On the Stellar Content of the Carina Dwarf Spheroidal Galaxy¹

Dwarf galaxies play a fundamental role in several astrophysical problems. Current cosmological simulations predict dwarf satellite populations significantly larger than the number of dwarfs observed near giant spirals like the Milky Way and M31. This discrepancy is called the "missing satellite problem" and is a challenge to the currently most popular cosmological model: the Λ Cold Dark Matter paradigm (Klypin et al. 1999; Moore et al. 2006; Madau et al. 2008). However, it has not yet been established whether this discrepancy is due to limitations in the theoretical modeling or to observational bias at the faint end of the dwarf galaxy luminosity function (Klimentowski et al. 2009; Koposov et al. 2009; Kravtsov 2010).

Moreover, detailed analyses of the distribution of dwarf galaxies in multidimensional parameter space indicate that they follow very tight relations. In particular, Prada & Burkert (2002) suggested that Local Group (LG) dwarf galaxies show strong correlations (fundamental line—FL) among mass-to-light (M/L) ratio, surface brightness, and metallicity. They explained the correlation between M/L ratio and metallicity using a simple chemical enrichment model in which the hot metal-enriched gas, at the end of the star-formation epoch, is lost via continuous galactic winds (Mac Low & Ferrara 1999). In a recent investigation, Woo & Chiueh (2009) reached a similar conclusion, i.e., that LG dwarf galaxies follow a one-parameter relation driven by the total stellar mass. They also found that dwarf spheroidals (dSphs) appear to be, at fixed stellar mass, systematically more metal-rich than dwarf irregulars (dIs, see also Mateo (1998)). Futhermore, the FL of dIs in a five-dimensional parameter (total space mass, surface brightness, rotation velocity, metallicity, star-formation rate) is linear and tight, and extends the scaling relations of giant late-type galaxies into the low-mass regime. On the other hand, the FL of dSphs as defined in a four-dimensional parameter space (total mass, surface brightness, rotation velocity, metallicity) is also linear and very tight, but it does not lie on an extrapolation of the scaling relations of giant early-type galaxies. Iron and heavy-element abundances, the spread in these quantities, and the star-formation history (SFH) of LG dwarf galaxies, are crucial observational constraints on the physical mechanisms responsible for the scaling relations described above, and on their implications for different cosmological models (Orban et al. 2008). Spectroscopic investigations based on high-resolution spectra of bright red giants (RGs) in several LG dSphs indicate a large spread in iron abundance (Shetrone et al. 2001, 2003). This is also supported by metallicity estimates from the Ca II triplet method (Battaglia et al. 2008). In this context, the Carina dSph is particularly relevant: (i) It is relatively close and its central density is modest (ρ = 0.17 M_⊙ pc^-3, Mateo 1998) so it is easily resolved with ground-based telescopes (Mould & Aaronson 1983). (ii) It is the LG dSph most clearly showing multiple, widely separated star-formation episodes (Mighell 1990, 1997; Smecker-Hane et al. 1994, 1996; Hurley-Keller et al. 1998; Hernandez et al. 2000; Harbeck et al. 2001; Dolphin 2002; Rizzi et al. 2003), with a tail that might extend to less than one Gyr ago (Monelli et al. 2003). (iii) It is also a benchmark to constrain the pulsation properties of old (RR Lyrae, Saha et al. 1986) and intermediate-age (dwarf Cepheids, Mateo et al. 1998; anomalous Cepheids, Dall'Ora et al. 2003) variable stars. (iv) High-resolution spectra are available for a sample of 10 bright RGs. The mean metallicity is [Fe/H] = -1.69 with σ = 0.51 dex (Koch et al. 2008). Independent measurements by Shetrone et al. (2003), using high-resolution spectra for five bright RGs, provided a mean metallicity of [Fe/H] = -1.64 and σ = 0.2 dex. (iv) Medium-resolution calcium triplet measurements are also available for a large sample (437) of RG stars (Koch et al. 2006); their metallicity distribution shows a peak at [Fe/H] = -1.72 ± 0.01 (metallicity scale by Carretta & Gratton 1997, hereafter CG97) and the metallicity ranges from ∼-2.5 to ∼-0.5 dex. Using the same spectra, but different selection criteria (364 candidate Carina stars, Helmi et al. 2006) found the same peak in the metallicity distribution ([Fe/H] = -1.7 ± 0.1 dex) and metallicities ranging from ∼-2.3 to ∼-1.3 dex (see their Fig. 2).

Deep, accurate photometric investigations indicate an old (13 Gyr) stellar population and several distinct intermediate-age populations (2–6 Gyr) together with a blue plume of younger stars. The occurrence of both old and intermediate-age populations is also indicated by two distinct samples of core helium-burning stars: an old, wide HB and a separate red clump (RC). The shape and thickness of the RGB (ΔB - V ∼ 0.02 for 20.5 ≲ V ≲ 20.75 mag) suggest that the old and intermediate-age stellar populations have a minimal spread in chemical composition. This seems inconsistent with the spectroscopic observations that indicate a metallicity distribution possibly reaching extreme values near -3.1 and +0.1 dex (scale Zinn & West 1984, hereafter ZW84), or near -2.8 and -0.2 dex (scale CG97).

Despite the criticisms raised by Koch et al. (2006), conclusions based on purely photometric indicators are hampered by only two major factors: (i) A decrease in photometric precision when moving from bright to faint RG stars might produce a spread in color mimicking a spread in metal content. In particular, Carina's stellar density is very low, requiring that photometry be obtained with multichip CCD cameras or with many different pointings of a single CCD. A high degree of consistency in the photometric zero-points across the field is essential to avoid spurious broadening of the RG and main-sequence (MS) loci. (ii) Due to its low surface density, large extent, and low Galactic latitude (-22°), the Carina sample is contaminated by foreground field stars. Their colors, unfortunately, are similar to the Carina RGB, once again mimicking a broadening in color due to a spread in age/metallicity.

Our group is involved in a long-term investigation of the stellar populations in the Carina dSph. In particular, we have assembled our own and archival UBVI images covering the entire body of the galaxy. In this investigation we focus on comparing Carina with old and intermediate-age template clusters. We also present a new method for estimating the metal content of complex stellar systems from the difference in color between the middle of the RR Lyrae instability strip and the peak of RC stars.

The data were collected in various observing runs between 1992 December and 2005 January. They include images from three telescopes: the CTIO 1.5 m telescope with a single Tektronix2K CCD, the CTIO 4 m Blanco telescope with the Mosaic II camera, and the ESO/MPG 2.2 m telescope with the Wide Field Imager camera (both proprietary and archival data). Details of the observations will be presented in a future paper. The data presented here represent 4152 individual CCD images with essentially complete coverage of the central regions of Carina (≈40^' × 55^', lacking only those regions obliterated by bright foreground stars), and some sampling of the galaxy's halo out to an extreme radius of ∼108^'. They were obtained in four photometric bands (B and I with the 1.5 m telescope, and UBVI with both the 4 m and 2.2 m telescopes).

The data were reduced using the DAOPHOT/ALLFRAME package (Stetson 1987, 1994). Individual point-spread functions (PSFs) were produced for each chip of every exposure, using semiautomated routines to select bright, isolated PSF stars. Subsets of the data from each observing run were first reduced separately. Finally, everything was merged together and a single run of ALLFRAME was adopted to reduce all the images of the center of Carina, with separate reductions for nonoverlapping outlying fields. Because of the multiple chips and pointings required to cover the galaxy, no star appears in every image; any given star may have up to 17 calibrated measurements in U, 156 in B, 207 in V, and 70 in I. A total of 205,338 individual stars were catalogued and measured. Among these, 72,595 have photometric measurements in all four filters, and 129,230 have at least V and either B - V or V - I. The remainder, either extremely faint or located near the periphery of our coverage, have astrometry and instrumental magnitudes only.

The Carina data were contained within 206 individual data sets, where a data set is essentially the totality of data obtained with one C-CD from one night of observing. These, along with 1331 data sets from other nights and other telescopes, were calibrated to the current version of Stetson's (2000) photometric standard system, which is believed to be equivalent to that of Landolt (1992) to well under 0.01 mag in each of B, V, and I (see also Stetson 2005). The U photometric bandpass is more problematic, and in this article we will employ the U-band data only for qualitative and relative comparisons.

For a robust discrimination of candidate Carina members from foreground field stars and background galaxies we adopted the U - V, B - I color-color diagram (C-CD). The top panel of Figure 1 shows that this is effective for distinguishing candidate Carina and field stars. Carina RGs (U - V≥0.6, B - I≥1.4) attain, at fixed B - I, bluer U - V colors than field stars. The difference is mainly caused by a difference in the mean metallicity between Carina and the more metal-rich Galactic disk stars, but the difference in surface gravity between galaxy giants and foreground dwarfs may also play a role. In contrast, hot horizontal-branch (HB) stars and relatively young MS stars in Carina (B - I ≲ 0.4) are relatively free from field star contamination, since they are bluer than the thick disk and halo turnoffs; compact background galaxies represent the principal source of contamination among these bluer objects. Fortunately, this C-CD is also a good diagnostic to separate candidate Carina stars from blue background galaxies since the latter show, at fixed B - I, systematically bluer U - V colors than stars of any luminosity class or metallicity (apart from white dwarfs, which are not an issue here).

Zoom In Zoom Out Reset image size

Based on the above empirical evidence, we performed a series of tests to identify candidate Carina stars using various CMDs together with the C-CD. The solid-outlined box in the top panel of Figure 1 shows the final selection, while the bottom panels show the V, B - V CMD of the total sample (left), the candidate Carina stars (middle), and the candidate nonmember objects (right). Note that after the selection there remain some field stars with colors similar to Carina's RGB (V ≤ 21.5, B - V ∼ 0.4–0.6; middle panel). However, RG stars are distributed along a very narrow sequence over the entire magnitude range (18 ≤ V ≤ 23), and the number of field stars lying near that RGB in both color and magnitude is appreciably reduced. The same statement applies to the intermediate-age red clump (RC: V ∼ 20.5, B - V ∼ 0.65) and to the old HB (V ∼ 20.75, B - V ∼ 0.65). These evolutionary phases are characterized by very narrow distributions in either magnitude or color or both. The improvement compared to data already available in the literature is due to smaller photometric uncertainties (σ_B-V ≤ 0.02 for V ≤ 21, see error bars in Fig. 1). Plain evolutionary arguments (Monelli et al. 2003; Zoccali et al. 2003) suggest that small dispersions in magnitude and color in these evolutionary phases imply a negligible spread in chemical composition. This finding supports the results based on B,V,I photometry of Carina RG stars obtained by Rizzi et al. (2003). The presence of a few stars in the right panel with magnitudes and colors typical of red HB (V ∼ 20.75, B - V ∼ 0.4) and faint SGB stars (V ∼ 23, B - V ∼ 0.5) demonstrates that our C-CD selection is imperfect—especially at intermediate colors—and some real Carina members have probably been erroneously rejected. However, in these photometric selections it is more important to keep probable nonmembers out than to keep possible members in (Calamida et al. 2009).

To provide robust constraints on Carina's stellar content, we compared it with stellar systems characterized by different ages and metal abundances. In particular, to constrain the old stellar population we selected three globular clusters (GCs) - M79 = NGC 1904, M55 = NGC 6809, and M53 = NGC 5024—with metal abundances ranging from -1.64 ± 0.15 to -2.02 ± 0.15, and low foreground reddenings (E(B - V) ≤ 0.09, see Table 1). The optical BVI photometry of these GCs is also on the Stetson (2005) system. The accuracy of the photometry is certainly better than our knowledge of the foreground reddening toward these clusters and Carina. The V, B - V (left panels) and V, B - I (middle panels) CMDs of the selected GCs are shown in Figure 2, together with their individual ridge lines (red curves). We also selected three intermediate-age clusters (IACs) belonging to the Small Magellanic Cloud (SMC)—Kron 3, NGC 339, Lindsay 38—with similar ages and iron contents ranging from -1.08 ± 0.12 to -1.59 ± 0.10 dex (see Table 1 in Glatt et al. 2009). The photometry for these clusters is from Advanced Camera for Surveys on Hubble Space Telescope data transformed into the V, I Johnson-Cousins bands using relations by Siranni et al. (2005). These clusters are not included in the homogenous photometry project (the original sources of the data are listed in Table 1), so we are unable to independently confirm the precision of the photometry or the accuracy of the calibrations. Data plotted in the right panels of Figure 2 display the V, V - I CMDs for these clusters together with their ridge lines and the outlines of their RCs.

Zoom In Zoom Out Reset image size

To compare the GCs with the old stellar populations in Carina, we arbitrarily shifted their ridge lines in magnitude and color to provide a good fit with the SGB (23.25 ≲ V ≲ 23.75, 0.45 ≤ B - V ≤ 0.60) in Carina. Then we inferred the difference in distance and in reddening between Carina (μ = 20.15, E(B - V) = 0.03; Dall'Ora et al. 2003; Pietrzynski et al. 2009) and the individual GCs. The distances and the cluster reddenings we found following this approach agree, within the errors, with similar estimates available in the literature (see columns [4] and [5] in Table 1). The top and the middle left panels of Figure 3 show that the GCs M79 and M55 provide a poor match to Carina, since hot HB stars are systematically brighter than Carina HB stars. On the other hand, both the GB and the HB of the more metal-poor GC (M53) match Carina quite well. The same outcomes apply if the comparison is performed using the V, B - I CMDs (see the middle panels in Fig. 3). The extinction in the BVI bands was estimated using the empirical relations provided by McCall (2004).

Zoom In Zoom Out Reset image size

The same approach was followed to compare the IACs with Carina stellar populations. The ridge lines were shifted in magnitude and color to provide a good fit to the lower envelope of intermediate-age SGB (22.5 ≲ V ≲ 23.0, 0.45 ≲ V - I ≲ 0.80) stars in Carina. The distances that we estimate are larger than those estimated by Glatt et al. (2009) (see Table 1) using isochrones. However, the current distances agree, within the errors, with SMC distances available in the literature (Romaniello et al. 2009). The same outcome applies for the cluster reddenings (see the right panels in Fig. 3). This comparison between Carina and the IACs revealed the following: (i) the color extents of the SGBs of the the two more metal-rich clusters (Kron 3 and NGC 339; i.e., the length of the horizontal sequence between the upper MS and the lower GB) are smaller than the color range covered by Carina's intermediate-age SGB stars; (ii) the slope of the RGB in the two more metal-rich clusters (Kron 3, NGC 339) is shallower than the observed slope of Carina's RGB; (iii) the ridge line of Lindsay 38 is a fairly good match to Carina's intermediate-age population; (iv) the RC contour of Kron 3 agrees with Carina's RC, while for the other two clusters, the clumps are either slightly redder and brighter (NGC 339) or marginally fainter (Lindsay 38).

These comparisons indicate that Carina's old population is relatively metal-poor, and its spread in metallicity is modest: at most 0.2–0.3 dex. Carina's intermediate-age population seems to have the same metallicity as Lindsay 38, or perhaps slightly lower. The spread in metal abundance in this stellar component also seems very limited, since the RGBs of the two more metal-rich IACs are considerably redder than the red envelope of Carina's RGB. The lack of accurate photometry for metal-poor ([Fe/H] < -1.6 dex) IACs with well-populated RGBs prevents us from providing more firm constraints on Carina's intermediate-age population. These findings do not depend on the adopted metallicity scale, and if we adopt instead iron abundances on the Carretta et al. (2009, hereafter C09) metallicity scale (see column [3] in Table 1), the metallicity range inferred for Carina is very similar.

For a new estimate of the metal abundance of Carina, we decided to use a new indicator. Evolutionary prescriptions for old and intermediate-age stellar structures indicate that the difference in color between the center of the RR Lyrae instability strip and the RC stars is strongly correlated with iron abundance. The correlation is caused by the mild dependence of the color of the instability strip on iron abundance and by the significant dependence of the color of RC stars when moving from metal-poor to metal-rich compositions (see Fig. 4). We estimated the predicted difference in color between RC stars (at the beginning of central helium burning) and the center of RR Lyrae strip (log T_eff = 3.85) from the large set of scaled-solar evolutionary models (Pietrinferni et al. 2004) available in the BaSTI database.¹⁹ We adopted evolutionary prescriptions that cover a wide range of iron abundances (-2.27 ≤ [Fe/H] ≤ +0.06 dex), and an age range of 1–6 Gyr for the intermediate-age population and 13 Gyr for the old population. Note that the instability strip is minimally affected by cluster age when moving from 7–8 to 12–13 Gyr. In this context two robust evolutionary predictions need to be underlined. (i) When moving from metal-poor to metal-rich compositions, the intermediate-age RC becomes redder than the old RGB for [Fe/H] = -1.79 dex (see the top panel of Fig. 4). Current evolutionary models predict that a change in metallicity from [Fe/H] = -1.96 to [Fe/H] = -1.49 produces a change in V magnitude of RC stars of 0.03 mag, but a change of 0.12 mag in the B - V color. (ii) The spread in magnitude of HB stars at the mean color of the RR Lyrae strip increases, as expected, with the spread in metallicity. Note that a change in metallicity from [Fe/H] = -1.96 to [Fe/H] = -1.49 implies a change in V magnitude of HB stars of 0.10 mag, but a change smaller than 0.01 mag in the B - V color of the instability strip (see the bottom panel of Fig. 4).

Zoom In Zoom Out Reset image size

Finally, we performed a linear regression between metallicity and the RC - HB difference in three different colors:

These relations (see also right panels in Fig. 5) indicate that the difference in the B - I color is significantly more sensitive to [Fe/H] than the B - V color. The V - I is the least sensitive of all and shows a stronger nonlinear trend in the metal-poor ([Fe/H] ≤ -1.5) regime. We estimated the difference in color between the RC and the middle of the RR Lyrae instability strip (see the left panels in Fig. 5). We found Δ B - V H B R C = 0.37 ± 0.05 , and , where the errors account for uncertainties in the photometry and in the estimate of the mean color of the RR Lyrae strip. We applied these relations and found the following mean metallicities: , ) and ). We adopted the estimate based on B - I, since it is the most precise.

Zoom In Zoom Out Reset image size

As a further test to constrain the spread in metallicity of the Carina subpopulations, we computed a series of synthetic CMDs changing the spread both in age and in metal content. The set of evolutionary models used for these numerical simulations are the same used to derive the new metallicity indicator (Pietrinferni et al. 2004). The code adopted to compute the synthetic CMDs has already been discussed by Pietrinferni et al. (2004) and by Cordier et al. (2007). We only briefly mention the initial parameters adopted to compute the synthetic CMDs. The initial mass function was modeled using a power law with a Salpeter exponent (α = -2.35), and we assumed a fraction of unresolved binaries of the order of 10% with a minimum mass ratio for the binary systems equal to 0.7. The synthetic CMDs were constructed assuming two star-formation events, with a mean age of 2.5–4 Gyr for the intermediate-mass and of 11–13 Gyr for the low-mass subpopulation (Monelli et al. 2003). We also assumed that the two star-formation episodes include the same fraction of stars. Moreover, we assumed mean metallicities of [Fe/H] = -1.5 and of -2.0 dex for the intermediate-age and the old burst, respectively. In accordance with these findings, we assumed the same spread in metallicity—± 0.1 dex—for the intermediate and the old star-formation episode. Finally, to mimic actual observations, synthetic photometric errors were added using a Gaussian in the B and V bands with a dispersion equal to 0.02 mag. The assumptions adopted to simulate the Carina observations are crude, but a detailed analysis is beyond the aims of this investigation. We plan to study the Carina star-formation history in a forthcoming paper (Monelli et al. 2010, in preparation). Data plotted in the top panel of Figure 6 show that the aforementioned assumptions provide evolutionary features that are in reasonable agreement with the observations (see Fig. 2). In particular, both RC and HB stars are clearly separated in magnitude and in color from each other and from RG stars. Moreover, we selected two bins along the RGB for 20.75 ≤ m_v ≤ 21.25 and 21.50 ≤ m_v ≤ 22.00 and we found that the spread in B - V color is ∼0.06 mag. This value is quite similar, within the errors, to the spread in B - V color of the observed CMD at the same magnitude intervals, i.e., 0.06 mag. To constrain the impact of a spread in metal content, we used the same evolutionary ingredients adopted to construct the synthetic CMD, but we assumed for each of the two stellar components a spread in metallicity of 0.2 dex. The spread in B - V color at the same magnitude intervals is slightly larger than observed, namely 0.08 mag.

Zoom In Zoom Out Reset image size

Finally, we adopted for the intermediate-age stellar component alone a spread in metallicity of 0.5 dex. Data plotted in the bottom panel of Figure 6 clearly show that such a spread in metallicity causes the RC stars to attain colors that are redder than the RG stars. The spread in B - V color at the same magnitude intervals is a factor of 2 larger than observed, i.e., 0.12 mag. Moreover, the synthetic CMD suggests the occurrence of subgiant and giant branch stars with colors that are redder than the subgiant of the old subpopulation. The observed CMD does not show these features.

The comparison with empirical calibrators further supports the contention that Carina hosts predominantly metal-poor stellar populations, in agreement with the spectroscopic investigations. However, the limited range in color covered by the RC stars appears incompatible with the spectroscopic results: the spread in iron abundance of the intermediate-age population is either small, or is counterbalanced by some other—as yet unrecognized—variable. The same conclusion applies for the old population, based on the limited magnitude spread among the old HB stars. These conclusions are also supported by synthetic CMDs constructed assuming a different spread in metallicity between the intermediate and the old subpopulations. The simulations indicate a spread in metallicity of the two subpopulations smaller than 0.2 dex.

It is a real pleasure to thank an anonymous referee for positive comments on the results of this investigation, and for a suggestion. We also thank A. Grazian for several useful discussions concerning the intrinsic color of galaxies as a function of the redshift, and K. Glatt for sending us her photometric catalogs of the three SMC clusters. This project was supported by Monte dei Paschi di Siena (PI: S. Degl'Innocenti), and PRIN-INAF2006 (PI: M. Bellazzini). H. A. S. thanks the National Science Foundation for support under grant AST0607249.

• Figure 1
• Figure 2
• Figure 3
• Figure 4
• Figure 5