THE C+N+O ABUNDANCE OF ω CENTAURI GIANT STARS: IMPLICATIONS FOR THE CHEMICAL-ENRICHMENT SCENARIO AND THE RELATIVE AGES OF DIFFERENT STELLAR POPULATIONS^*

Omega Centauri (ω Cen) is one of the most intriguing globular clusters (GCs) of the Galaxy. At odds with the majority of GCs, which are mono-metallic, it shows large star-to-star metallicity variations up to more than 1 dex (e.g., Norris et al. 1996; Suntzeff & Kraft 1996). At the same time, it shares with the most mono-metallic GCs the presence of a Na–O anticorrelation, which is present across almost the entire metallicity range (Marino et al. 2011a; Johnson & Pilachowski 2010). These two facts suggest unique complexity in the cluster's star formation history and its chemical evolution.

The complexity of ω Cen also manifests itself in its color–magnitude diagram (CMD) with the presence of multiple red-giant branches (RGBs), multiple sub-giant branches (SGBs; Lee et al. 1999; Pancino et al. 2000), and an extended horizontal branch (Villanova et al. 2007; Cassisi et al. 2009; D'Antona et al. 2010; Bellini et al. 2010). The main sequence (MS) is double (Anderson 1998), and a third, less-populated MS has been discovered by Bedin et al. (2004) and associated with the most metal-rich population. The bluer MS (bMS) has a higher metallicity than the red MS (rMS; Piotto et al. 2005). So far, the only way to account for these observations is to assume the bMS to be highly enhanced in He (Norris 2004; Bedin et al. 2004; Piotto et al. 2005). Direct evidence in these directions has been recently detected from the analysis of the He i 10830 transition on ω Cen RGB stars (Dupree et al. 2011). The correlation of the He line detection with [Fe/H], Al, and Na supports the assumption that helium is enhanced in the bMS.

Due to the observational scenario, being more complex than in any other GC, it has often been suggested (e.g., Bekki & Freeman 2003) that ω Cen may be the remnant of a now-dissolved dwarf galaxy, once similar to the Sagittarius dwarf (with its central GC M54) now being torn apart by the Galactic tidal field. In any case, a successful description of the ω Cen star formation history should be able to explain both the Na–O anticorrelation at the various metallicities and the general rise in slow-process (s-process) elements as a function of metallicity (Norris & Da Costa 1995; Johnson & Pilachowski 2010; Smith et al. 2000; Marino et al. 2011a). The presence of the Na–O anticorrelation implies that ω Cen, similar to mono-metallic GCs, has experienced enrichment from high-temperature H-burning processed material. At the same time, an additional physical mechanism should be present to produce s-process elements. In the Sun, the s-process elements' abundance is mainly due to two components: the main component (attributed to low-mass AGB stars of 1.5–3 M_☉; Busso et al. 1999) and the weak component (attributed to massive stars; see Raiteri et al. 1993 and references therein). At solar metallicity, this last component mainly produces the lighter s-process nuclei, at most up to Sr and Kr (Raiteri et al. 1993). In the case of ω Cen, if we assume that the s-elements are produced in the strong component, i.e., in a population of less massive asymptotic-giant-branch (AGB) stars (Busso et al. 1999; Ventura et al. 2009), we have to face a timescale discrepancy that makes it hard to solve the entire puzzle (D'Antona et al. 2011). Indeed, these AGBs have longer lifetimes than the more massive AGB stars, which are presumed responsible for the He enrichment and for Na–O anticorrelations. If, alternatively, the s-process elements are produced in the weak component by massive stars that explode as Type II supernovae, it remains to be seen whether the models can produce elements as heavy as Ba and La in stellar environments characterized by the chemical abundances observed in ω Cen.

An important ingredient to understand the star formation history of this poorly understood GC is the determination of the relative ages of its stellar populations. Numerous studies based on CMD analysis combined with metallicity distribution of turnoff (TO) and SGB stars have yielded conflicting results, suggesting age differences from less than 2 Gyr (Lee et al. 2005; Sollima et al. 2005; Calamida et al. 2009), up to 5 Gyr (Villanova et al. 2007). Rapid-formation scenarios wherein the entire cluster could have formed within a few times 10⁸ years have recently been suggested by D'Antona et al. (2011) and Valcarce & Catelan (2011).

Theoretical isochrones show that a variation of the total C+N+O abundance can have an effect on the GC ages obtained from CMD fitting. Therefore, the relative age dating of the stellar populations hosted in the cluster would be severely affected by the presence of C+N+O differences among the ω Cen sub-populations (Cassisi et al. 2008; Ventura et al. 2009; Pietrinferni et al. 2009; D'Antona et al. 2009).

In an effort to shed light on these issues, in this paper we measure the overall C+N+O abundance in 77 RGB stars of ω Cen, which span nearly its entire metallicity range. The layout of this paper is as follows. In Section 2 we describe the data analysis; results are presented in Section 3, and their impact on the chemical enrichment scenario and the age measurements are discussed in Sections 4 and 5, respectively; Section 6 is a summary of our results.

Our data set consists in a sample of 77 RGB stars observed with the FLAMES/GIRAFFE HR4 setup (program: 082.D-0424A). We also have, from Marino et al. (2011a), additional spectra for the same stars with different GIRAFFE setups, from which we have derived abundances for Fe, Na, O, and n-capture element La (75% s-process in the solar system; Simmerer et al. 2004). In Marino et al. (2011a) we provide measurements also for Ba abundances, but here we prefer to use only La as representative of n-capture elements because Ba measurements are more uncertain, since the only analyzed Ba transition is a blend (see Marino et al. 2011a). We refer the reader to this paper for a more detailed description of the sample and the data reduction. Here, we analyze, for 77 stars of the Marino et al. (2011a) sample, chemical abundances for carbon and nitrogen.

The HR4 setup covers the spectral range from ∼4188 to ∼4392 Å, providing a resolution R ∼ 20,000. The typical signal-to-noise ratio of the final combined spectra at the central wavelength of the spectral range is ∼80.

Our abundance analysis used the local thermodynamic equilibrium analysis code MOOG (Sneden 1973). Carbon was measured from spectral synthesis of the G bandheads (A²Δ–X²Π) near 4314 and 4323 Å. The nitrogen abundance was derived from synthesis of the CN blue-system (B²Σ–X²Σ) bandhead at ∼4215 Å. The synthesis line list for the blue CN band is described in Hill et al. (2002). The line list for the CH band was provided by B. Plez (2008, private communication). Synthetic spectra were employed by using Castelli & Kurucz (2004) models for effective temperatures (T_eff), gravities (log g), metallicities, and microturbulences (ξ_t) determined in Marino et al. (2011a). In computing the C abundance we used the previously determined O contents (Marino et al. 2011a), and for N, both observed C and O abundances needed to be employed. As an example, we show in Figure 1 the spectral synthesis for the CH and CN bands, and the O line for star 248664. For more details on the atmospheric-parameter determinations, and a discussion on related uncertainties, we refer to Marino et al. (2011a). Here, we focus on the C and N measurements.

Zoom In Zoom Out Reset image size

The error analysis for C and N was performed by varying T_eff, log g, metallicity, and ξ_t for several stars, and redetermining the abundances. The parameters were varied by ΔT_eff = ±100 K, Δlog g = ±0.20, Δξ_t = ±0.10 km s⁻¹, and metallicity by 0.10 dex. In addition, the derived C abundance is dependent on the O abundance and therefore so is the N abundance, and the circulation of correlated errors tends to mimic the effects of CNO-cycle burning (e.g., Iben & Renzini 1983). Indeed, in the red giant atmospheres carbon is mainly present in the CO molecules, and in the molecular equilibrium an overestimation in oxygen produces an overestimation of carbon from the CH band, and vice versa. In the CN bands, a carbon overestimation causes nitrogen to be underestimated. In the treatment of the internal errors for carbon, we then varied the oxygen abundance by ±0.10 and repeated the spectrum synthesis to determine the error introduced by uncertainties in the O abundances. These errors were then added in quadrature to the errors introduced by atmospheric parameters and yielded an overall error of ∼ ±0.10 dex to the [C/Fe] abundances. For nitrogen we varied both O and C, the latter by ±0.10 dex. We added these errors in quadrature with those introduced by the model atmosphere and estimated the internal uncertainty of the [N/Fe] values to be ∼0.20–0.30.

Carbon and nitrogen abundances for ω Cen RGB stars were determined by Brown & Wallerstein (1993; 6 stars), Norris & Da Costa (1995; 40 stars), and Stanford et al. (2010; 33 stars). A proper comparison with these studies cannot be made because there are no stars in common; however, we note that our C and N values span a range similar to that of Brown & Wallerstein (1993) and Stanford et al. (2010). Possible systematic differences in C could not be excluded between our stars and those of Stanford et al., while systematically lower N abundances (of ∼0.50 dex) were found by Norris & Da Costa (1995).

In Marino et al. (2011a), we derived the O abundance from the line at 6300 Å by comparing our observations with synthetic spectra that had solar C and N. Here, we redetermine oxygen by adopting in the synthesis the measured N and C abundances. These improved [O/Fe] measures are in good agreement with the previous measurements, with only few stars differing by more than 0.1 dex from the values in Marino et al. (2011a).

Figure 2 plots the abundances of C, N, and O (Table 1) measured in this paper and the Na measurements from Marino et al. (2011a). We observe large star-to-star variations in all of these elements, as has already observed in all the GCs studied to date. Carbon spans a range from [C/Fe] ∼−0.9 to [C/Fe] ∼+0.6. Most stars are C-depleted, but there is a population of stars with [C/Fe] > 0.0. Nitrogen spans almost 2 dex, from [N/Fe] ∼ 0.0 to stars with [N/Fe] ∼ 1.8 dex. There is only one star (214842) that is enhanced in both C and N (Figure 2, panel c).

Zoom In Zoom Out Reset image size

Our results show the existence of well-defined patterns among C, N, O, and Na. A positive correlation is present between O and C, and between Na and N, and N anticorrelates with O (Figure 2, panels (a), (b), (d)). A C–N anticorrelation is not obvious (panel c), but it can be more easily recognizable by grouping stars on the basis of different Fe content, as will be discussed in Section 3.1. Such chemical patterns indicate the occurrence of high-temperature H-burning involving the CNO and NeNa cycles, and are consistent with prediction in Ventura & D'Antona (2009).

In the following sections, we present the pattern of C, N, O abundances with [Fe/H] (Section 3.1) and with respect to the position of stars on the Na–O anticorrelation (Section 3.2) studied by Marino et al. (2011a).

3.1. CNO Pattern with [Fe/H]

Before discussing in detail the C, N, O patterns, we recall that ω Cen exhibits a large spread in [Fe/H], ranging from ∼ − 2.0 to ∼ − 0.7 dex. The iron distribution is multi-modal with distinct peaks at [Fe/H] ∼−1.75, −1.60, −1.45, −1.00, and there is a broad distribution of stars extending up to [Fe/H] ∼−0.70 (Marino et al. 2011a). Although the existence of these multiple populations is clear, observational errors often prevent us from definitively establishing which individual stars belong to which population. Keeping this in mind, Marino et al. (2011a) defined six groups of stars with different iron abundances to investigate the Na–O anticorrelation for different metallicity groups (see their Figure 8).

In the lower panels of Figure 2, we perform a similar analysis for C versus N in the same metallicity bins defined in Marino et al. (2011a). Unfortunately, we do not have C and N measurements for the most metal-rich group, with [Fe/H] >−0.90, of Marino et al. (2011a). We see that the C–N anticorrelation is present across the entire range of metallicities. The dashed-black line in each panel (that is the same as in panel c) has been traced to arbitrarily separate the C-poor/N-rich population from the C-rich/N-poor one. The red dot-dashed line is the "best-fit" line of constant [C/Fe]+[N/Fe] for the mid-metallicity stars, that has been taken equal to 0.80. It is worth noting that when we compare the relative location along the C–N anticorrelation with this line, it becomes evident that the average C abundance increases from lower to higher metallicities. The shape of the C–N anticorrelation for different metallicities is similar to that of the Na–O anticorrelation (Figure 8 of Marino et al. 2011a), with the most metal-rich groups hosting a larger fraction of N-rich (Na-rich, O–C-poor) stars. This behavior leads few stars, belonging to the metal richer group (e.g., 214842), to be enhanced both in C and N.

In the upper panels of Figure 3, we plot the individual carbon, nitrogen, and oxygen abundances against iron. While [N/Fe] exhibits a mild correlation with [Fe/H], there is no clear correlation either between [C/Fe] and [Fe/H] or between [O/Fe] and [Fe/H]. A significant spread in light elements is present at all metallicities with [C/Fe], [N/Fe], and [O/Fe] spanning large ranges (more than 1 dex).

Zoom In Zoom Out Reset image size

If, however, we instead plot the total [C+N+O] against [Fe/H], we find a correlation with a Pearson coefficient of 0.71. This is shown in panel (d) for the abundance relative to iron and in panel (e) for the total abundance. The red points show the averages over bins in [Fe/H] defined above. The vertical error bars correspond to the error in the mean related to the observed dispersion and the horizontal bars denote the boundaries of the [Fe/H] bins.

The representative error bar shown in the lower right of panels (d) and (e) comes from adding the individual C, N, and O errors in quadrature, and indicates that the error in the total [(C+N+O)/Fe] should be quite large (∼0.3 dex). When we estimated the errors in these abundances, we made the assumption that the abundances were statistically independent. Given, though, that the C, N, and O abundances are correlated with each other, a more realistic estimate of the error might be derived empirically from the rms of the [(C+N+O)/Fe] within each metallicity bin. In this case, we find an rms ranging from 0.10 to 0.15 dex, and stars with intermediate metallicity (−1.65 ⩽ [Fe/H] <−1.05) have the largest rms.

The bottom panels of Figure 3 show the trends we have derived for ω Cen in the context of other clusters. M22 has stars with a range of metallicities (Marino et al. 2009), and these stars are shown as the green points, which happen to follow the ω Cen trend almost perfectly. On the other hand, the mono-metallic clusters for which we have data (NGC 6397, NGC 6752, M4, and 47 Tuc) do not follow this trend and, in general, show no discernible trend of [(C+N+O)/Fe] with metallicity.

3.2. CNO, Na, and La Abundances in the Na–O Anticorrelation

To investigate how the studied chemical abundances behave for stars occupying a different location on the Na–O plane, we divided stars into two groups based on their location along the Na–O trend (Marino et al. 2011a). In normal mono-metallic clusters different generations of stars can be segregated in this way, and the Na–O groupings tend to have different C and N abundances, as can be seen in the case of M4 (Marino et al. 2008, see their Figure 10). The situation in ω Cen has an added level of complexity, since the large variation in metallicity does not allow us to simply identify two (or more) populations via Na and O. Both Johnson & Pilachowski (2010) and Marino et al. (2011a) have shown that stars of both first and second generations are present across a large range of metallicities. Note that here we extend the same nomenclature used for mono-metallic GCs to ω Cen and name first and second generation Na-poor/O-rich and Na-rich/O-poor stars, respectively, whatever their [Fe/H]. Of course, in the case of ω Cen this designation should not be taken literally, as stars of both groups are present in a large range in metallicity. We will therefore first follow the chemical patterns for the elements affected by p-captures (C, N, O, Na), n-capture elements (La), and C+N+O for the first generation alone, and then will examine the second generation.

In Figure 4, we represent in different colors our selected first- and second-generation stars on the Na–O anticorrelation from Marino et al. (2011a). The Na-poor/O-rich first-generation (green open triangles) and the Na-rich/O-poor second-generation stars (magenta open squares) have been arbitrarily selected by the dashed line. Similarly to that done in the lower panels of Figure 2, stars have been represented in different panels depending on their [Fe/H] bin. In each panel, the gray crosses represent the entire sample. The so selected first- and second-generation stars have been plotted in an O–La plane, as shown in Figure 5 (left panel). It is worth noting that in first-generation stars [O/Fe] abundances correlate with [La/Fe], with the Pearson coefficient equal to 0.73. To quantify this correlation we have determined the mean [O/Fe] abundances for three La intervals spanned by first-generation stars, whose sizes are indicated by the horizontal dark-green lines in Figure 5. The mean [O/Fe] values for the different Fe and La intervals are listed in Table 2, together with the rms and associated errors. In first stellar generation, some hints for a correlation may be present also between [O/Fe] and [Fe/H], as shown in the middle panel of Figure 5. In this case, the O rise is more difficult to be claimed over observational errors and the mean [O/Fe] values in the selected metallicity bins (shown in dark-green in the middle panel of Figure 5) agree within a 3σ values.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 2. Oxygen Mean Content for First Generation Stars in the Fe Bins of Marino et al. (2011a) and for Three Intervals in [La/Fe]

	[Fe/H] Range						[La/Fe] Range
	< − 1.90	−1.90/− 1.65	−1.65/− 1.50	−1.50/− 1.05	−1.05/− 0.95	> − 0.95	< − 0.30	−0.30/0.30	>0.30
[O/Fe]	0.41	0.45	0.50	0.62	0.64	⋅⋅⋅	0.35	0.47	0.61
±	0.03	0.02	0.03	0.05	0.16	⋅⋅⋅	0.01	0.02	0.03
σ	0.13	0.13	0.10	0.18	0.23	⋅⋅⋅	0.08	0.10	0.15
First-generation stars	20	48	17	14	3	0	33	38	31

Download table as:  ASCIITypeset image

We suggest some caution with the O trend in first-generation stars, as we cannot fully exclude that it may be due to some unidentified systematics with iron. Johnson & Pilachowski (2010) do not find a trend of [O/Fe] with [Fe/H], but claim instead to find weak trends (∼0.1 dex) in other α element ratios, such as [Ca/Fe] and [Si/Fe], which however are even weaker than our trend in [O/Fe]. Note that such trend is detectable only among first-generation stars, as oxygen has been depleted in second-generation stars by p-capture reactions.

As shown in the right panel of Figure 5, in first-generation stars O increases in concert with the total CNO. At odds with first-generation stars, the second generation ones (represented in magenta) do not appear to show any correlation with either La, Fe, and CNO, and occupy a spread region in all these abundance planes. Lanthanum abundance ratios follow a similar pattern for first- and second-generation stars as shown in Figure 6. This implies that the material out of which second-generation stars have formed was not exposed to neutron sources (no additional n-captures besides those experienced by first-generation stars) but only to p-captures. If AGB stars were responsible for the p-capture processing, this must have taken place in stars experiencing negligible third dredge-up, otherwise the material would have been further enriched in s-process elements.

Zoom In Zoom Out Reset image size

In Figure 7, we summarize our results for C, N, O, and Na abundance ratios, which have been plotted as a function of [(C+N+O)/Fe], [Fe/H], and [La/Fe], for the first- and second-generation stars. Carbon; as well as O, increases as a function of the total CNO and [La/Fe], suggesting that C and O evolve in a similar way (though their errors are correlated).

Zoom In Zoom Out Reset image size

At odds with C and O, neither N nor Na show any evidence for correlation with the total CNO abundance in first-generation stars. However, the minimum values for these two elements may increase slightly with both Fe and La. N and Na display a much stronger trend with either the overall CNO, iron, and lanthanum for second-generation stars, i.e., N and Na are higher for stars with higher CNO/Fe/La.

All the observed abundance pattern suggests that C and O from one side, and N and Na form the other, have undergone a similar processing in the evolution of the cluster. In addition, the first and second generation, selected on the basis of their Na and O content, appear to show well-defined individual chemical trends along the overall observed range in metallicity.

Finally, the CNO increase among first-generation stars is driven by the rise of O (and partly also of C), rather than N. As C and O increase with metallicity, more N could be produced by CNO cycling, and so in O-poor/Na-rich second-generation stars both [N/Fe] and [Na/Fe] increase with metallicity. Figure 8 shows CNO versus Fe separately for first- and second-generation stars, demonstrating that the increase of CNO with Fe is common to both generations. The small offset (<∼0.1 dex) between the two generations may be real (due to part of CNO elements having been processed to heavier nuclei by p-captures), but we cannot fully exclude a slight systematic underestimate of N among second-generation stars.

Zoom In Zoom Out Reset image size

The interpretation of the photometric and spectroscopic evidences on the stellar populations of ω Cen in terms of its formation and evolution is posing formidable difficulties. In this section, we try to extract as much as possible from what the data themselves apparently demand.

Although there is photometric evidence for at least six different stellar populations in ω Cen (e.g., Bellini et al. 2010), we have distinguished between a first and a second generation as comprised of those stars that either do not or do show evidence for additional p-capture processing at high temperature, respectively. Thus, the first generation may represent a less complex case to interpret in terms of chemical evolution, with the second generation possibly arising from the AGB ejecta of the former one, as generally entertained.

4.1. Formation Scenarios

4.2. Chemical Enrichment

The C+N+O abundance has a strong impact on the location of isochrones at the TO and SGB level, and therefore on age determinations. In order to estimate the effects of the large variation of the overall CNO among ω Cen's stellar populations on the age measurement, we have calculated stellar models by including a variety of assumptions on the C+N+O abundance and spanning a metallicity range suitable for the ω Cen stellar populations. The adopted physical inputs and numerical assumptions are the same as in Pietrinferni et al. (2009), which presents evolutionary computations accounting for a CNO enhancement of a factor of ∼2 relative to the "standard" α-enhanced mixture. A fully consistent set of α-enhanced isochrones with no CNO enhancement for various iron abundances (Pietrinferni et al. 2006), allows us to compare, in a fully homogeneous theoretical framework, CNO-enhanced and not-CNO-enhanced isochrones.

As expected, we find that, for a fixed TO and SGB brightness, the CNO-enhanced isochrones provide younger ages than isochrones corresponding to a canonical α-enhanced mixture (Pietrinferni et al. 2006): as a rule-of-thumb we found the ∂age/∂[CNO] ∼ −3.3 Gyr dex⁻¹, regardless of the iron content. By accounting for the observed [CNO/Fe] abundance in ω Cen, we have estimated the difference between the age one would measure for each population by adopting appropriate CNO-enhanced isochrones and that obtained by using "normally" α-enhanced isochrones, i.e., neglecting the measure CNO enhancement. The overall effect is to decrease the age estimate for the higher metallicity stars. The age decrease ranges from ∼0.5 Gyr for stars with [Fe/H] ∼−1.8 up to ∼2 Gyr for the most metal-rich stars with [Fe/H] ∼−1.0 (Figure 9), as listed in Table 3.

Zoom In Zoom Out Reset image size

Table 3. CNO Contents and Estimated Age Differences for the ω Cen [Fe/H] Groups

[Fe/H]	[(C+N+O)/Fe]	log(C+N+O)	Δage
			(Gyr)
−1.97 < [Fe/H] <−1.90	0.24 ± 0.04	7.43 ± 0.05	−0.8 ± 0.1
−1.90 < [Fe/H] <−1.65	0.34 ± 0.03	7.69 ± 0.04	−1.1 ± 0.1
−1.65 < [Fe/H] <−1.50	0.47 ± 0.02	8.03 ± 0.03	−1.6 ± 0.1
−1.50 < [Fe/H] <−1.05	0.48 ± 0.04	8.24 ± 0.05	−1.6 ± 0.1
−1.05 < [Fe/H] <−0.95	0.81 ± 0.12	8.94 ± 0.11	−2.7 ± 0.4

Download table as:  ASCIITypeset image

This result confirms that the measurement of relative ages of stellar populations in GCs requires accurate measurements of the overall C+N+O abundance. Indeed, it is well known that the double SGBs detected in NGC 1851, M22, and 47 Tucanae (Milone et al. 2008, Milone et al. 2011; Anderson et al. 2009; Piotto 2009) can be interpreted in quite different ways, depending on the CNO content of the single populations. In particular, the faint SGB can be associated either with a stellar population significantly older than the stars in the bright SGB (up to 1 Gyr), or to a second generation, almost coeval with the brighter SGB (age differences of 100–200 Myr), but with enhanced CNO (Cassisi et al. 2008; Ventura et al. 2009). This scenario has received further support from recent findings in M22 by Marino et al. (2009, 2011b): this GC hosts two stellar groups with different s-element and CNO abundances, and, in addition, the two groups of CNO-rich and CNO-poor stars also have different iron abundances, in close analogy with ω Cen.

In NGC 1851, results on possible internal variations of C+N+O are contradictory (Yong et al. 2009; Villanova et al. 2010). However, theoretical models, properly accounting for photometric signatures of the chemical peculiarities observed in the cluster sub-populations as provided by Sbordone et al. (2011), suggest that in NGC 1851 only a C+N+O-enhanced second generation can satisfy all the observational constraints given by CMDs in various photometric bands (Milone et al. 2008; Han et al. 2009).

Many studies have attempted to reconstruct the evolutionary history of ω Cen, by determining relative ages among the different stellar sub-groups, but none has been able to account for variations in the overall C+N+O abundance. Considering the complexity of its SGB morphology and of the metallicity distribution, and different assumptions regarding He enhancement among ω Cen populations, the relative ages for the various stellar populations as measured in different studies are quite different, ranging from a small or null age dispersion, as proposed by Sollima et al. (2005), to an age spread of the order of 2–3 Gyr (Hilker et al. 2004; Stanford et al. 2006) to 4 or more Gyr (Hughes & Wallerstein 2000; Hilker & Richtler 2000; Villanova et al. 2007).

It is beyond the scope of this paper to evaluate in detail how C+N+O abundances change the age dating obtained in the multiple cited works. However, we can identify some general trends. By using the same α-enhanced isochrones adopted here, Villanova et al. (2007) found a large age difference (∼5 Gyr) among the ω Cen populations. Specifically, they identify four groups of stars: (1) an old metal-poor group ([Fe/H] ∼−1.7), (2) an old population of metal-rich stars ([Fe/H] ∼−1.1), (3) a group of metal-intermediate young stars ([Fe/H] ∼−1.4), about 1–2 Gyr younger than the old components, and (4) a young metal-poor group ([Fe/H] ∼−1.7) up to ∼5 Gyr younger than the old populations. When accounting for the CNO overabundance and using appropriate CNO-enhanced isochrones, the age estimate of the metal-rich component in Villanova et al. should be decreased by ∼2 Gyr. As a consequence, the metal-rich population is no longer coeval with the most metal-poor population, but is slightly younger. Similarly, due to its enhanced CNO content, the intermediate metallicity ([Fe/H] ∼−1.4) population would be younger by ∼1 Gyr relative to the estimates given by these authors. On the other hand, the age difference between the two groups of old and young metal-poor stars should not change as they have the same iron and CNO content.

Hilker et al. (2004) and Stanford et al. (2006) suggest that ω Cen has experienced a prolonged star formation of ∼2–4 Gyr but, at odds with Villanova et al. (2007), with the most metal-rich stars being younger. Similar conclusions have been reached by Hughes & Wallerstein (2000) and Hilker & Richtler (2000), who estimated that the more metal-rich stars were younger than the metal-poor ones by ⩾3 Gyr, and 3–6 Gyr, respectively. In the case of Hilker et al. (2004) and Stanford et al. (2006), we predict that CNO effects on the relative-age estimate could increase the duration of the star formation in ω Cen up to ∼4–6 Gyr.

Sollima et al. (2005) suggest that the global star formation history of ω Cen should be confined to within ∼2 Gyr, with the metal-poor population ([Fe/H] ∼−1.7) representing the first and the major star formation burst. Our results suggest that the ages of the metal-intermediate and metal-rich stars given by these authors could be overestimated by ∼0.6 and ∼1.5 Gyr, with respect to the metal-poor population. In any case, given their uncertainties of 2 Gyr, these corrections do not change the main result of their work, e.g., the suggestion that ω Cen stellar populations formed in a short time period.

As mentioned above, the goal of this paper is not to do a detailed analysis of previous age determinations. As discussed before there are contradicting results in recent literature. Our aim is simply to show that there is significant variation in the C+N+O content among ω Cen's populations that may easily change these datations. In addition to that caveat, it is worth noting that He enhancement is likely present in the O-poor, metal-rich stars. This occurrence should be taken into account when measuring the relative ages via isochrone fitting among the various sub-populations, since He affects the SGB shape (but not the luminosity level).

In this paper, we have presented C, N, O abundances for 77 ω Cen RGB stars in the metallicity range ∼ −2.0 < [Fe/H] <−0.9. We have coupled our results with the ones of our previous work on Na–O–La abundances presented in Marino et al. (2011a). From our analysis we found:

• 1.  
  a correlation between the total CNO and the iron abundance, with the most metal-rich population being enhanced by ∼0.5 dex in [(C+N+O)/Fe] relative to the most metal-poor one;
• 2.  
  O and C grows with [Fe/H] and [La/Fe] in the O-rich/Na-poor stars;
• 3.  
  stars selected on the basis of their position on the Na–O plane, show defined chemical patterns in their light elements C, N, O, and Na as a function CNO, Fe, and La, that allow us to distinguish between a first and a second generation of stars, each possibly resulting from a series of separate bursts of star formation; and
• 4.  
  [La/Fe] correlates tightly with [Fe/H], following almost precisely the same trend regardless of the O/Na abundances.

In an attempt to make sense of these observational trends we have explored two (speculative) scenarios for the formation and evolution of this most puzzling object. In one option the system has evolved monolithically, i.e., remaining chemically homogeneous at each time. Thus, within less than ∼30–40 Myr a series of bursts of star formation each enriched in iron and α elements by core-collapse supernovae from the previous burst(s) would have been followed each by a secondary burst of star formation originating from material ejected by AGB stars, heavily processed by p-captures. The formation of the whole stellar population inhabiting the cluster today would have taken less than ∼2 × 10⁸ yr. Alternatively, rather than being sequential in time, such primary and secondary bursts of star formation would have taken place separately in space, before their products could merge at the bottom of the potential well. In both cases, secondary bursts would be fed via cooling flows made of AGB ejecta, leading to more centrally concentrated second generations, which indeed appears to be so in this (and other) clusters (Sollima et al. 2005; Bellini et al. 2009; Johnson & Pilachowski 2010). We admit that both scenarios require a series of very contrived and ad hoc assumptions, yet the extraordinary complexity of ω Cen may not admit simple solutions.

The CNO abundance affects the determination of the relative ages of cluster subpopulations via isochrone fitting of the TO SGB region. In light of our results, we have discussed this issue in the case of ω Cen by comparing isochrones with standard and with enhanced CNO, with the latter ones giving younger ages for the same TO luminosity. Although the determination of relative ages of the ω Cen sub-populations is beyond the aims of our study, we argue that a trend in CNO/Fe can help reducing large age spread among the various sub-populations, as found by some studies in the literature.

We thank the anonymous referee whose suggestions have significantly improved this work. A.P.M., G.P., S.C. and A.A. are founded by the Ministry of Science and Technology of the Kingdom of Spain (grant AYA 2010-16717). A.P.M. and A.A. are also founded by the Instituto de Astrofísica de Canarias (grant P3-94).