New Subgrouping of Multiple Stellar Populations in NGC 2808 Based on Low-resolution Spectroscopy

Numerous and dedicated observations made during the last two decades have established that most globular clusters (GCs) host multiple stellar populations (Lee et al. 1999; Gratton et al. 2012; Renzini et al. 2015; Bastian & Lardo 2018, and references therein). They usually show abundance variations in light elements with little or no dispersion in heavy elements (see, e.g., Carretta et al. 2009a). The abundance dispersions in such "normal" GCs are generally explained as a self enrichment by asymptotic giant branch (AGB) stars (D'Antona & Caloi 2004; D'Ercole et al. 2010; D'Antona et al. 2016), fast-rotating massive stars (Decressin et al. 2007b, 2007a; Krause et al. 2013), massive interacting binaries (de Mink et al. 2009; Bastian et al. 2013), or supermassive stars (Denissenkov & Hartwick 2014; Gieles et al. 2018). A few GCs showing abundance variations in heavy elements, however, have been discovered, such as ω Centauri and M22 (Lee et al. 1999, 2009; Marino et al. 2009; Johnson & Pilachowski 2010). Because heavy-element dispersions indicate an enrichment by supernovae (SNe), such "peculiar" GCs would have been more massive in the past to retain SNe ejecta (Timmes et al. 1995; Baumgardt et al. 2008; Silich & Tenorio-Tagle 2017).

Among the "normal" GCs, NGC 2808 is one of the most studied GCs due to its intriguing features on the color–magnitude diagram (CMD). The red giant branch (RGB) does not show the discreteness on the optical or near-infrared CMD as other GCs without significant spread in heavy elements. Multimodalities, however, are observed along both the main sequence (MS; Piotto et al. 2007) and the horizontal branch (HB) including the extreme-blue HB (see, e.g., Dalessandro et al. 2011). From these features, the presence of helium-enhanced multiple populations has been postulated (Lee et al. 2005; Piotto et al. 2007), and the helium variation has indeed been confirmed by spectroscopy (Pasquini et al. 2011; Marino et al. 2014). In addition, Hubble Space Telescope (HST) ultraviolet photometry and high-resolution spectroscopy have revealed the presence of at least five subpopulations with different light-element abundances (Milone et al. 2015; Carretta 2015). Note that a GC with more than five subpopulations is a rare case, except for ω Centauri and NGC 7089 (see, e.g., Milone et al. 2017). In spite of these extreme features, NGC 2808 has been considered as a "normal" GC because its stars do not show clear signs of heavy elements spreads, which is also supported by the recent low-resolution spectroscopic study using MUSE (Latour et al. 2019). Husser et al. (2020), using the same spectra, however, reported an unexpected Ca ii triplet metallicity variation in this GC, suggesting a possibility of a little spread in heavy elements.

In order to further investigate light and heavy elements abundance patterns in NGC 2808, in this paper, we measured spectral indices for CN, CH, and Ca ii H&K lines as well-known tracers of N, C, and Ca abundances, respectively, for the sample of 91 RGB stars in this GC. These indices are widely used in the multiple population studies (Norris & Smith 1983; Suntzeff & Smith 1991; Smith et al. 1996; Cohen 1999; Harbeck et al. 2003; Pancino et al. 2010; Smolinski et al. 2011; Simpson et al. 2017; Gerber et al. 2020, and references therein), including our previous works (Han et al. 2015; Lim et al. 2015, 2016, 2017).

The spectroscopic observations were performed with the Wide Field Reimaging CCD Camera (WFCCD) mounted on the du Pont 2.5 m telescope at Las Campanas Observatory (LCO). The spectroscopic target stars were selected from the photometric data obtained at the Cerro Tololo Inter-American Observatory (CTIO) 4 m Blanco telescope and reduced as described in Han et al. (2009). For the multiobject spectroscopy, a total of seven multislit masks were designed, each of them contains ∼21 slits of 12 width. The observations were made in Febuary 2012 and April 2016. The HK grism, which provides a central wavelength of 3700 Å with a dispersion of 0.8 Å pixel⁻¹, was used in order to securely detect CN, CH, and Ca ii H&K lines near 4000 Å. At least two science images, together with three flats and an arc lamp frame for calibration, were obtained for each mask. Exposure times for science frames are longer than 1500 s depending on the brightness grouping of stars. We used the modified version of the WFCCD reduction package and IRAF ⁴ for the data reduction and radial-velocity (RV) measurements as employed in previous studies (Prochaska et al. 2006; Lim et al. 2015, 2016, 2017). The RV distribution of our sample stars is roughly Gaussian with a peak value of ∼92 km s⁻¹, which is comparable to but somewhat smaller than the value of ∼101 km s⁻¹ listed in Harris (1996). This is probably because the typical uncertainty of RV values from our low-resolution measurement appears to be relatively large (∼30 km s⁻¹) compared to those from high-resolution spectroscopy. After the rejection of low signal-to-noise ratio (S/N) samples (<8 at 3900 Å) and nonmember stars (RV > 2.5σ of the mean velocity of the GC), 91 stars were finally used for our analysis, which are identified in Figure 1. The typical S/N for the selected sample stars is ∼17 at 3900 Å.

Zoom In Zoom Out Reset image size

Following Harbeck et al. (2003) and Lim et al. (2015), we calculated the S(3839), CH(4300), and HK' indices to measure the strengths of CN, CH, and Ca ii H&K lines of each target stars in NGC 2808, respectively. The definitions for these indices are where F_3861−3884, for example, is the integrated flux from 3861–3884 Å. The measurement error of each index was estimated from Poisson statistics in the flux measurements (Vollmann & Eversberg 2006). We also measured δ indices (δCN, δCH, and δHK') as the difference between the original index and the least-square fitting lines of the full sample in this GC, in order to compare chemical abundances of stars excluding the effects of magnitude (gravity and temperature). The measured indices and the errors are listed in Table A1.

We first define subpopulations based on the CN–CH diagram. For the RGB stars in NGC 2808, the δCN and δCH indices are anticorrelated as shown in the left panel of Figure 2. This CN–CH anticorrelation is one of the well-known features in low-resolution spectroscopic studies for "normal" GCs (Suntzeff & Smith 1991; Smith et al. 1996; Cohen 1999; Harbeck et al. 2003; Pancino et al. 2010; Smolinski et al. 2011; Lim et al. 2015, and references therein). NGC 2808 follows this general trend as expected from the homogeneity in iron abundance among its member stars (Gratton et al. 2011; Carretta 2015). High-resolution spectroscopic studies found a similar anticorrelation between Na and O abundances for "normal" GCs, which is widely used to define subpopulations in GCs. For instance, Carretta et al. (2009b) defined P (Na-poor/O-rich stars), I (Na-/O-intermediate stars), and E (Na-rich/O-poor stars) subpopulations on the Na–O plane (see also Carretta 2014). In a similar manner, we have divided four subpopulations along the CN–CH sequence, CN-weak/CH-strong G1, CN-intermediate/CH-strong G2, CN-strong/CH-intermediate G3, ⁵ and CN-strong/CH-weak G4. These four subpopulations are indentified in the $\delta \mathrm{CN}\mbox{--}\delta \mathrm{CH}$ histogram as shown in the right panel of Figure 2. Qualitatively, these four peaks are consistent with those reported by Norris (1987) from CN photometry. Our spectroscopic study is based on $\delta \mathrm{CN}\mbox{--}\delta \mathrm{CH}$, and, therefore, the G3 and G4 subpopulations are more clearly separated. Examples of observed spectra for each subpopulation are shown in Figure 3.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure 4 shows the measured spectral indices (CN, CH, and HK') as functions of V magnitude obtained from our CTIO photometry. The strengths of these indices increase as V magnitude decreases, which is due to the effects of temperature and surface gravity. In order to compare chemical abundances of stars without the magnitude effects, we calculated the differences in mean values of the δ indices between subpopulations, which are summarized in Table 1. The top-right and middle-right panels of Figure 4 show that the δCN and δCH indices have significant spreads that are clearly larger than typical measurement errors. The δCN enhancements of G2, G3, and G4 compared to G1 are 0.25, 0.4, and 0.45, respectively. Unlike δCN index, δCH decreases toward the later generations. The mean values of G3 and G4 are decreased by 0.045 and 0.095 compared to G1, respectively; while the difference in δCH between G1 and G2 is negligible. The δCN and δCH indices, in other words, are not linearly anticorrelated, but the relation between them has a "banana" shape (see the left panel of Figure 2), like the well-known Na–O anticorrelation. When the G2, G3, and G4 are combined to G2+ as in our previous works ⁶ , the differences in δCN and δCH indices between G1 and G2+ are ∼0.36 and ∼0.05, respectively. This difference in δCN is somewhat larger compared to other GCs showing CN–CH anticorrelation. For instance, G1 and G2+ in both NGC 288 and NGC 362 are separated by ∼0.31 in δCN. When the G4 is excluded, the difference between G1 and "G2+G3" in NGC 2808 becomes more comparable to other GCs (from ∼0.36 to ∼0.32). Interestingly, the G4 of NGC 2808 shows a "jump" in δHK' index, while G2 and G3 show no appreciable separations from G1. The difference in δHK' index between G1 and G4 is 0.045 (see the bottom right panel of Figure 4), which is significant at 5.0σ level. The abundance pattern that G4 shows (i.e.,δCN increased, δCH decreased, δHK' increased) is very unusual in the sense that most CN-enhanced and CH-depleted subpopulation in a "normal" GC does not show an enhancement in δHK' index. In "peculiar" GCs, on the other hand, the metal-enriched later generation, traced by δHK' index, shows enhancements in both δCN and δCH indices (Han et al. 2015; Lim et al. 2015, 2017). We note that subpopulations showing a similar abundance pattern (Na increased, O decreased, Ca increased) have been reported by Carretta & Bragaglia (2021) in eight GCs, although the definition of Ca excess in that study is the excess of [Ca/H] at given [Mg/H] with respect to field stars. We also note that the spread in δHK' index, caused by G4, in NGC 2808 is qualitatively consistent with the result of the recent spectroscopic study by Husser et al. (2020), which reported an unexpected variation of Ca ii triplet line strengths in this GC. In Section 5.2, we discuss the probable cause of the spread in Ca line strengths observed in this GC.

Zoom In Zoom Out Reset image size

A high-resolution spectroscopic survey with FLAMES/VLT provides high-quality and homogeneous spectroscopic data set for the elemental abundances of 140 RGB stars in NGC 2808 (Carretta et al. 2003, 2004, 2006, 2018; Carretta 2006, 2014, 2015; see also Gratton et al. 2011; Marino et al. 2014). We cross-matched our low-resolution data with this high-resolution data set, which results in the common sample of 24 stars. Figure 5 (Figure 6) shows the relations, for the common sample stars, between δCN (δCH) and abundances of key elements in the multiple population studies, where the abundance ratios are adopted from Carretta (2015) and Carretta et al. (2018). The δCN is anticorrelated with O and Mg, while it is correlated with Na and Al. The δCH shows the opposite patterns, i.e., anticorrelated with Na and Al, and correlated with O and Mg. These trends are in good agreement with previous studies for GCs (Sneden et al. 1992; Smith et al. 1996; Marino et al. 2008; Lim et al. 2016, 2017). In particular, Smith et al. (2013) and Smith (2015) noticed a nonlinear relation between CN and Na (and O). The relation between δCN and Na, together with that between δCH and O (Mg), further implies that our grouping of G1, G2, G3, and G4 would correspond to subpopulations defined with [Na/O] or [Na/Mg] in high-resolution studies. The relations between $\delta \mathrm{CN}\mbox{--}\delta \mathrm{CH}$ and [Na/O], [Na/Mg] are shown in Figure 7. It shows that our G1, G2, G3, and G4 subpopulations, which are defined based on $\delta \mathrm{CN}\mbox{--}\delta \mathrm{CH}$, are well matched, respectively, with P1, P2, I1+I2, and E subpopulations defined from the [Na/O] and [Na/Mg] (Carretta 2015). Despite this general agreement, we can see some notable differences in population division between the two studies. As the histograms show, our $\delta \mathrm{CN}\mbox{--}\delta \mathrm{CH}$ index can more clearly separate G1 (P1) and G2 (P2) than does [Na/O] or [Na/Mg]. It is mainly due to the fact that G1 (P1) and G2 (P2) show a larger difference in δCN compared to that in Na. On the other hand, G3 (I1+I2) and G4 (E) are more separated in [Na/O] and [Na/Mg] than $\delta \mathrm{CN}\mbox{--}\delta \mathrm{CH}$, which is because G3 (I1+I2) and G4 (E) show larger differences in O and Mg compared to that in δCH.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure 8 shows a relation between our δCN and [CN/Fe] ⁷ adopted from Carretta et al. (2018). It shows that our G1 and G2 are clearly identified by δCN, whereas they have the same [CN/Fe]. The [CN/Fe] seems to work in separating subpopulations only beyond G2. It implies that, for the earlier subpopulations, δCN measured from ∼3900 Å is more sensitive to abundance variations than is [CN/Fe] obtained from the near-IR region (see also Harbeck et al. 2003; Pancino et al. 2010). On the other hand, G3 appears to have a higher value of [CN/Fe] than G4. Since our population division between G3 and G4 is not as efficient as that from high-resolution spectroscopy (see Figure 7), this is probably because our G3 is polluted by G4 stars.

Zoom In Zoom Out Reset image size

In this work, we have measured CN, CH, and HK' indices for the 91 RGB stars in NGC 2808 and confirmed the large spreads in CN and CH. Along the anticorrelated sequence based on CN and CH indices, we have divided 4 subpopulations, G1, G2, G3, and G4. A "jump" in δHK' index between G1 and G4 has been detected. In addition, we found that δCN is more efficient in dividing G1 and G2 compared to [Na/Fe]. More detailed aspects of these three points are discussed below.

5.1. Comparison of Population Division

5.2. Intrinsic Ca Spread in NGC 2808

In Section 3, we showed a "jump" of Ca ii H&K line strength at G4. A variation in Ca ii triplet line strength is also reported by Husser et al. (2020). The spread of Ca line strength in a GC is usually interpreted as an intrinsic star-to-star variation in Ca abundance (Suntzeff 1980; Norris & Freeman 1982; Han et al. 2015; Lim et al. 2015, 2017), a sign of SN II enrichment. Since Carretta (2015) provides Ca abundances, we can verify whether the G4 (the most enriched population) is indeed enhanced in Ca or not. Therefore, we compared relative differences between the most enriched subpopulation (G4 or E) and the combination of the less enriched subpopulations (G1+G2+G3 or P1+P2+I1+I2) in δHK', [Ca/H], and [Ca/Fe], because G1 (P1), G2 (P2), and G3 (I1+I2) have indistinguishable δHK' and [Ca/H]. The left panel of Figure 9 clearly shows an enhancement in δHK' index for G4 compared to the previous generations. A similar trend is observed in the [Ca/H] as shown in the middle panel. The relative difference in [Ca/H] between the subpopulation E and the less enriched subpopulations is 0.026 dex with 3.2σ significance. This enhancement in [Ca/H] at the subpopulation E (G4) has been explained by assuming a proton-capture reaction under extreme conditions with low metallicity, extremely high temperature (T > 150 MK), and/or increased cross section (Ventura et al. 2012; Carretta 2015; Carretta & Bragaglia 2021).

Zoom In Zoom Out Reset image size

Figure 9 shows, however, that, unlike [Ca/H], the subpopulation E does not show any statistically significant enhancement in [Ca/Fe] compared to the other subpopulations. ¹⁰ This may indicate that the enhancement in δHK' or [Ca/H] in the most extreme subpopulation may not be entirely due to the actual enhancement in Ca. One possible explanation for this is the depletion of H caused by significant He enhancement. From the photometric studies of MS and HB stars in NGC 2808, it is already reported that the He mass fraction in the most extreme subpopulation (G4) approaches ∼0.4 (Lee et al. 2005; Piotto et al. 2007; Pasquini et al. 2011; Marino et al. 2014). This strong He enhancement would lead to the depletion of H in the stellar atmosphere, and, therefore, could explain up to 0.08 dex enhancement in [Ca/H]. An extreme Mg depletion can be another candidate for the large enhancement in δHK' of G4. Cohen et al. (2010) measured Ca ii triplet line strength for RGB stars in NGC 2419, and reported ∼0.2 dex spread of Ca abundance in this GC. Mucciarelli et al. (2012), however, argued that this spread from Ca ii lines is not due to the Ca or Fe abundance spread, but mainly due to the large depletion in Mg abundance. They pointed out the role of Mg in the opacity of cool stars as important electron donors to form the H⁻ ion. A depletion of Mg, therefore, leads to a decrease of the opacity and the electron pressure, which results in stronger Ca ii lines at fixed Ca and Fe abundances. Ca and Fe abundances obtained from Ca i and Fe i lines, respectively, indeed do not show any intrinsic spread in this GC (Mucciarelli et al. 2012). The same effect is observed in NGC 5824 (Da Costa et al. 2014; Roederer et al. 2016; Mucciarelli et al. 2018), another GC showing a large dispersion in Mg abundance. Like NGC 2419 or NGC 5824, NGC 2808 is also one of the GCs which host a strong Mg-depleted subpopulation (Carretta 2015; Mucciarelli et al. 2015), although the spread of Mg in NGC 2808 is relatively small (∼0.5 dex) compared to those in NGC 2419 (∼2 dex) and NGC 5824 (∼1 dex). Therefore, it is possible that a strong Mg depletion can contribute to the "jump" of Ca ii H&K line strength observed at G4 in NGC 2808.

5.3. The Large Difference in CN Between G1 and G2

Since the CN strength is a well-known tracer of N abundance in studies of GCs (see, e.g., Smith et al. 1996), the large difference in CN strength between G1 and G2 would imply a N enhancement in G2 compared to G1. The large difference in N is also supported by the HST photometry for this GC (Milone et al. 2015). As described in Section 5.1, Milone et al. (2015) compared Δ(F336W-F438W) photometric index, which is sensitive to N abundance, with [Na/Fe] in their Figure 8. The comparison shows that there is a large and discrete difference in Δ(F336W-F438W) between population B (P1) and C (P2), whereas they are overlapped in [Na/Fe]. They estimated that Δ[N/Fe] of population C (P2) with respect to B (P1) is ∼0.5 dex, which is calculated from the comparison of observed and synthetic color, and this value is much larger than Δ[Na/Fe] (∼0.2 dex) between them.

The larger difference in [N/Fe] between G1 and G2 compared to that in [Na/Fe] is also explained in the chemical evolution model of Kim & Lee (2018), which is developed to reproduce the Na–O anticorrelation generally observed in GCs. In this model, the chemical abundances of later generations are successively enriched by accumulated gas with processed materials ejected by winds of massive stars (WMSs) and AGB stars of previous generations. Since the N yields of WMSs and AGB stars are larger than Na yields, this model naturally predicts that N would be more enhanced than Na in G2 compared to G1 (see Figure 10). For subsequent generations (after G3), we do not always find a higher N enhancement than Na when compared to the previous generation. This is probably because abundances of subsequent generations depend more on the adopted star formation history. Salaris et al. (2020) has recently shown that the initial difference in [N/Fe] between G1 and G2 would be even larger if the effect of the first dredge up is properly accounted for, while this effect on [Na/Fe] is expected to be negligible (Karakas & Lattanzio 2014). More detailed quantitative comparison with a chemical evolution model, therefore, should include this effect.

Zoom In Zoom Out Reset image size

The study of multiple stellar populations with CN band is not restricted to GCs, but can be extended to the Milky Way (MW) bulge. The multiple stellar population model has been suggested to explain the double red clump (RC) phenomenon observed in the CMD of the MW bulge (Lee et al. 2015). This model predicts chemical abundance differences, such as N and Na, between the two RCs. Since, as shown in this work, CN band strength is more efficient in separating G1 and G2 than is Na abundance, we expect a larger difference between bright and faint RC stars when the CN index is employed as a tracer of multiple populations. Indeed, we have detected a sign of the difference in CN band strength between the stars in the two RC regimes (Lee et al. 2018). The CN survey for the bulge RC stars in various galactic longitude and latitude fields is in progress (S. Hong et al. 2021 in preparation), and will give a new insight on the structure and formation of the MW bulge.

We are greatful to the referee for a number of helpful suggestions. We also thank to the staff of LCO for support during the observations. Support for this work was provided by the National Research Foundation of Korea (2017R1A2B3002919). D.L. gratefully acknowledges funding by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)—Project-ID 138713538—SFB 881 ("The Milky Way System", subproject A03, A05)." S.I.H. acknowledges support provided by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (Nos. 2018R1D1A1A02085433 and 2020R1I1A1A01052358).

Facility: Du Pont (WFCCD). -

Software: IRAF, WFCCD Reduction package (Prochaska et al. 2006; Lim et al. 2015).