THE CARINA PROJECT. VII. TOWARD THE BREAKING OF THE AGE–METALLICITY DEGENERACY OF RED GIANT BRANCH STARS USING THE C_{U, B, I} INDEX

The photometric data are already presented in Bono et al. (2010). In particular, we focus here on the central region only, where U, B, V, and I data overlap over an area of ≈40' × 40'. The left panel of Figure 1 shows the (V, B−I) CMD of 20,240 stars selected with the following criteria: (1) they are measured in the four bands, and (2) they are bona fide Carina stars, with the removal of most of the possible Galactic field stars and foreground galaxies following the procedure explained in Bono et al. (2010), and based on the position of the sources in the (U−V), (B−I) color–color plane. The plot demonstrates a nicely cleaned CMD showing well-defined evolutionary sequences, in particular the two well-separated turnoffs and SGBs representing the old and intermediate-age populations, merging into the thin RGB. The few remaining field stars populate an almost vertical sequence at color 1.2 < (B − I) < 1.6 mag (cyan symbols for the brightest ones). These stars are virtually impossible to clean, even using a color–color plane, because their position is almost indistinguishable from the Carina turnoff (TO) stars. Note that most of the stars bluer than the field sequence and brighter than V = 20 mag are Anomalous Cepheids of Carina (Coppola et al. 2013). The orange crosses mark the faint part of the asymptotic giant branch (AGB) star sequence.

Zoom In Zoom Out Reset image size

The right panel of the same figure shows the (V, c_{U, B, I}) pseudo-CMD of Carina. As discussed in detail by Monelli et al. (2013), the features in the (V, c_{U, B, I}) diagram are reversed compared to a standard CMD, at least in the metallicity range covered by the Carina stars. The c_{U, B, I} of MS stars gets larger (i.e., less negative) as the stars get brighter, and after the turnoff the stars evolve to lower index values along the two SGBs. In this plane, the two RGBs do not totally merge in a single narrow sequence, as happens in standard optical CMDs. Although some degree of overlap is present for −1.70 < c_{U, B, I} <−1.6 mag, it is possible to follow the two RGB sequences that remain well-defined and run almost vertically and parallel from the SGB to the tip of the RGB. In particular, the RGB stars with more negative c_{U, B, I} are a sequitur of the SGB of the old population, while the more densely populated, higher c_{U, B, I} side of the RGB is the sequitur of the intermediate-age population. Only at the brightest magnitudes (V ≲ 19 mag) do the two sequences diverge slightly. We stress that the photometric error is of the order of a few hundredths of magnitude along the whole RGB, and therefore cannot account for the observed spread in the c_{U, B, I} index. Note that the contamination due to field stars is negligible across the RGB, while in this plane the majority of AGB stars overlie the RGB sequences. For this reason, the latter have been removed from the following analysis.

To better characterize these two RGB sequences, we followed the procedure outlined in Figure 2. Bona fide RGB stars are shown as black open circles (left and central panels), while the red line, drawn by eye, is intended to represent the right edge of their distribution. Our selection includes only stars in the magnitude range 17.5 < V < 22 mag. Nevertheless, we verified that the following analysis is minimally affected, within 1σ, by both the exact location of the envelope and the limit magnitude. The RGB was rectified with respect to the red line, which is the origin of the abscissas in the central panel. This shows, for each star, the distance in color Δc_{U, B, I} from the red line. Finally, the right panel presents the distribution of the Δc_{U, B, I} (gray histogram). The shape of the global distribution is best fitted by the sum (black line) of the two superimposed Gaussian curves (green and magenta lines), whose peaks are separated by ∼0.08 mag. This indicates that a simple photometric index such as the c_{U, B, I} is able to largely separate, for the first time, the two main components along the RGB of Carina. In particular, we conclude that stars with c_{U, B, I} <−1.70 mag belong to the old population, while stars with less negative index values are predominantly intermediate-aged.

Zoom In Zoom Out Reset image size

We verified this finding by considering only kinematically selected stars, adopting the sample of RGB stars from Fabrizio et al. (2011). This is shown in Figure 3, which is a replica of Figure 2 with 413 RGB stars with a radial velocity consistent with that of Carina superimposed. By adopting the selection from Fabrizio et al. (2011) and retaining stars with radial velocities within 4σ from the mean value, in the range +180 to +260 km s⁻¹, the occurrence of a double peak in the pseudo-color distribution due to the two main Carina populations is confirmed.

Zoom In Zoom Out Reset image size

In this section, we investigate whether a connection exists between the two sequences identified in the RGB and the chemical compositions of member stars. To do this, we use three independent samples of spectroscopic data available in the literature (see Figure 4). We adopt two samples of high-resolution measurements of Fe lines from Fabrizio et al. (2012; R ∼ 40,000, large circles) and Lemasle et al. (2012; R ∼ 22,000, triangles). The two samples include 44 and 35 stars for a total of 68 red giants (11 in common). The entire sample of stars with high-resolution spectra has at least one photometric measurement in the four adopted filters, namely, V and UBI (see Figure 4). We also included red giant stars for which Fe abundances were determined by Koch et al. (2006) using medium-resolution (R ∼6000) spectra (small circles in Figure 4) in the CaT region. Out of the initial catalog by Koch et al. (2006, see their Table 6), including 1197 stars, we selected 212 objects with an [Fe/H] abundance and that obey the same photometric and radial velocity selection criteria adopted in Figures 1 and 2. We note that in this analysis, we adopt the [Fe/H] abundance from Koch et al. (2006) and the radial velocity from Fabrizio et al. (2011).

Zoom In Zoom Out Reset image size

The analysis was performed by splitting all the spectroscopic samples according to their c_{U, B, I} index. In particular, green and magenta symbols account for stars with c_{U, B, I} <−1.70 mag and c_{U, B, I} >−1.70 mag, respectively. For comparison, the right panel presents the (V, B−I) classical CMD, where the two samples of stars selected according the c_{U, B, I} index are mixed. Figure 5 presents the metallicity distribution for the three samples. The sample from Koch et al. (2006) has been further split in two, separating the brightest (bottom left, V < 19 mag, 71 stars) from the faintest (bottom right, V ⩾ 19 mag, 141 stars) ones, to account for the change in the RGB morphology previously discussed (Figure 1). The plot discloses that the stars selected on the sequence with more negative c_{U, B, I} (green histogram), identified as the RGB of the old population, are on average more metal-poor than the stars on the other sequence (magenta), dominated by the intermediate-age population. The different independent samples provide a very good overall agreement, and in particular the sample of bright stars from Koch et al. (2006) notably agree with the high-resolution samples. In the case of the faint CaT stars, the overlap between the two histograms increases, possibly due to the increase in the observational errors. The mean metallicity of each sample is reported in Table 1. We find that the mean metallicity of the old population is ≈0.5 dex more metal-poor than the intermediate-age one.¹³ Assuming the Fabrizio et al. (2012) data, which include stars with both Fe i and Fe ii measurements, we find that the mean metallicities are [Fe/H] =−2.32 ± 0.08 dex and [Fe/H] =−1.82 ± 0.04 dex for the old and the intermediate-age group, respectively. While previous investigations (see, e.g., Lemasle et al. 2012) have suggested a similar chemical evolution for Carina, we stress that our approach is independent of isochrone fitting. Identification of RGB stars as members of the old or the intermediate-age population does not require previous knowledge of the metallicity.

Zoom In Zoom Out Reset image size

The cut adopted to split the two quoted populations, (c_{U, B, I} = −1.73) mag, is, as noted by the anonymous referee, arbitrary. Mild changes in the adopted cut can move objects from the old sample into the intermediate-age one and vice versa. To constrain, on a quantitative basis, the impact of the adopted cut on the mean metallicity of the two samples, we performed the selection by adopting c_{U, B, I} =−1.67 and c_{U, B, I} =−1.73 mag. The difference was fixed according to the photometric error of the c_{U, B, I} index at V ∼ 20.5 (σ_UBI = 0.03 mag). The mean metallicities listed in Table 1 further support that the criterion adopted to split the two samples minimally affects the current conclusions. In particular, the systematic difference between the mean metallicity of the old and the intermediate-age population is systematically, on average, larger than 0.5 dex. This result is an independent demonstration that the color spread observed in c_{U, B, I} is intrinsic. In fact, the c_{U, B, I} index and the metallicity used here have been measured in an independent way. If the RGB spread observed in c_{U, B, I} was entirely due to photometric errors (either statistical or systematic), a star with small (large) c_{U, B, I} would have the same probability of being either metal-rich or metal-poor. Moreover, this result is supported by three independent sets of spectroscopic measurements.

Figure 6 presents the iron abundance as a function of the Δc_{U, B, I} parameter previously defined. The top panels show the trend for the two high-resolution samples, while the lower panels present the medium-resolution ones. The plots disclose a clear trend: the lower the Δc_{U, B, I}, the lower the Fe abundance. We fit the data with a linear function in the form f = ax + b, reporting in each panel the values of both the zero point (b) and the slope (a), as well as the Spearman coefficient (r), which suggests that the correlation is significant. The four samples provide a general good agreement, in particular concerning the bright portion of the RGB where both high and medium resolutions are present. The subsample of faint stars presents a steeper correlation. This can be easily explained, taking into account that the RGB morphology in the c_{U, B, I} plane changes as a function of magnitude. In the upper part, the two RGB sequences, where most of the high-resolution stars are, tend to diverge. This pushes the stars on the old RGB to lower c_{U, B, I}, making the relation less steep. For magnitudes fainter than V ∼ 19 mag, the two RGB are closer, and this is reflected in a steeper relation between Fe and Δc_{U, B, I}.

Zoom In Zoom Out Reset image size

In Section 2.1, we mentioned that the cut at c_{U, B, I} =−1.70 mag to split the old and the intermediate-age is approximately equivalent to a cut at Δc_{U, B, I} =−0.1 mag. The anonymous referee noted that the two quoted cuts are equivalent for the high-resolution sample by Lemasle et al. (2012; top right panel of Figure 6) and for the faint medium-resolution sample (bottom right panel of Figure 6). On the other hand, there are objects in the bright magnitude sample and in the high-resolution sample by Fabrizio et al. (2012; left panels of Figure 6) moving across the boundary of the Δc_{U, B, I} distribution. To take account of the role that the selection criterion has on the correlation between the Δc_{U, B, I} parameter and the iron abundance, we performed the same test, cutting the two populations at Δc_{U, B, I} =−0.1 mag. Data plotted in Figure 7 show that only a few stars near the edge are classified in a different population. Therefore, this means that we cannot reach a firm conclusion concerning the nature (old versus intermediate-age) of the objects that are located across the boundary (c_{U, B, I} =−1.70 mag, Δc_{U, B, I} = −0.1 mag) adopted to split the two main populations. However, the above results further support evidence that the correlation is marginally dependent on the criterion adopted to split the two populations.

Zoom In Zoom Out Reset image size

Finally, we stress that this analysis is not meant to find an absolute calibration of the c_{U, B, I} index as a function of metallicity. Nevertheless, the correlation supports the connection between the c_{U, B, I} index and the Fe content of the Carina RGB stars.

To assess whether the theoretical models are able to reproduce the observed color distribution of RGB stars, in the following we perform a comparison with theoretical models from the BaSTI¹⁴ database (Pietrinferni et al. 2004). Figure 8 shows a comparison in the (V, c_{U, B, I}) plane with scaled-solar isochrones of three different metallicities (Z = 0.0001, 0.0003, 0.0006), and ages ranging from 4 to 13 Gyr, in steps of 3 Gyr. The figure discloses that the color range covered by the selected isochrones (≈0.03 mag at V = 21 mag) is significantly smaller than the observed one in the (V, c_{U, B, I}) plane. Interestingly, in the upper part of the RGB (V ≲ 19.5 mag), the effect of metallicity is to split the isochrones, in agreement with the observed increasing separation of the Carina RGB sequences, discussed in Figure 1. For fainter magnitudes, it is not possible to disentangle the isochrones of different metallicity, at least in the range typical of the Carina Fe content. The effect of age along the RGB is largely negligible in this plane for both the old and the intermediate-age population. Overall, we conclude that current models based on a standard heavy-element distribution do not account for the spread in c_{U, B, I} observed in the Carina RGBs.

Zoom In Zoom Out Reset image size

These results do not substantially change when using α- or He-enhanced models (see also Cassisi et al. 2013a, 2013b). Current theoretical framework does not allow us to fully explain the behavior of the two sequences in the c_{U, B, I} plane by accounting for iron, age, α elements, and He differences alone. We remark that this cannot be due to a residual shortcoming in the adopted color–T_eff transformations because we are using model predictions in a strictly differential way.

Figure 9 shows how stars of the two sequences are distributed in different CMDs, namely U−I, B−I, and V−I (top row, from left to right). The two sequences appear overlapped and mixed, and therefore indistinguishable, in B−I, while they are slightly segregated in V−I and substantially more so in U−I. However, the two sequences are swapped in two planes: in U−I the green symbols (old stars) are concentrated on the blue edge of the RGB, while the magenta circles (intermediate-age population) are located closer to the red edge. The opposite occurs for the V−I color. This is evident in the lower panels, where we show the fiducial lines of the two samples. To quantify the split, we calculate the color difference at a magnitude level of I = 20 mag—that is, roughly two magnitudes below the tip of the RGB. As expected, we find that the separation is negligible in B−I, and only slightly larger in V−I, on the order of 0.01 mag, which is statistically significant given the typical photometric error at this magnitude level. The largest separation is found in the U−I color, of the order of −0.08 mag.

Zoom In Zoom Out Reset image size

It is instructive to compare the behavior of the RGBs in Carina with similar observations of multiple RGBs in GCs. In several GCs, the RGB and the MS of first-population stars are redder than the second-population RGB and MS in B−I and V−I colors, and the color separation increases for increasing color baseline (that is, Δ(B − I) > Δ(V − I) > 0). However, the RGBs can merge together or even change their relative positions when using a near-UV filter (Δ(U − I) < 0). On the other hand, if far-ultraviolet (FUV) filters are adopted, the separation between the two RGBs increases again and is significantly larger in FUV − I than in visual colors (see Milone et al. 2012a, 2012b).

This has been explained as being due to the detailed abundances of the various light elements (e.g., C, N, Na, O) and helium. In particular, the colors of the first-generation stars in GCs are well reproduced assuming primordial He abundance and an O-rich/N-poor composition that is similar to that of halo-field stars of the same metallicity. On the other hand, the colors of the second-population stars agree with a composition in which N and Na are enhanced—along with increased He—while C and O are depleted. Sbordone et al. (2011) have demonstrated that such abundance variations mostly affect the CN, NH, and CH bands in the UV part of the spectrum, and this translates into different colors for the different populations when UV filters are used. These effects can be well reproduced by stellar models if the appropriate chemical mixtures are adopted (Sbordone et al. 2011).

Results shown in Figure 9 demonstrate that the RGB behavior of stellar populations in the Carina dSph and in GCs differs significantly. Carina is indeed an intrinsically more complicated system than a typical GC, due to the bursty star formation history and the large age and metallicity spread present in each population. Multiple populations in a GC formed on a much shorter timescale, and, in most cases, the internal chemical enrichment affected only the light elements involved in the high-temperature H-burning processes, while typically no spread in iron due to supernova pollution is found.

Moving to a more speculative scenario, a possible explanation for the empirical evidence is that light-element abundances could play a non-negligible role in determining the behavior of the multiple RGBs in the Carina dSph, in analogy to what is observed in GCs. The c_{U, B, I} index is very sensitive to light-element variations, which are measured in dwarf galaxies as well. However, it is not clear at present whether the chemical patterns are similar to that of GCs. In fact, the Na–O and C–N anti-correlations have not been observed for dSph galaxies in general, or for Carina in particular. Venn et al. (2012) report measurements of one O line, and also Na i determinations for a few stars, but no anti-correlation clearly stems from the data. On the other hand, more metal-poor stars, with [Fe/H] <−2 dex, seem to have lower Na content than more metal-rich ones (Venn et al. 2012, their Figure 12), in agreement with what was found in GCs. Overall, C and N abundances are yet not available. These can be crucial elements, as the overall C+N+O abundance significantly influences the luminosity of the SGB and hence affects the age determinations of the different populations (Cassisi et al. 2008; Ventura et al. 2009; Marino et al. 2012). Measurements of the overall C+N+O content of intermediate-age and old stars in Carina are mandatory to properly estimate the age of its stellar populations and thus to properly calibrate the chemical enrichment history.

Following the referee's suggestion, we show in Figure 10 a comparison with literature values, in particular for the metallicity (top panel) and the age (bottom). The top panel shows the V–c_{U, B, I} diagram, with the two groups of stars selected in Section 2.2 highlighted in green and magenta. The starred symbols show the 35 stars by Lemasle et al. (2012), while the squares show the 44 stars by Fabrizio et al. (2012; 11 objects in common) color-coded as labeled. The above data indicate that the most metal-poor stars attain the lowest c_{U, B, I} values, and fall on the RGB sequence identified by the green circles. On the other hand, stars more metal-rich than [Fe/H] ∼−2 attain larger c_{U, B, I} values, and all but three fall on the other sequence identified by the magenta circles.

Zoom In Zoom Out Reset image size

The bottom panel of Figure 10 shows once again the two groups of stars selected in Section 2.2 in green and in magenta together with the 35 stars by Lemasle et al. (2012). The spectroscopic sample by Fabrizio et al. (2012) was not plotted in this panel, because the authors did not provide an individual age estimate. Interestingly, we found that the most metal-poor stars also have the largest age (saturated at 15 Gyr). This implies that all but one of these objects are located in the lowest c_{U, B, I} sequence, which we identify with the RGB of the old population. Similarly, all of the intermediate-age stars (t < 8 Gyr) cluster in the region of the higher c_{U, B, I} stars. On the other hand, starred symbols plotted in green with estimated ages between ∼8 and ∼11 Gyr attain a c_{U, B, I} index ranging from ∼ − 1.75 to ∼ − 1.42 mag. In this context it is worth mentioning that the current comparison does not allow us to reach a firm conclusion concerning the objects with ages ranging from 8 to 11 Gyr. The reasons are manifold. The uncertainty associated with the individual ages of RG stars is at least of the order of 2 Gyr (Lemasle et al. 2012). The uncertainty on the individual iron abundance is at least of the order of 0.1 dex (Fabrizio et al. 2012). The sample is limited and these objects are located across the boundary adopted to split the old and the intermediate-age population.

Figure 11 shows the same [Fe/H] abundance (top) and age (bottom) as a function of the Δc_{U, B, I} index, with the same color code adopted for metallicity in the top panel of Figure 10. The above figures suggest a segregation, in the sense that stars with more negative Δc_{U, B, I} are systematically older and more metal-poor than those with a higher index. This further strengthens the use of c_{U, B, I} and Δc_{U, B, I} to separate stars from different populations. However, this result should be cautiously treated, since other parameters together with age and iron content are necessary to fully characterize the properties of the Carina stars (see Figure 8 and Section 3.2). This means that the comparison between isochrones and the CMD along the RGB might be affected by possible systematic effects. The anonymous referee noted that the correlation between age and metallicity is a thorny problem for the Carina dSph, since the two main episodes of star formation show a well-defined separation of the order of 3–4 Gyr. According to the recent literature, the age distribution of the two main stellar populations in Carina cover a broad range in age (Mighell 1990, 1997; Smecker-Hane et al. 1994, 1996; Hurley-Keller et al. 1998; Hernandez et al. 2000; Dolphin 2002; Rizzi et al. 2003; Monelli et al. 2003; Bono et al. 2010; Stetson et al. 2011; Lemasle et al. 2012). The difference is mainly due to the adopted stellar isochrones and to the diagnostic adopted to constrain the age (MSTO versus red giants). Unfortunately, we still lack firm theoretical and empirical constraints to assess whether the Carina star formation episodes were affected by the interaction with the Galaxy (Monelli et al. 2003; Pasetto et al. 2011; Fabrizio et al. 2011; Lemasle et al. 2012). However, the above uncertainties do not affect our conclusions, since we are only using relative ages to separate the old and the intermediate-age stellar populations.

Zoom In Zoom Out Reset image size

The quoted problems do not apply to other satellite dSphs, such as Sculptor, showing a well-defined age–metallicity relation (de Boer et al. 2012; Starkenburg et al. 2013). This stellar system appears as a fundamental benchmark to further constrain the use of the c_{U, B, I} index to identify distinct stellar populations.