The Bulge Metallicity Distribution from the APOGEE Survey

The results presented in this paper are for a sample of 7545 giant stars observed in the APOGEE survey between July 2011 and July 2014 that have estimated distances that place them within 4.5 kpc from the Galactic Center (GC; ${d}_{\mathrm{GC}}\leqslant \ 4.5$ kpc). This choice was motivated by sampling the edges of the long bar (Wegg et al. 2015). The stars are distributed in 83 APOGEE pointings of typically 100 stars each, as illustrated in Figure 1. The individual circular fields, ranging from 1 to 3 degrees in diameter, are identified by their Galactic coordinates (l, b) in degrees, e.g., 00+02. Five fields toward the central Galaxy—the APOGEE field centered on the Galactic Center (000+00), a BRAVA field (000−05), Baade's Window (001−04), and the Sagittarius fields, SGRC and SGRCM-4—were excluded due to their special target selection criteria. The observations cover both the outer bulge ($| l|$ or $| b| \gt 4^\circ$) and importantly, the poorly studied low-latitude bulge ($| b| \lt 4^\circ$). Of the 83 fields, 16 lie in the inner bulge, $| (l,b)| \leqslant \ 4^\circ$, while 29 outer fields (up to $l=30^\circ$) have $| b| \leqslant \ 2^\circ$. Previously such low-latitude regions had very few stars observed with high-resolution spectroscopy: only a few dozen M-type giants with $| b| \leqslant 2^\circ$ and a few hundred G-type and K-type giants. The vast majority of the sample is at distances between 4 and 12 kpc from the Sun, and suffers line-of-sight extinctions between 0.2 and 1 mag.

Zoom In Zoom Out Reset image size

The APOGEE H-band spectra, acquired with a cryogenically cooled, multi-object spectrograph (Wilson et al. 2012) coupled to the Sloan Foundation 2.5 m telescope (Gunn et al. 2006) at the Apache Point Observatory, in New Mexico and recorded by three HAWAII-2RG detectors, span the wavelength range 1.51–1.69 μm. The instrument has 300 input fibers and, in the standard APOGEE configuration, approximately 230 fibers are assigned to science targets²¹ and 70 are reserved for calibration: 35 targeting hot stars to record the telluric absorption pattern plus 35 sky fibers. To achieve Nyquist sampling at the shortest wavelengths, multiple exposures are taken while dithering the detector array by half a pixel (see Nidever et al. 2015, for further details).

For the bulge fields, stars were selected from the 2MASS Point Source Catalog (Skrutskie et al. 2006) by color and magnitude, largely adopting ${[J-{K}_{{\rm{s}}}]}_{0}\geqslant \ 0.5$ and $7\leqslant H\,\leqslant 11$, although a few fields were considered part of the APOGEE disk sample and slightly different selection criteria, with deeper integrations, were applied for them. The specified color cut was adopted to minimize the contamination by foreground dwarf stars (most prominent at ${[J-{K}_{{\rm{s}}}]}_{0}\lt 0.8$), retaining potential low-metallicity giants in the sample. The faint limit of $H\lt 11$ was set to ensure a minimum signal-to-noise ratio (S/N) of 100 per pixel for any inner Galaxy fields where single, approximately 1 hr visits were used (as opposed to the 3 visits norm for APOGEE fields; for more details, see Zasowski et al. 2013). Only ∼687 of the 7545 stars presented here have ${\rm{S}}/{\rm{N}}\lt 100$ and all have ${\rm{S}}/{\rm{N}}\geqslant \ 50$. Reddening corrections were estimated by combining near- and mid-IR photometry (2MASS, IRAC, and WISE), using the RJCE method (Majewski et al. 2011) and the Indebetouw et al. (2005) extinction law.

Raw data were processed with APOGEE's custom data reduction pipeline (Nidever et al. 2015), following a standard procedure: pixel dither combination, spectral extraction, wavelength calibration, sky emission and telluric contamination correction, and (when applicable) visit combination. All the spectra have been publicly released as part of the SDSS Data Release 12 (DR12; Alam et al. 2015).

The APOGEE Stellar Parameter and Abundance Pipeline (ASPCAP; García Pérez et al. 2016) was employed to determine stellar metallicities ([Fe/H]) simultaneously with the other atmospheric parameters ${T}_{\mathrm{eff}}$, $\mathrm{log}g$, [C/Fe], [N/Fe], and [α/Fe]. ASPCAP relies upon ${\chi }^{2}$ minimization to match each star's entire APOGEE spectrum to a library of precomputed, LSF-convolved (FWHM resolving power $R\equiv \lambda /\delta \lambda$ ∼ 22,500), and normalized synthetic spectra (Shetrone et al. 2015; Zamora et al. 2015). The microturbulence was tied to the surface gravity value by the relation ${\xi }_{{\rm{t}}}=2.478-0.325\,\mathrm{log}g$, derived from the analysis of a subsample of APOGEE data. The final ASPCAP metallicities were calibrated to well-known values of a sample of globular and open cluster stars (Holtzman et al. 2015). Based on the dispersion around the calibration values, the typical metallicity accuracy is estimated to be about 0.12 dex. However, the precision of the measurements is significantly better, typically about 0.05 dex (Holtzman et al. 2015). This precision is usually enhanced when abundance ratios such as [O/Fe] are considered. In fact, Nidever et al. (2014) found, for red clump stars in the thin disk, a spread in [α/Fe] at any given [Fe/H] between 0.02 and 0.04 dex, for high S/Ns, and Bertrán de Lis et al. (2016) found that stars in open clusters with similar temperatures showed consistent [O/Fe] ratios to within 0.01 dex.

Our sample is dominated by cooler red giants, and consequently, our metallicities are slightly more uncertain than the bulk of the APOGEE sample, around 0.05–0.09 dex. Additional details about the APOGEE DR12 extracted parameters and abundances may be found in Holtzman et al. (2015).

To select stars with reliable ASPCAP parameters, the APOGEE_ASPCAPFLAG bitmask flag was used.²² For the present study, stars were removed from our sample if any of the following were true: unreliable ${T}_{\mathrm{eff}}$, $\mathrm{log}g$, or [Fe/H]; large differences between photometric and spectroscopic ${T}_{\mathrm{eff}}$ estimates; large ${\chi }^{2}$ values; low ${\rm{S}}/{\rm{N}};$ and indications of rapid rotation from broadened line profiles. The final sample stars have parameters inside the ranges −0.4 ≲ $\mathrm{log}g$ ≲ 4.0, $3600\,\leqslant$ ${T}_{\mathrm{eff}}$ $\leqslant \,5500$ K, and −2.7 ≲ [Fe/H] ≲ +0.6, which match those where the DR12 atmospheric parameters are calibrated. Known cluster members, as reflected in the APOGEE_TARGET1 and APOGEE_TARGET2 flags, were removed from the sample. In addition, stars with a radial velocity dispersion (vscatter) larger than 1 km s⁻¹ were excluded, since that is usually an indication of binarity.

Our results, illustrated in Figure 6, present a metal-rich bulge at low heights ($| Z| \leqslant 0.50$ kpc). This has been suggested by previous studies, but those were based on 2D maps (l, b), without spatial resolution along the line of sight. Our 3D maps show significant variations in metallicity with position within the bulge. The side of the bulge closest to the Sun ($Y\lt 0$) appears to be metal-rich ([Fe/H] ∼ +0.2), while the more distant parts ($Y\gt 0$) seem to be more metal-poor ([Fe/H] ≲ −0.2).

Figure 7 collapses the information on that axis to offer a different perspective of the median metallicity as a function of Galactocentric distance (${R}_{\mathrm{GC}}=\sqrt{{X}^{2}+{Y}^{2}}$, with ${R}_{\mathrm{GC}}$ set to negative values at $Y\lt 0$). Changing the sign of ${R}_{\mathrm{GC}}$ depending on whether a location is before or beyond the GC. The Galactic center is very useful to separately consider the more distant regions, which are more prone to systematic effects (see Section 3). The part of the bulge that is closer to the Sun seems to be more homogeneous in metallicity, although, low-metallicity regions are also observed. The far/distant side of the bulge exhibits, in general, lower metallicities, but those low-metallicity regions can be followed, at intermediate Galactic longitudes ($l\sim 15^\circ$), by regions of higher metallicities.

Zoom In Zoom Out Reset image size

The APOGEE survey was conducted from the northern hemisphere, but did manage to observe some lower latitude regions of the southern Galactic hemisphere, albeit with overall poorer statistics, a situation now being remedied by data acquisition with the southern-hemisphere-based spectrograph of APOGEE-2. In general, data from the northern Galactic latitudes appear fairly similar to those obtained from regions located south of the Galactic plane.

As discussed in previous sections, the observed variation in metallicity with heliocentric distance is suggestive of biases in the stellar sample due to the cool limits of the model atmospheres used in the spectral analysis. The Galactic bar can contribute to the observed variations; however, its effect should have a marked dependence on Galactic longitude, which is not observed, and a symmetry in metallicity with respect to the bar location would be expected, which is not apparent in the data, probably because of our sample selection (See Section 6). The parts of the bulge closer to the Sun are the ones less affected by sample biases, and we will focus on those for the reminder of the paper.

Far from the midplane ($| Z| \geqslant +0.75$ kpc), Figure 7 shows a more homogeneous bulge dominated by stars with relatively low metallicity ([Fe/H] ≲ −0.5 on average). Interestingly, some locations show super-solar metallicity, and a couple quite low metallicities ([Fe/H] $\lt \,-1$). APOGEE's increased coverage of the low-latitude bulge allows us to firmly establish the presence of a vertical metallicity gradient, consistent with the findings of Zoccali et al. (2008), Gonzalez et al. (2013), and Ness et al. (2013). The gradient is no longer evident on the distant part of the bulge, but, as discussed in the previous section, our sample lacks metal-rich stars in those regions, especially closer to the plane where the extinction is stronger.

Figure 8 shows the median metallicity in the region located in the part of the bulge closer to the Sun ($-5\leqslant {R}_{\mathrm{GC}}\leqslant 0$) as a function of distance from the plane. The slope of the vertical gradient is not constant, but it appears to be the steepest at intermediate heights from the Galactic plane, at $+0.50\leqslant | Z| \leqslant +1.00\,\mathrm{kpc}$, with changes in metallicity of approximately −0.2 dex in 0.25 kpc. These regions have a sparser coverage in the southern Galactic hemisphere than in the northern one; however, the results for both are fully consistent.

Zoom In Zoom Out Reset image size

An inner flattening of the metallicity gradient was suggested in earlier studies (e.g., Ramírez et al. 2000; Rich et al. 2012; Babusiaux et al. 2014), which had data for only a few (l, b) locations. We do not only confirm the flattening, but also show the presence of a transition region at intermediate heights, and a flattening beyond $| Z| \gt 1\,\mathrm{kpc}$.

There are only a few studies of the bulge MDF that include low-latitude regions, and these have been restricted to a narrow range in (l, b) (e.g., Rich et al. 2007, 2012; Gonzalez et al. 2015; Zoccali et al. 2017). Babusiaux et al. (2014) had observations in fields at $b=0^\circ$, but their results were based on optical spectra acquired at a lower spectral resolution (R = 6500). The APOGEE database now permits the most complete and accurate study of the distribution of individual metallicities for stars in the Galactic midplane and inner bulge.

The metallicity distributions at different projected Galactocentric radii and heights, in spatial bins of 2 kpc and 0.25 kpc respectively, are shown in Figure 9. Only spatial bins including more than 30 stars are presented. Normalized MDFs are displayed in 0.15 dex metallicity bins, which are twice as large as the typical metallicity uncertainty for stars in the sample.

Zoom In Zoom Out Reset image size

In the midplane, corresponding to the bottom row of panels in Figure 9, stars from low to super-solar metallicities are observed ([Fe/H] ∼ −1 to +0.5), but the metal-rich stars are the dominant components. This is particularly true in the regions on the bulge quadrants closer to the Sun, at ${R}_{\mathrm{GC}}\lt 0$, which we trust to be the ones less affected by sample biases. This metallicity range is very similar to that reported by Gonzalez et al. (2015), but it does reach significantly lower metallicities than in the studies by Rich et al. (2007, 2012), most likely due to a smaller sample size that makes them miss the rare very low-metallicity stars. On the other hand, stars of lower metallicity are major contributors far from the plane ($| Z| \gt 0.75$). Note the presence of a significant metal-poor contribution in the central regions, as previously seen in APOGEE data for the GC by Schultheis et al. (2015), and reported earlier for a few inner fields by Babusiaux et al. (2014).

The detection of different metallicity distributions in the inner Galaxy can be interpreted in terms of density variations in multiple overlapping metallicity components (e.g., disk, bar, classical bulge, and inner halo), as suggested by Ness et al. (2013), rather than a bulk change in the overall population metallicity. For each of the selected regions in Figure 9, the distribution of the metallicities contains information about such components.

Ness et al. (2013) concluded that a minimum sample of ∼500 ARGOS survey stars were required to detect multiple metallicity components. The minimum number of stars per bin must be lower for the APOGEE sample, due to its greater metallicity precision: 0.05–0.09 dex versus 0.13 dex for ARGOS. Lindegren & Feltzing (2013) found that the minimum sample size required for resolving two different chemical distributions separated by r times the standard deviation (i.e., the measurement uncertainty) could be approximated by the expression ${N}_{\min }\simeq \ \exp (0.6+13\,{r}^{-0.8})$. This means that two populations whose metallicities differ by 0.32 dex could be resolved only with a sample of ∼500 stars measured with the precision of the ARGOS observations, while only ∼60 would be required at the typical precision of the APOGEE bulge sample. In our analysis, we typically have more than 100–200 stars per spatial bin, and, for the most part, we find a smooth variation of the distributions across neighboring regions.

The metallicity distributions often exhibit multiple peaks, and vary with position. A three Gaussian (3G) decomposition of the MDFs based on a maximum likelihood estimator and an analysis of jackknife samples (Bovy et al. 2011) returned components at four different metallicities, marked with vertical lines in Figure 9: +0.32 (metal-rich), +0.00 (intermediate metallicity), and −0.46 and −0.83 (metal-poor). It should be noted that the separation of the multiple components is larger than the uncertainties in their fitted mean metallicities ($\sigma \leqslant 0.15$ dex). The four different values are related to the four distinct centers found from the various 3G decompositions, with generally three of these four discriminated at each location.

There is an indication of a metal-poor component at [Fe/H] = −1.22 far from the plane ($| Z| \gt +1.00$ kpc), which may be connected to the stellar halo (Allende Prieto et al. 2014). The fraction of very metal-poor stars detected in this study is larger than that found in the ARGOS survey: 1.47% versus 0.07%, respectively, for a metallicity of [Fe/H] $\leqslant \,-1.5$. Most of the ARGOS metal-poor stars are seen at high Galactic latitudes ($b\geqslant 6^\circ$). Possible explanations for the greater numbers of the current work include a larger presence of this population at low heights and/or a metallicity bias.

The components peaking at [Fe/H] ∼ −0.83 and −0.46, more prominent far from the plane, may be related to the thick disk (see, e.g., Lee et al. 2011). For a more appropriate comparison in terms of homogeneity and proximity, compare with disk values in Hayden et al. (2014). The contribution of the component centered at [Fe/H] = −0.46 is significant in the central regions of the low bulge ($0\leqslant \ | Z| \leqslant \ +0.25$ kpc). In fact, the fraction of metal-poor stars ([Fe/H] $\leqslant \,-0.3$) along the midplane ranges from 7% at the nearest location to 41% in the GC. Such differences are large in comparison to the noise.

The components at metallicities ∼0.2–0.3 resemble the distributions reported for the central parts of the thin disk (see, e.g., $R\sim 5\,\mathrm{kpc}$ in Figure 7 of Hayden et al. 2014). This may be indicative of a bulge with a disk origin. The solar and super-solar metallicity components have a larger contribution at low $| Z|$, becoming the major contributors (especially the most metal-rich) on the side closer to the Sun. Interestingly, the solar-metallicity component extends to 3 kpc in radius (cylindrical coordinates). This component is present at low heights, independent of the set of heliocentric distance estimates employed (estimates other than those of the current work were also investigated); however, it is not visible in the most central regions. More uncertain is its vertical extent, whose contribution extends significantly beyond the intermediate heights depending on the adopted set.

The values of the metallicity components found in this study ($+0.32$, $+0.00$, −0.46, and −0.83) are consistent with those previously reported in the literature. The first and third most metal-rich components are in good agreement with the high-spectral-resolution results from GIBS (+0.26 and −0.31; Gonzalez et al. 2015) and the Gaia-ESO survey (+0.18 and −0.50; Rojas-Arriagada et al. 2014). The agreement with ARGOS (+0.10, −0.28, −0.68, and −1.18; Ness et al. 2013) is slightly worse, possibly due to differences with their metallicity scale. Based on a common stellar sample, their estimates in the super-solar metallicity regime are lower than the APOGEE values, and that explains some of the different components identified. However, in the low-metallicity regime, there are discrepancies between the identified populations that cannot be explained by a metallicity offset. Some differences with the study of microlensed dwarfs (Bensby et al. 2017) are also observed. We note that both Ness et al. (2013) and Bensby et al. (2017) cover different parts of the bulge than this study. Both studies find more peaks than those obtained in our 3G decompositions, despite the smaller stellar sample in Bensby et al. (2017). A metal-rich component has also been detected in the midplane by Babusiaux et al. (2014; +0.20). The good agreement demonstrated with various literature studies as well as the small derived uncertainties in the metallicity decomposition offer further support for the distributions we identify in the APOGEE data.

The MVG simulation consists of a boxy bulge that evolved from an exponential disk (Q = 1.5, scale-length of 1.29 kpc, and scale-height of 0.225 kpc) embedded in a live dark matter halo, and that suffered from instabilities and bar buckling (see Martinez-Valpuesta & Gerhard 2011). The resulting bar has a half-length of 4.5 kpc and an orientation of 25° between the bar major axis and the Sun-GC axis. Metallicity was added to the simulation by assigning a radial metallicity gradient ([Fe/H] $\,=0.6-0.4\,{R}_{\mathrm{GC}});$ R_GC in units of kiloparsecs) to the initial disk, which was chosen to reproduce the vertical metallicity variations observed by Gonzalez et al. (2013). According to the model, low-metallicity stars from the outer disk are mapped inwards to high latitudes producing the vertical gradients. The simulation snapshot we adopted was taken after the system was relaxed, $t\sim 1.9\,\mathrm{Gyr}$.

The simulation shows a metal-rich inner bulge elongated along the bar and surrounded by a metal-poor disk (see Figure 10) due to the initial setup. Low-metallicity stars come from the outer disk (no thick disk is included). The model does not show the asymmetry in the metallicity distributions we observe between the quadrants closer to the Sun and those beyond the Galactic center, in line with our conclusion that those are the result of a bias in our sample. On the other hand, we would have expected to observe the symmetry around the bar position shown by the simulations, but we do not.

Zoom In Zoom Out Reset image size

The simulation cannot reproduce the high metallicities observed in the solar neighborhood nor of the inner disk. Furthermore, Hayden et al. (2014) show a quite flat radial gradient for the thin disk near the bulge at $| Z| \lt 0.25\,\mathrm{kpc}$. This is in contrast with the larger gradient adopted in MVG. Milder metallicity gradients, such as those observed near the Sun, may reproduce better the high metallicities we observe in the midplane.

This model consists of a mixture of multiple stellar populations: bar, thin and thick disks, and halo. Specific properties are assigned as follows:

• 1.  
  A thin disk with ages from 0 to 10 Gyr, with an age–metallicity relation from Haywood (2008) in the solar neighborhood, and a radial metallicity gradient of −0.07 dex kpc⁻¹. Its scale-length has been constrained from a study of 2MASS star counts presented in Robin et al. (2012).
• 2.  
  A bar with an age of 8 Gyr, an average solar metallicity, and no gradients. The shape of the bar has been determined from 2MASS color–magnitude diagrams (Robin et al. 2012).
• 3.  
  A thick disk having two epochs of star formation at ages of 10 and 12 Gyr. Its characteristics have been determined in Robin et al. (2014). The mean metallicities are −0.5 and −0.8, respectively, and no metallicity gradients are assumed.
• 4.  
  A stellar halo, with an age of 14 Gyr, a mean metallicity of −1.5, and no metallicity gradient.

The kinematics for each population are computed mainly as described in Robin et al. (2003) for the thin and thick disks, and for the stellar halo, as given after the updates on the age velocity dispersion relation coming from the fit to RAVE and Gaia TGAS data (Robin et al. 2017). For the bar, the full 3D velocity field is computed using an N-body model from Debattista et al. (2006), scaled to fit BRAVA's data (Gardner et al. 2014; Robin et al. 2014).

This model has been observationally constrained, but in that exercise no APOGEE data were used. In the comparison below, APOGEE data are simulated by applying the selection criteria introduced during the survey targeting process. The number of targets in each field are selected exactly as was done for actual APOGEE observations. Further cuts are applied to remove regions of the $\mathrm{log}g$-versus-${T}_{\mathrm{eff}}$ plane compromised by ASPCAP's limitations.

The sample extracted from this model is therefore restricted to $4000\,\leqslant$ ${T}_{\mathrm{eff}}$ (K) $\leqslant \,4500$. Cuts in distances are not applied to avoid introducing uncertainties associated with the observed distance estimates. However, the high ${T}_{\mathrm{eff}}$ cut provides a natural culling of most of the foreground giants.

Observed and simulated MDFs are compared in three latitude bins in Figure 11. The APOGEE observations are overall well fitted by the simulations. Nonetheless, there are some differences, e.g., the super-solar metallicity contribution is overestimated in the model. The variation in metallicity as a function of Galactic latitude is produced by the different proportions of the populations included in the model, distorted by the APOGEE selection function. In these simulations, the dominant populations are the thin disk and the bar at low latitudes, the thick disk at high latitudes, and a combination of both components in between. There is no need to include a specific bulge component to reproduce the observed distributions.

Zoom In Zoom Out Reset image size

Our observed metal-rich components and the component at −0.46 would be associated with the bar+thin disk and the thick disk, respectively, in the model. The inner Galaxy shows a vertical transition from metal-rich to metal-poor brought by a changeover from a region dominated by a bar+thin disk to a thick-disk one, in line with our data.

Still, the thick disk would have a significant concentration in the central regions and would be the main sampled population at the far side of the bulge. This is caused by our target and field selection, and the shortest scale-length of the thick disk in comparison with that of the thin disk (Bensby et al. 2011; Bovy et al. 2012).

Note that the chemodynamical model of Portail et al. (2017) suggested that stars with metallicities as low as [Fe/H] ∼ −0.5 are strongly barred. A boxy/peanut-like structure was also assigned to stars of such low metallicities in Ness et al. (2016). Though, both studies are based on a metallicity grouping based on the ARGOS components. The ARGOS and APOGEE surveys are not necessarily on the same metallicity scale (Schultheis et al. 2017); therefore, the question is whether the findings of Ness et al. (2016) and Portail et al. (2017) are robust enough for the analysis assumptions; e.g., APOGEE versus their assumed ARGOS metallicity grouping.

The interpretation of the solar-metallicity component in terms of the BGM model is more challenging. In the simulation, the bar stops at 3.5 kpc from the GC (the thin and thick disks extend beyond), while the observed solar-metallicity component extends farther. An association of this component with the bar is not straightforward, because the bar and thin disk may not be chemically distinct. However, should the association be confirmed (e.g., using kinematics), our observations would give further support to the existence of a long bar (∼4.5 kpc), for which additional recent support has been offered (Wegg et al. 2015).

The BGM model has indications of a very metal-poor component (old thick disk and halo) everywhere, but with a limited contribution at low heights and large Galactocentric radii, consistent with our non-detections.