Assembling the Milky Way Bulge from Globular Clusters: Evidence from the Double Red Clump

Red clump (RC) stars are the metal-rich counterparts of core-He-burning horizontal-branch (HB) stars, and are ubiquitously observed in the color–magnitude diagram (CMD) of the Milky Way bulge. In the high-latitude ($| {b}| \gt 6^\circ$) field of the bulge, the RC is surprisingly split into two groups, with a magnitude difference of ∼0.5 between the bright RC (bRC) and the faint RC (fRCs; see Figure 1). Based on the latitude and longitude dependence of this double RC feature and the presence of an X-shaped structure in some extragalactic bulges, it was originally interpreted as evidence for a giant X-shaped structure in the Milky Way bulge that originated from the buckling instability of a bar (McWilliam & Zoccali 2010; Nataf et al. 2010; Li & Shen 2012; Wegg & Gerhard 2013). According to this scenario, the foreground (background) arm of the X-structure would be observed as a bRC (fRC) to an observer on Earth. While it is well established that the low-latitude field of the bulge is dominated by a bar population, this led to the consensus that even the high-latitude field of the bulge has a "pseudo-bulge" characteristic. However, a completely different interpretation has been suggested (Lee et al. 2015; Lee & Jang 2016; Joo et al. 2017), motivated by the multiple stellar population phenomenon observed in globular clusters (GCs; Lee et al. 1999; Gratton et al. 2012; Renzini et al. 2015; Bastian & Lardo 2017, and references therein). In this model, the bRC comprises He-enhanced second- and later-generation (G2) stars, while the fRC is composed of He-normal first-generation (G1) stars.⁴ In GCs, He-enhanced G2 stars are also enhanced in N and Na (Gratton et al. 2012; Marino et al. 2014).

Zoom In Zoom Out Reset image size

In GCs with multiple populations, the CN-band traces N abundance, and therefore it has been widely used in the search for stars that originated from GCs in the halo and the inner Milky Way (Martell et al. 2011; Schiavon et al. 2017). This chemical tagging based on CN-band strength is now well established and can also be used in the investigation of the origin of the double RC phenomenon. The purpose of this Letter is to report our discovery that the stars in the two RC regimes show significant difference in the CN-band strength, supporting the multiple population origin of the double RC phenomenon.

We have observed CN-band strengths for 478 RC and red giant branch (RGB) stars in the high-latitude bulge field, (l, b) ∼ (−1°, −85), where the double RC is most prominent. Figure 1 shows our sample stars in RC and RGB regimes in the $J-K$ versus K CMD obtained from 2MASS All-Sky Point Source Catalog (Skrutskie et al. 2006). Note that our RC zones in the CMD are mixed with background RGB stars, where the RC and RGB stars are spectroscopically indistinguishable (Boberg et al. 2016; Masseron & Hawkins 2017). We will therefore compare the mean difference in CN-band strength between the two RC zones, which can be then converted to the difference between only the genuine RC stars. In order to minimize the effect from evolutionary mixing, we have restricted the RGB stars to within ±1.2 V mag (±2.0 K mag) from the RC level.

Our low-resolution spectroscopy was carried out with the same instrument setup and data reduction procedures used for our previous investigations of multiple populations in GCs (Lim et al. 2015, 2017). The observations were performed with the du Pont 2.5 m telescope at Las Campanas Observatory in 2016 June and 2017 April and June. We used the Wide Field Reimaging CCD (WFCCD) instrument with HK grism providing a 25' × 25' field-of-view with a dispersion of 0.8 Å/pixel and a spectral coverage 3500–5000 Å. For the program field, 16 multi-slit masks were generated, each of which includes about 35 slits of 12 width. Typically, three 1800 s science exposures, three flats, and arc lamp frames were taken for each mask. The raw data were reduced using IRAF⁵ and the modified version of WFCCD reduction package in the usual manner (Prochaska et al. 2006; Lim et al. 2015). In addition, we performed radial velocity correction with the IRAF RV package. Figure 2 shows two typical examples of spectra from our observations, each representing stars in the bRC and fRC regimes, respectively.

Zoom In Zoom Out Reset image size

The strengths of CN, CH molecular bands and Ca H&K lines were estimated from the commonly used spectral indices, CN(3839), CN(4142), CH(4300), and HK' (Norris & Freeman 1979; Harbeck et al. 2003; Lim et al. 2015). Following the same procedures that we adopted for RGB stars in GCs in our previous studies, we measured these indices individually for each spectrum as the ratio of absorption strength to the nearby continuum. The measurement errors were calculated assuming Poisson statistics in the flux measurements. The final indices for a star were obtained by taking the error-weighted mean of the indices measured from each exposure, excluding low-quality spectra with S/N at 3900 Å < 5. In addition, stars with large measurement error or those abnormally far from the mean of the distribution are rejected to a confidence level of 99.7%. A total of 478 stars (176 for bRC; 206 for fRC; 96 for RGB) were selected for our analysis, which are identified in Figure 1. We also measured delta indices, δCN(3839), δCN(4142), δCH(4300), and δHK', as the difference between the original index and the least-square lines for RGB stars, to see the difference of chemical composition without the effects of effective temperature and surface gravity. Finally, we calculated the mean differences of δ-indices between the stars in the bRC and fRC groups. The confidence level of the difference was estimated from the error of the mean for each group.

As illustrated in Figure 3, the two scenarios for the origin of the double RC predict completely different distributions of fRC and bRC stars in the magnitude versus CN index diagram. In the case of an X-shaped bulge scenario, the difference in magnitude between the two RCs is due to the difference in distance, and the stars in the bRC would be identical with those in the fRC in terms of chemical composition. We therefore expect only a horizontal displacement of bRC compared to fRC in this diagram, with no difference in the peak position of the CN index distribution between the stars in the bRC and fRC regimes. On the other hand, in the multiple population model, the stars in the bRC would be G2 enhanced in N, He, and Na, while those in the fRC are G1 without these enhancements. In this scenario, the stars in the bRC are brighter because of enhanced He (Lee et al. 2015; Joo et al. 2017), and are CN-strong because of N enhancement. Therefore, the stars in the bRC and fRC regimes are predicted to show a marked difference in CN index distribution. Note further that the direction of skewness of the distribution, for both bRC and fRC regimes, respectively, is just the opposite of that predicted in the X-shaped bulge scenario.

Zoom In Zoom Out Reset image size

The two CN indices measured, CN(3839) and CN(4142), are presented in Figure 4 in the K magnitude versus CN index diagram. It is most obvious from the upper panels that stars in the bRC regime are systematically CN-enhanced compared to those in the fRC regime, just as expected in the multiple population scenario. The mean difference in the CN(3839) index between the stars in the bRC and fRC regimes is measured to be ΔCN(3839) = 0.150 ± 0.023 mag, which is significant at the 6.4σ level. For the delta index (δCN), we find more than 5σ difference in δCN(3839) between the stars in bRC and fRC regimes (ΔδCN = 0.125 ± 0.023 mag). A KS test also confirms that the probability is extremely low (0.00001%) for the two samples to be drawn from the same distribution. When the relative fractions of background RGB stars (66% and 78% for bRC and fRC, respectively), which would belong to both G1 and G2, are taken into account (see Figure 1), this observed quantity would correspond to ΔδCN ≈ 0.458 mag for the mean difference between only genuine RC stars in the two RC zones. We note that this difference is comparable to those observed in GCs between G1 and G2 (Lim et al. 2015, 2017). In addition, the CN index distribution is observed to be skewed toward higher CN with a peak at lower CN for fRC, while it is skewed toward lower CN with a peak at higher CN for the bRC. This is fully consistent with the multiple population scenario, but is opposite to the predicted distribution of the X-shaped bulge scenario.

Zoom In Zoom Out Reset image size

A similar result is obtained from CN(4142), further strengthening the difference in δCN between the two RC samples. Unlike the behavior of the CN-band, only very little difference is observed for the CH-band (ΔδCH = 0.012 ± 0.005 mag; see the upper panels of Figure 5). This is qualitatively consistent with the trend observed in GCs, although, in the metal-rich regime, CN-strong stars are not observed to be extremely CH-weak as in metal-poor GCs (Martell et al. 2011; Boberg et al. 2016). Furthermore, our observations show that the difference in heavy element abundance, as traced by Ca, is negligible (ΔδHK' = 0.009 ± 0.010 mag; see the bottom panels of Figure 5) between the stars in the two RC zones. This chemical pattern (CN-strong, CH-weak, and Ca-weak) is exactly the behavior that we would expect from G2 stars with GC origin.⁶ One may argue that some difference in CN-band strength might be observed in the X-shaped bulge scenario as well, because the line of sight would cross different distances above the Galactic plane for the foreground and background arms of the X-structure. However, even if the X-shaped structure follows a shallow metallicity gradient observed in the bulge (∼0.04 dex/°; Fragkoudi et al. 2017), the predicted difference in CN index from this effect alone is estimated to be almost negligible (∼5%) compared to the observed difference between the two RC regimes.

Zoom In Zoom Out Reset image size

It is now well established that a significant number of CN-strong (and relatively CH-weak) stars can form only in a GC environment, and these N-rich G2 stars are also enhanced in He and Na (Martell et al. 2011; Gratton et al. 2012, and references therein). Therefore, our result is direct evidence that the double RC is due to the multiple population phenomenon, and has little to do with an X-shaped structure. Stellar evolution models naturally predict super-He-rich stars to be placed on the bRC in the metal-rich population like the bulge (Lee et al. 2015; Lee & Jang 2016; Joo et al. 2017), which is also seen in the metal-rich bulge GC Terzan 5 (Ferraro et al. 2009; Joo et al. 2017). Chemical evolution models also suggest a large difference in He abundance between G2 and G1 in the metal-rich regime (Kim & Lee 2017), which is mostly due to the strong metallicity dependence of He yields from the winds of massive stars (Maeder 1992).

The astonishing point from our result is an apparent similarity in the population ratio of G1 and G2 between our high-latitude bulge field and GCs.⁷ This would indicate that a significant population of stars in the Milky Way bulge were assembled from disrupted proto-GCs or subsystems similar to GCs in terms of chemical evolution. Despite many uncertainties, it appears that GCs played a far more important role in the formation of the bulge than the outer halo, where only ∼3% of the stars show G2 characteristics (Martell et al. 2011). Chemical tagging using the Apache Point Observatory Galactic Evolution Experiment (APOGEE) survey has also uncovered a large population of N-rich stars in the inner Galaxy (Schiavon et al. 2017). While this result is qualitatively consistent with our discovery of CN-enhanced stars in the bulge, a quantitative comparison of the two results is complicated because their sample is dominated by the bar population (with G1 characteristics) in low-latitude fields. As the current view on the 3D mapping of the Milky Way bulge is based on the previous interpretation of the double RC phenomenon (Wegg & Gerhard 2013), our new result calls for a major revision in our quest to understand the structure of the Milky Way bulge. Our result would suggest that the Milky Way has a composite bulge with a bar embedded in a bulge component of GC origin. In this paradigm, the variation of the RC luminosity function with Galactic longitude and latitude is well explained in the multiple population scenario, which is also not inconsistent with the observed kinematics (Lee et al. 2015; Joo et al. 2017). Note that the stellar density in the claimed faint X-shaped structure from the WISE residual map (Ness & Lang 2016) is way too low to be observed as the double RC (López-Corredoira 2017; Han & Lee 2018).

Because the classical bulge component of the Milky Way bulge has long been considered as the nearest early-type galaxy (ETG) for which we can resolve into stars, our discovery in the Milky Way bulge would further imply that ETGs would similarly consist of stars having GC origins. In this respect, it is interesting to find that most massive ETGs and their GCs are Na- and CN-enhanced (Worthey 1998; van Dokkum & Conroy 2010). Further investigations with Gaia trigonometric parallax distances and high-resolution spectroscopy for the bulge stars in the double RC would undoubtedly help to strengthen our new results and the implications drawn from them.

We are grateful to John Norris for many helpful discussions and suggestions. We also thank the anonymous referee for constructive comments, and the staff of LCO for their support during the observations. This work was supported by the National Research Foundation of Korea (grants 2017R1A6A3A11031025, 2017R1A2B3002919, and 2017R1A5A1070354).