., Arnett & Thielemann 1985; Thielemann & Arnett 1985; Woosley & Weaver 1995). Stars forming out of gas that has been primarily polluted by Type II SNe should have [O/Fe]\sim+0.4 and exhibit increasing [Na/Fe] and [Al/Fe] abundances with increasing metallicity. However, these elements can also be produced (or destroyed) in lower mass stars that reach internal temperatures high enough to activate the proton–capture ON, NeNa, and MgAl cycles. If this processed material is mixed to the surface, then some stars may return gas to the ISM that is O–poor and Na/Al–rich compared to the material ejected by Type II SNe. This scenario is believed to occur in the RGB and AGB phases of low and intermediate mass (\lesssim8 M{}_{\rm\sun}) stars (e.g., Sweigart & Mengel 1979; Cottrell & Da Costa 1981; Denisenkov & Denisenkova 1990; Langer et al. 1993; Ventura & D’Antona 2009; Karakas 2010), but also in the cores of massive, rapidly rotating main sequence stars (e.g., Decressin et al. 2007).

Figures 1518 highlight the distinct light element abundance patterns found in several different stellar populations. Examination of these trends indicates that although \omega Cen shares some abundance patterns with other globular cluster, Galactic disk, halo, bulge, and nearby dwarf galaxy stars, it differs from all of these both in the extent of its star–to–star abundance variations and its individual abundance ratios. Approximately half of the RGB–MP stars have O, Na, and Al abundances that are consistent with those found in similar metallicity halo, and to a lesser extent, dwarf galaxy stars. The chemical composition of these stars is believed to be primarily a result of Type II SN enrichment, and the chemical similarities among these populations is not unexpected. It seems likely that \omega Cen would have had considerable interaction with the primordial gas that formed the Galactic halo, and it has been shown that reproducing the cluster’s metallicity distribution function is only possible in an open box scenario (e.g., Ikuta & Arimoto 2000; Romano et al. 2007). However, the remaining RGB–MP stars exhibit [O/Fe], [Na/Fe], and [Al/Fe] abundances that are significantly different than those found in metal–poor halo and dwarf galaxy stars. In particular, the “enhanced” RGB–MP stars are O–poor and Na/Al–rich. Similar chemical compositions are only found in some monometallic globular cluster stars (e.g., see reviews by Kraft 1994; Gratton et al. 2004). Interestingly, the number of stars in \omega Cen that are O–poor and Na/Al–rich increases to 60–95\% at higher metallicities. The RGB–Int2+3, and especially the RGB–a, stars have [O/Fe], [Na/Fe], and [Al/Fe] ratios that differ significantly even from individual globular clusters by at least a factor of two. Figure 18 shows that this is true even when considering [O/Na], [O/Al], and [Na/Al] ratios instead of [X/Fe]. Our data indicate that the \omega Cen stars at [Fe/H]\gtrsim–1.3 experienced an additional enrichment process that is not observed in any other stellar system studied so far, but the combined populations of M54 and the Sagittarius dwarf galaxy may share some similar trends (Carretta et al. 2010).

The light element abundance patterns in other globular clusters are typically believed to be the result of high temperature proton–capture nucleosynthesis operating in an environment where a combination of the ON, NeNa, and MgAl cycles are or were active. Material that has been processed through these proton–capture cycles is expected to exhibit a deficiency in [O/Fe] concurrent with supersolar [Na/Fe] and [Al/Fe] ratios, which should naturally lead to O–Na and O–Al anticorrelations along with a Na–Al correlation. In Figures 1921, we plot [O/Fe], [Na/Fe], and [Al/Fe] against each other for the major \omega Cen populations described in \lx@sectionsign4.1. We find that the O, Na, and Al abundance relations found in the RGB–MP, RGB–Int1, and RGB–Int2+3 populations are consistent with the abundance patterns that are characteristic of high temperature proton–capture processing. Furthermore, the impact of proton–capture nucleosynthesis appears to increase as a function of increasing metallicity. Both the extent of the light element variations and the percentage of stars that are O–poor and Na/Al–rich increases monotonically with [Fe/H]. However, the same O–Na, O–Al, and Na–Al relations are not observed in the RGB–a population. Instead, the RGB–a, as well as a few RGB–Int2+3, stars exhibit a rather uniform composition that is moderately O–poor ([O/Fe]\sim–0.15), very Na–rich ([Na/Fe]\sim+0.78), and is unlike any of the more metal–poor \omega Cen stars.

A common interpretation of the light element abundance trends in monometallic globular clusters is that the O–rich, Na/Al–poor stars represent the first generation of stars formed from the ejecta of Type II SNe, and the O–poor, Na/Al–rich stars represent a subsequent generation formed from gas that had been chemically enriched by intermediate mass AGB stars or some other polluting source in which the ON, NeNa, and/or MgAl cycles were active (e.g., D’Ercole et al. 2008; Carretta et al. 2009b). The first generation stars are often referred to as “primordial” stars, and the enriched populations are referred to as either “intermediate” or “extreme”, depending on the level of O–depletion and Na–enrichment (e.g., Carretta et al. 2009a; but see also Lee 2010 for a different interpretation).

The \omega Cen data can be divided into similar subpopulations. Here we follow a similar definition to that used in Carretta et al. (2009a) where the primordial component is defined as having [O/Fe]\geq0 and [Na/Fe]\leq+0.1, the intermediate component includes stars with [O/Fe] ratios satisfying the relation [O/Fe]\geq[0.62([Na/Fe])–0.65], and the extreme component consists of the remaining most O–poor stars. Monometallic globular clusters typically consist of \sim20–40\% of stars belonging to the primordial component, \sim30–80\% in the intermediate component, and \lesssim20\% in the extreme component (e.g., Carretta et al. 2009a). As can be seen in Figures 1921, the RGB–MP stars follow the general trend observed in monometallic globular clusters with a primordial:intermediate:extreme distribution of 50\%:43\%:7\%, respectively. The RGB–Int1 population contains roughly an equal proportion of primordial, intermediate, and extreme abundance stars with a distribution of 30\%:32\%:38\%. However, the RGB–Int2+3 and RGB–a stars contain far more extreme abundance stars than are found in any globular cluster with distributions of 11\%:15\%:74\% and 5\%:14\%:81\%, respectively. The large number of intermediate and extreme abundance stars indicates that \omega Cen likely experienced a similar enrichment process to that in monometallic globular clusters during each round of star formation, and it is interesting to note that the populations expected to be He–rich exhibit the largest fraction of extreme abundance stars. While there is a clear delay in the onset of whichever mechanism drives the O–poor, Na/Al–rich abundance phenomenon, it is worth noting that we find a very low incidence of carbon stars161616While we do not provide explicit carbon abundance measurements in this paper, the possible carbon stars listed in Figure 1, Figure 3, and Table 2 were identified by visual inspection of their spectra. However, all three of the possible carbon stars identified here that also overlap with the van Loon et al. (2007) survey (LEID 32059, 41071, and 52030) are confirmed carbon stars based on the presence of strong C{}_{\rm 2} bands in their spectra. (<2\%; see Figure 1) despite the large population of O–poor stars. Unfortunately, we cannot distinguish between in situ carbon stars and those formed from mass transfer, but the frequency of carbon stars on the giant branch is consistent with the expected binary fraction of \sim3–4\% (Mayor et al. 1996).

5.2.1 Enrichment by Pollution and in situ Processing

Although we have identified the major light element abundance trends for \omega Cen, the information so far has only led us to conclude that proton–capture nucleosynthesis has likely played a significant role in the cluster’s chemical enrichment. Further examination is required in order to understand the possible location(s) where these processes are or were active. The comparatively small star–to–star dispersion in [X/Fe] exhibited by the heavy \alpha and Fe–peak elements (see Figure 10) indicates that the >1 dex variations observed for [O/Fe], [Na/Fe], and [Al/Fe] are not due to incomplete mixing of SN ejecta, as is suspected for [Fe/H]<–3 halo stars (e.g., McWilliam 1997). Previous studies have found that many of the light element abundance patterns exhibited by monometallic globular cluster stars, which are subsequently shared by many \omega Cen stars, may be best explained by proton–capture nucleosynthesis operating at temperatures near 70\times10{}^{\rm 6} K (e.g., Langer et al. 1997; Prantzos et al. 2007). If at least part of the abundance patterns found in \omega Cen and other globular cluster stars are due to pollution from external sources, then the currently favored production mechanisms are: (1) hot bottom burning in >5 M{}_{\sun} AGB stars (e.g., Ventura & D’Antona 2009; Karakas 2010), (2) hydrogen shell burning in now extinct but slightly more massive RGB stars (Denissenkov & Weiss 2004), and (3) core hydrogen burning in rapidly rotating massive stars (Decressin et al. 2007).

While massive, rapidly rotating stars and extinct \sim0.9–2 M{}_{\sun} RGB stars may also reproduce many of the observed light element trends, presently there are no detailed theoretical yields spanning a fine grid of metallicities similar to those available for intermediate mass AGB stars. Furthermore, the time scale of pollution from extinct low mass RGB stars is at least 2–3 times longer than the estimated age spread among the different \omega Cen populations, but this does not rule out possible mass transfer pollution from these objects. Additionally, the massive, rapidly rotating star scenario is expected to produce a continuum of polluted stars with varying He abundances (Renzini 2008), which is inconsistent with the singular Y=0.38 value that seems required to fit the blue main sequence (e.g., Piotto et al. 2005). Romano et al. (2010) also point out that if the winds from massive main sequence stars are also responsible for the anomalous light element abundance variations in the current generations of \omega Cen stars, it is not clear why the He enrichment was delayed until higher metallicities. However, Renzini (2008) and Romano et al. (2010) find that intermediate mass AGB stars may provide a reasonable explanation for the high He content in some stars, in addition to the average behavior of [Na/Fe], and to a lesser extent [O/Fe], in \omega Cen. Therefore, we will only consider the AGB pollution scenario here, but we caution the reader that several qualitative and quantitative hurdles remain in order for AGB pollution to be a viable explanation of light element variations in globular clusters (e.g., Denissenkov & Herwig 2004; Denissenkov & Weiss 2004; Fenner et al. 2004; Ventura & D’Antona 2005; Bekki et al. 2007; Izzard et al. 2007; Choi & Yi 2008).

In Figure 22, we plot our derived O, Na, and Al abundances as a function of [Fe/H], and overplot the metallicity dependent theoretical yields from Type II SNe, as well as, 3–6 M{}_{\sun} AGB stars. While the \omega Cen stars with chemical compositions similar to the Galactic disk and halo appear well bounded by production from Type II SNe, the enhanced stars at least qualitatively follow the general trends predicted by production from >5 M{}_{\sun} AGB stars. In particular, the depletion of oxygen concurrent with the rise in sodium and decline in the maximum [Al/Fe] ratio with increasing metallicity are all consistent with the predicted patterns exhibited by material that has been processed via hot bottom burning in >5 M{}_{\sun} AGB stars. However, the theoretical AGB yield curves shown in Figure 22 do not include lifetime estimates for the polluting AGB stars, and one could envision sliding the various curves along the abscissa to account for age differences among the different populations. In other words, plots similar to Figure 22 lend insight into whether the abundance trends are possibly consistent with AGB pollution, but numerical chemical evolution models are required to fully constrain which mass ranges have contributed to the chemical composition of stars in a given population.

Despite this limitation, we can use Figure 22 to elicit some constraints. We find that while 5–6 M{}_{\sun} AGB ejecta are generally consistent with the abundance trends observed at all metallicities, 3–4 M{}_{\sun} AGB stars likely did not contribute significantly to \omega Cen’s chemical enrichment until about [Fe/H]=–1.3. This is most evident by examining the [O/Al] and [Na/Al] ratios in Figure 22. The <5 M{}_{\sun} AGB stars produce [O/Fe] and [Na/Fe] ratios that are too high and [Al/Fe] ratios that are too low to fit the data, even when diluted with SN or >5 M{}_{\sun} AGB ejecta. Figures 1921 also support the rejection of 3–4 M{}_{\sun} AGB ejecta, which originate from AGB stars of comparable metallicity, from contributing significantly to the chemical composition of stars with [Fe/H]<–1.3. However, Figures 1922 do not rule out that <5 M{}_{\sun} AGB stars with [Fe/H]\lesssim–1.5 impacted enrichment of the RGB–Int2+3 and RGB–a populations. The [Na/Fe] and [Al/Fe] yields from metal–poor AGB stars are mostly consistent with the trends observed in the intermediate and most metal–rich \omega Cen giants, but it seems that an additional mechanism may be required to explain the [O/Fe] abundances. Note that our conclusions are not drastically altered if we adopt the theoretical AGB yields from Karakas (2010)171717This statement is based on using the average mass fraction data from Tables A2–A6 in Karakas (2010)., which uses mixing length theory for convection, instead of the Ventura & D’Antona (2009) yields, which use the full spectrum of turbulence theory for convection and are shown in Figures 1922. Unfortunately, Karakas (2010) does not provide yield information for metallicities between [Fe/H]=–2.3 and –0.7, which makes direct comparison with \omega Cen difficult because most stars fall in the missing range.

One of the most puzzling aspects concerning the abundance patterns of light elements in \omega Cen is the strongly bimodal distribution at intermediate metallicities (see Figure 11). If ISM pollution was driven by AGB stars, then it is unclear why (1) only the RGB–MP stars exhibit a continuous distribution of [O/Fe], [Na/Fe], and [Al/Fe] abundances and (2) more than 70\% of the more metal–rich stars have envelope material that has experienced significant proton–capture processing. As can be seen in Figures 1922, the [O/Fe] yields from AGB stars are by far the most inconsistent with our data, but the full mass range of AGB stars may reproduce the [Na/Fe] and [Al/Fe] abundances at nearly all metallicities. Depleting the oxygen abundance from [O/Fe]=+0.4 to [O/Fe]<–0.4 via hot bottom burning in AGB stars is generally not achieved for any mass or metallicity range. However, D’Ercole et al. (2010) showed that including the ejecta of “super–AGB” (>6.5 M{}_{\sun}) stars may reproduce the super O–poor ([O/Fe]<–0.4) abundances found in some globular clusters under the assumption that the massive AGB stars deplete to [O/Fe]\approx–1. Despite this, it is unlikely that more massive AGB stars are the culprits behind the large contingent of super O–poor \omega Cen stars because one would have to assume an IMF strongly weighted toward \sim5–9 M{}_{\sun} stars in order to produce so many super O–poor stars. Note that this is not as much of a problem in monometallic globular clusters because the number of super O–poor stars is <20\% (e.g., Carretta et al. 2009a). It seems that invoking some degree of in situ proton–capture processing is required to explain the observed abundance patterns of \omega Cen stars with [Fe/H]\gtrsim–1.6, in order to avoid unrealistic requirements such as IMFs strongly weighted toward intermediate mass stars or forming a majority of the RGB–Int1, RGB–Int2+3, and RGB–a stars almost entirely out of a narrow mass range of AGB stars.

A key assumption when considering in situ processing in low mass RGB stars is that the material being enriched near the hydrogen burning shell must be able to mix into the convective envelope and be brought to the surface. In stars with normal helium abundances, it is not believed that this can occur until the hydrogen burning shell erases the molecular weight barrier left behind by the convective envelope after first dredge–up (e.g., see review by Salaris et al. 2002). However, some or all of the intermediate metallicity stars in \omega Cen are thought to be quite He–rich, and D’Antona & Ventura (2007) found that stars with Y=0.35–0.40 should contain a much more shallow molecular weight gradient that might not inhibit deep mixing. Instead, deep mixing in He–rich stars might be active over a wide range of luminosities on the giant branch, which would be consistent with our observation that the degree of light element enrichment is not strongly correlated with luminosity. These authors also find that reproducing the abundance patterns exhibited by the super O–poor stars can be achieved by in situ mixing if the RGB stars are already polluted by the ejecta of intermediate mass AGB stars. In their scenario, in situ mixing should decrease the envelope [O/Fe] ratio by up to a factor of 10 while only increasing the [Na/Fe] ratio by about 0.2 dex. While the evolution of [Al/Fe] is not reported by D’Antona & Ventura (2007), we can speculate that the enhancement in [Al/Fe] is smaller than that experienced by [Na/Fe] given the higher temperatures required to convert Mg to Al.

As mentioned above, the proposed deep mixing scenario only works if the intermediate metallicity RGB stars in \omega Cen formed from material that was already enriched by hot bottom burning in intermediate mass AGB stars. Our current data set does not provide direct evidence of this, but we may look to the behavior of silicon as a proxy indicator because {}^{\rm 28}Si can be produced through leakage from the MgAl cycle at temperatures >65\times10{}^{\rm 6} K (e.g., Yong et al. 2005; Carretta et al. 2009b). In Figure 23, we plot our [X/Fe] abundances as a function of [Fe/H] color coded by the primordial, intermediate, and extreme abundance components described above. While most of the \alpha and Fe–peak elements do not display any particular dependence on light element abundance, the RGB–MP and RGB–Int1 extreme component stars exhibit silicon enhancements of nearly 0.3 dex compared to the primordial and intermediate component stars. Furthermore, in Figure 24 we plot [O/Fe], [Na/Fe], and [Al/Fe] versus [Si/Fe], [Ca/Fe], and [Ti/Fe] and find that only silicon shows any semblance of a correlation with O, Na, and Al, as is indicated by the respective Pearson correlation coefficients shown in Figure 24. This suggests that silicon may have undergone an additional production process not experienced by the heavier \alpha elements. The existence of an Al–Si correlation concurrent with an O–Si anticorrelation suggests that the O–poor stars were likely polluted by material that had been processed at temperatures exceeding \sim65\times10{}^{\rm 6} K. These conditions are reached during hot bottom burning in intermediate mass AGB stars, but not in the hydrogen burning shells of low mass RGB stars.

Since the \omega Cen stars likely satisfy the prerequisites needed for in situ mixing to occur, we may attribute a large portion of the [O/Fe], and to a lesser extent the [Na/Fe], variations to this process. The relatively small number of RGB–MP stars that are super O–poor suggests that the helium content had not yet been significantly increased in the cluster to allow the formation of He–rich stars. In fact, there are very few super O–poor stars at [Fe/H]<–1.7. The radial segregation of O–poor stars (see Figure 9) is also consistent with the idea that additional time was needed to increase the cluster He content, and may indicate that He–rich gas was preferentially funneled into the cluster core, as is suggested in the models by D’Ercole et al. (2008). We find that the light element abundance trends in the intermediate metallicity and RGB–a stars are consistent with an AGB pollution plus in situ mixing scenario. In these stars, the high [Na/Fe] and [Al/Fe] abundances are consistent with production in comparable metallicity or more metal–poor AGB stars because in situ mixing is not expected to significantly increase [Na/Fe] or [Al/Fe] in He–rich RGB stars that are already O–poor and Na/Al–rich (D’Antona & Ventura 2007). Additionally, the increasing minimum [O/Fe] abundance at [Fe/H]\gtrsim–1 is consistent both with the increase in the [O/Fe] yields for >5 M{}_{\sun} AGB stars and the fact that in situ mixing should produce less advanced proton–capture processing at higher metallicities. This is due primarily to the lower temperatures achieved in the interiors of more metal–rich stars, but may also occur if the He mass fraction in the RGB–a stars is smaller than in the RGB–Int2+3 stars, which could lead to more shallow mixing. Lastly, we note that because the [Na/Fe] and [Al/Fe] abundances do not share the same correlation as [O/Fe] with radial location, it may be the case that some stars producing high Na and Al yields do not necessarily produce large He yields.

One of the most important effects of including in situ mixing in the chemical enrichment picture is that it reduces the necessity for AGB stars to account for all abundance patterns, and also increases the mass range of available AGB polluters to more than just those with favorable yields. The inferred high He–content of the blue main sequence population and the observed large increase in s–process enrichment in this cluster indicate that the large degree of ISM pollution required for the above scenario to work is not unreasonable. A detailed comparison of the {}^{\rm 24}Mg, {}^{\rm 25}Mg, and {}^{\rm 26}Mg isotopes may be particularly illuminating in order to investigate whether hot bottom burning, in situ processing, or both played an active role in shaping the abundance patterns of \omega Cen giants. Note that fluorine is also expected to be strongly depleted in the proposed deep mixing scenario. Furthermore, a large sample spectroscopic abundance analysis of stars at lower luminosities could provide an interesting test for the impact of in situ mixing.

5.2.2 Oxygen Abundances and a Possible Connection to the Blue Main Sequence

The discovery and subsequent detailed analyses of \omega Cen’s blue main sequence (Anderson 1997, 2002; Bedin et al. 2004; Norris 2004; Piotto et al. 2005; Sollima et al. 2007; Bellini et al. 2009b) have revealed that this population represents \sim30\% of all main sequence stars, is preferentially located near the cluster center, and perhaps most importantly is more metal–rich than the dominant red main sequence. As mentioned in \lx@sectionsign1, the commonly accepted reason for the existence of the blue main sequence is that these stars are significantly more He–rich than the dominant population of more metal–poor stars. In fact, Norris (2004) and Piotto et al. (2005) find that the blue main sequence is best fit with an extreme helium abundance of Y\approx0.38. One of the most interesting characteristics of the blue main sequence is that it is well detached from the red main sequence in color, and almost no stars are found in between the two sequences (e.g., see Bedin et al. 2004; their Figure 1). From this and the information above, we might expect the current RGB stars that were once part of the blue main sequence to be chemically conspicuous, preferentially located near the cluster center, and more metal–rich that the dominant stellar population.

Examination of the abundance patterns in the RGB–Int1 and RGB–Int2+3 stars reveals that the [O/Fe] ratio stands out as a possible indicator of which stars once belonged to the blue main sequence. At intermediate metallicities, a majority of the stars have [O/Fe]\leq0, and the radial distribution of these stars shows that more than 90\% are located inside 10\arcmin from the cluster center while only 70\% of those with [O/Fe]>0 are located in the same range (see Figure 9). However, in order for the [O/Fe] abundance to be considered as a chemical tracer of the blue main sequence it must qualitatively and quantitatively agree with the observed trends of blue main sequence stars. In Figure 25, we plot the number ratio of O–poor to O–rich stars out to \sim15\arcmin, and also plot the measured number ratios of blue to red main sequence stars from Bellini et al. (2009b). Although we are plotting an indirect measurement of the ratio of blue to red main sequence stars with N{}_{\rm O-poor}/N{}_{\rm O-rich} and the Bellini et al. (2009b) data represent direct measurements, we find that the two trends are in reasonable agreement. Both data sets indicate that the majority of O–poor (blue main sequence) stars are inside \sim5\arcmin of the cluster center, and the relative ratio of O–poor/O–rich (blue/red main sequence) stars decreases at larger radii. Note that Sollima et al. (2007) come to a similar conclusion when considering stars located at \sim7–23\arcmin from the cluster center.

Since the relative ratios of O–poor to O–rich stars follow those observed for the blue and red main sequences, we may expect the absolute number of O–poor stars to also be consistent with that of the blue main sequence stars due to our high completion percentage (see Figure 2). As mentioned previously, it is estimated that the blue main sequence constitutes \sim25–35\% of all main sequence stars, and we find in agreement with this estimate that 27\% of all RGB stars in our sample are O–poor. Additionally, Piotto et al. (2005) showed that the blue main sequence is best fit by a metallicity similar to that of the RGB–Int1 and RGB–Int2+3 populations. We find that at least 65\% of the O–poor stars in our sample are located in the appropriate metallicity range. This percentage may in fact be somewhat larger if we consider that (1) very few O–poor stars are found at [Fe/H]<–1.7, (2) the average [Fe/H] abundance error is roughly \pm0.1 dex, and (3) the boundary between the RGB–MP and RGB–Int1 populations is not uniquely defined. However, we would still find that \sim20–30\% of intermediate metallicity stars are O–rich. Note that a significant number of O–rich, intermediate metallicity stars (i.e., stars in the correct metallicity range that would not lie on the blue main sequence) would be consistent with the observation by Sollima et al. (2006) that many RR Lyrae stars with [Fe/H]\sim–1.2 have standard helium abundances.

In any case, we have demonstrated that the O–poor giants are spatially similar, found mostly in the same metallicity range, and are present in nearly identical proportions to those found on the blue main sequence. It is not entirely clear why the [O/Fe] ratio in the giants shares a similar sensitivity to radial location and metallicity with the blue main sequence stars, but we speculate that the oxygen deficient stars are connected with the blue main sequence through helium enrichment. That is, the He–rich main sequence stars are pushed blueward on the color–magnitude diagram, and the He–rich giants experience in situ mixing that strongly depletes oxygen without similarly large increases in sodium and aluminum. Comparison between the 7770 Å oxygen triplet line strengths in blue and red main sequence stars would provide a direct confirmation of our hypothesis. Although we have invoked in situ mixing to explain the very large O–depletion in these stars, note again that the scenario proposed by D’Antona & Ventura (2007) requires that these star were already somewhat O–poor.

5.3 Neutron–Capture Processing

While the isotopes of most elements lighter than about zinc are produced primarily through charged particle reactions, the isotopes of elements beyond the Fe–peak are mostly produced through neutron–capture reactions. Neutron–capture nucleosynthesis is believed to proceed through two main channels: (1) the s–process where the neutron–capture rate is slow compared to the \beta–decay rate of unstable nuclei and (2) the r–process where the neutron–capture rate is fast compared to the \beta–decay rate of unstable nuclei (e.g., see recent review by Sneden et al. 2008). The large difference in neutron fluxes required for the two processes points to different operational environments. The main component of the s–process is widely believed to be active in thermally pulsing low and intermediate mass AGB stars, but theoretical models indicate that most of the s–process element production is probably constrained to stars in the range \sim1.3–3 M{}_{\sun} (e.g., Busso et al. 1999; Herwig 2005; Straniero et al. 2006). AGB stars \lesssim1.3 M{}_{\sun} have envelope masses that are too small for third dredge–up to occur, and more massive AGB stars are only believed to experience a few third dredge–up episodes. Conversely, the exact location(s) where the r–process operates is (are) not well defined, but significant circumstantial evidence suggests an explosive origin associated with core collapse SNe (e.g., Mathews & Cowan 1990; Cowan et al. 1991; Wheeler et al. 1998; Arnould et al. 2007; Sneden et al. 2008).

The solar system abundances indicate that \sim70–75\% of lanthanum is produced via the s–process and more than 95\% of europium is produced by the r–process (e.g., Sneden et al. 1996; Bisterzo et al. 2010). Therefore, we adopt lanthanum as an s–process indicator and europium as an r–process indicator. As can clearly be seen in Figure 17, the average [La/Fe] ratio increases by more than a factor of three between the RGB–MP and intermediate metallicity populations. Similar increases are not found for any other elements in our sample. This indicates that the s–process has played a significant role in the chemical evolution of \omega Cen, and is a dominant process at [Fe/H]\gtrsim–1.6. Comparison with the other stellar populations plotted in Figure 17 shows that the level of s–process enrichment was far greater in \omega Cen. It is interesting to note that the [La/Fe] ratio does not continue to increase beyond [Fe/H]\geq–1.5 despite the fact that s–process production appears to peak in the metallicity range –1.5\lesssim[Fe/H]\lesssim–0.8, at least for the “standard” {}^{\rm 13}C pocket (e.g., see Bisterzo et al. 2010, their Figure 8). This may indicate that a large fraction of the gas was swept out of the cluster through interaction with the Galaxy before low mass AGB stars with [Fe/H]\gtrsim–1.5 had a chance to contribute to \omega Cen’s chemical enrichment. Comparison between the RGB–Int2+3 and RGB–a stars shows that the average [La/Fe] ratio decreases by \sim0.2 dex (see Figure 12) for higher metallicities, but is still significantly enhanced compared to the Galactic disk and bulge trends. This suggests that s–process production still continued at high metallicities, but the rate of production did not exceed that of iron.

For the RGB–MP stars, the average [Eu/Fe] abundances are similar to those found in halo, dwarf galaxy, and individual globular cluster stars. At higher metallicities, the average [Eu/Fe] abundance of \omega Cen stars actually decreases while the average [La/Fe] abundance shows a significant increase. In fact, many intermediate metallicity \omega Cen stars have [Eu/Fe] abundances that are lower than those found in halo and dwarf galaxy stars, and are especially Eu–deficient compared to globular cluster stars. However, the average [Eu/Fe] abundance increases again at [Fe/H]\gtrsim–1.2 toward values similar to those found in the Galactic disk and bulge. The cause of the decrease in [Eu/Fe] at intermediate metallicities, and the low [Eu/Fe] abundances in general, is not entirely clear. It is believed that \sim8–10 M{}_{\sun} SNe may produce a large portion of the r–process elements (e.g., Mathews & Cowan 1990), but other processes such as neutron star and black hole mergers may be important as well (e.g., see review by Sneden et al. 2008 and references therein). It may be the case that either the IMF did not favor a large number of stars in the 8–10 M{}_{\sun} range or that one or more of the typical r–process production mechanisms was not active at “normal” levels in the intermediate metallicity range. Interestingly, the metallicity range at which the [Eu/Fe] abundance is lowest is also where Cunha et al. (2002; 2010) find low [Mn/Fe] and [Cu/Fe] values. Since Cunha et al. (2010) attributes the low [Cu/Fe] and [Mn/Fe] abundances to metallicity dependent SN yields, we can speculate that the low [Eu/Fe] values might also be due to a related effect. Although manganese and europium are produced through different processes, their production may be tied to similar progenitor objects and/or environments.

Despite the obvious differences in [La/Fe] and [Eu/Fe] abundances for \omega Cen stars compared to those in other populations, the ratio of these elements provides a better diagnostic for analyzing the impact of the s– and r–processes. In the Galactic disk and halo, the [La/Eu] ratio slowly increases with metallicity, and this is believed to be primarily due to the longer time scales required for low and intermediate mass stars to evolve into AGB stars (e.g., Simmerer et al. 2004). Dwarf galaxies also tend to exhibit an increase in s–process elements at higher metallicities, but are typically more s–process enhanced than similar metallicity halo and disk stars (e.g., Geisler et al. 2007). In contrast, most globular clusters follow the disk/halo trend and are generally r–process rich (e.g., Gratton et al. 2004). Figure 18 plots the [La/Eu] ratio as a function of [Fe/H] for these populations and also illustrates the relatively rapid transition in \omega Cen from being r–process to s–process dominated. Many stars in the RGB–MP population exhibit [La/Eu] ratios that are identical to those found in halo, dwarf galaxy, and globular cluster stars. However, almost all of the more metal–rich stars have [La/Eu]>0, and many of these stars have [La/Eu] ratios matching those expected for pure s–process production. Note that proper accounting of hyperfine structure for both the La and Eu lines has revised our [La/Eu] ratios downward, at least for the most La–rich stars, from those found in Johnson et al. (2009). The new results are consistent with the more metal–rich stars forming from gas that was already heavily polluted with s–process elements, but does not require surface pollution from mass transfer. This is in agreement with the results from Stanford et al. (2010), which suggest that the strontium abundances (a light s–process element) in several \omega Cen stars are the result of primordial pollution rather than surface accretion.

Although AGB stars of about 1.3–8 M{}_{\sun} may be able to produce s–process elements, Smith et al. (2000) used the [Rb/Zr] ratio to show that AGB stars between \sim1.5–3 M{}_{\sun} were likely the dominant s–process enrichment sources in \omega Cen. Since these stars have lifetimes of 3\times10{}^{\rm 8}–2\times10{}^{\rm 9} years, the time delay between the formation of RGB–MP stars and subsequent generations had to be at least this long. This delay is consistent with the estimated 2–4 Gyr age range of \omega Cen stars (e.g., Stanford et al. 2006), and is also consistent with the time required for >4–5 M{}_{\sun} AGB stars to have polluted the ISM, as seems required to explain at least part of the light element abundance trends.

In addition to analyzing the behavior of elements produced exclusively through neutron–capture processes, we can also examine how neutron–capture nucleosynthesis may have affected the abundances of lighter elements. In Figure 26, we plot multiple elements as a function of lanthanum abundance. As expected, the [Ni/Fe] and [Eu/Fe] ratios do not exhibit any correlation with [La/Fe]. This confirms our assumption that europium is produced almost exclusively through the r–process, and that nickel, along with other Fe–peak elements, is not significantly affected by the s–process. Additionally, we find that all other elements exhibit a mild correlation with [La/Fe]. Given the strong enhancement in lanthanum, it is not surprising that the lighter elements might also be mildly affected. Unfortunately, it is difficult to disentangle the production of these elements from other sources. We suspect that much of the correlation between the heavy \alpha elements and lanthanum may be due to the combined effects of Type II SN and AGB s–process production overlapping in the same metallicity regime. In particular, the largest increase in [La/Fe] occurs at the transition between the RGB–MP and RGB–Int1 populations. The elevated [Si/Fe] and [Ca/Fe] ratios concurrent with an increase in [Fe/H] strongly suggests that Type II SNe were the major producers of these elements. As mentioned in \lx@sectionsign5.1, the increase in Si and Ca abundances may be the result of metallicity dependent Type II SN yields rather than additional production from the s–process. At present, we do not have a definitive explanation for the increase in [Ti/Fe] or its correlation with [La/Fe]. However, we point out that the stable isotope {}^{\rm 50}Ti is a neutron magic nucleus, and it has been predicted that the helium shell of thermally pulsing AGB stars may exhibit a large {}^{\rm 50}Ti/{}^{\rm 48}Ti ratio (e.g., Gallino et al. 1994). If at least some AGB stars that eject large amounts of s–process elements also eject material with a high {}^{\rm 50}Ti/{}^{\rm 48}Ti ratio, then this may provide an explanation for the Ti–La correlation.

It is interesting to note that while the O–rich stars exhibit a correlation with [La/Fe], the same relation appears to be mostly absent from the O–poor stars. This supports the idea that the depletion of oxygen is driven by an additional process, such as in situ mixing, that does not alter the [La/Fe] ratio. Although we find that nearly all of the O–poor stars also have [La/Fe]>–0.2, we point out that this may be mostly related to the fact that the O–poor (He–rich?) stars formed at a time when the average lanthanum abundance was already becoming significantly enhanced in the cluster ISM. Also note that we do not find any correlation between lanthanum abundance and radial location in the cluster, which does not match the observed trend for the O–poor stars (see Figure 9). We therefore conclude that the simultaneous rise in the number of O–poor and La–rich stars are not due to the exact same mechanisms. However, we believe that both phenomena are at least in some way related to pollution from low and/or intermediate mass AGB stars.

5.4 Final Remarks

The data presented here and in previous analyses indicate that \omega Cen experienced a unique chemical enrichment history. The occurrence of at least 4–5 discrete star formation episodes spanning >1–2\times10{}^{\rm 9} years seems required to rectify the breadth of the main sequence turnoff, the metallicity distribution function, and the large enhancement of s–process elements. Despite \omega Cen’s rather extensive chemical enrichment, the most metal–poor stars ([Fe/H]<–2) exhibit abundance patterns and star–to–star dispersions that are nearly identical to those found in similar metallicity halo and dwarf galaxy stars. These signatures strongly suggest a rapid enrichment time scale in which only massive stars had time to contribute to \omega Cen’s chemical composition. Additionally, at least half of the RGB–MP stars exhibit abundance trends that are consistent with the metal–poor halo, and the heavy \alpha element trends seem to indicate that the initial chemical enrichment occurred on a time scale that was sensitive to Type II SNe of different masses. However, a significant portion of the RGB–MP stars have [O/Fe], [Na/Fe], and [Al/Fe] abundances that are unlike stars found in the halo, and are instead more similar to those found in monometallic globular clusters. A clear delay in the presence of O–poor, Na/Al–rich stars until [Fe/H]\sim–1.7 suggests that new generations significantly polluted by the ejecta of \lesssim8 M{}_{\sun} stars did not form until about the same time as the second major episode of star formation. Furthermore, the neutron–capture data indicate that at least 1 Gyr had to have elapsed between the formation of the RGB–MP and RGB–Int1 populations.

Since a majority of RGB–MP stars have abundance patterns matching those predicted for Type II SN pollution, it seems likely that \omega Cen was able to retain and mix a significant percentage of SN ejecta at early times in the cluster’s evolution. However, at intermediate metallicities \omega Cen’s overall chemistry experienced a dramatic shift that strongly deviates from trends observed in the Galactic halo and most dwarf galaxies. The products of proton– and neutron–capture nucleosynthesis began to dominate the chemical composition of progressively more metal–rich stars, despite obvious contributions from Type II SNe. The significant pollution of intermediate metallicity stars by O–poor (He–rich?), Na/Al–rich, and s–process enhanced gas is undoubtedly the result of the RGB–MP stars evolving and enriching the cluster ISM. In order for pollution to occur at the levels observed in \omega Cen, the cluster must not have strongly interacted with the Galaxy until after at least the formation of the RGB–Int1 population. Otherwise, it is likely that the gas would have been removed by ram pressure stripping. The radial concentration of the RGB–Int2+3 and RGB–a stars near the cluster core indicates that enriched gas was funneled toward the cluster center and/or the central region was the only location where the escape velocity was large enough to retain gas ejected by SNe or AGB stars. The rapid decline in the relative number of “primordial” composition stars in the RGB–Int2+3 and RGB–a populations may be evidence that \omega Cen began to lose mass at [Fe/H]\gtrsim–1.3. Significant mass loss from the cluster may also help explain the minimal impact Type Ia SNe have played in \omega Cen’s chemical enrichment.

6 SUMMARY

We have measured chemical abundances of O, Na, Al, Si, Ca, Sc, Ti, Fe, Ni, La, and Eu for 855 RGB stars in the globular cluster \omega Cen. The abundances were obtained using moderate resolution (R\approx18,000), high S/N (>100) spectra obtained with the Hydra multifiber spectrograph on the Blanco 4m telescope at CTIO. The data set covers more than 80\% of stars with V\leq13.5, more than 90\% of stars with V\leq13.0, and samples the full breadth of the giant branch to include the most metal–poor and most metal–rich stars in the cluster. Similarly, we have achieved a completion fraction of \sim50–100\% at radii extending out to \sim24\arcmin from the cluster center. All abundances were determined using either equivalent width or spectrum synthesis analyses along with the inclusion of blended molecular lines, hyperfine structure, and/or isotope broadening when appropriate. An empirical hyperfine structure correction for the 6774 Å La II line is also provided.

We find in agreement with past photometric and spectroscopic studies that \omega Cen contains multiple, discrete stellar populations with large star–to–star abundance variations for all elements. The metallicity distribution function contains five peaks centered at [Fe/H]=–1.75, –1.50, –1.15, –1.05, and –0.75. However, for the analysis we have combined the [Fe/H]=–1.15 and –1.05 peaks into a single population. The (now four) stellar populations are identified as the RGB–MP ([Fe/H]\leq–1.6), RGB–Int1 (–1.6<[Fe/H]\leq–1.3), RGB–Int2+3 (–1.3<[Fe/H]\leq–0.9), and RGB–a ([Fe/H]>–0.9, which constitute 61\%, 27\%, 10\%, and 2\% of our sample, respectively. The metallicity distribution function also exhibits a sharp cutoff at the metal–poor end such that only 2\% of the stars in our sample have [Fe/H]<–2. The RGB–MP and RGB–Int1 populations appear to be uniformly mixed in the cluster, but the RGB–Int2+3 and RGB–a stars are preferentially located near the cluster core. Additionally, almost 90\% of the most metal–poor stars ([Fe/H]<–2) reside within 5\arcmin of the cluster center.

The abundance trends exhibited by the heavy \alpha (Si, Ca, and Ti) and Fe–peak elements (Sc and Ni) are generally well described by production from Type II SNe at all metallicities. That is, the \alpha elements are typically enhanced at [\alpha/Fe]\approx+0.3 and [Sc,Ni/Fe]\approx0. While the Fe–peak element [X/Fe] ratios and star–to–star variations remain mostly constant over \omega Cen’s full metallicity range, the heavy \alpha elements show a more complicated morphology. Over the metallicity range spanned by the RGB–MP, there is a noticeable decrease in the average [Si/Fe] and [Ca/Fe] abundances with increasing [Fe/H], but the average [Ti/Fe] abundance remains essentially constant. However, the decrease in [Si/Fe] is a stronger function of [Fe/H] than for [Ca/Fe]. The average [X/Fe] ratios for all three heavy \alpha elements increase with metallicity between the RGB–MP and RGB–Int1 populations and remains mostly enhanced at higher metallicities. It seems that many of these abundance trends may be driven by mass and/or metallicity dependent Type II SN yields and a new round of star formation creating the RGB–Int1 and subsequent populations. The simultaneous rise in [Ti/Fe] at [Fe/H]\gtrsim–1.6 may be driven by a different production mechanism because theoretical Type II SN yields do not predict a large increase in titanium with either progenitor mass or metallicity.

Although some previous analyses have suggested that Type Ia SNe may have become significant contributors to \omega Cen’s chemical enrichment at [Fe/H]>–1, we do not find particularly strong evidence supporting this claim. In the more metal–rich RGB–Int2+3 and RGB–a populations, we find that the average [\alpha/Fe] abundances remain elevated above the level found in disk and dwarf galaxy stars of similar metallicity. While there does appear to be a decrease in [Ca/Fe] at [Fe/H]>–1, this may be attributed to metallicity dependent Type II SN yields. Additionally, the strong rise in the average [Na/Fe] ratio for the RGB–Int2+3 and RGB–a stars seems inconsistent with Type Ia SNe production. The maximum [O/Fe] abundance also begins to decrease at [Fe/H]\gtrsim–1.2; however, this element, as well as Na and Al, may be altered by in situ mixing or pollution from sources other than Type II or Ia SNe. Therefore, the [X/Fe] ratios for these light elements may not directly trace SN production or even reflect a star’s original composition. We cannot explicitly rule out that Type Ia SNe have contributed to \omega Cen’s chemical enrichment, but it seems that their involvement has been mostly limited.

Unlike the heavy \alpha and Fe–peak elements, the light elements (O, Na, and Al) exhibit >0.5 dex star–to–star abundance variations at all metallicities. Although roughly half of the RGB–MP stars exhibit light element abundance patterns that are consistent with those found in similar metallicity halo and dwarf galaxy stars, the remaining RGB–MP stars, as well as >70\% of more metal–rich stars, show light element abundance patterns that are more similar to those found in individual globular clusters (i.e., O–poor and Na/Al–rich). Interestingly, the presence of these stars is a strong function of metallicity, and the [X/Fe] distribution functions are bimodal at intermediate metallicities. While very few O–poor, Na/Al–rich stars are found at [Fe/H]<–1.7, the majority of stars in the RGB–Int1 and subsequent populations exhibit these characteristics. We find that many of the metallicity dependent light element trends can be at least qualitatively reproduced by hot bottom burning in intermediate mass AGB stars. This is evidenced by the pervasive O–Na and O–Al anticorrelations and concurrent Na–Al correlation present in all stars with [Fe/H]\lesssim–1. Interestingly, the RGB–a stars no longer exhibit the light element relations and instead appear to have a roughly uniform composition. In any case, the light element trends in stars with [Fe/H]\lesssim–1 are similar to what is found in monometallic globular clusters, but the relative fraction of O–poor, Na/Al–rich stars in \omega Cen at [Fe/H]>–1.6 is significantly larger than those found in other globular clusters. Since a wide mass range of AGB stars seem able to reproduce the observed [Na/Fe] and [Al/Fe] trends but only a narrow range eject O–poor material, we conclude that the [Na/Fe] and [Al/Fe] ratios in the “enhanced” \omega Cen stars may be explained solely by pollution from intermediate mass AGB stars. However, the strongly depleted [O/Fe] ratios in many stars appear to require an additional process. Interestingly, we find a low incidence of carbon stars (<2\%) in our sample despite the large population of O–poor giants.

It seems that some degree of in situ processing must be invoked in order to interpret the large population of O–poor stars. We find an interesting parallel between O–poor giants and blue main sequence stars that may explain at least part of this phenomenon. The two populations share strikingly similar radial locations, metallicities, and number fractions. In particular, the O–poor and blue main sequence stars are both predominantly found inside \sim10\arcmin from the cluster center, are mostly found at intermediate metallicities, and constitute \sim30\% of the RGB and main sequence by number. Since the blue main sequence stars are believed to be He–rich, it seems likely that the O–poor stars may also be He–rich. Previous theoretical analyses of He–rich, globular cluster RGB stars predict that significant in situ mixing can occur more easily in He–rich compared to He–normal stars. Furthermore, it is predicted that the surface [O/Fe] abundance may be significantly depleted, but [Na/Fe] (and presumably [Al/Fe]) should be mostly unaffected. However, this scenario assumes that the O–poor, Na/Al–rich stars were already polluted by material that was moderately processed by proton–capture nucleosynthesis before ascending the RGB. The observed O–Si anticorrelation and Al–Si correlation may support this scenario. These relations can naturally arise due to leakage from the MgAl cycle at temperatures exceeding \sim65\times10{}^{\rm 6} K; temperatures this high are achieved in hot bottom burning conditions but not in the interiors of low mass RGB stars. If we assume that the observed O–poor stars are also He–rich and therefore more apt to experience in situ deep mixing, then this may explain why only the [O/Fe] ratio is correlated with radial location. Since the Na and Al abundances do not strongly correlate with radial location like O, this may be an indication that the stars responsible for producing the high Na and Al abundances do not necessarily produce high He yields as well.

A majority of RGB–MP stars have [La/Fe], [Eu/Fe], and [La/Eu] ratios indicating that the r–process was the primary production mechanism early in \omega Cen’s history. This is similar to what is found in metal–poor halo, globular cluster, and dwarf galaxy stars, and is consistent with a rapid formation time scale of the RGB–MP population. However, the [La/Fe], [Eu/Fe], and [La/Eu] abundance patterns indicate that the s–process became the dominant neutron–capture production mechanism at [Fe/H]>–1.6, and was active at a level above that observed in any other stellar population to date. In fact, almost no stars with [Fe/H]>–1.6 have [La/Fe]<0, and many stars in the intermediate metallicity populations exhibit [La/Eu] ratios suggesting pure s–process production. However, proper accounting of hyperfine structure in determining both La and Eu abundances has revised our [La/Eu] ratios downward from those in Johnson et al. (2009), and we now find that surface pollution from mass transfer is not generally required to explain the stars with large [La/Eu] ratios. Interestingly, the typical [Eu/Fe] abundances in the RGB–Int1 and RGB–Int2+3 stars are well below those observed in similar metallicity halo and globular cluster stars. This suggests that typical r–process production mechanisms may have been suppressed in \omega Cen.

While we find that both the [Ni/Fe] and [Eu/Fe] ratios are independent of a star’s [La/Fe] abundance, all other elements exhibit a mild correlation with [La/Fe]. Since the <3 M{}_{\sun} AGB stars believed to produce most of the s–process elements in \omega Cen are also predicted to produce some light elements, the correlation with La is not entirely unexpected. Interestingly, the O–rich stars show a correlation with [La/Fe], but the O–poor stars do not. This suggests that the O–depletion phenomenon is driven by an additional process, such as in situ mixing, that does not alter the envelope [La/Fe] ratio. With regard to the heavy \alpha elements, we suspect that the correlation with La may be due to the combined effects of Type II SNe producing \alpha elements and low/intermediate–mass AGB stars producing s–process elements at approximately the same time. This is supported by the observation that the rise in s–process and \alpha elements occurs in the same metallicity range. We do not have a definitive explanation for the correlation between [Ti/Fe] and [La/Fe] because the [Ti/Fe] ratio is not believed to be significantly enhanced in Type II SNe. However, we point out that the stable isotope {}^{\rm 50}Ti is a neutron magic nucleus and that the He shell of thermally pulsing AGB stars are predicted to exhibit large {}^{\rm 50}Ti/{}^{\rm 48}Ti ratios. Therefore, if at least some AGB stars that eject large amounts of s–process elements also eject material with a high {}^{\rm 50}Ti/{}^{\rm 48}Ti ratio, this may explain both the rise [Ti/Fe] at [Fe/H]\gtrsim–1.6 and the Ti–La correlation.

This publication makes use of data products from the Two Micron All Sky Survey, which is a joint project of the University of Massachusetts and the Infrared Processing and Analysis Center/California Institute of Technology, funded by the National Aeronautics and Space Administration and the National Science Foundation. This research has made use of NASA’s Astrophysics Data System Bibliographic Services. Support of the College of Arts and Sciences and the Daniel Kirkwood fund at Indiana University Bloomington for CIJ and CAP is gratefully acknowledged. We would like to thank Bob Kraft and Chris Sneden for many helpful discussions, Katia Cunha for sending an electronic version of her paper in advance of publication, and TalaWanda Monroe for her assistance in obtaining these observations. We would also like to thank Frank and Janet Winkler and the CTIO staff for their generous hospitality. We also thank the referee for his/her careful reading and thoughtful comments that led to improvement of the manuscript.

Appendix A Empirical Lanthanum 6774 Å Hyperfine Broadening Correction

The 6774 Å La II line is often measurable in the spectra of [Fe/H]\gtrsim–2 RGB and AGB stars, but accurate La abundance determinations from this line can be hampered by hyperfine broadening if the EW exceeds \sim50 mÅ. Unfortunately, we are not aware of any publicly available linelists that include log gf values for the individual hyperfine components of the 6774 Å line. However, the spectra used here and in Johnson et al. (2009) provide EW measurements of the 6774 Å line and spectrum synthesis abundance determinations from the 6262 Å line in 85 giants. Since the 6262 Å abundance determinations properly account for hyperfine broadening, we can use these data to derive an empirical correction factor for EW–based abundance measurements that use the 6774 Å line. In Figure 27, we plot [La/Fe]{}_{\rm syn}–[La/Fe]{}_{\rm EW} (\Delta[La/Fe]{}_{\rm EW}) as a function of EW. The least–squares fit to the data gives the empirical correction factor as,

 \Delta[La/Fe]_{EW}=[(5.0\times 10^{-6})(EW^{2})]-[(0.0068)(EW)]+0.1084(\sigma=% 0.07), (A1)

where the EW is measured in units of mÅ. This relation is qualitatively expected because it shows that a straight–forward EW analysis will overestimate the La abundance by an increasingly larger amount as one moves up the curve–of–growth to larger EWs. Since this is an empirical correction, it is difficult to predict how the relation might change outside the T{}_{\rm eff} (3800–5000 K), log g (\lesssim2), and metallicity (–2.5\lesssim[Fe/H]\lesssim–0.5) regime of our sample.

Facilities: CTIO

References

• Alonso et al. (1999) Alonso, A., Arribas, S., & Martínez-Roger, C. 1999, A&AS, 140, 261
• Alonso et al. (2001) Alonso, A., Arribas, S., & Martínez-Roger, C. 2001, A&A, 376, 1039
• Anders & Grevesse (1989) Anders, E., & Grevesse, N. 1989, Geochim. Cosmochim. Acta, 53, 197
• Anderson (1997) Anderson, A. J. 1997, Ph.D. Thesis
• Anderson (2002) Anderson, J. 2002, Omega Centauri, A Unique Window into Astrophysics, 265, 87
• Anderson et al. (2009) Anderson, J., Piotto, G., King, I. R., Bedin, L. R., & Guhathakurta, P. 2009, ApJ, 697, L58
• Andrievsky et al. (2008) Andrievsky, S. M., Spite, M., Korotin, S. A., Spite, F., Bonifacio, P., Cayrel, R., Hill, V., & François, P. 2008, A&A, 481, 481
• Arnett & Thielemann (1985) Arnett, W. D., & Thielemann, F.-K. 1985, ApJ, 295, 589
• Arnould et al. (2007) Arnould, M., Goriely, S., & Takahashi, K. 2007, Phys. Rep., 450, 97
• Bedin et al. (2004) Bedin, L. R., Piotto, G., Anderson, J., Cassisi, S., King, I. R., Momany, Y., & Carraro, G. 2004, ApJ, 605, L125
• Bekki & Norris (2006) Bekki, K., & Norris, J. E. 2006, ApJ, 637, L109
• Bekki et al. (2007) Bekki, K., Campbell, S. W., Lattanzio, J. C., & Norris, J. E. 2007, MNRAS, 377, 335
• Bellini et al. (2009) Bellini, A., et al. 2009a, A&A, 493, 959
• Bellini et al. (2009) Bellini, A., Piotto, G., Bedin, L. R., King, I. R., Anderson, J., Milone, A. P., & Momany, Y. 2009b, A&A, 507, 1393
• Bensby et al. (2003) Bensby, T., Feltzing, S., & Lundström, I. 2003, A&A, 410, 527
• Bensby et al. (2005) Bensby, T., Feltzing, S., Lundström, I., & Ilyin, I. 2005, A&A, 433, 185
• Bisterzo et al. (2010) Bisterzo, S., Gallino, R., Straniero, O., Cristallo, S., Kappeler, F. 2010, MNRAS, 404, 1529
• Blackwell & Shallis (1977) Blackwell, D. E., & Shallis, M. J. 1977, MNRAS, 180, 177
• Boehm-Vitense (1979) Boehm-Vitense, E. 1979, ApJ, 234, 521
• Bragaglia et al. (2010) Bragaglia, A., Carretta, E., Gratton, R., D’Orazi, V., Cassisi, S., & Lucatello, S. 2010, arXiv:1005.2659
• Brewer & Carney (2006) Brewer, M.-M., & Carney, B. W. 2006, AJ, 131, 431
• Brown & Wallerstein (1993) Brown, J. A., & Wallerstein, G. 1993, AJ, 106, 133
• Busso et al. (1999) Busso, M., Gallino, R., & Wasserburg, G. J. 1999, ARA&A, 37, 239
• Calamida et al. (2005) Calamida, A., et al. 2005, ApJ, 634, L69
• Calamida et al. (2009) Calamida, A., et al. 2009, ApJ, 706, 1277
• Cannon & Stobie (1973) Cannon, R. D., & Stobie, R. S. 1973, MNRAS, 162, 207
• Carretta et al. (2009) Carretta, E., et al. 2009a, A&A, 505, 117
• Carretta et al. (2009) Carretta, E., Bragaglia, A., Gratton, R., & Lucatello, S. 2009b, A&A, 505, 139
• Carretta et al. (2010) Carretta, E., et al. 2010, ApJ, 714, L7
• Castellani et al. (2007) Castellani, V., et al. 2007, ApJ, 663, 1021
• Castelli et al. (1997) Castelli, F., Gratton, R. G., & Kurucz, R. L. 1997, A&A, 318, 841
• Cavallo et al. (2004) Cavallo, R. M., Suntzeff, N. B., & Pilachowski, C. A. 2004, AJ, 127, 3411
• Charbonnel et al. (1998) Charbonnel, C., Brown, J. A., & Wallerstein, G. 1998, A&A, 332, 204
• Choi & Yi (2008) Choi, E., & Yi, S. K. 2008, MNRAS, 386, 1332
• Cohen (1981) Cohen, J. G. 1981, ApJ, 247, 869
• Cottrell & Da Costa (1981) Cottrell, P. L., & Da Costa, G. S. 1981, ApJ, 245, L79
• Cowan et al. (1991) Cowan, J. J., Thielemann, F.-K., & Truran, J. W. 1991, Phys. Rep., 208, 267
• Cunha et al. (2002) Cunha, K., Smith, V. V., Suntzeff, N. B., Norris, J. E., Da Costa, G. S., & Plez, B. 2002, AJ, 124, 379
• Cunha et al. (2010) Cunha, K., Smith, V. V., Bergemann, M., Suntzeff, N. B., & Lambert, D. L. 2010, ApJ, 717, 333
• D’Antona & Ventura (2007) D’Antona, F., & Ventura, P. 2007, MNRAS, 379, 1431
• D’Cruz et al. (2000) D’Cruz, N. L., et al. 2000, ApJ, 530, 352
• D’Ercole et al. (2008) D’Ercole, A., Vesperini, E., D’Antona, F., McMillan, S. L. W., & Recchi, S. 2008, MNRAS, 391, 825
• D’Ercole et al. (2010) D’Ercole, A., D’Antona, F., Ventura, P., Vesperini, E., & McMillan, S. L. W. 2010, MNRAS, 999
• Decressin et al. (2007) Decressin, T., Meynet, G., Charbonnel, C., Prantzos, N., & Ekström, S. 2007, A&A, 464, 1029
• Denisenkov & Denisenkova (1990) Denisenkov, P. A., & Denisenkova, S. N. 1990, Soviet Astronomy Letters, 16, 275
• Denissenkov & Herwig (2004) Denissenkov, P. A., & Herwig, F. 2004, ApJ, 612, 1081
• Denissenkov & Weiss (2004) Denissenkov, P. A., & Weiss, A. 2004, ApJ, 603, 119
• Dinescu et al. (1999) Dinescu, D. I., Girard, T. M., & van Altena, W. F. 1999, AJ, 117, 1792
• Edvardsson et al. (1993) Edvardsson, B., Andersen, J., Gustafsson, B., Lambert, D. L., Nissen, P. E., & Tomkin, J. 1993, A&A, 275, 101
• Fenner et al. (2004) Fenner, Y., Campbell, S., Karakas, A. I., Lattanzio, J. C., & Gibson, B. K. 2004, MNRAS, 353, 789
• Ferraro et al. (2004) Ferraro, F. R., Sollima, A., Pancino, E., Bellazzini, M., Straniero, O., Origlia, L., & Cool, A. M. 2004, ApJ, 603, L81
• Freeman & Rodgers (1975) Freeman, K. C., & Rodgers, A. W. 1975, ApJ, 201, L71
• Fulbright (2000) Fulbright, J. P. 2000, AJ, 120, 1841
• Fulbright et al. (2007) Fulbright, J. P., McWilliam, A., & Rich, R. M. 2007, ApJ, 661, 1152
• Gallino et al. (1994) Gallino, R., Raiteri, C. M., Busso, M., & Matteucci, F. 1994, ApJ, 430, 858
• Gehren et al. (2004) Gehren, T., Liang, Y. C., Shi, J. R., Zhang, H. W., & Zhao, G. 2004, A&A, 413, 1045
• Geisler et al. (2007) Geisler, D., Wallerstein, G., Smith, V. V., & Casetti-Dinescu, D. I. 2007, PASP, 119, 939
• Girardi et al. (2007) Girardi, L., Castelli, F., Bertelli, G., & Nasi, E. 2007, A&A, 468, 657
• Gnedin et al. (2002) Gnedin, O. Y., Zhao, H., Pringle, J. E., Fall, S. M., Livio, M., & Meylan, G. 2002, ApJ, 568, L23
• Gratton et al. (1999) Gratton, R. G., Carretta, E., Eriksson, K., & Gustafsson, B. 1999, A&A, 350, 955
• Gratton et al. (2004) Gratton, R., Sneden, C., & Carretta, E. 2004, ARA&A, 42, 385
• Harris (1996) Harris, W. E. 1996, AJ, 112, 1487
• Herwig (2005) Herwig, F. 2005, ARA&A, 43, 435
• Hilker & Richtler (2000) Hilker, M., & Richtler, T. 2000, A&A, 362, 895
• Hill et al. (2002) Hill, V., et al. 2002, A&A, 387, 560
• Hinkle et al. (2000) Hinkle, K., Wallace, L., Valenti, J., & Harmer, D. 2000, Visible and Near Infrared Atlas of the Arcturus Spectrum 3727-9300 Å ed. Kenneth Hinkle, Lloyd Wallace, Jeff Valenti, and Dianne Harmer. (San Francisco: ASP) ISBN: 1-58381-037-4
• Hughes & Wallerstein (2000) Hughes, J., & Wallerstein, G. 2000, AJ, 119, 1225
• Ikuta & Arimoto (2000) Ikuta, C., & Arimoto, N. 2000, A&A, 358, 535
• Ivans et al. (1999) Ivans, I. I., Sneden, C., Kraft, R. P., Suntzeff, N. B., Smith, V. V., Langer, G. E., & Fulbright, J. P. 1999, AJ, 118, 1273
• Ivans et al. (2001) Ivans, I. I., Kraft, R. P., Sneden, C., Smith, G. H., Rich, R. M., & Shetrone, M. 2001, AJ, 122, 1438
• Izzard et al. (2007) Izzard, R. G., Lugaro, M., Karakas, A. I., Iliadis, C., & van Raai, M. 2007, A&A, 466, 641
• Johnson et al. (2005) Johnson, C. I., Kraft, R. P., Pilachowski, C. A., Sneden, C., Ivans, I. I., & Benman, G. 2005, PASP, 117, 1308
• Johnson et al. (2008) Johnson, C. I., Pilachowski, C. A., Simmerer, J., & Schwenk, D. 2008, ApJ, 681, 1505
• Johnson et al. (2009) Johnson, C. I., Pilachowski, C. A., Michael Rich, R., & Fulbright, J. P. 2009, ApJ, 698, 2048
• Karakas (2010) Karakas, A. I. 2010, MNRAS, 403, 1413
• Kayser et al. (2006) Kayser, A., Hilker, M., Richtler, T., & Willemsen, P. G. 2006, A&A, 458, 777
• Kraft (1994) Kraft, R. P. 1994, PASP, 106, 553
• Kupka et al. (2000) Kupka, F. G., Ryabchikova, T. A., Piskunov, N. E., Stempels, H. C., & Weiss, W. W. 2000, Baltic Astronomy, 9, 590
• Langer et al. (1993) Langer, G. E., Hoffman, R., & Sneden, C. 1993, PASP, 105, 301
• Langer et al. (1997) Langer, G. E., Hoffman, R. E., & Zaidins, C. S. 1997, PASP, 109, 244
• Lawler et al. (2001) Lawler, J. E., Bonvallet, G., & Sneden, C. 2001a, ApJ, 556, 452
• Lawler et al. (2001) Lawler, J. E., Wickliffe, M. E., den Hartog, E. A., & Sneden, C. 2001b, ApJ, 563, 1075
• Lecureur et al. (2007) Lecureur, A., Hill, V., Zoccali, M., Barbuy, B., Gómez, A., Minniti, D., Ortolani, S., & Renzini, A. 2007, A&A, 465, 799
• Lee et al. (1999) Lee, Y.-W., Joo, J.-M., Sohn, Y.-J., Rey, S.-C., Lee, H.-C., & Walker, A. R. 1999, Nature, 402, 55
• Lee et al. (2005) Lee, Y.-W., et al. 2005a, ApJ, 621, L57
• Lee et al. (2005) Lee, J.-W., Carney, B. W., & Habgood, M. J. 2005b, AJ, 129, 251
• Lee et al. (2009) Lee, J.-W., Kang, Y.-W., Lee, J., & Lee, Y.-W. 2009, Nature, 462, 480
• Lee (2010) Lee, J.-W. 2010, MNRAS, 405, L36
• Magain (1984) Magain, P. 1984, A&A, 134, 189
• Majewski et al. (2000) Majewski, S. R., Ostheimer, J. C., Kunkel, W. E., & Patterson, R. J. 2000, AJ, 120, 2550
• Mandushev et al. (1991) Mandushev, G., Staneva, A., & Spasova, N. 1991, A&A, 252, 94
• Marcolini et al. (2007) Marcolini, A., Sollima, A., D’Ercole, A., Gibson, B. K., & Ferraro, F. R. 2007, MNRAS, 382, 443
• Marino et al. (2008) Marino, A. F., Villanova, S., Piotto, G., Milone, A. P., Momany, Y., Bedin, L. R., & Medling, A. M. 2008, A&A, 490, 625
• Marino et al. (2010) Marino, A. F., Piotto, G., Gratton, R., Milone, A. P., Zoccali, M., Bedin, L. R., Villanova, S., & Bellini, A. 2010, IAU Symposium, 268, 183
• Mashonkina et al. (2000) Mashonkina, L. I., Shimanskiĭ, V. V., & Sakhibullin, N. A. 2000, Astronomy Reports, 44, 790
• Mathews & Cowan (1990) Mathews, G. J., & Cowan, J. J. 1990, Nature, 345, 491
• Matteucci et al. (1993) Matteucci, F., Raiteri, C. M., Busson, M., Gallino, R., & Gratton, R. 1993, A&A, 272, 421
• Mayor et al. (1996) Mayor, M., Duquennoy, A., Udry, S., Andersen, J., & Nordstrom, B. 1996, The Origins, Evolution, and Destinies of Binary Stars in Clusters, 90, 190
• McCall (2004) McCall, M. L. 2004, AJ, 128, 2144
• McDonald et al. (2009) McDonald, I., van Loon, J. T., Decin, L., Boyer, M. L., Dupree, A. K., Evans, A., Gehrz, R. D., & Woodward, C. E. 2009, MNRAS, 394, 831
• McWilliam & Rich (1994) McWilliam, A., & Rich, R. M. 1994, ApJS, 91, 749
• McWilliam (1997) McWilliam, A. 1997, ARA&A, 35, 503
• McWilliam & Smecker-Hane (2005) McWilliam, A., & Smecker-Hane, T. A. 2005, ApJ, 622, L29
• Meylan et al. (1995) Meylan, G., Mayor, M., Duquennoy, A., & Dubath, P. 1995, A&A, 303, 761
• Milone et al. (2008) Milone, A. P., et al. 2008, ApJ, 673, 241
• Milone et al. (2010) Milone, A. P., et al. 2010, ApJ, 709, 1183
• Moretti et al. (2009) Moretti, A., et al. 2009, A&A, 493, 539
• Nomoto et al. (1997) Nomoto, K., Iwamoto, K., Nakasato, N., Thielemann, F.-K., Brachwitz, F., Tsujimoto, T., Kubo, Y., & Kishimoto, N. 1997, Nuclear Physics A, 621, 467
• Nomoto et al. (2006) Nomoto, K., Tominaga, N., Umeda, H., Kobayashi, C., & Maeda, K. 2006, Nuclear Physics A, 777, 424
• Norris & Da Costa (1995) Norris, J. E., & Da Costa, G. S. 1995, ApJ, 447, 680
• Norris et al. (1996) Norris, J. E., Freeman, K. C., & Mighell, K. J. 1996, ApJ, 462, 241
• Norris et al. (1997) Norris, J. E., Freeman, K. C., Mayor, M., & Seitzer, P. 1997, ApJ, 487, L187
• Norris (2004) Norris, J. E. 2004, ApJ, 612, L25
• Origlia et al. (2003) Origlia, L., Ferraro, F. R., Bellazzini, M., & Pancino, E. 2003, ApJ, 591, 916
• Paltoglou & Norris (1989) Paltoglou, G., & Norris, J. E. 1989, ApJ, 336, 185
• Pancino et al. (2000) Pancino, E., Ferraro, F. R., Bellazzini, M., Piotto, G., & Zoccali, M. 2000, ApJ, 534, L83
• Pancino et al. (2002) Pancino, E., Pasquini, L., Hill, V., Ferraro, F. R., & Bellazzini, M. 2002, ApJ, 568, L101
• Pancino et al. (2003) Pancino, E., Seleznev, A., Ferraro, F. R., Bellazzini, M., & Piotto, G. 2003, MNRAS, 345, 683
• Pancino et al. (2007) Pancino, E., Galfo, A., Ferraro, F. R., & Bellazzini, M. 2007, ApJ, 661, L155
• Piotto et al. (2005) Piotto, G., et al. 2005, ApJ, 621, 777
• Piotto et al. (2007) Piotto, G., et al. 2007, ApJ, 661, L53
• Piotto (2009) Piotto, G. 2009, arXiv:0902.1422
• Prantzos et al. (2007) Prantzos, N., Charbonnel, C., & Iliadis, C. 2007, A&A, 470, 179
• Press et al. (1992) Press, W. H., Teukolsky, S. A., Vetterling, W. T., & Flannery, B. P. 1992, Cambridge: University Press, —c1992, 2nd ed.,
• Prochaska & McWilliam (2000) Prochaska, J. X., & McWilliam, A. 2000, ApJ, 537, L57
• Ralchenko et al. (2008) Ralchenko, Y., Kramida, A.E., Reader, J., and NIST ASD Team (2008).NIST Atomic Spectra Database (version 3.1.5). National Institute of Standards and Technology, Gaithersburg, MD.
• Ramírez & Cohen (2002) Ramírez, S. V., & Cohen, J. G. 2002, AJ, 123, 3277
• Reddy et al. (2003) Reddy, B. E., Tomkin, J., Lambert, D. L., & Allende Prieto, C. 2003, MNRAS, 340, 304
• Reddy et al. (2006) Reddy, B. E., Lambert, D. L., & Allende Prieto, C. 2006, MNRAS, 367, 1329
• Reijns et al. (2006) Reijns, R. A., Seitzer, P., Arnold, R., Freeman, K. C., Ingerson, T., van den Bosch, R. C. E., van de Ven, G., & de Zeeuw, P. T. 2006, A&A, 445, 503
• Renzini (2008) Renzini, A. 2008, MNRAS, 391, 354
• Rey et al. (2004) Rey, S.-C., Lee, Y.-W., Ree, C. H., Joo, J.-M., Sohn, Y.-J., & Walker, A. R. 2004, AJ, 127, 958
• Richer et al. (1991) Richer, H. B., Fahlman, G. G., Buonanno, R., Fusi Pecci, F., Searle, L., & Thompson, I. B. 1991, ApJ, 381, 147
• Romano et al. (2007) Romano, D., Matteucci, F., Tosi, M., Pancino, E., Bellazzini, M., Ferraro, F. R., Limongi, M., & Sollima, A. 2007, MNRAS, 376, 405
• Romano et al. (2010) Romano, D., Tosi, M., Cignoni, M., Matteucci, F., Pancino, E., & Bellazzini, M. 2010, MNRAS, 401, 2490
• Salaris et al. (2002) Salaris, M., Cassisi, S., & Weiss, A. 2002, PASP, 114, 375
• Samland (1998) Samland, M. 1998, ApJ, 496, 155
• Sbordone et al. (2007) Sbordone, L., Bonifacio, P., Buonanno, R., Marconi, G., Monaco, L., & Zaggia, S. 2007, A&A, 465, 815
• Schaller et al. (1992) Schaller, G., Schaerer, D., Meynet, G., & Maeder, A. 1992, A&AS, 96, 269
• Shetrone & Keane (2000) Shetrone, M. D., & Keane, M. J. 2000, AJ, 119, 840
• Shetrone et al. (2001) Shetrone, M. D., Côté, P., & Sargent, W. L. W. 2001, ApJ, 548, 592
• Shetrone et al. (2003) Shetrone, M., Venn, K. A., Tolstoy, E., Primas, F., Hill, V., & Kaufer, A. 2003, AJ, 125, 684
• Simmerer et al. (2004) Simmerer, J., Sneden, C., Cowan, J. J., Collier, J., Woolf, V. M., & Lawler, J. E. 2004, ApJ, 617, 1091
• Skrutskie et al. (2006) Skrutskie, M. F., et al. 2006, AJ, 131, 1163
• Smith et al. (2000) Smith, V. V., Suntzeff, N. B., Cunha, K., Gallino, R., Busso, M., Lambert, D. L., & Straniero, O. 2000, AJ, 119, 1239
• Smith et al. (2002) Smith, V. V., Terndrup, D. M., & Suntzeff, N. B. 2002, ApJ, 579, 832
• Sneden (1973) Sneden, C. 1973, ApJ, 184, 839
• Sneden et al. (1991) Sneden, C., Kraft, R. P., Prosser, C. F., & Langer, G. E. 1991a, AJ, 102, 2001
• Sneden et al. (1991) Sneden, C., Gratton, R. G., & Crocker, D. A. 1991b, A&A, 246, 354
• Sneden et al. (1996) Sneden, C., McWilliam, A., Preston, G. W., Cowan, J. J., Burris, D. L., & Armosky, B. J. 1996, ApJ, 467, 819
• Sneden et al. (1997) Sneden, C., Kraft, R. P., Shetrone, M. D., Smith, G. H., Langer, G. E., & Prosser, C. F. 1997, AJ, 114, 1964
• Sneden et al. (2004) Sneden, C., Kraft, R. P., Guhathakurta, P., Peterson, R. C., & Fulbright, J. P. 2004, AJ, 127, 2162
• Sneden et al. (2008) Sneden, C., Cowan, J. J., & Gallino, R. 2008, ARA&A, 46, 241
• Sollima et al. (2005) Sollima, A., Ferraro, F. R., Pancino, E., & Bellazzini, M. 2005, MNRAS, 357, 265
• Sollima et al. (2006) Sollima, A., Borissova, J., Catelan, M., Smith, H. A., Minniti, D., Cacciari, C., & Ferraro, F. R. 2006, ApJ, 640, L43
• Sollima et al. (2007) Sollima, A., Ferraro, F. R., Bellazzini, M., Origlia, L., Straniero, O., & Pancino, E. 2007, ApJ, 654, 915
• Sollima et al. (2009) Sollima, A., Bellazzini, M., Smart, R. L., Correnti, M., Pancino, E., Ferraro, F. R., & Romano, D. 2009, MNRAS, 396, 2183
• Stanford et al. (2006) Stanford, L. M., Da Costa, G. S., Norris, J. E., & Cannon, R. D. 2006, ApJ, 647, 1075
• Stanford et al. (2010) Stanford, L. M., Da Costa, G. S., & Norris, J. E. 2010, ApJ, 714, 1001
• Straniero et al. (2006) Straniero, O., Gallino, R., & Cristallo, S. 2006, Nuclear Physics A, 777, 311
• Suntzeff & Kraft (1996) Suntzeff, N. B., & Kraft, R. P. 1996, AJ, 111, 1913
• Sweigart et al. (1974) Sweigart, A. V., Mengel, J. G., & Demarque, P. 1974, A&A, 30, 13
• Sweigart & Mengel (1979) Sweigart, A. V., & Mengel, J. G. 1979, ApJ, 229, 624
• Thevenin (1990) Thevenin, F. 1990, A&AS, 82, 179
• Thielemann & Arnett (1985) Thielemann, F. K., & Arnett, W. D. 1985, ApJ, 295, 604
• Timmes et al. (1995) Timmes, F. X., Woosley, S. E., & Weaver, T. A. 1995, ApJS, 98, 617
• Tomkin et al. (1992) Tomkin, J., Lemke, M., Lambert, D. L., & Sneden, C. 1992, AJ, 104, 1568
• Tsujimoto & Shigeyama (2003) Tsujimoto, T., & Shigeyama, T. 2003, ApJ, 590, 803
• van de Ven et al. (2006) van de Ven, G., van den Bosch, R. C. E., Verolme, E. K., & de Zeeuw, P. T. 2006, A&A, 445, 513
• van Leeuwen et al. (2000) van Leeuwen, F., Le Poole, R. S., Reijns, R. A., Freeman, K. C., & de Zeeuw, P. T. 2000, A&A, 360, 472
• van Loon et al. (2007) van Loon, J. T., van Leeuwen, F., Smalley, B., Smith, A. W., Lyons, N. A., McDonald, I., & Boyer, M. L. 2007, MNRAS, 382, 1353
• Venn et al. (2004) Venn, K. A., Irwin, M., Shetrone, M. D., Tout, C. A., Hill, V., & Tolstoy, E. 2004, AJ, 128, 1177
• Ventura & D’Antona (2005) Ventura, P., & D’Antona, F. 2005, A&A, 431, 279
• Ventura & D’Antona (2009) Ventura, P., & D’Antona, F. 2009, A&A, 499, 835
• Villanova et al. (2007) Villanova, S., et al. 2007, ApJ, 663, 296
• Villanova et al. (2010) Villanova, S., Carraro, G., Scarpa, R., & Marconi, G. 2010, New A, 15, 520
• Wheeler et al. (1998) Wheeler, J. C., Cowan, J. J., & Hillebrandt, W. 1998, ApJ, 493, L101
• Woolley (1966) Woolley, R. R. 1966, Royal Observatory Annals, 2, 1
• Woosley & Weaver (1995) Woosley, S. E., & Weaver, T. A. 1995, ApJS, 101, 181
• Yong et al. (2005) Yong, D., Grundahl, F., Nissen, P. E., Jensen, H. R., & Lambert, D. L. 2005, A&A, 438, 875
• Yoshii et al. (1996) Yoshii, Y., Tsujimoto, T., & Nomoto, K. 1996, ApJ, 462, 266
• Zhang et al. (2008) Zhang, H. W., Gehren, T., & Zhao, G. 2008, A&A, 481, 489
• Zucker et al. (1996) Zucker, D., Wallerstein, G., & Brown, J. A. 1996, PASP, 108, 911