A Chemical Composition Survey of the Iron-complex Globular Cluster NGC 6273 (M19)^*

In Johnson et al. (2015b), we identified an intrinsic metallicity spread in NGC 6273, and noted the existence of several stars redder than the formal RGB that could belong to an even more metal-rich component. Since the previous observations were restricted to the color range 0.7 ≤ J–K${}_{{\rm{S}}}$ ≤ 1.0 on the upper RGB, we expanded the target selection criteria for the new observations to include stars in the color range of 0.6 ≤ J–K${}_{{\rm{S}}}$ ≤ 1.3. The new observations also span luminosities from the HB to the RGB-tip, and range from 053 to 1398 in projected distance from the cluster center (see Figure 1). However, stars closer to the cluster center were given higher priorities in the target ranking process. All coordinates and photometry for the target selection process were taken from the Two Micron All Sky Survey (2MASS; Skrutskie et al. 2006) database.

Zoom In Zoom Out Reset image size

In order to efficiently obtain a large number of high-resolution spectra, we employed the Michigan/Magellan Fiber System (M2FS; Mateo et al. 2012) and MSpec multi-object spectrograph mounted on the Magellan–Clay 6.5 m telescope. In single order mode, M2FS is capable of placing 256 12 fibers on targets across a nearly 30' field of view. However, additional orders can be observed simultaneously using a cross-disperser, at the expense of fewer targets. We utilized both options for this project. The first setup operated in single order mode and was optimized to observe the 8542 and 8662 Å near-infrared Calcium ii Triplet (CaT) lines. These data provided radial velocities and CaT metallicities for 466 stars, and permitted an investigation into the full spatial, color, and metallicity extent of NGC 6273. The second setup ("Bulge_GC1" filter) included 6 consecutive orders, spanned 6120–6720 Å, allowed for up to 48 fibers to be allocated per configuration, and was used to obtain radial velocities and detailed chemical abundances for 82 stars. As can be seen in Figure 1, both data sets spanned broad color and radial distance ranges, but the CaT data extended to fainter stars.

Both instrument setups utilized a four amplifier slow readout mode and were binned 2 × 1 (spatial×dispersion). The CaT and Bulge_GC1 observations were taken with the 180 μm (widest) and 125 μm slits, respectively. However, both setups yielded approximately the same resolving power of R ≡ λ/Δ λ ≈27,000, based on an examination of the ThAr wavelength calibration spectra. The two CaT fields were observed for a total of 10,200 s, and the two Bulge_GC1 fields were observed for a total of 21,600 s. A summary of the observation dates, instrument configurations, and integration times is provided in Table 1.

For data reduction, we followed the procedures outlined in Johnson et al. (2015b see their Section 2.3). Briefly, we used standard IRAF¹³ tasks to apply the bias correction, trim the overscan regions, correct for dark current, and combine the individual amplifier images from each CCD into single images. The IRAF dohydra task was used for aperture identification and tracing, flat-field correction, scattered light removal, wavelength calibration, cosmic-ray removal, and spectrum extraction. For the CaT data, we did not apply any corrections for fringing beyond the flat-field correction. A master sky spectrum was created for each exposure by combining the individual sky fiber spectra. The target spectra were then sky corrected using the skysub routine. Finally, the individual extracted spectra for each star were co-added separately, normalized with the continuum routine, and corrected for telluric absorption lines using the telluric task. Typical signal-to-noise ratios (S/Ns) ranged from about 20–100 per pixel for the CaT data and 30–100 per pixel for the Bulge_GC1 data.

We supplemented the M2FS CaT data set with additional observations of 300 RGB stars taken with the VLT FLAMES–GIRAFFE instrument. The data were downloaded from the European Southern Observatory (ESO) Science Archive Facility under request number 210062.¹⁴ The FLAMES observations spanned a broad range of magnitudes, but were generally fainter than the M2FS data. However, the spatial coverage between the two data sets was similar (see Figure 1). Note that we have only included stars for which we could identify a 2MASS source within 2'' of the coordinates provided in the image headers.

All of the FLAMES–GIRAFFE observations were obtained using the HR21 setup, which provides R ≈ 18,000 spectra from 8482 to 9000 Å. However, we only analyzed the region spanning 8500–8700 Å, which is similar to the M2FS CaT data and includes the same 8542 and 8662 Å CaT features. The observations were taken via six configurations, each with an integration time of 2445 s. Most stars were observed in two configurations, but not always with the same fiber each time. A small number of stars were observed in three or more configurations, and a few were observed only once. A summary of the observation dates for each configuration is provided in Table 1.

The data were primarily reduced using the GIRAFFE Base-Line Data Reduction Software (girBLDRS¹⁵ ) package. The girBLDRS suite was used to carry out basic CCD processing tasks (e.g., bias correction and overscan trimming) and also the more advanced multi-fiber tasks we performed with dohydra for the M2FS data (see Section 2.1). Similar to the M2FS CaT data, we did not apply any further corrections for fringing beyond the flat-field correction. The sky subtraction, continuum normalization, and spectrum combining were carried out with the same IRAF routines as used for the M2FS data. However, since the FLAMES data were obtained over the course of several weeks to months, we applied the heliocentric velocity corrections provided in the image headers before combining the multiple exposures. The final S/N values are comparable to those of the M2FS CaT data.

NGC 6273 is known to have a broad RGB and a peculiar HB morphology that is similar to ω Cen (Piotto et al. 1999; Momany et al. 2004; Brown et al. 2010, 2016; Han et al. 2015). Therefore, in support of our spectroscopic observations, we have obtained new HST Wide Field Camera 3 UVIS channel (WFC3/UVIS) data centered on NGC 6273 that includes the F336W, F438W, F555W, and F814W filters. The observations were split into a series of short and long exposures, taken over the course of four orbits, that ranged in duration from 10 to 685 s. A post-flash of 2.0–4.7 s was included for all exposures, and the BLADE = A option was set for all of the 10 s exposures to minimize shutter-induced vibration (see Section 6.11.4 of the WFC3 handbook¹⁶ ). A summary of the filter choices, integration times, and observation dates is provided in Table 1.

The basic data reductions were carried out by the Space Telescope Science Institute's WFC3 pipeline, but we only performed analyses on the CTE-corrected flc images. All photometry was obtained using the DOLPHOT¹⁷ (Dolphin 2000) package and its associated WFC3 module. The DOLPHOT parameters closely followed the values recommended by Williams et al. (2014) and provided by the DOLPHOT/WFC3 documentation for point sources in crowded fields. No special attempt was made to recover saturated stars; however, only a small number of the brightest stars, predominantly in the F814W filter, were lost due to saturation.

As noted by several previous authors (Racine 1973; Harris et al. 1976; Piotto et al. 1999; Davidge 2000; Valenti et al. 2007; Brown et al. 2010; Alonso-García et al. 2012), differential reddening is a significant concern along lines of sight near NGC 6273. Previous work estimated that the cluster has E(B – V) = 0.31–0.47 mag and ΔE(B – V) ∼ 0.2–0.3 magnitudes. We observe a similar reddening range of ΔE(B – V) = 0.36 magnitudes using corrections kindly provided by A. Milone (2016, private communication; see also Milone et al. 2012 for an outline of the dereddening procedure) via the F336W and F814W data sets. Additionally, we find that adopting an absolute color excess of E(B – V) = 0.37 mag places the coolest HB stars at approximately the correct F555W magnitude, assuming a distance of 9 kpc (Piotto et al. 1999).¹⁸ Further details regarding the photometric analysis, including the dereddening procedure, will be provided in a future publication. However, in Figure 2, we show the smoothed reddening map of the WFC3 field, and include several dereddened color–magnitude diagrams with the radial velocity members identified.

Zoom In Zoom Out Reset image size

Many globular clusters have been shown to rotate with amplitudes of the order of a few km ${{\rm{s}}}^{-1}$ (e.g., Côté et al. 1995; Lane et al. 2009, 2010a; Bellazzini et al. 2012; Bianchini et al. 2013; Kacharov et al. 2014; Kimmig et al. 2015; Lardo et al. 2015). In Figure 4, we investigated net rotation in NGC 6273 by following a standard technique in which the average radial velocity is calculated for stars on either side of an imaginary line passing through the cluster center. The bisecting line is rotated east through west in $10^\circ$ increments, and the velocity differences are plotted as a function of position angle. The resulting data can be fit with a sinusoidal function of the form

$$\begin{eqnarray}&&{\rm{\Delta }}\langle {V}_{r}\rangle ={A}_{\mathrm{rot}.}\sin (\mathrm{PA}+{\rm{\Phi }}),\end{eqnarray} \tag{ 1 }$$

where ${A}_{\mathrm{rot}.}$ is twice the actual projected rotation amplitude, Φ = 270°—PA${}_{{\rm{o}}}$, and PA${}_{{\rm{o}}}$ is the angle of maximum rotation. Bellazzini et al. (2012) argue that the projected ${A}_{\mathrm{rot}.}$ value should be a reasonable estimate for the true maximum rotation amplitude, and we have adopted their interpretation here.

Zoom In Zoom Out Reset image size

For NGC 6273, we find a clear rotation signature with A${}_{\mathrm{rot}.}$ = 3.83 ± 0.12 km ${{\rm{s}}}^{-1}$ and PA${}_{{\rm{o}}}$ = 126° ± 2°. We calculated the rotation profile using various angular bin sizes and found that, while ${A}_{\mathrm{rot}.}$ only varied by a few tenths of a km ${{\rm{s}}}^{-1}$, the PA${}_{{\rm{o}}}$ value could change by ∼15°. Therefore, we follow Bellazzini et al. (2012) and have adopted the conservative 1σ uncertainties of ±0.5 km ${{\rm{s}}}^{-1}$ for ${A}_{\mathrm{rot}.}$ and ±30° for PA${}_{{\rm{o}}}$. Compared to the large globular cluster samples presented in Bellazzini et al. (2012), Kimmig et al. (2015), and Lardo et al. (2015), NGC 6273 exhibits relatively strong rotation. NGC 6273's large ${A}_{\mathrm{rot}.}$ value is consistent with other clusters having similar metallicity and mass (e.g., ω Cen; see Figures 11 and 19 in Bellazzini et al. 2012 and Lardo et al. 2015, respectively).

In Figure 5, we also investigated the change in velocity dispersion as a function of the projected radial distance from the cluster center. As expected, we find that the velocity dispersion decreases from at least 10 km ${{\rm{s}}}^{-1}$ inside 1' to less than 5 km ${{\rm{s}}}^{-1}$ outside 5'. We also estimated the cluster's central velocity dispersion (σ_o) using simple Plummer models (Plummer 1911) of the form

$$\begin{eqnarray}&&{\sigma }^{2}(r)=\displaystyle \frac{{\sigma }_{o}^{2}}{\sqrt{1+{\left(\tfrac{r}{{r}_{h}}\right)}^{2}}},\end{eqnarray} \tag{ 2 }$$

where ${r}_{{\rm{h}}}$ is the Plummer scale radius.²⁰ We fit two models: (1) one with both σ_o and ${r}_{{\rm{h}}}$ varied as free parameters and (2) one with σ_o varied as a free parameter and ${r}_{{\rm{h}}}$ held fixed. For the latter case, we assumed the half-light radius was approximately equal to the half-mass radius and adopted a half-light radius of 132 (Harris 1996; 2010 revision). The resulting fit provided σ_o = 10.98 ± 0.40 km ${{\rm{s}}}^{-1}$. For the former case, we found σ_o = 10.35 ± 0.69 km ${{\rm{s}}}^{-1}$ and r${}_{{\rm{h}}}$ = 167 ± 041.

Zoom In Zoom Out Reset image size

However, we regard these values as lower limits of the true central velocity dispersion because the measured velocity dispersion for the bin closest to the cluster core is sensitive to the adopted bin size. For example, when the first bin contains stars with projected radial distances of 02–10, as is done in Figure 5, the dispersion is ∼10 km ${{\rm{s}}}^{-1}$, but if we change the range to 02–07, then the dispersion increases to ∼12 km ${{\rm{s}}}^{-1}$. Furthermore, a simple Plummer model assumes spherical symmetry, but NGC 6273 is relatively elliptical in shape (White & Shawl 1987; Chen & Chen 2010). Additional velocity measurements inside ∼02–05 and the application of more sophisticated models are likely to find a true σ_o > 12 km ${{\rm{s}}}^{-1}$. We estimate that the cluster's true A${}_{\mathrm{rot}.}$/σ_o ratio is ∼0.30–0.35, which is typical for massive elliptical metal-poor globular clusters (e.g., see Bellazzini et al. 2012; Kacharov et al. 2014; Kimmig et al. 2015; Lardo et al. 2015).

The model atmosphere parameters effective temperature (T${}_{\mathrm{eff}}$), surface gravity (log(g)), metallicity ([Fe/H]), and microturbulence (${\xi }_{\mathrm{mic}.}$) were determined spectroscopically for all radial velocity member stars observed with the Bulge_GC1 setup. A spectroscopic determination of especially ${\text{}}{T}_{\mathrm{eff}}$ and log(g) is preferred over photometric measurements for NGC 6273 because of the cluster's large and variable reddening (see Section 2.3). We followed the general analysis procedures outlined in Johnson et al. (2015b), which includes the use of the 1D local thermodynamic equilibrium (LTE) line analysis code MOOG²¹ (Sneden 1973; 2014 version). In particular, we solved for ${\text{}}{T}_{\mathrm{eff}}$ by enforcing excitation equilibrium with the Fe i lines and solved for surface gravity by adjusting log(g) until the Fe i and Fe ii lines provided the same abundance. In the few instances where only Fe i could be measured, we assigned stars a log(g) value that was compatible with other cluster members of similar temperature and metallicity. Microturbulence was measured by adjusting ${\xi }_{\mathrm{mic}.}$ until the derived log (Fe i) abundance was independent of line strength. Finally, the metallicity of each model was set as the average of [Fe i/H] and [Fe ii/H].

In order to generate the models, we interpolated within the available grid of ATLAS9 model atmospheres²² (Castelli & Kurucz 2004). For most stars, we used the α-enhanced models in order to compensate for the difference between [Fe/H] and [M/H]. However, a small number of stars in our sample have [α/Fe] ∼ 0, and for those stars we used the scaled-solar models. For every star, we started with a base-line model of ${\text{}}{T}_{\mathrm{eff}}=4500$ K, log(g) = 1.20 cgs, [Fe/H] = −1.60 dex, and ${\xi }_{\mathrm{mic}.}$ = 1.70 km ${{\rm{s}}}^{-1}$, and iteratively solved for all four parameters simultaneously.

Lind et al. (2012) showed that, for some stars, departures from LTE can have a significant impact on the model atmosphere parameters derived by spectroscopic methods. However, the impact on stars in the temperature, gravity, and metallicity regime probed here is likely to be small. Additionally, the relative effects due to departures from LTE should be mostly negligible within a small parameter space (e.g., Wang et al. 2016), and we have attempted to empirically cancel out large non-LTE and 3D model atmosphere deficiencies by performing a differential analysis relative to Arcturus. Therefore, we have not applied any non-LTE corrections to our data. We note also that Dupree et al. (2016) showed the addition of a chromosphere may alter the derived abundances for some elements. However, since we lack the spectral coverage necessary for constraining a chromospheric model, our model atmosphere parameters and abundances are based only on radiative/convective equilibrium models. The final model atmosphere parameters for all member stars derived from the Bulge_GC1 data are provided in Table 4.

Table 4.  Model Atmosphere Parameters for NGC 6273 Members

Star Name	T${}_{\mathrm{eff}}$	log(g)	[Fe/H]	${\xi }_{\mathrm{mic}.}$
(2MASS)	(K)	(cgs)	(dex)	(km ${{\rm{s}}}^{-1}$)
17022227−2613433c	4400	1.15	−1.67	1.90
17022817−2616426	4675	1.75	−1.66	1.95
17022912−2617443c	4325	0.80	−1.80	1.80
17023087−2618515	4575	1.10	−1.86	1.95
17023192−2614177c	4325	1.35	−1.49	2.10
17023225−2614521	4500	1.15	−1.71	1.85
17023338−2617104c	4675	1.50	−1.63	1.85
17023342−2616165	4575	1.65	−1.51	1.75
17023346−2616375	4700	1.95	−1.42	1.80
17023388−2607556	4600	1.15	−1.78	2.05
17023394−2616196	4250	0.20	−1.94	2.05
17023435−2616386	4200	0.70	−1.77	1.90
17023459−2615560c	4250	0.85	−1.85	1.90
17023460−2616038	4625	1.25	−1.72	1.90
17023517−2616130	4400	1.30	−1.39	1.65
17023523−2617058	4350	1.20	−1.52	1.70
17023529−2613089c	4500	1.40	−1.63	1.95
17023551−2616175	⋯	⋯	⋯	⋯
17023583−2616444	4775	1.70	−1.70	1.80
17023589−2615218	4775	1.90	−1.56	1.85
17023595−2615342c	4350	1.55	−1.22	2.30
17023618−2616576	4800	1.90	−1.55	1.70
17023685−2616454c	4650	1.60	−1.78	1.90
17023694−2615130	4900	2.15	−1.48	1.70
17023720−2614581a	4900	2.15	−1.54	1.70
17023723−2617063	4400	1.25	−1.64	1.80
17023728−2617024	4500	1.05	−1.83	1.75
17023744−2615306	4650	1.55	−1.78	1.35
17023783−2615095c	⋯	⋯	⋯	⋯
17023898−2618010	4650	1.65	−1.48	1.50
17023916−2616500	4300	1.00	−1.71	1.70
17023938−2619361	4550	1.15	−1.71	1.90
17023943−2615343	4575	1.10	−1.70	1.95
17023946−2615017a	4800	2.00	−1.49	1.50
17023956−2617202c	4850	2.10	−1.45	1.80
17023984−2617360a	4500	1.50	−1.41	1.80
17023993−2616370c	⋯	⋯	⋯	⋯
17024016−2615588	⋯	⋯	⋯	⋯
17024032−2617400	4700	1.95	−1.44	1.80
17024041−2617149	4550	1.25	−1.70	1.80
17024104−2616507b	4600	1.25	−1.74	1.65
17024128−2616015	⋯	⋯	⋯	⋯
17024132−2613517a	⋯	⋯	⋯	⋯
17024153−2621081	4250	0.85	−1.53	2.10
17024165−2617033b	4550	1.25	−1.90	1.75
17024173−2616245	4225	1.10	−1.57	1.95
17024226−2615137	4500	1.40	−1.59	2.10
17024242−2615557	4425	1.40	−1.60	1.85
17024289−2615274a	4650	1.10	−1.70	1.85
17024371−2620183a	4500	1.20	−1.73	1.70
17024377−2615526c	4475	1.50	−1.42	1.90
17024412−2616495	4475	1.20	−1.51	1.90
17024416−2615177b	4800	1.95	−1.44	2.00
17024472−2615190	⋯	⋯	⋯	⋯
17024566−2615124a	4775	2.40	−1.09	2.00
17024625−2610100	4400	0.75	−2.00	1.90
17024627−2614484c	⋯	⋯	⋯	⋯
17024838−2615546	4250	0.70	−1.54	2.00
17025033−2615582a	4575	2.00	−1.27	1.80

Notes.

^aObserved in Johnson et al. (2015b), the Bulge_GC1 setup, and the M2FS Calcium Triplet setup. ^bObserved in Johnson et al. (2015b) and the Bulge_GC1 setup. ^cObserved in the Bulge_GC1 and M2FS Calcium Triplet setups.

A machine-readable version of the table is available.

Download table as:  DataTypeset image

The abundances of Si i, Ca i, Cr i, Fe i, Fe ii, and Ni i were obtained by measuring the EW of individual lines selected by Johnson et al. (2015b) to be relatively free of contamination from significant blends and residual telluric features. On average, the Si i, Ca i, Cr i, Fe i, Fe ii, and Ni i abundances were based on the measurement of 2, 5, 2, 33, 4, and 4 absorption lines, respectively. However, we only measured the abundances of these elements from the Bulge_GC1 spectra. We utilized the same EW measuring code, line list, and solar reference abundances described in Johnson et al. (2015b see their Section 3.2 and their Table 2), and also used the same abfind driver in MOOG to calculate the final abundance ratios. The [Si i/Fe], [Ca i/Fe], [Cr i/Fe], [Fe i/H], [Fe ii/H], and [Ni i/Fe] abundances for every cluster member observed in the Bulge_GC1 setup are provided in Tables 5–6.

The abundances of Na i, Mg i, Al i, La ii, and Eu ii were obtained by using the synth driver in MOOG to fit synthetic spectra to the observations. The synthetic spectra were calculated using the line list developed for Johnson et al. (2015b), which is tuned to reproduce the Arcturus spectrum near the lines of interest and includes the updated CN line list from Sneden et al. (2014). We preferred to use spectrum synthesis rather than an EW analysis for these elements because their abundances are more sensitive to blending, contamination from other features, and/or broadening effects. For example, the Na and Al lines can have significant contamination from nearby atomic features and molecular CN, especially in the more metal-rich stars. Additionally, the Mg triplet near 6319 Å contains very weak lines, and the nearby continuum can be affected by a shallow but broad Ca i autoionization feature. The La and Eu lines are also relatively weak, but are further affected by hyperfine structure broadening. The Eu lines also contain a mixture of transitions from the ${}^{151}$Eu and ${}^{153}$Eu isotopes, for which we assumed the ${}^{151}$Eu:${}^{153}$Eu Solar System ratio of 47.8%:52.2% (Lawler et al. 2001).

The final [Na/Fe], [Mg/Fe], [Al/Fe], [La/Fe], and [Eu/Fe] abundances derived for cluster members observed with the Bulge_GC1 setup are provided in Tables 5–6. All atomic parameters and solar reference abundances are available in Johnson et al. (2015b; their Table 2).

In addition to the [Fe/H] abundances derived from the EW measurements of individual Fe i and Fe ii lines, we measured [Fe/H] in a larger sample of stars using the 8542 and 8662 Å CaT lines. These strong lines have been shown to be sensitive to a star's metallicity and relatively insensitive to a star's age or [α/Fe] abundance, in a variety of environments (e.g., Armandroff & Da Costa 1991; Olszewski et al. 1991; Idiart et al. 1997; Rutledge et al. 1997; Cole et al. 2004; Carrera et al. 2007; Battaglia et al. 2008; Da Costa 2016b). Although several CaT metallicity calibrations exist (e.g., Starkenburg et al. 2010; Saviane et al. 2012; Carrera et al. 2013; Vásquez et al. 2015), we followed the technique outlined in Yong et al. (2016) that utilizes the Mauro et al. (2014) calibration.

As noted by Yong et al. (2016), the Mauro et al. (2014) calibration has two significant advantages for NGC 6273: (1) the luminosity component of the calibration depends on a star's ${K}_{{\rm{S}}}$ magnitude, rather than V magnitude, which is much less affected by differential reddening; and (2) the significantly flatter slope of the summed EW (ΣEW) versus K${}_{{\rm{S}}}$(HB)–K${}_{{\rm{S}}}$ relation reduces the effects of photometric, distance, and reddening uncertainties on the derived [Fe/H] values. Additionally, we note that 2MASS provides uniform ${K}_{{\rm{S}}}$ photometry for our entire sample, but uniform V magnitudes are not yet available for all stars. However, since most CaT–metallicity relations may only be reliable down to the luminosity level of the HB (e.g., Da Costa et al. 2009), we did not determine CaT metallicities for stars fainter than the HB. This cut-off primarily affected the FLAMES CaT sample.

The Mauro et al. (2014) calibration requires a measurement of the summed 8542 Å and 8662 Å CaT EWs, defined as

$$\begin{eqnarray}&&{\rm{\Sigma }}\mathrm{EW}={\mathrm{EW}}_{8542}+{\mathrm{EW}}_{8662},\end{eqnarray} \tag{ 3 }$$

and the value ${K}_{{\rm{S}}}$(HB)–K${}_{{\rm{S}}}$, where ${K}_{{\rm{S}}}$(HB) is the magnitude of the HB. Following Yong et al. (2016), we have adopted ${K}_{{\rm{S}}}$(HB) = 12.85 mag (Valenti et al. 2007). The EWs for each line were fit using a function that is the sum, rather than the convolution, of a Gaussian and Lorentzian profile. Using Equation (7) and following Mauro et al. (2014), we adopted their relation,

$$\begin{eqnarray}&&{\rm{\Sigma }}\mathrm{EW}=-0.385[{K}_{S}(\mathrm{HB})-{K}_{S}]+{W}^{\prime },\end{eqnarray} \tag{ 4 }$$

to solve for the reduced EW (W${}^{\prime }$). The [Fe/H] values for each star were then determined using Equation (8) and the cubic calibration from Mauro et al. (2014) for the Carretta et al. (2009c) metallicity scale:

$$\begin{eqnarray}[\mathrm{Fe}/{\rm{H}}] & = & -4.61+1.842\langle W^{\prime} \rangle -0.4428\langle W^{\prime} {\rangle }^{2}\\ & & +0.04517\langle W^{\prime} {\rangle }^{3}.\end{eqnarray} \tag{ 5 }$$

The individual EWs, ΣEW, and ${W}^{\prime }$ values for all NGC 6273 members are provided in Table 7.

Table 7.  Calcium Triplet Metallicity Data

Star Name	EW${}_{8542}$	EW${}_{8662}$	$\sum \mathrm{EW}$	W'	[Fe/H]	Δ[Fe/H]
(2MASS)	(Å)	(Å)	(Å)	(Å)	(dex)	(dex)
M2FS Calcium Triplet Members
17015056−2616256	⋯	⋯	⋯	⋯	⋯	⋯
17021380−2613223	1.99	1.45	3.44	3.17	−2.01	0.10
17021778−2616058	2.11	1.47	3.58	3.21	−1.99	0.12
17022040−2616289b	2.54	1.77	4.31	3.50	−1.89	0.13
17022227−2613433c	3.04	2.32	5.36	4.24	−1.61	0.11
17022395−2614538b	2.63	2.02	4.65	3.90	−1.74	0.11
17022413−2619124	2.37	1.73	4.10	3.77	−1.79	0.10
17022442−2616495	3.29	2.46	5.75	4.47	−1.51	0.10

Notes.

^aObserved in Johnson et al. (2015b), the Bulge_GC1 setup, and the M2FS Calcium Triplet setup. ^bObserved in Johnson et al. (2015b) and the M2FS Calcium Triplet setup. ^cObserved in the Bulge_GC1 and M2FS Calcium Triplet setups. ^dObserved in the M2FS Calcium Triplet and FLAMES Calcium Triplet setups.

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image

A comparison between the observations of Yong et al. (2016) and our CaT data set revealed 27 stars in common. For this subset, the Yong et al. (2016) [Fe/H] values are on average 0.06 dex more metal-rich than ours, but the metallicities from both studies are well-correlated (see Figure 6). Similarly, we found 50 stars in our sample that were observed in both the CaT and Bulge_GC1 setups, and a comparison of the derived [Fe/H] values is provided in Figure 6. The [Fe/H] measurements from both data sets are relatively well-correlated, but the CaT data are 0.12 dex more metal-rich, on average. Therefore, the final CaT-based [Fe/H] abundances provided in Table 7, and used throughout the rest of the paper, have been shifted by −0.12 dex in order to place the CaT and Bulge_GC1 data sets on the same scale.

Zoom In Zoom Out Reset image size

For the reasonably high S/N Bulge_GC1 region spectra analyzed here, the dominant sources of internal abundance uncertainties are related to the line-to-line abundance scatter from uncertain log(gf) values, small profile fitting and/or continuum placement errors, and model atmosphere parameter uncertainties. The standard error of the mean provides a reasonable estimate of the abundance errors due to line list and profile fitting uncertainties, and for this data set we find a typical measurement uncertainty of 0.05 dex (σ = 0.03 dex) in log (X).

In order to estimate the uncertainties in ${\text{}}{T}_{\mathrm{eff}}$ and log(g), we provide a comparison of the spectroscopically derived parameters with those expected from Dartmouth isochrones (Dotter et al. 2008) with ages of 12 Gyr, [α/Fe] = +0.4 dex, and [Fe/H] = −1.75, −1.50, and −1.20 dex in Figure 7. The isochrones with different [Fe/H] are included because of the metallicity spread detected in the cluster (Han et al. 2015; Johnson et al. 2015b; Yong et al. 2016; see also Section 5.1). Figure 7 shows that the derived temperature and surface gravity values are in good agreement with those predicted by the isochrones. Specifically, we find the average differences between the spectroscopic and isochrone temperature (ΔT${}_{\mathrm{eff}}$) and surface gravity (Δlog(g)) values to be −8 K and +0.01 cgs, respectively, and do not detect any significant trends as a function of temperature, gravity, or metallicity. The dispersions in ΔT${}_{\mathrm{eff}}$ and Δlog(g) are found to be 92 K and 0.17 cgs, respectively. Therefore, we have adopted 100 K and 0.15 cgs as the typical model atmosphere uncertainties for ${\text{}}{T}_{\mathrm{eff}}$ and log(g). For the model atmosphere metallicity, we have adopted an uncertainty of 0.10 dex based on the combined measurement errors of [Fe i/H] and [Fe ii/H]. Additionally, we estimate the typical ${\xi }_{\mathrm{mic}.}$ uncertainty to be 0.10 km ${{\rm{s}}}^{-1}$ based on the scatter and fitting uncertainties present in plots of log (Fe i) versus log(EW/λ).

Zoom In Zoom Out Reset image size

The abundance uncertainty values (Δ[X/Fe] or Δ[Fe/H]) were determined by rerunning MOOG and changing each model atmosphere parameter by the estimated uncertainties listed previously. Only one parameter was changed per run while the other values were held fixed. To speed up the analysis, we converted abundances to EWs for the elements measured by spectrum synthesis using the ewfind driver in MOOG. The total internal abundance uncertainties listed in Tables 5–6 were determined by adding the model atmosphere error terms, plus the random measurement uncertainties, in quadrature.

For the CaT data, we estimated the abundance uncertainties by analyzing the correlation between the 8542 and 8662 Å EWs. In other words, we used the strong correlation between EW${}_{8542}$ and EW${}_{8662}$ to predict the EW of each line based on the other one. The predicted EWs were then propagated through Equations (8)–(9), and the difference between these values and the [Fe/H] abundance given in Table 7 was taken as the measurement error. Using this method, we found an average Δ[Fe/H] = 0.15 dex (σ = 0.07 dex). We note that this value is similar to the fitting uncertainty of Equation (9) (Mauro et al. 2014). A typical [Fe/H] uncertainty of 0.15 dex is also similar to the 1σ scatter (0.21 dex) between [Fe/H] values determined from the CaT and Bulge_GC1 data.

The color and CaT abundance spreads observed by Piotto et al. (1999) and Rutledge et al. (1997) provided some of the first evidence that NGC 6273 may host stars with different metallicities. More recently, high-resolution spectroscopic measurements from Johnson et al. (2015b) showed that the cluster contains stars with [Fe/H] ranging from −1.80 to −1.30 dex, and also found that the cluster hosts at least two distinct populations separated in [Fe/H] by ∼0.25 dex. Similarly, Han et al. (2015) used narrow-band hk photometry to clearly show that the cluster's sub-giant branch and RGB are split into two sequences with different compositions. Yong et al. (2016) also reported CaT metallicities ranging from [Fe/H] = −1.84 to −0.70 dex, further indicating the presence of a large metallicity spread in the cluster.

The Johnson et al. (2015b) and Yong et al. (2016) spectroscopic results are based on the analysis of only 18 and 44 RGB stars observed in the Bulge_GC1 and CaT spectral regions, respectively. Therefore, we add here [Fe/H] measurements for 51 RGB members in the Bulge_GC1 region and 191 RGB members in the CaT region (see Tables 4, 6, and 7). For the Bulge_GC1 data, we find a full range of [Fe/H] = −2.00 to −1.09 dex, an average $\langle [\mathrm{Fe}/{\rm{H}}]\rangle =-1.61$ dex, a dispersion (${\sigma }_{[\mathrm{Fe}/{\rm{H}}]}$) of 0.18 dex, and an interquartile range (IQR) of 0.24 dex. Similarly, the CaT data exhibit a full range of [Fe/H] = −2.22 to −0.56 dex, an average $\langle [\mathrm{Fe}/{\rm{H}}]\rangle =-1.73$ dex, ${\sigma }_{[\mathrm{Fe}/{\rm{H}}]}$ = 0.24 dex, and an IQR of 0.27 dex. A comparison between the Bulge_GC1 and CaT metallicity distributions is shown in Figure 8, and both data sets provide evidence that NGC 6273 harbors an intrinsic metallicity spread.

Zoom In Zoom Out Reset image size

Further examination of Figure 8 also indicates that NGC 6273 may host distinct populations with different [Fe/H], rather than just a broadened distribution. Specifically, Figure 8 suggests that at least three major components may exist: (1) a "metal-poor" population ([Fe/H] ≤ −1.65); (2) a "metal-intermediate" population (−1.65 < [Fe/H] ≤ −1.35); and a "metal-rich" tail ([Fe/H] > −1.35), and that these components constitute 46% ± 8%, 48% ± 8%, and 6% ± 4% of our total Bulge_GC1 data set, respectively. We find the average metallicities of the metal-poor, metal-intermediate, and metal-rich populations to be $\langle [\mathrm{Fe}/{\rm{H}}]\rangle =-1.77$ dex (σ = 0.08 dex), $\langle [\mathrm{Fe}/{\rm{H}}]\rangle =-1.51$ dex (σ = 0.07 dex), and $\langle [\mathrm{Fe}/{\rm{H}}]\rangle =-1.22$ dex (σ = 0.09 dex), respectively. The clustering of stars near [Fe/H] = −1.75 and −1.50 is consistent with the [Fe/H] abundances and split RGB sequences derived by Johnson et al. (2015b) and Han et al. (2015), and the presence of a metal-rich tail extending up to at least [Fe/H] ≈ −1 matches the findings of Yong et al. (2016).

Figure 9 indicates that a radial metallicity gradient may exist in the cluster such that the metal-intermediate stars are more centrally concentrated than the metal-poor stars. Although the metal-rich stars observed with the Bulge_GC1 setup all reside inside 3' of the cluster center (see also Figure 3), the sample size is too small to draw any strong conclusions about this population's radial distribution. For the two dominate populations, a difference in their radial distributions is only observed at projected distances ≳15 from the cluster center, and a two-sided Kolmogorov–Smirnov test indicates that we do not have enough evidence to reject the null hypothesis that the two data sets are drawn from the same radial distribution. However, we note that the radial range where the distributions may differ is within ∼1–3 half-mass radii, which is the region that Vesperini et al. (2013) estimate the local population mixtures may closely match the global ratios. Interestingly, if the radial segregation of stars with different metallicities is confirmed from larger sample sizes, then NGC 6273 would share a similar metallicity gradient morphology with ω Cen (e.g., Norris et al. 1996; Suntzeff & Kraft 1996; Rey et al. 2004; Bellini et al. 2009; Johnson & Pilachowski 2010). Such a gradient would contrast with NGC 1851, where Carretta et al. (2010b) found the metal-poor stars to be the most centrally concentrated.

Zoom In Zoom Out Reset image size

Spectroscopic observations have indicated that several clusters, including ω Cen (e.g., Norris & Da Costa 1995; Johnson & Pilachowski 2010; Marino et al. 2011a), NGC 5286 (Marino et al. 2015), M2 (Yong et al. 2014), M54 (Carretta et al. 2010a), Terzan 5 (Origlia et al. 2013; Massari et al. 2014), NGC 1851 (Yong & Grundahl 2008; Carretta et al. 2011; Lim et al. 2015), and M22 (Pilachowski et al. 1982; Da Costa et al. 2009; Marino et al. 2009, 2011b), may host multiple populations with distinct [Fe/H] ratios. However, recent studies by Mucciarelli et al. (2015b) and Lardo et al. (2016) claim that at least some of these [Fe/H] spreads are spurious detections driven by a disparity between [Fe i/H] and [Fe ii/H]. Similarly, Ivans et al. (2001), Lapenna et al. (2014), and Mucciarelli et al. (2015c) found that [Fe/H] determinations for RGB and AGB stars can differ systematically by >0.1 dex, and that mixing RGB and AGB stars in a sample can produce an artificial metallicity spread. A common thread connecting these issues is the method by which a star's surface gravity is determined (spectroscopic versus photometric). Specifically, spectroscopic determinations may produce optimal log(g) values that correspond to masses that are systematically too low (<0.5 ${M}_{\odot }$ in many cases). Since we utilize a spectroscopic surface gravity method and find that NGC 6273 shares many chemical and morphological characteristics with clusters such as M2 and M22, for which intrinsic [Fe/H] spreads are contested, it is prudent to examine alternative lines of evidence that may support or refute NGC 6273 possessing an intrinsic metallicity spread.

Both spectroscopy and photometry unambiguously agree that ω Cen possesses discrete RGB populations with different [Fe/H], and in Figure 10 we directly compare the spectra of NGC 6273 and ω Cen stars that have physical parameters typical of those in the metal-poor, metal-intermediate, and metal-rich groups. The ω Cen temperature and gravity parameters were determined entirely from photometric methods (see Johnson & Pilachowski 2010), assuming masses of 0.8 ${M}_{\odot }$, and therefore should avoid the potential spectroscopic gravity problems noted above. As can be seen in Figure 10, the nearly identical Fe i and Fe ii line profiles suggest that the NGC 6273 and ω Cen stars share similar compositions across a wide [Fe/H] range. We note also that the CaT [Fe/H] distribution shown in Figure 8 for NGC 6273 closely matches the extended metallicity distribution found in ω Cen (e.g., Norris et al. 1996; Suntzeff & Kraft 1996).

Zoom In Zoom Out Reset image size

As shown in Figure 2, the broad color dispersion along the upper RGB provides some evidence that NGC 6273 may harbor an intrinsic metallicity spread. We investigate this further in Figure 11 by examining the upper RGB regions of the (F336W)${}_{{\rm{o}}}$ versus (F336W–F555W)${}_{{\rm{o}}}$ and (F555W)${}_{{\rm{o}}}$ versus (F438W–F555W)${}_{{\rm{o}}}$ color–magnitude diagrams and identifying the Bulge_GC1 spectroscopic targets with different [Fe/H]. Both color–magnitude diagrams indicate that the two dominant metallicity groups tend to separate on the upper RGB. The (F336W)${}_{{\rm{o}}}$ versus (F336W–F555W)${}_{{\rm{o}}}$ plot in particular suggests that the brightest ∼0.5 mag of the RGB-tip may split into at least two sequences with different [Fe/H], which is similar to the result found by Marino et al. (2015) for NGC 5286. However, the F336W and F438W filters, and by extension the F336W–F555W and F438W–F555W colors, can be sensitive to both a star's overall metallicity and its C+N+O abundances. Therefore, the color–magnitude diagrams shown in Figure 11 are consistent with an intrinsic metallicity spread, but a detailed examination of the cluster's CNO (and also He) abundances is required in order to fully confirm this result. Interestingly, the two metal–rich stars in Figure 11 are located at colors that are bluer than might be expected from their metallicities alone. We note that similar observations have been found for the equivalent "s-poor/Fe-rich" stars in NGC 5286 (Marino et al. 2015) and M2 (Yong et al. 2014). The bluer colors for these stars may reflect lower atmospheric opacities driven by different light element compositions and perhaps lower [α/Fe] ratios, at least for NGC 6273.²³

Zoom In Zoom Out Reset image size

Figure 11 also shows that very few of our Bulge_GC1 targets are on the AGB, indicating that the measured metallicity spread is not caused by systematic differences in the RGB and AGB abundance scales. In fact, the few AGB stars in our sample appear to belong to both the metal-poor and metal-intermediate populations, and the two metal-rich stars could also belong to a more metal-rich AGB sequence. Therefore, we regard the combined evidence of separate sub-giant and RGB sequences observed by Han et al. (2015) with the hk filter, the RGB color dispersions seen in Figures 2 and 11 here, the large metallicity spreads detected previously by Johnson et al. (2015b) and Yong et al. (2016), and the spectroscopic data presented here as strong evidence that NGC 6273 possesses an intrinsic metallicity spread.

5.3.1. Alpha Element Abundances

The α elements Mg, Si, and Ca are largely produced during hydrostatic and explosive carbon, neon, and oxygen burning in massive stars (e.g., Woosley & Weaver 1995). In environments where chemical enrichment has been dominated by the products of core-collapse supernovae (SNe), one tends to find stars with [α/Fe] abundances that are enhanced by about a factor of two to three over the solar ratio (e.g., see the review by McWilliam 1997). In contrast, longer enrichment timescales may produce stars with lower [α/Fe] ratios as SNe Ia begin to contribute larger amounts of Fe-peak elements than α elements (Tinsley 1979).

As can be clearly seen in Gratton et al. (2004; their Figure 4), nearly all Galactic globular clusters have [α/Fe] ∼ 0.2–0.4 dex. Furthermore, the star-to-star scatter of [α/Fe] within a given cluster is typically <0.1 dex, which suggests that the products of core-collapse SNe were well-mixed. Only a small number of clusters, such as Ruprecht 106, Terzan 7, and Palomar 12, are known to have abnormally low [α/Fe] ratios (e.g., Pritzl et al. 2005), and all three of these clusters are thought to have extragalactic/accretion origins (e.g., Cohen 2004; Law & Majewski 2010; Villanova et al. 2013). Therefore, one does not normally expect to find stars with enhanced and depleted [α/Fe] ratios within a single globular cluster, beyond the well-known proton-capture nucleosynthesis variations (see Section 5.3.2). In fact, only the massive iron-complex clusters ω Cen, NGC 6273, M54, M2, and Terzan 5 show any evidence of hosting stars with different [α/Fe] ratios (Pancino et al. 2002; Origlia et al. 2003, 2011, 2013; Carretta et al. 2010a; Johnson & Pilachowski 2010; Yong et al. 2014; Johnson et al. 2015b).

Figure 12 and Table 8 show the [Mg/Fe], [Si/Fe], [Ca/Fe], and averaged [α/Fe] patterns of NGC 6273's various populations. In agreement with Johnson et al. (2015b), we find that most stars in NGC 6273 have elevated α element abundances, but that the average [Mg/Fe] and [Si/Fe] ratios may decrease slightly as a function of increasing metallicity. The [Si/Fe] abundances in particular may show additional substructure, and we find some evidence that the average [Si/Fe] abundances of the "α-enhanced" metal-intermediate stars may be lower than those of the metal-poor and metal-rich groups. We note that a similar change in the [Si/Fe] abundances with [Fe/H] has been observed in ω Cen (Johnson & Pilachowski 2010), which suggests that this trend could be the signature of a particular self-enrichment mode in massive clusters. However, the [Ca/Fe] abundances show no significant trends as a function of [Fe/H], and the typical dispersion within each sub-population is ∼0.1 dex.

Zoom In Zoom Out Reset image size

Table 8.  Mean Composition Properties

Statistic	[Fe/H]	[Na/Fe]	[Mg/Fe]	[Al/Fe]	[Si/Fe]	[Ca/Fe]	[Cr/Fe]	[Ni/Fe]	[La/Fe]	[Eu/Fe]
	(dex)	(dex)	(dex)	(dex)	(dex)	(dex)	(dex)	(dex)	(dex)	(dex)
Metal-poor Population
Average	−1.77	0.30	0.40	0.74	0.35	0.25	0.00	−0.03	0.15	0.39
σ	0.08	0.19	0.08	0.34	0.10	0.09	0.10	0.10	0.24	0.15
Metal-intermediate Population
Average	−1.51	0.26	0.34	0.75	0.23	0.22	0.02	−0.06	0.47	0.36
σ	0.07	0.20	0.10	0.27	0.15	0.11	0.13	0.10	0.28	0.12
Metal-rich Population
Average	−1.22	0.19	0.17	0.46	0.11	0.16	0.02	−0.11	0.57	0.42
σ	0.09	0.34	0.16	0.41	0.21	0.21	0.21	0.21	0.40	0.21

Download table as:  ASCIITypeset image

In a previous analysis, Johnson et al. (2015b) discovered that the most metal-rich star in their sample exhibited low [X/Fe] ratios for several species, including the α elements. The new data presented here indicate that not all metal-rich stars have low [α/Fe], but we find at least five "low-α" stars that have approximately solar [Mg/Fe], [Si/Fe], and [Ca/Fe] abundances. As can be seen in Figure 12, all five low-α stars have [Fe/H] > −1.5 dex. Additionally, the specific frequency of low-α stars increases with metallicity such that these stars constitute 9% (3/33) of the metal-intermediate population and 50% (2/4) of the metal-rich population. However, we caution that the measured specific frequency values are likely affected by small number statistics and should be confirmed with additional observations.

Although we noted above that ω Cen, M54, M2, and Terzan 5 also contain stars with higher [Fe/H] and lower [α/Fe], none of these clusters exactly matches the pattern of NGC 6273. For example, the α-poor stars in M2 and Terzan 5 are exclusively found in the most metal-rich populations, but neither cluster appears to contain α-enhanced and α-poor stars at the same metallicity. Although Johnson & Pilachowski (2010; see their Figure 10) found several stars with high and low [Si/Fe] and [Ca/Fe] abundances across a broad range of [Fe/H] in ω Cen, follow-up observations are required to confirm that this pattern matches what is found in NGC 6273.

Interestingly, the M54 cluster and Sagittarius nuclear field star system may provide the closest example to what is observed in NGC 6273 (see Carretta et al. 2010a). In this system, the metal-poor cluster M54 contains a metallicity spread but only α-enhanced stars. In contrast, the surrounding galaxy field stars are generally more metal-rich and have lower [α/Fe]. Therefore, if NGC 6273 formed in the core of a system similar to the Sagittarius dwarf galaxy, then the cluster may have been able to accrete a small number of metal-rich, α-poor field stars from its progenitor population. Alternatively, the low-α stars in NGC 6273 may have been preferentially polluted by the ejecta of SNe Ia, perhaps in a scenario similar to that discussed in D'Antona et al. (2016). However, such a scenario would have to be able to produce low-α stars with different [Fe/H] but otherwise similar compositions (see Sections 5.3.2 and 5.3.3), and may even have to occur multiple times in clusters like NGC 6273.

5.3.2. Light Element Abundances

As mentioned in Section 1, globular clusters show clear light element abundance variations that extend beyond the effects of first dredge-up and are a result of high temperature (>40 MK; Langer et al. 1993, 1997; Prantzos et al. 2007) proton-capture burning. Since these effects are observed in main-sequence and evolved RGB stars (e.g., Gratton et al. 2001), we know that the gas from which present day cluster stars formed was polluted by a previous generation of more massive stars. Although the exact nucleosynthesis sources remain a mystery, the observed effects include anti-correlations among the element pairs C–N, O–N, O–Na, O–Al, and Mg–Al and correlations of C–O, N–Na, and Na–Al (e.g., Sneden et al. 2004). He enhancements are also likely found in stars with low O/Mg and high Na/Al (e.g., Bragaglia et al. 2010a, 2010b; Dupree et al. 2011). For this paper, we adopt the common nomenclature that "first generation" stars are those with compositions similar to metal-poor halo field stars (i.e., lower He, N, Na, and Al abundances; higher C, O, and Mg abundances) and "second generation" stars are those with enhanced He, N, Na, and Al abundances and depleted C, O, and possibly Mg abundances.

The Mg–Al anti-correlation is only found in a handful of the most massive clusters, but may be particularly useful for identifying discrete populations (e.g., Carretta 2014, 2015). Since the full Mg–Al cycle is activated at a higher temperature than the O–N and Ne–Na cycles, the presence (or not) of a Mg–Al anti-correlation provides important insight into the burning temperatures achieved by the pollution sources. Similarly, a few of the most massive clusters also exhibit abundance variations that extend to elements as heavy as Si, K, and Sc, which is likely a byproduct of even higher temperature proton-capture burning (Yong et al. 2005; Carretta et al. 2009b, 2013, 2014; Johnson & Pilachowski 2010; Cohen & Kirby 2012; Mucciarelli et al. 2012, 2015a; Ventura et al. 2012; Carretta 2015; Roederer & Thompson 2015). Notably, many of these clusters share similar properties with NGC 6273, such as extended blue HBs.

In Figure 13 and Table 8, we compare the [Na/Fe], [Mg/Fe], [Al/Fe], and [α/Fe] abundances of the three different metallicity groups in NGC 6273. Similar to the results of Johnson et al. (2015b), we find that both [Na/Fe] and [Al/Fe] vary by about factors of 5 and 10, respectively, and that clear Na–Al correlations are independently present in the metal-poor, metal-intermediate, and metal-rich populations. Therefore, NGC 6273 shares a common feature observed in other iron-complex clusters: each population with a unique metallicity was able to generate its own independent spread of light element abundances that closely resembles the patterns exhibited by monometallic clusters.

Zoom In Zoom Out Reset image size

The Na–Al correlation in Figure 13 shows a paucity of stars near [Al/Fe] ∼ 0.6 dex. If we adopt this cut-off as the discriminator between first and second generation stars, then we find that approximately two-thirds of the cluster stars can be classified as second generation stars. The metal-poor and metal-intermediate populations each favor second generation stars with first:second generation ratios of 35%:65% and 28%:72%, respectively. On the other hand, the metal-rich population has a ratio of 75%:25%, but this measurement is based on only four stars. Therefore, the numerical dominance of second generation stars in NGC 6273 fits a common trend observed in many Galactic globular clusters (e.g., Carretta et al. 2009c, see their Figure 10). Similarly, Figure 9 shows that the second generation stars in NGC 6273 are more centrally concentrated than the first generation stars, which again matches a pattern observed in many iron-complex and monometallic clusters (e.g., Lardo et al. 2011). We note also that an additional paucity of stars may be present near [Na/Fe] ∼ 0.35 dex, which may further distinguish the most Na/Al-rich stars. These stars, which constitute ∼33% of our sample, are likely equivalent to the "extreme" population found by Carretta et al. (2009c) in several clusters, the "E" population of NGC 2808 (Milone et al. 2015b), and the faint sub-giant branch stars of 47 Tuc (Marino et al. 2016). We note that a similarly high fraction of very Na/Al-rich stars is also found in ω Cen (Johnson & Pilachowski 2010; Marino et al. 2011a) and M54 (Carretta et al. 2010a).

For [Mg/Fe] and [Al/Fe], Figure 13 shows that the behavior of the element pair may change for stars of different metallicity in NGC 6273. The metal-poor component shows no correlation between [Mg/Fe] and [Al/Fe], but the metal-intermediate population shows evidence of a Mg–Al anti-correlation for stars with [Al/Fe] < 1.0 dex. A similar Mg–Al anti-correlation may also be present for the metal-rich stars, but the sample size (four stars) is too small to draw any clear conclusions. Since ${}^{24}$Mg is only significantly depleted at temperatures ≳65 MK (e.g., see Prantzos et al. 2007; their Figure 2), the different Mg–Al relations for the metal-poor and metal-intermediate stars suggest that the gas from which each population's second generation stars formed was processed at different temperatures. However, we did not find any residual correlations between Mg or Al and the heavier elements like Si, which indicates that the pollution source(s) responsible for the Mg–Al anti-correlation in NGC 6273 likely did not reach temperatures high enough to significantly activate the ${}^{27}$Al(p,γ)${}^{28}$Si reaction related to the Mg–Al chain.

An examination of Mg–Al trends in the iron-complex clusters ω Cen, M54, M2, M22, NGC 1851, and NGC 5286 revealed that only ω Cen (Norris & Da Costa 1995; Smith et al. 2000; Da Costa et al. 2013), M54 (Carretta et al. 2010a), and NGC 1851 (Carretta et al. 2011, 2012) exhibit evidence of Mg–Al anti-correlations. Unlike NGC 6273, none of these clusters show evidence that the presence of a Mg–Al anti-correlation depends on a population's metallicity. However, we note that in ω Cen the metal-intermediate and metal-rich stars exhibit clear changes in their light element patterns (e.g., Norris & Da Costa 1995; Johnson & Pilachowski 2010; Marino et al. 2011a). For example, the very O-poor/Na-rich stars that dominate by number at higher metallicity are not found at [Fe/H] ≲ −1.8, and O and Na are actually correlated in the most metal-rich stars. Additionally, for M54, Carretta et al. (2010a) found that the light element variations are more extended for the metal-rich stars than the metal-poor population. Therefore, NGC 6273, ω Cen, and M54 provide evidence that a cluster's enrichment signature can change with time, and that multiple pollution sources may be able to produce chemical patterns that are similar for some element pairs (e.g., Na–Al) but not others (e.g., Mg–Al).

Interestingly, Figure 13 shows that the metal-intermediate stars with [Al/Fe] > 1.0 dex have [Mg/Fe] ∼ 0.4 dex, rather than the [Mg/Fe] ∼ 0.0 dex abundances that might be expected. The reason for this discrepancy is not immediately clear, but we note that many similar metallicity clusters have stars with [Mg/Fe] ∼ 0.4 dex and [Al/Fe] ∼ 1.0 dex (e.g., Carretta et al. 2010b; see their Figure 6). Additionally, we note that Norris & Da Costa (1995) found that intermediate metallicity stars in ω Cen could have [Al/Fe] > 1.0 dex but [Mg/Fe] could range from about 0.6 dex (no Mg–Al anti-correlation) to 0.0 dex (clear Mg–Al anti-correlation; see also Carretta et al. 2010a, their Figure 18). In this context, an examination of the ${}^{24}$Mg, ${}^{25}$Mg, and ${}^{26}$Mg abundances in NGC 6273, similar to the analysis of Da Costa et al. (2013) in ω Cen, could be particularly illuminating. However, we also caution that the 6319 Å Mg i lines used here are relatively weak, especially in stars with intrinsically low [Mg/Fe], so the exact shape of the Mg–Al anti-correlation should be confirmed with additional analyses.

Finally, we note that all of the low-α stars have [Na/Fe] and [Al/Fe] compositions that are consistent with those of first generation stars. Although the sample size of low-α stars is small, a simulation of 10${}^{5}$ random draws from our α-enhanced population indicated that there is only about a 0.05% chance that we would randomly draw five stars that have [Na/Fe] < 0.25 dex and [Al/Fe] < 0.50 dex. Therefore, we speculate that the low-α population may have been unable to form second generation stars. The pattern of low [Na/Fe] and [Al/Fe] abundances in the low-α stars of NGC 6273 mirrors the composition differences found by Carretta et al. (2010a) when comparing the M54 cluster and Sagittarius galaxy field stars. The similar composition patterns of the low-α stars in NGC 6273 and the Sagittarius field stars strengthens the idea that NGC 6273 may have accreted its low-α stars from a surrounding field population that was once part of a now dispersed dwarf galaxy.

5.3.3. RGB Versus AGB Abundance Patterns

As noted by Gratton et al. (2010) and many previous authors (e.g., Mallia 1978; Norris et al. 1981; Suntzeff 1981; Smith & Norris 1993; Pilachowski et al. 1996b; Ivans et al. 1999; Sneden et al. 2000), some globular clusters may contain RGB and AGB populations with different light element abundances. Specifically, RGB stars that evolve onto the HB with masses ≲0.55 M${}_{\odot }$, presumably those with the highest He, N, Na, and Al abundances and lowest C, O, and Mg abundances, may not ascend the AGB and instead end their lives as AGB-manqué stars (e.g., Greggio & Renzini 1990). As a result, we expect to find that the light element abundance distributions of AGB stars should exhibit a paucity of second generation stars when compared to the RGB ratios.

Renewed interest in this field has produced somewhat conflicting results with the missing second generation fraction ranging from 100% (Campbell et al. 2013; MacLean et al. 2016) to only a few percent (Johnson & Pilachowski 2012; García-Hernández et al. 2015; Johnson et al. 2015a; Lapenna et al. 2016; Wang et al. 2016). However, the growing consensus is that only the most extreme second generation stars probably fail to ascend the AGB.

Since Figure 14 shows that NGC 6273 contains a very extended blue HB, and that ∼30% of the cluster's HB stars have masses ≲0.55 M${}_{\odot }$, we investigate here whether any second generation stars may have failed to ascend the AGB. We restrict the comparison to only the targets shown in Figure 11 since these are the only stars in our sample that can be reliably assigned to either the RGB or AGB sequences. Although the sample sizes are small (9 AGB; 28 RGB), we find similar first:second generation ratios of 24%:76% and 14%:86% for the RGB and AGB samples, respectively. However, further inspection of the [Na/Fe] and [Al/Fe] distributions in Figure 15 reveals that the AGB sample does not contain stars with [Na/Fe] > 0.5 dex nor [Al/Fe] > 1.0 dex. In other words, only the most Na/Al-rich, and presumably He-enhanced, stars may have failed to ascend the AGB.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

It is possible that the paucity of extreme Na/Al-rich AGB stars is a product of our small sample size. To investigate this, we performed ${10}^{5}$ random draws of an equivalent AGB sample from the RGB distribution, and we found about a 7% chance that the missing Na/Al-rich AGB stars could be due to the small sample size. Interestingly, the missing Na/Al-rich AGB stars account for ∼30% of the RGB sample, which is comparable to the fraction of extreme HB and blue hook stars found on the HB (see Figure 14 and Section 5.4). Therefore, we conclude that the NGC 6273 RGB stars with [Na/Fe] > 0.5 dex and [Al/Fe] > 1.0 dex likely evolve to become extreme HB or blue hook stars and fail to ascend the AGB.

5.3.4. Fe-peak Element Abundances

The Fe-peak elements Cr and Ni are largely produced in the late burning stages of massive stars, but include some production by SNe Ia as well (e.g., Timmes et al. 1995). Within a single globular cluster, the star-to-star scatter in [Ni/Fe] and [Cr/Fe] is typically ≲0.1 dex (e.g., Gratton et al. 2004, see their Figure 2). Similarly, the average [Cr/Fe] and [Ni/Fe] ratios are about solar across the entire metallicity range spanned by clusters in the Galaxy.

In Figure 16 and Table 8, we show the abundance patternsof [Cr/Fe] and [Ni/Fe] for NGC 6273. Overall, we find $\langle [\mathrm{Cr}/\mathrm{Fe}]\rangle =0.01$ dex (σ = 0.12 dex) and $\langle [\mathrm{Ni}/\mathrm{Fe}]\rangle \,=-0.05$ dex (σ = 0.11 dex), which is in agreement with Johnson et al. (2015b). An examination of Figure 16 shows that the metal-poor, metal-intermediate, and metal-rich stars all exhibit nearly identical [Cr/Fe] and [Ni/Fe] abundances and dispersions. However, we note that several (but not all) of the low-α stars have [Cr, Ni/Fe] ≲ −0.2 dex, similar to what is found in some clusters associated with the Sagittarius dwarf galaxy. A detailed examination of key Fe-peak elements that are sensitive to nucleosynthesis processes operating in different environments, such as Mn, Co, Zn, and Cu (e.g., Nomoto et al. 2006), may provide additional insight into whether the stars with low [α/Fe], [Cr/Fe], and [Ni/Fe] have similar origins.

Zoom In Zoom Out Reset image size

5.3.5. Neutron-capture Element Abundances

Most of the stable isotopes heavier than the Fe-peak are produced either by the r-process over short timescales or by the s-process over much longer timescales (e.g., see the review by Sneden et al. 2008). As a result, old globular clusters tend to have heavy element compositions that are dominated by r-process nucleosynthesis, which is evidenced by their characteristically low [La/Eu] ratios (e.g., see Gratton et al. 2004, their Figure 6). Although the Galactic globular cluster system exhibits a trend of increasing s-process contributions at higher [Fe/H] (e.g., James et al. 2004), a small number of clusters, such as M4 (e.g., Ivans et al. 1999), deviate from this trend and exhibit significantly higher [La/Eu] ratios. In these clusters, the gas from which their stars formed likely experienced additional, but uniform, pollution from a previous generation of ∼1.5–4 ${M}_{\odot }$ AGB stars (e.g., Busso et al. 1999).

As mentioned in Section 1, one of the "chemical tags" of iron-complex clusters is that they exhibit clear correlations between [Fe/H] and the products of s-process enrichment. All iron-complex clusters for which the heavy elements have been measured contain populations of Fe/s-poor and Fe/s-rich stars with similar Ba and La enhancements (Johnson et al. 2015b; Marino et al. 2015). As a result, merger scenarios seem unlikely for every case because each cluster would have had to form from the coalescence of populations with nearly identical Fe/s-poor and Fe/s-rich compositions (but see also Gavagnin et al. 2016). Instead, we regard the combination of [Fe/H] and s-process enhancements as a sign that iron-complex clusters were able to sustain extended star formation and self-enrichment, and that the time frame was long enough for low and intermediate mass AGB stars to contribute to the composition of the more metal-rich stars.

Figure 16 and Table 8 show a clear increase in [La/Fe] with [Fe/H] for NGC 6273, in agreement with the results of Johnson et al. (2015b). Therefore, we confirm that NGC 6273 possesses the same s-process enrichment profiles as other iron-complex clusters. We also find for Eu that the cluster average is about [Eu/Fe] = 0.4 dex, regardless of a star's metallicity. This suggests that massive stars were largely responsible for the increase in [Fe/H] within the cluster, and that the production rate of Fe and Eu was approximately constant. In Figure 17, we update the analysis of Johnson et al. (2015b) with a sample size that is ∼3× larger and confirm that the rise in [La/Fe], and thus the [La/Eu] ratio, with metallicity is due to almost pure s-process enrichment. In fact, if we assume that the most La-poor stars represent the initial pure r-process composition of the cluster, a simple dilution model shows that nearly all of the stars can be accounted for by adding ∼90% s-process material and ∼10% r-process material to the initial r-process composition. The constant r-process contribution is qualitatively in agreement with the [Eu/Fe] observations of Figure 16 because some level of r-process enrichment is required to maintain the cluster's overall Eu enhancement at higher [Fe/H].

Zoom In Zoom Out Reset image size

Interestingly, the low-α stars in Figure 16 either have [La/Fe] ∼ 0.6 dex and [Eu/Fe] ∼ 0.7 dex or [La/Fe] ∼ 0.0 dex and [Eu/Fe] ∼ 0.2 dex. Although the origin of these stars is not clear, it is tempting to speculate that two different formation channels may exist (e.g., in situ versus accretion). We note in particular that low-α stars with high [La/Fe] and [Eu/Fe] are found in the Sagittarius field (e.g., McWilliam et al. 2013), albeit at higher [Fe/H]. The existence of these stars further strengthens the idea that at least some of the low-α stars in NGC 6273 could have been accreted from a surrounding field population. The low-α stars with lower [La/Fe] and [Eu/Fe] are perhaps a bigger puzzle, but they could have been formed in situ and preferentially enriched by SNe Ia or massive stars with peculiar enrichment signatures. However, Figure 17 shows that all of the low-α stars have about the same [La/Eu] ratios, and may even fall on a separate enrichment sequence. In any case, the simple dilution model shown in Figure 17 suggests that the low-α stars experienced significant r-process enrichment compared to a majority of the α-enhanced metal-intermediate and metal-rich cluster stars.

NGC 6273 has long been known to exhibit a peculiar HB morphology that includes a very extended blue HB, a clear gap near temperatures of ∼20,000 K, and a large population of blue hook stars (Piotto et al. 1999; Brown et al. 2001, 2010; Momany et al. 2004). We confirm these features with new HST color–magnitude diagrams in Figure 14, and find that NGC 6273's HB includes several distinct groups.²⁴ Although a detailed examination of each HB group is beyond the scope of this paper, we draw attention to NGC 6273's large blue hook population in the context of its complex formation history.

Blue hook stars are among the hottest core He burning stars in old globular clusters, and are thought to form when stars reach the RGB-tip with masses low enough to delay the core He flash until after a star reaches the white dwarf cooling sequence (e.g., D'Cruz et al. 1996; Moehler et al. 2000; Brown et al. 2010). The presence of blue hook stars is known to correlate with cluster mass (Rosenberg et al. 2004; Dieball et al. 2009; Brown et al. 2010, 2016), which we illustrate in Figure 18 by showing that both the ratio of blue hook to canonical HB stars $\left(\tfrac{{N}_{\mathrm{BHk}}}{{N}_{\mathrm{HB}}}\right)$ and the raw number of blue hook stars (N${}_{\mathrm{BHk}}$) is higher in the more massive clusters. However, He enhancement is also likely tied to blue hook formation (e.g., D'Antona et al. 2002; Tailo et al. 2015).

Zoom In Zoom Out Reset image size

Although present day cluster mass and the level of He-enrichment strongly correlate with the presence of blue hook stars, neither parameter nor a combination of the two parameters seems adequate to completely predict blue hook formation. For example, He-enrichment scenarios (e.g., D'Antona et al. 2010) are presently unable to explain the significant carbon enhancements that are found in He-enhanced blue hook stars (Moehler et al. 2007, 2011; Latour et al. 2014), and Figure 18 shows that clusters with similar absolute magnitudes (proxies for masses) can have vastly different blue hook populations. To illustrate this point, we note that the iron-complex clusters NGC 6273 and M2 differ by only 0.1 mag in ${M}_{{\rm{V}}}$, have similarly extended blue HB morphologies, exhibit comparable light and heavy element abundance variations, have similar average metallicities and ages, and have total HB counts that agree to within $0.5 \%$, but M2 has a $\tfrac{{N}_{\mathrm{BHk}}}{{N}_{\mathrm{HB}}}$ ratio of 0.006 (3 blue hook stars) whereas NGC 6273 has $\tfrac{{N}_{\mathrm{BHk}}}{{N}_{\mathrm{HB}}}=0.224$ (∼120 blue hook stars).²⁵ Furthermore, dynamical and binary star evolutionary processes may be ruled out as explanations because the blue hook stars in clusters with large $\tfrac{{N}_{\mathrm{BHk}}}{{N}_{\mathrm{HB}}}$ ratios, including NGC 6273, do not exhibit radial gradients (e.g., Bedin et al. 2000; Brown et al. 2010). Therefore, additional parameters must play a role in producing blue hook stars.

Interestingly, the three objects in Figure 18 that contain >100 blue hook stars and have $\tfrac{{N}_{\mathrm{BHk}}}{{N}_{\mathrm{HB}}}$ > 0.10 are the iron-complex clusters NGC 6273, M54, and ω Cen. All three clusters have about the same average metallicity, have large metallicity spreads, and exhibit extreme variations in light element, heavy element, and (most likely) He abundances. However, at least ω Cen and M54 are particularly noteworthy because these clusters are strongly suspected to have extragalactic origins (e.g., Bekki & Freeman 2003; Mackey & van den Bergh 2005). The similar chemical pattern and HB morphology that NGC 6273 shares with ω Cen and M54 suggest that NGC 6273 may have also been accreted by the Milky Way. If these clusters are all remnants of dwarf galaxy systems, then it is reasonable to assume that each cluster has experienced significant mass loss. Therefore, a cluster's formation environment and initial mass may play critical roles in forming large populations of blue hook stars, and the different blue hook populations of NGC 6273 and M2 could be explained if NGC 6273 was initially much more massive than M2 and/or formed in a different environment. In this context, we note that NGC 2419²⁶ and NGC 2808 would also be candidates that may have formed with much larger initial masses, and their higher and lower $\tfrac{{N}_{\mathrm{BHk}}}{{N}_{\mathrm{HB}}}$ ratios compared to NGC 6273 could be driven by their lower and higher respective metallicities. At least for NGC 2419, there are also some indications that the cluster may have an extragalactic origin (Mackey & van den Bergh 2005).