Detailed Chemical Abundances in NGC 5824: Another Metal-Poor Globular Cluster with Internal Heavy Element Abundance Variations
Ian U. Roederer, Mario Mateo, John I. Bailey III, Meghin Spencer, Jeffrey D. Crane, Stephen A. Shectman
aa r X i v : . [ a s t r o - ph . S R ] O c t Mon. Not. R. Astron. Soc. , 1–26 (2015) Printed 23 October 2015 (MN L A TEX style file v2.2)
Detailed Chemical Abundances in NGC 5824: AnotherMetal-Poor Globular Cluster with Internal Heavy ElementAbundance Variations ⋆ Ian U. Roederer , † , Mario Mateo , John I. Bailey III , Meghin Spencer ,Jeffrey D. Crane , Stephen A. Shectman Department of Astronomy, University of Michigan, 1085 S. University Avenue, Ann Arbor, MI 48109, USA Carnegie Observatories, 813 Santa Barbara Street, Pasadena, CA 91101, USA Joint Institute for Nuclear Astrophysics and Center for the Evolution of the Elements, USA
23 October 2015
ABSTRACT
We present radial velocities, stellar parameters, and detailed abundances of 39 ele-ments derived from high-resolution spectroscopic observations of red giant stars in theluminous, metal-poor globular cluster NGC 5824. We observe 26 stars in NGC 5824using the Michigan/Magellan Fiber System (M2FS) and two stars using the Magel-lan Inamori Kyocera Echelle (MIKE) spectrograph. We derive a mean metallicity of[Fe/H] = − ± ± α - or Fe-group abundance ratios. Twenty-five of the 26 stars exhibit a n -capture en-richment pattern dominated by r -process nucleosynthesis ( h [Eu/Fe] i = +0.11 ± h [Ba/Eu] i = − ± s -process en-hancement ([Ba/Fe] = +0.56 ± ± s -process enhancement via mass-transferfrom a binary companion. The Pb and other heavy elements produced by the s -processsuggest a timescale of no more than a few hundred Myr for star formation and chemicalenrichment, like the complex globular clusters M2, M22, and NGC 5286. Key words: globular clusters: individual (NGC 5824) – nuclear reactions, nucleosyn-thesis, abundances – stars: abundances
The number of globular clusters with spectroscopically-confirmed metallicity dispersions has grown dramati-cally over the last decade. These are generally amongthe most massive clusters, and many are also metal-poor. This list includes ω Cen (e.g., Norris & Da Costa1995; Smith et al. 2000; Johnson & Pilachowski 2010; ⋆ This paper includes data gathered with the 6.5 meter MagellanTelescopes located at Las Campanas Observatory, Chile. † E-mail: [email protected]
Marino et al. 2011), M2 (Yong et al. 2014b), M19(Johnson et al. 2015b), M22 (e.g., Da Costa et al. 2009;Marino et al. 2009), M54 (Carretta et al. 2010b), NGC 1851(Yong & Grundahl 2008; Carretta et al. 2010c, 2011),NGC 3201 (Gonzalez & Wallerstein 1998; Simmerer et al.2013), NGC 5286 (Marino et al. 2015), and Ter 5(Origlia et al. 2011, 2013; Massari et al. 2014). Manyof these clusters exhibit multiple sequences on the sub-giant branch on the color-magnitude diagram when usingbroadband optical photometry (e.g., Piotto et al. 2012),but NGC 3201 does not. To this list we add M92, where ahigh-precision differential analysis by Langer et al. (1998) c (cid:13) Roederer et al. revealed one star with a metallicity significantly higher thanother cluster members. Clusters that are not traditionallythought to contain internal metallicity spreads may exhibitsmall variations ( < i lines may yieldspurious metallicity results. Other contradicting results ex-ist, too. Cohen et al. (2010) reported a spread in [Ca/H]in NGC 2419 based on the Ca ii near-infrared triplet.Mucciarelli et al. (2012) argued that this spread was causedby changes in the continuous opacity induced by severe Mgdepletions in some stars, rather than a dispersion in Caabundances. Intrinsic abundance dispersions based on theCa ii triplet have also been found in M22 by Da Costa et al.(2009). Mucciarelli et al. (2015c) reported that no metallic-ity dispersion exists in M22, but that result does not have aphysical explanation since Mg shows no depletions in M22.While ongoing work seeks to re-examine the small metal-licity dispersions ( ∼ ∼ n -capture elements, provide another perspectiveinto the complex formation histories of globular clusters.In metal-poor clusters with no metallicity dispersion, the n -capture elements appear to have been produced predom-inantly, if not exclusively, by some form of rapid n -captureprocess ( r -process; e.g., Gratton, Sneden, & Carretta 2004;Roederer et al. 2010b). This indicates a rapid enrichmenttimescale from core-collapse supernovae or neutron-starmergers. In the class of complex clusters, the n -capture el-ements show star-to-star variations in all cases where theyhave been studied ( ω Cen, M2, M19, M22, NGC 1851, andNGC 5286). The heavy elements in these clusters were pro-duced by each of the slow n -capture process ( s -process) andthe r -process, and often the proportions of r - and s -processmaterial vary within each cluster. The r -process materialis thought to have been present in the gas from which allstars formed. The s -process material is thought to have beenproduced in other stars within the cluster, ejected into thecluster ISM, and incorporated into the stars observed today(e.g., Smith et al. 2000). However, the chemical enrichment scenarios to explain these signatures are neither simple noruniform from one cluster to another.Recent observations by Saviane et al. (2012) andDa Costa et al. (2014) revealed that NGC 5824 may alsobe a member of the class of complex globular clusters,similar to M2, M22, or NGC 5286. This southern clus-ter ( α = 15:03:58.6, δ = − th most-luminous cluster around the Milky Way ( M V = − ∼ × M ⊙ (McLaughlin & van der Marel 2005). NGC 5824 is an outerhalo cluster located 32.1 kpc from the Sun and 25.9 kpc fromthe Galactic center. There is moderate reddening along theline of sight, E ( B − V ) = 0.14, and NGC 5824 lies west ofthe Galactic bulge ( ℓ = 332.6 ◦ ) and north of the Galacticplane ( b = +22.1 ◦ ). NGC 5824 does not show any split orbroadened sequences in broadband optical color-magnitudediagrams (Piotto et al. 2002), but its blue horizontal branchis well-populated. Grillmair et al. (1995) identified light be-yond the tidal radius of NGC 5824, but Carballo-Bello et al.(2014) did not detect any photometric signatures of an ex-tended stellar halo around NGC 5824. Low-resolution spec-troscopy of the Ca ii triplet obtained by Da Costa et al. re-vealed a possible metallicity spread at the ≈ α -elements, Fe-group elements, and n -capture elements. Wedescribe our observations in Section 2, the radial velocitymeasurements in Section 3, the equivalent width (EW) mea-surements and line list in Sections 4 and 5, and the detailsof the abundance analysis in Sections 6 and 7. We presentour abundance results in Section 8, weigh the evidence fora metallicity spread in Section 9, and discuss all other ele-ments in Section 10. We summarize our conclusions in Sec-tion 11. Our first set of observations were made using the Michi-gan/Magellan Fiber System (M2FS) and MSpec dou-ble spectrograph (Mateo et al. 2012; Bailey et al. 2012)mounted on the Nasmyth platform at the 6.5 m LandonClay Telescope (Magellan II) at Las Campanas Observatory,Chile. M2FS uses fiber plug plates to achieve high multiplex-ing capability over a 30’-diameter field of view. We observed50 candidate members of NGC 5824 and 11 blank sky posi-tions. Three observations were made on 2015 April 15, 2015April 19, and 2015 April 20, with a total integration time of11.5 h.Our observations were made in HiRes mode with 95 µ mentrance slits. This setup delivers spectral resolving power R ≡ λ/ ∆ λ ∼ ≈ c (cid:13) , 1–26 bundances in NGC 5824 Table 1.
Magnitudes, S/N Estimates, Velocities, and Model Atmosphere Parameters for Stars Observed inNGC 5824Star V S/N pix − V r T eff log g v t [M/H](4570 ˚A) (km s − ) (K) (km s − )11001198 16.35 51 − − − − − − − − − − − − − − − − − − − − − a . . . . . . . . . . . .42007331 16.82 28 − − − − a . . . . . . . . . . . .42008343 16.49 41 − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − − a Velocity derived from a single, noisy measurement only; uncertainty is likely to be 1.5 km s − or greater. spectral resolution varies by ≈ λ ≈ python routines written by J. I. B. Standard iraf routines were used to perform all other tasks (flatfielding,extraction, wavelength calibration, spectra co-addition, ve- c (cid:13) , 1–26 Roederer et al.
Figure 1.
The M2FS spectrum of star 21000267. locity shifting, and continuum normalization). The M2FSobservations of NGC 5824 were made during dark time.Sky contamination was found to be negligible, so no skysubtraction was performed. Table 1 lists the V magnitudes(from Da Costa et al.) and S/N ratios per pixel in the co-added M2FS spectra. Our proposed M2FS integration timeswere shortened considerably due to poor weather during theM2FS block run in April, 2015. We measure radial velocitiesfor all targets observed (Section 3), but a reliable chemicalanalysis (Section 7) is only possible for the brightest targets.After an initial analysis of the M2FS spectra, two starswith nearly identical stellar parameters but differing lev-els of n -capture elements were selected for additional ob-servations. Stars 42009955 and 61005163 were observed on2015 June 16 using the Magellan Inamori Kyocera Echelle(MIKE) spectrograph (Bernstein et al. 2003), which is alsomounted on the Nasmyth platform at Magellan II (but notsimultaneously with M2FS). These spectra were taken withthe 0 . ′′ × . ′′ R ∼ R ∼ ≈ − on the blue and red arms, respectively. The blue and red armsare split by a dichroic at ≈ ≈ CarPy
MIKE data reduction pipeline(see, e.g., Kelson 2003). Coaddition and continuum normal-ization were performed within iraf . The total integrationtimes were 4.2 h and 3.3 h for stars 42009955 and 61005163,respectively. The S/N ratios of the co-added MIKE spectrarange from ≈ ≈ Figure 2 illustrates a portion of the region of overlap be-tween the M2FS and MIKE spectra for one of the stars incommon, 61005163. The differences, resampled to the M2FSpixel spacing, are shown in the bottom of each panel in Fig-ure 2. Qualitatively, the agreement between the spectra ap-pears excellent, and we now attempt to quantify the degreeof similarity.Figure 3 shows a histogram of the residuals of the nor-malized M2FS and MIKE spectra for each of the two stars.Only wavelengths from 4430 to 4630 ˚A are included. Themean difference is − ≈ ≈ ≈ ≈ ≈
20 per cent of thespectrum. This leaves (0 . − . − . ) / ≈ We measure radial velocities for each M2FS observationusing the iraf
FXCOR task to cross-correlate, order-by-order, against a template. We use one high-S/N spectrumof 21000267 as the template. We establish the zeropoint ofthis spectrum by using SPLOT to measure the wavelengthsof 51 Fe i lines; we then compare with the laboratory wave-lengths presented by Nave et al. (1994). The velocity zero-point of the template is accurate to better than 0.2 km s − .The statistical uncertainties associated with each cross-correlation relative to the template vary with S/N, rangingfrom 0.4 km s − for the brightest targets to 1.3 km s − forthe faintest ones. We estimate these uncertainties based onrepeat observations of each star. We estimate the absolutevelocity uncertainty to be ≈ − , based on observa-tions of the metal-poor standard star HD 122563 taken withthe same M2FS configuration in February, 2015. c (cid:13) , 1–26 bundances in NGC 5824 Figure 2.
Portions of the M2FS and MIKE spectra of star61005163. The M2FS spectrum is shown by the black line, andthe MIKE spectrum is shown by the bold gray line. The thin grayline at the bottom of each panel shows the residual between thetwo spectra on the same scale.
Figure 3.
Residuals in the normalized flux between the M2FSand MIKE spectra of stars 42009955 and 61005163. The meandifference and standard deviation are printed and illustrated bythe red curve.
Table 2.
Equivalent Width Measurements for Lines Detectedin the M2FS SpectraStar Species λ EW(˚A) (m˚A)11001198 Mg i i i i i We report heliocentric radial velocities, V r , for each starin Table 1. We calculate heliocentric corrections using the iraf RVCORRECT task. The velocities in Table 1 repre-sent a weighted mean of our three observations. The weightsroughly correspond to the S/N obtained in each set of ob-servations, with a weight of 1 for the first and third epochsand a weight of 3 for the second epoch. Our radial velocitiesdiffer by only 1.9 ± − ( σ = 9.1 km s − ) from thosereported by Da Costa et al. (2014).The mean systemic radial velocity for NGC 5824, basedon 50 stars, is − ± − ( σ = 4.9 km s − ).This agrees with the values measured by Da Costa et al.(2014), − ± − , and Dubath et al. (1997), − ± − . The dispersion of our velocity mea-surements, 4.9 km s − , is considerably smaller than the corevelocity dispersion derived from the integrated-light spectraof Dubath et al., 11.1 ± − . This may be expectedsince our targets mostly sample the outer regions of the clus-ter (0.9 < r h <
14, where r h = 0.45 arcmin is the half-lightradius given by Harris 1996).We measure radial velocities from the MIKE spectraof stars 42009955 and 61005163 by cross-correlating a late-type metal-poor template against the echelle order contain-ing the Mg i b lines, as described in Roederer et al. (2014a).The MIKE velocity zeropoint is reliable to ≈ − (e.g., Roederer & Kirby 2014). The heliocentric radial veloc-ities measured from the MIKE spectra of stars 42009955 and61005163, − ± − and − ± − , arein agreement with those measured two months earlier usingM2FS, − ± − and − ± − . Ourdata offer no evidence for radial velocity variations for eitherstar. We measure EWs from the spectra using a semi-automatedroutine, ew.pro , that fits Voigt (or Gaussian) line pro-files to continuum-normalized spectra. As discussed inRoederer et al. (2014a), all fits are presented to the user forapproval, modification, or rejection. The EWs measured inthe M2FS spectra are listed in Table 2, and the EWs mea-sured in the MIKE spectra are listed in Table 3. The full listof lines examined in the M2FS spectra is shown in Table 4.EWs measured from MIKE spectra are useful to demon- c (cid:13) , 1–26 Roederer et al.
Table 3.
Equivalent Width Measurements for Lines Detectedin the MIKE SpectraSpecies λ E.P. log gf Ref. EW EW42009955 61005163(˚A) (eV) (m˚A) (m˚A)Li i i − i i i gf value and HFS; (13) Ruffoni et al. 2014; (14) NIST,using HFS from Kurucz & Bell 1995; (15) Wood et al. 2014a;(16) Roederer & Lawler 2012; (17) Bi´emont et al. 2011; (18)Ljung et al. 2006; (19) NIST, using HFS/IS from McWilliam1998 when available; (20) Lawler, Bonvallet, & Sneden 2001,using HFS from Ivans et al. 2006; (21) This study; (22)Roederer, Marino, & Sneden 2011; (23) Lawler et al. 2009;(24) Li et al. 2007, using HFS from Sneden et al. 2009;(25) Ivarsson, Litz´en, & Wahlgren 2001, using HFS fromSneden et al. 2009; (26) Den Hartog et al. 2003, using HFS/ISfrom Roederer et al. 2008 when available; (27) Lawler et al.2006, using HFS/IS from Roederer et al. 2008 when available;(28) Lawler et al. 2001b, using HFS/IS from Ivans et al.2006; (29) Roederer et al. 2012b; (30) Den Hartog et al.2006; (31) Lawler et al. 2001a; (32) Wickliffe, Lawler, & Nave2000; (33) Lawler et al. 2008; (34) Lawler et al. 2007; (35)Bi´emont et al. 2000, using HFS/IS from Roederer et al. 2012b;(36) Nilsson et al. 2002. strate the integrity of EWs measured from M2FS spec-tra. Roederer et al. (2014a) found that EWs measured us-ing ew.pro were statistically identical among spectra ob-tained with three different spectrographs—MIKE, the TullCoud´e Spectrograph, and the High Resolution Spectro-graph (HRS). (The Tull and HRS instruments are at Mc-Donald Observatory, Texas). Bedell et al. (2014) performedan analysis of EWs measured from high-quality asteroid-reflected solar spectra taken with MIKE and ESPaDOnS.(ESPaDOnS is at the the Canada-France-Hawaii Telescope.)They found abundance differences at the ≈ ± σ = 7.9 m˚A), which is not significant, indi-cating that the M2FS EWs are also consistent with externalscales. Table 4.
List of Lines Examined in M2FS SpectraSpecies Wavelength E.P. log gf Ref.(˚A) (eV)Mg i − i − i − ii − i Figure 4.
Comparison of EWs measured in M2FS and MIKEspectra of stars 42009955 and 61005163. The dashed line marksthe one-to-one correspondence.
Tables 3 and 4 include references for the atomic data. Weprivilege log gf values from recent laboratory studies when-ever possible, since these investigations frequently deliver ≈ ≈ gf values of a givenspecies whenever possible, to minimize any systematic off-sets from one study to another. Our goal in making thesechoices is to minimize the impact of systematics arising fromthe set of lines examined. As we show in Section 8.1, thisgoal can be achieved in most cases. c (cid:13) , 1–26 bundances in NGC 5824 Table 5.
List of Hyperfine Component Patterns for Heavy El-ements Sorted by Species and WavelengthSpecies Isotope Wavelength E.P. log gfa (˚A) (eV)Ba ii
137 4553.9980 0.000 − ii
137 4554.0000 0.000 − ii
137 4554.0010 0.000 − ii
135 4554.0020 0.000 − ii
135 4554.0030 0.000 − a The log gf values of each isotope are normalized to the totallog gf value for the transition. Hyperfine splitting (HFS) structure and isotope shifts(IS) are included in the spectrum synthesis when thesedata are available. We adopt an isotopic ratio for C/ C of 4. For Cu, we adopt the solar isotopic ratio, Cu/ Cu = 2.24. We adopt r -process isotopic fractionsfor Ba, Nd, Sm, Eu, and Pb using the values presented inSneden, Cowan, & Gallino (2008). For star 61005163, we in-stead adopt an appropriate mixture of r - and s -process iso-topic fractions (see Section 10.3). At the request of the ref-eree, we have included the HFS and IS component patternsfor lines of n -capture elements studied in this work. Theseare presented in Table 5. We emphasize that these data sim-ply echo those given in the references to Table 3.The La ii line at 4522.37 ˚A was not covered by thelaboratory study of Lawler, Bonvallet, & Sneden (2001), butit is the only reliable La ii line that appears in our M2FSspectra. We use a high-resolution spectrum of the r -process-rich standard star BD+17 3248, a red giant with [Fe/H] = − gf valuefor this line. Our examination of five other La ii lines inBD+17 3248 yields log gf = − ± We calculate effective temperatures, T eff , using the de-reddened V − I color- T eff relation from Alonso et al.(1999), assuming E ( B − V ) = 0.14 (Zinn 1980;Reed, Hesser, & Shawl 1988) and the Cardelli et al. (1989)extinction coefficients. We transform the Da Costa et al.(2014) V − I colors, measured in the standard Johnson-Cousins system, to the Johnson system using the relationgiven by Bessell (1979). A second step is frequently em-ployed by fitting a relationship between a magnitude (of-ten K ) and T eff , which reduces the impact of photometricerrors and differential reddening (e.g., Carretta et al. 2007;Roederer & Thompson 2015). This approach assumes thatthe RGB has no intrinsic spread at a given color. This as- sumption may be invalid for NGC 5824 since an internalmetallicity spread may be present, so we do not employ thissecond step when calculating T eff . Monte Carlo error prop-agation calculations indicate that the statistical uncertain-ties on T eff from all sources of error (intrinsic scatter in thecolor- T eff relation, photometric error, and uncertainty in thereddening) are ≈
145 K. The optional second step—fittingand interpolating a relationship between K and T eff —yieldsstatistical uncertainties ≈ g , usingthe relation log g = 4 log( T eff ,⋆ )+log( M/M ⊙ ) − . M bol , ⊙− M bol ,⋆ ) − .
61. The constant 10.61 is derived from the so-lar values given in Cox (2000), the quantity M bol ,⋆ is givenby BC V + m V − ( m − M ), and BC V is given by eq. 18 ofAlonso et al. (1999). We assume masses of 0.8 ± ⊙ for all stars. We adopt the distance modulus given by Harris(1996), m − M = 17.94 ± g values are ≈ T eff and mass es-timates.We interpolate model atmospheres from the atlas9 grid of α -enhanced models (Castelli & Kurucz 2003), usingan interpolator provided by A. McWilliam (2009, privatecommunication). We derive abundances using a recent ver-sion of the moog line analysis software (Sneden 1973), withupdates described in Sobeck et al. (2011). We determine mi-croturbulent velocities, v t , by requiring no trend between theline strength, parameterized by log(EW/ λ ), and abundancederived from Fe i lines. We estimate a statistical uncertaintyof ≈ − in v t for a fixed set of T eff and log g . We setthe model metallicity, [M/H], equal to the Fe abundance de-rived from Fe ii lines and adjust the microturbulent velocityand metallicity iteratively. We also iteratively cull Fe i and ii lines giving abundances more than 2 σ from the mean.We consider the model parameters to have converged whensubsequent adjustment of v t is less than 0.05 km s − , ad-justment of the model [M/H] is less than 0.03 dex, and allFe i and ii lines agree to within 2 σ .Systematic sources of uncertainty are undoubtedlylarger than the statistical errors noted here, but they havelittle effect on the relative abundances of stars in the sameevolutionary state. We discuss the impact of these uncer-tainties in Section 9.2.The adopted model atmosphere parameters are listedin Table 1 for stars with moderate or high S/N levels and T eff > ⊙ (Gratton et al. 2010), we estimate a temperature of 3890 K,which is nearly 200 K cooler than the next-coolest giant. Ourmethod may not be applicable to stars this cool without ad-ditional modification, and we aim to preserve a differentialquality to the analysis. We do not analyze the abundancesin star 21002944 and will not discuss it further. c (cid:13) , 1–26 Roederer et al.
Most abundances are derived using an adaptation of thebatch mode capabilities of moog , forcing theoretical EWsto match the observed ones by adjusting the abundance.Other abundances are derived by matching synthetic to ob-served spectra. We adopt the latter method when lines aretoo blended to measure reliable EWs or when HFS or IScould affect the derived abundance. In a few cases, we re-port 3 σ upper limits on the abundance using a version of theformula presented on p. 590 of Frebel et al. (2008), which isderived from equation A8 of Bohlin et al. (1983).We assume local thermodynamic equilibrium (LTE)holds for the line-forming layers of the atmosphere for allspecies except Na i (Lind et al. 2011) and K i (Takeda et al.2002). Both of these species—and O, discussed below—canonly be detected in the MIKE spectra. The non-LTE cor-rections for the Na i lines at 5682, 5688, 6154, and 6160 ˚Arange from − − i line at 7698 ˚A are ≈ − i i ] line at 6300 ˚A, which isexpected to form under LTE, is weak and cannot be reliablymeasured even after removing the telluric lines.No C or N lines are detected in our M2FS spectra. Inthe two MIKE spectra, C abundances are derived from theCH A ∆ − X Π G band ( ≈ B Σ − X Σ violet band( ≈ Table 6 lists the mean abundances derived for each elementin each star observed with M2FS. Tables 7 and 8 list themean abundances derived from the MIKE spectra of stars42009955 and 61005163. The abundances in Tables 7 and8 reflect the non-LTE corrections discussed in Section 7.We use standard definitions of elemental abundances andratios in these tables. For element X, the logarithmic abun-dance is defined as the number of X atoms per 10 hydrogenatoms, log ǫ (X) ≡ log ( N X /N H )+ 12.0. For elements X andY, [X/Y] is the logarithmic abundance ratio relative to thesolar ratio on the Asplund et al. (2009) abundance scale,defined as log ( N X /N Y ) − log ( N X /N Y ) ⊙ , using like ion-ization states; i.e., neutrals with neutrals and ions with ions.Abundances or ratios denoted with the ionization state in-dicate the total elemental abundance as derived from tran-sitions of that particular state.Four sets of uncertainties are listed in these tables. The Table 9.
Mean Abundances in NGC 5824 Determined fromM2FS SpectraRatio Species Mean Std. err. Std. dev. N stars [Fe/H] i − ii − i − i +0.20 0.02 0.10 25[Sc/Fe] ii − i − ii +0.29 0.02 0.08 25[V/Fe] i − i − ii +0.13 0.03 0.15 25[Mn/Fe] i − i − i − ii − ii − ii − ii − ii +0.12 0.03 0.16 24[Eu/Fe] ii +0.11 0.02 0.12 25[Dy/Fe] ii +0.16 0.04 0.17 22Note—Star 61005163 has been excluded from the data presentedin this table. statistical uncertainty, σ stat , is given by equation A17 ofMcWilliam et al. (1995). This includes uncertainties in theEW, line profile fitting, log gf values, and non-LTE correc-tions, if any. The total uncertainty, σ tot , is given by equa-tion A16 of McWilliam et al. This includes the statisticaluncertainty and reflects uncertainties in the model atmo-sphere parameters. The other two uncertainties, σ neut and σ ions , are useful when constructing abundance ratios amongdifferent elements. We recommend that σ neut for element Abe added in quadrature with σ stat for element B when com-puting the ratio [A/B] when B is derived from lines of theneutral species. Similarly, we recommend using σ ions insteadof σ neut when element B is derived from lines of the ionizedspecies. These latter two sets of uncertainties are omittedfor the [Fe/H] ratios.Table 9 reports the weighted mean abundances of allelements studied in the M2FS spectra of NGC 5824. Star61005163, whose anomalous abundance pattern will be dis-cussed in great detail in later sections, is omitted from themeans presented in Table 9. Figure 5 compares the abundance ratios derived from M2FSand MIKE spectra for the two stars in common, 42009955and 61005163. In general, there is superb agreement. Themost significant disagreement occurs for the [Mg/Fe] ratio,where the M2FS ratio is lower by 0.43 ± i line, at 4571.10 ˚A, iscovered in the M2FS spectra. This line originates from theground level of the Mg i atom, whereas all other Mg i linestypically studied in metal-poor giants originate from levelsat ≈ ≈ c (cid:13)000
Mean Abundances in NGC 5824 Determined fromM2FS SpectraRatio Species Mean Std. err. Std. dev. N stars [Fe/H] i − ii − i − i +0.20 0.02 0.10 25[Sc/Fe] ii − i − ii +0.29 0.02 0.08 25[V/Fe] i − i − ii +0.13 0.03 0.15 25[Mn/Fe] i − i − i − ii − ii − ii − ii − ii +0.12 0.03 0.16 24[Eu/Fe] ii +0.11 0.02 0.12 25[Dy/Fe] ii +0.16 0.04 0.17 22Note—Star 61005163 has been excluded from the data presentedin this table. statistical uncertainty, σ stat , is given by equation A17 ofMcWilliam et al. (1995). This includes uncertainties in theEW, line profile fitting, log gf values, and non-LTE correc-tions, if any. The total uncertainty, σ tot , is given by equa-tion A16 of McWilliam et al. This includes the statisticaluncertainty and reflects uncertainties in the model atmo-sphere parameters. The other two uncertainties, σ neut and σ ions , are useful when constructing abundance ratios amongdifferent elements. We recommend that σ neut for element Abe added in quadrature with σ stat for element B when com-puting the ratio [A/B] when B is derived from lines of theneutral species. Similarly, we recommend using σ ions insteadof σ neut when element B is derived from lines of the ionizedspecies. These latter two sets of uncertainties are omittedfor the [Fe/H] ratios.Table 9 reports the weighted mean abundances of allelements studied in the M2FS spectra of NGC 5824. Star61005163, whose anomalous abundance pattern will be dis-cussed in great detail in later sections, is omitted from themeans presented in Table 9. Figure 5 compares the abundance ratios derived from M2FSand MIKE spectra for the two stars in common, 42009955and 61005163. In general, there is superb agreement. Themost significant disagreement occurs for the [Mg/Fe] ratio,where the M2FS ratio is lower by 0.43 ± i line, at 4571.10 ˚A, iscovered in the M2FS spectra. This line originates from theground level of the Mg i atom, whereas all other Mg i linestypically studied in metal-poor giants originate from levelsat ≈ ≈ c (cid:13)000 , 1–26 bundances in NGC 5824 Table 6.
Mean Abundances in Individual Stars, As Derived from the M2FS Spec-traSpecies Star N lines log ǫ [Fe i /H] σ stat σ tot σ neut σ ions Fe i − i − i − i − i − Table 7.
Mean Abundances in Star 42009955, As Derived from the MIKESpectrumSpecies N lines log ǫ [X/Fe] a σ stat σ tot σ neut σ ions Fe i
117 +5.11 − ii
10 +5.48 − i < − < − − i < +7.00 < +0.70 . . . . . . . . . . . .Na i i i i i i
12 +4.21 +0.26 0.11 0.22 0.12 0.20Sc ii i
17 +2.53 − ii
13 +3.27 +0.33 0.05 0.10 0.16 0.09V i
10 +1.38 − ii i
14 +2.96 − ii i − i i − i − i i < +1.45 < +1.32 . . . . . . . . . . . .Sr ii ii − − ii i − ii − − ii − − ii − − ii − − ii − − ii − ii − ii − ii − ii − ii − ii − i − ii < − < +0.36 . . . . . . . . . . . . a [Fe/H] is given for Fe i and Fe ii c (cid:13) , 1–26 Roederer et al.
Table 8.
Mean Abundances in Star 61005163, As Derived from the MIKESpectrumSpecies N lines log ǫ [X/Fe] a σ stat σ tot σ neut σ ions Fe i
124 +5.21 − ii
10 +5.58 − i < − < − − i i − i i < +4.37 < +0.21 . . . . . . . . . . . .Si i i i
12 +4.32 +0.27 0.11 0.22 0.12 0.20Sc ii i
15 +2.62 − ii
13 +3.35 +0.32 0.05 0.10 0.17 0.09V i
10 +1.50 − ii i
14 +3.11 − ii i − i i − i − i i < +1.25 < +1.02 . . . . . . . . . . . .Sr ii ii − ii i ii ii − ii
18 +0.13 +0.47 0.05 0.09 0.17 0.08Pr ii − ii − ii − ii − ii − ii − ii − ii − ii − i ii a [Fe/H] is given for Fe i and Fe ii tent with expectations that the ground level of Mg i mayexperience substantial overionization relative to the LTElevel populations (Asplund 2005). The non-LTE calculationsof Shimanskaya, Mashonkina, & Sakhibullin (2000) confirmthis, although they do not extend to stars cool enough tojustify applying corrections to our data. We derive Mg abun-dances from three other Mg i lines in the MIKE spectra. TheMg i line at 4571 ˚A also gives abundances lower than theseother three lines by 0.41 dex, on average, confirming thatour M2FS measurements are not uniquely in error. We cau-tion that the [Mg/Fe] ratios reported in Tables 6 and 9 arelikely to be lower by ≈ i lines. The [Ni/Fe] ratios are both mildly different in thesetwo sets of spectra, where the M2FS ratio is lower than theMIKE ratio by 0.19 ± i lines, at 4470.48and 4604.99 ˚A, have been used in both sets of spectra of bothstars. The abundances derived from them agree to 0.07 dexor better. Abundances derived from these lines in the MIKEspectra are lower by 0.18 dex, on average, than the abun-dances derived from the five other Ni i lines in each star.This indicates that our M2FS Ni abundance determinationsare themselves not in error, just as we have found in thecase of Mg. No non-LTE calculations are available for Nilines, to the best of our knowledge; however, the results ofWood et al. (2014a) suggest that any non-LTE effects on c (cid:13) , 1–26 bundances in NGC 5824 Figure 5.
Comparison of abundance ratios derived from M2FSand MIKE spectra for the two stars in common. Red circles indi-cate star 42009955, and blue squares indicate star 61005163. Thedotted line marks a difference of zero. neutral Ni are minimal, at least in metal-poor subgiants.The cause of the Ni offset is unclear at present. We cautionthat the [Ni/Fe] ratios reported in Tables 6 and 9 are likelyto be lower by ≈ i lines. We compare our derived metallicities, inferred from individ-ual Fe ii lines in 26 stars in NGC 5824, with those calculatedby Da Costa et al. (2014), inferred from a calibration of theCa ii near-infrared triplet EWs. Da Costa et al. did not re-port metallicities for individual stars, so we follow their pro-cedure using the calibration given by Saviane et al. (2012).We estimate uncertainties on the Da Costa et al. metallici-ties using a Monte Carlo approach.Figure 6 illustrates the metallicity distribution for starsin NGC 5824. The hatched dark blue histogram representsmetallicities derived by us. The black line represents thefull sample of 108 stars examined by Da Costa et al. (2014),and the light blue histogram represents the subset of 26 starsin common to both studies. Our mean metallicities derivedfrom Fe i and Fe ii are different by 0.44 dex, but both havea standard deviation of 0.08 dex. The metallicities derivedfrom Fe ii lines are in much better agreement with thoseof Da Costa et al. and an external metallicity scale (Sec-tion 9.4), so we prefer these values.A small offset in [Fe/H] is present between our workand that of Da Costa et al. (2014). For the 26 stars in com-mon, our [Fe/H] values derived from Fe ii lines are higherby 0.08 ± ± ± < −
1, including many used by Saviane et al.Da Costa et al. note that the uncertainty in the mean metal-licity derived for NGC 5824 using their Ca triplet calibra-
Figure 6.
The metallicity distribution function of NGC 5824.The dark blue hatched histogram represents metallicities derivedby us using Fe i lines (top panel) or Fe ii lines (bottom panel) in26 stars. The unfilled black histogram represents metallicities forthe full sample of 108 stars examined by Da Costa et al. (2014).The filled light blue histogram represents the subset of 26 starsfrom Da Costa et al. that are common to our study. tion, [Fe/H] = − ± Figure 7 compares our results with those of Da Costa et al.(2014). The metallicities derived by Da Costa et al. and usare inferred from independent spectral features. We mightexpect to find a correlation in Figure 7 if [Ca/Fe] is rela- c (cid:13) , 1–26 Roederer et al.
Figure 7.
Comparison of metallicities derived in our study (usingFe ii lines) with those inferred from Ca ii triplet EWs measured byDa Costa et al. (2014). The dashed line represents a one-to-onecorrelation. tively constant and there is an intrinsic metallicity disper-sion within NGC 5824 that exceeds the observational un-certainties. The p -value for the linear correlation coefficientis 0.52, affirming our visual impression that no significantcorrelation is found.Figure 8 illustrates our observed metallicity distribu-tions in NGC 5824, which are equivalent to the hatched bluehistograms in Figure 6. Our metallicities have standard de-viations of 0.08 dex derived from either Fe i or Fe ii lines.The black curves in Figure 8 illustrate normal distributionswith a mean metallicity of − i ) or − ii ) and standard deviations of 0.08 dex. We estimate theformal widths of the [Fe/H] distributions that would be ex-pected given errors in the input quantities used to calculate T eff and log g . We assume σT eff = 145 K, σ log g = 0.14 dex,and σv t = 0.1 km s − , as estimated in Section 6. Most of theerrors on the other input quantities in the log g calculationin Section 6 are systematic, so they affect all stars similarly.We ignore the errors in these quantities for now. For a givenstar, we use 1000 realizations of model atmosphere param-eters drawn from normal distributions with these standarddeviations, and we rederive the [Fe/H] ratios. The standarddeviations of the resulting [Fe/H] distributions vary some-what with T eff , but the median values are ≈ i and ii . This value is illustrated by the blue curvesin Figure 8.It is apparent from Figure 8 that the expected distribu-tions overestimate the observed widths by about a factor oftwo. This likely signals that we have overestimated the errorson the model atmosphere parameters. The largest source oferror that contributes to the width in the predicted [Fe/H]distribution is σT eff , which enters the calculation twice: oncethrough T eff itself, and once through 4log T eff in the log g for-mula. The Alonso et al. (1999) V − I color- T eff calibration Figure 8.
Observed metallicity distributions (gray shaded his-tograms) in NGC 5824. The top panel shows the results from Fe i lines, and the bottom panel shows the results from Fe ii lines.The bold black line represents a normal distribution with mean[Fe/H] = − i ) or [Fe/H] = − ii ) and standarddeviation 0.08, which is equivalent to the standard deviation ofthe derived metallicities for both Fe i and ii . The studded blueline represents a normal distribution with mean [Fe/H] = − i ) or [Fe/H] = − ii ) and standard deviation 0.15,which is equivalent to the formal error distributions predictedfor our data. All distributions are normalized to the areas of theobserved histograms. has an intrinsic scatter of 125 K. If we arbitrarily assumezero scatter in this calibration, σT eff decreases to 74 K (re-flecting quantities related to the photometry), and σ [Fe / H] decreases to 0.09 dex, close to the observed value.We compare the results with another T eff scale to assesswhether this arbitrary assumption is plausible. The V − I c color- T eff relation of Casagrande et al. (2010) has an intrin-sic scatter of only 59 K. This scale was calibrated usingdwarfs, but Casagrande et al. (2014) have shown that it isalso applicable to giants. The Casagrande et al. calibrationpredicts T eff values warmer than the Alonso et al. scale by85 K for these stars. The standard deviation of the residualsbetween the two scales is only 3 K, however. The excellentagreement of these two scales—zeropoint aside—offers re-assurance that the statistical errors on T eff may be smallerthan assumed.Our tests indicate that no reasonable assumptions forthe errors in T eff , log g , or v t can produce a predicted [Fe/H]distribution for either Fe i or ii narrower than the observed c (cid:13) , 1–26 bundances in NGC 5824 one. Furthermore, we have assumed during these calcula-tions that the measurement errors on the EWs are zero,which also would lead us to underestimate the widths of thepredicted [Fe/H] distributions. We conclude that there is nointernal metallicity dispersion among the stars observed inNGC 5824. We estimate a systematic uncertainty of 0.10 dexin the mean metallicity when sources of systematic uncer-tainty are considered.We do not find evidence in our data for the metal-licity spread reported by Da Costa et al. (2014), but thatstudy would also not have found evidence for a metallic-ity spread if they were restricted to the 26 stars studied byus. The dispersion in their metallicities of these 26 stars,0.06 dex, is equivalent to the expected error distribution.This value is smaller than the FWHM of the distributionshown in Figure 12 of Da Costa et al., 0.16 dex, or the inner-quartile range, 0.10 dex, reported based on their full sampleof 108 stars.Figure 9 illustrates the Da Costa et al. metallicities asa function of V magnitude. Open circles represent their fullsample, and filled circles represent stars in common withour sample. Stars that contribute to the tails of their metal-licity distribution have V >
17, where we have no stars incommon. One explanation is that this simply reflects theincreasing measurement errors for fainter stars. If, however,NGC 5824 has an intrinsic metallicity spread, the stars withhigher metallicities may not occupy the upper RGB, and ourobservations would be biased against them. Stars with lowermetallicities would generally not be fainter and should befound in our sample, so this explanation may only tell part ofthe story. The radial distributions of our sample and the fullDa Costa et al. sample are only mildly different, with theDa Costa et al. sample being slightly more extended thanours. The two-sided Kolmogorov-Smirnov test returns a p -value of 0.12 when applied to the cumulative radial distri-butions of these samples, and the Cramer-von Mises test re-turns a p -value of 0.044. If lower metallicity stars are foundat greater radii, on average, this could explain the tail tolower metallicities.Another partial explanation for the range of Ca ii tripletstrengths found by Da Costa et al. (2014) could be an opac-ity effect caused by the Mg-deficient stars. Mucciarelli et al.(2012) noted that the most Mg-deficient stars in NGC 2419were preferentially those with the highest [Ca/H] ratios. Mgis one of the dominant donors of electrons to form the H − ion, the primary source of continuous opacity in late-typemetal-poor stars. Mucciarelli et al. contend that the con-tinuous opacity decreases when Mg is depleted, leading tostronger Ca ii lines even though the Ca abundance is con-stant. The saturated Ca ii triplet lines are partially formedin layers of the atmosphere most sensitive to the electronpressure (see Figure 7 of Mucciarelli et al.), whereas lines onthe weak part of the curve of growth are minimally-affected.Figure 10 compares the Da Costa et al. metallicities derivedfrom Ca ii triplet lines with the [Mg/Fe] ratios derived by us.We estimate the significance of the correlation in Figure 10by calculating the p -values for the Spearman and Pearsoncoefficients from 1000 resamplings of the [Mg/Fe] and Ca ii triplet metallicity distributions given their observed errors.The median p -values are 0.40 and 0.37, respectively, indicat-ing that the correlation is not significant. Regardless, deple-tions in Mg could not fully explain the situation, as noted Figure 9.
Metallicities for the Da Costa et al. (2014) sample in-ferred from the Ca ii triplet EWs. Open circles denote all 108 starsin their sample, and filled circles mark the 26 stars in commonwith us. The red circle marks star 42009955, and the blue squaremarks star 61005163. in Section 1. Da Costa et al. (2009) discovered an intrinsicdispersion in the strengths of the Ca ii triplet in M22 redgiants, similar to that found in NGC 5824. However, no sig-nificant depletions in [Mg/Fe] or other electron donors arefound in M22 (Marino et al. 2011). No satisfactory explana-tion has been found for the Da Costa et al. result in M22,and the possibility of a metallicity spread in NGC 5824 isunresolved by our data. We use a differential approach to examine whether theremay be abundance variations between the two stars observedwith MIKE, 42009955 and 61005163. These two stars arehighlighted in Figures 9 and 10. There are 101 Fe i lines incommon, 10 Fe ii lines, 12 Ca i lines, 14 Ti i lines, 13 Ti ii lines, 14 Cr i lines, and 7 Ni i lines. The line-by-line differen-tials, in the sense of star 42009955 minus star 61005163, are δ [Ca/Fe] = − ± δ [Ti i /Fe] = − ± δ [Ti ii /Fe] = − ± δ [Cr/Fe] = +0.008 ± δ [Ni/Fe] = +0.007 ± δ [Ca/Ni] = − ± i lines, δ [Fe i /H] = − ± ii lines, δ [Fe ii /H] = − ± i lines may be a poor representation of the Fe abundance c (cid:13) , 1–26 Roederer et al.
Figure 10.
Our [Mg/Fe] ratios as a function of the[Fe/H] inferred from the Ca ii triplet (CaT) measurements ofDa Costa et al. (2014). The red circle marks star 42009955, andthe blue square marks star 61005163. in cool, metal-poor giants in globular clusters. Whateverthe cause of this effect, Mucciarelli et al. have shown thatit may be most pronounced in stars that show the s -processenhancement, like star 61005163 (see Section 10.3). Fe ii lines may be more reliable. We conclude that this differentialanalysis does not offer compelling evidence for metallicitydifferences between stars 42009955 and 61005163.Our referee notes, correctly, that this technique may notbe strictly valid if the He abundance or total abundance ofC, N, and O differs between these two stars. Sbordone et al.(2011) have shown that stars with the same color but dif-fering He or total C, N, and O abundances will separateinto different sequences using broadband optical photom-etry. Our method for deriving log g would then need tobe refined for the additional sequence. We cannot evaluatethe He abundances or total abundance of C, N, and O inthese two stars, and broadband optical photometry has notyet revealed multiple sequences on the red giant branch ofNGC 5824 (Piotto et al. 2002). Evidence suggests, however,that NGC 5824 may be a good candidate to search for thepresence of He variations (Section 10.1), which have beendetected in every cluster investigated (Milone 2015), so thismatter is unresolved at present. Koch & McWilliam (2008) have established a globular clus-ter metallicity scale based on a differential abundance anal-ysis with the K-giant Arcturus ( α Boo), whose temperatureis known to better than 30 K. Koch & McWilliam (2011)placed the metal-poor cluster NGC 6397 on this scale, find-ing [Fe/H] = − ± ± ii lines in common between three stars in NGC 6397 (stars 7230, 8958, and 13414)and our MIKE spectra of stars 42009955 and 61005163,we find a mean line-by-line offset of +0.23 ± − ± ± − ± ±
10 DISCUSSION10.1 Light Elements
Mg is the only light element that is sometimes found to varywithin globular clusters that is covered by our M2FS spectra.Nevertheless, the dispersion in the [Mg/Fe] ratios, 0.28 dex,far exceeds that of any other α - or Fe-group element. Theinner quartile range of [Mg/Fe] ratios is 0.35 dex, and thefull range is 1.1 dex. Recall (Section 8.1) that the [Mg/Fe]ratios derived from the M2FS spectra may be systematicallyunderestimated by ≈ ω Cen (e.g., Norris & Da Costa 1995; Smith et al.2000), M3 (Sneden et al. 2004; Johnson et al. 2005),M13 (Pilachowski et al. 1996; Shetrone 1996; Kraft et al.1997), M15 (Sneden et al. 1997; Carretta et al. 2009b),NGC 2419 (Cohen & Kirby 2012; Mucciarelli et al.2012), NGC 2808 (Carretta et al. 2009b; Carretta 2014),NGC 4833 (Carretta et al. 2014; Roederer & Thompson2015), NGC 6752 (Gratton et al. 2001; Yong et al. 2003),and possibly M62 (Yong et al. 2014a) and M92 (Shetrone1996). Some of these clusters are also suspected to con-tain groups of stars with substantial He enhancements( ω Cen: Piotto et al. 2005, King et al. 2012; M62: Milone2015; NGC 2419: Di Criscienzo et al. 2011; NGC 2808:Piotto et al. 2007, Milone et al. 2015a; NGC 6752:Milone et al. 2013). NGC 5824, too, is luminous, metal-poor, and shows a large internal [Mg/Fe] dispersion.The relatively large dispersion in [Mg/Fe] distinguishesNGC 5824 from other clusters with complex heavy-elementabundance patterns, where the [Mg/Fe] dispersions rangefrom 0.04 dex to 0.15 dex (M2, M19, M22, NGC 1851,NGC 5286; Carretta et al. 2011; Marino et al. 2011, 2015;Yong et al. 2014b; Johnson et al. 2015b). NGC 5824 maybe a good candidate to search for the presence of internalHe variations.Our MIKE spectra of stars 42009955 and 61005163 in-clude lines of O i , Na i , Mg i , and Al i . These are illustratedin Figure 11. Examination of the spectra reveal the usuallight element abundance variations found within globularclusters. Star 61005163 has relatively strong O i lines, weakNa i lines, strong Mg i lines, and non-detected Al i lines. Star42009955, in contrast, has non-detected O i lines, relativelystrong Na i lines, weak Mg i lines, and strong Al i lines.The abundances reported in Tables 7 and 8 quantitativelyconfirm these impressions.Si and K have also been shown to occasionallyparticipate in the high-temperature p -capture reactions c (cid:13) , 1–26 bundances in NGC 5824 Figure 11.
Selections of the MIKE spectra of stars 42009955 (thin red line) and 61005163 (bold blue line) around several O i , Na i ,Mg i , and Al i lines. that produce the light element abundance variations inglobular clusters (Yong et al. 2005; Carretta et al. 2009b,2013b; Cohen & Kirby 2012; Mucciarelli et al. 2012, 2015a;Ventura et al. 2012; M´esz´aros et al. 2015). These two ele-ments are only covered in the MIKE spectra. In both cases,the [Si/Fe] and [K/Fe] ratios are in agreement. These twostars show variations in the [O/Fe], [Na/Fe], [Mg/Fe], and[Al/Fe] ratios, so [Si/Fe] and [K/Fe] may not vary substan-tially within NGC 5824.CH and CN molecular features are detected in our twoMIKE spectra. [C/Fe] is subsolar and [N/Fe] is supersolarin both stars, which is consistent with the scenario whereC is depleted and converted to N as stars evolve up theRGB. We estimate the natal [C/Fe] ratios in these two starsusing the results of Placco et al. (2014), who provide a setof corrections based on stellar evolution models that relatethe current [C/Fe] ratio of a star on the RGB to the initial[C/Fe] ratio. For star 42009955, the observed [C/Fe] = − init ≈ − − init ≈ +0.44. Neitherof these initial [C/Fe] ratios would be considered “carbon- enhanced” by modern definitions (Aoki et al. 2007), whichrequire [C/Fe] init > +0.7.For completeness, we note that the Li i line at 6707 ˚Ais covered in our MIKE spectra. Li i is not detected in eitherstar. The upper limits are consistent with the levels of Li-depletion commonly found in other cluster and field stars onthe upper RGB (e.g., Gratton et al. 2000; Lind et al. 2009). Figure 12 compares the mean abundance ratios in NGC 5824(excluding star 61005163) with those in halo stars of similarmetallicity. The gray boxes represent the median ± one stan-dard deviation of 14 red giant stars in the solar neighbor-hood with − < [Fe/H] < − Z
28 (Ca to Ni) is in-distinguishable from that of the Galactic halo in the solarneighborhood. c (cid:13) , 1–26 Roederer et al.
Figure 12.
Comparison of abundance ratios in NGC 5824 anda sample of field red giants with similar metallicity. The blackpoints represent the NGC 5824 ratios, derived from the M2FSspectra, and the gray shaded boxes represent the median ± onestandard deviation for the field sample. Cu has been found to correlate with the s -process en-hancement found in other complex, low-metallicity clusters,like M2, M22, and NGC 5286. Zn also correlates with the s -process enhancement in M22. Cu i and Zn i lines are onlycovered in our MIKE spectra. One star, 42009955 shows no s -process enhancement, while the other, 61005163, does (seeSection 10.3). We find no significant differences among the[Cu/Fe] or [Zn/Fe] ratios in these two stars. Eight elements heavier than the Fe-group are detectable inour M2FS spectra, and 17 are detectable in our MIKE spec-tra. The n -capture abundance patterns are virtually identi-cal in 25 of the 26 stars examined, as shown in Figure 13.Each panel in Figure 13 shows the logarithmic abundancesin one star. For comparison, a template for the “main”component (Truran et al. 2002) of the r -process pattern isshown. Overall, the patterns observed in NGC 5824 are aclose match to the r -process pattern. The [Ba/Eu] ratio iscommonly used to quantify the relative contributions of the r - and s -process. In NGC 5824, h [Ba/Eu] i = − ± r -process-enhanced stars, − ± h [Eu/Fe] i = +0.11 ± ≈ +0.5 ± r -process material in a clus-ter. Sneden, Pilachowski, & Kraft (2000) found that the n -capture elements in globular cluster M92 are less abun-dant by ≈ n -capture elements. NGC 5824 may lieat the low end of this dispersion.Slight overabundances relative to this r -process patternare detectable for Sr ( Z = 38), Ba ( Z = 56), Ce ( Z = 58),and Nd ( Z = 60). It is well-known (e.g., McWilliam 1998;Johnson & Bolte 2002) that the ratios between the light(e.g., Sr) and heavy (e.g., Ba or Eu) n -capture elementsvary in low-metallicity field stars even when little or no s -process contributions are present. This phenomenon isalso observed in globular clusters (e.g., Yong et al. 2014a;Roederer & Thompson 2015), and NGC 5824 appears to be no exception in this regard. The Ba abundances are de-rived from a single, saturated line at 4554 ˚A, and they arehighly sensitive to the adopted v t value. Furthermore, theslight Ba, Ce, and Nd overabundances can be attributedto contributions from both the “weak” and “main” com-ponents of the r -process. We would also expect to seesignificant Pb enhancement if these overabundances comefrom s -process nucleosynthesis (e.g., Roederer et al. 2010b),since low-metallicity AGB stars are prodigious producersof Pb (e.g., Gallino et al. 1998; Van Eck et al. 2001). Thelow log ǫ (Pb/Eu) ratio observed in star 42009955 confirmsthat little or no s -process material is present. We concludethat the n -capture elements in these 25 stars were producedby some form of r -process nucleosynthesis. The r -processis thought to occur in explosive environments like core-collapse supernovae or neutron-star mergers, so this enrich-ment almost certainly occurred before the present-day starsof NGC 5824 formed.Figure 14 illustrates the relationships between heavy n -capture elements found in NGC 5824. The 25 stars withonly r -process products are shown with red circles. Theseform a well-defined locus in each panel, with the excep-tion of the Ba abundances, which we have previouslynoted are especially sensitive to the adopted v t values.Subtle correlations are apparent among some ratios, like[Ce/Fe] and [Nd/Fe]. This does not signal cosmic star-to-star dispersion within NGC 5824 (cf. Roederer 2011;Cohen 2011). Rather, these correlations are likely due torandom uncertainties in the model atmosphere parame-ters that impact the derived abundances similarly for eachelement (Roederer & Thompson 2015). We conclude thatthe n -capture abundance patterns within these 25 stars inNGC 5824 are effectively identical.The one exception, star 61005163, shows large excessesof Ba, La, Ce, and Nd in the M2FS spectrum. This star isshown by the blue square in Figure 14. Its [Sm/Fe], [Eu/Fe],and [Dy/Fe] ratios are indistinguishable from the other stars.These differences are immediately apparent from visual in-spection of the spectra. Figure 15 shows sections of theMIKE spectra of stars 42009955 and 61005163 surroundingsome of the n -capture lines. These two stars have identicalstellar parameters and metallicities, so differences in theirspectra can be attributed to unequal abundances. Most ofthe absorption lines are formed by Fe-group elements whoseabundances are identical in the two stars (Section 10.2), andtheir line strengths are identical, too. Lines of n -capture el-ements are marked, and these lines account for nearly all ofthe differences in the spectra.Figure 16 shows the n -capture abundance patternsderived from the MIKE spectra of stars 42009955 and61005163. The metal-poor ([Fe/H] = − r -process-enhanced standard star BD+17 3248 is shown for compari-son. The abundances of most n -capture elements are differ-ent between the two stars. No elements have a lower abun-dance in star 61005163 than in star 42009955. BD+17 3248is a good match for the elements with Z >
56 in star42009955, but not for these elements in star 61005163. Werefer to the abundance pattern in star 61005163 as the“ r + s ” pattern, and we refer to the abundance pattern instar 42009955 as the “ r -only” pattern.Following the procedure outlined inRoederer, Marino, & Sneden (2011) for the two n -capture c (cid:13) , 1–26 bundances in NGC 5824
40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 9040 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 9040 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 9040 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 9040 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 9040 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 9040 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 40 50 60 70 80 90Atomic Number−2−101 l og ε main r40 50 60 70 80 90 Figure 13.
Heavy element abundance patterns derived from M2FS spectra in all 26 stars examined. The red line marks a template forthe r -process abundance pattern (the star CS 22892–052; Sneden et al. 2003, 2009; Roederer et al. 2009), which is normalized to the Euabundance in each star.c (cid:13) , 1–26 Roederer et al.
Figure 14.
Abundances of heavy n -capture elements in NGC 5824. Red circles mark stars in the r -only group, and the blue squaremarks star 61005163. patterns in globular cluster M22, we subtract the abun-dance pattern in the r -only group from the abundancepattern in the r + s group to reveal the “ s -process resid-ual” pattern. This method assumes that the ( r -process)“foundation” is identical in all stars in NGC 5824, but itmakes no assumptions about the nucleosynthesis origins ofthe foundation. The difference between the two stars—the s -process residual—is illustrated in the bottom panel ofFigure 16. Note that the abundance differences are largelyinsensitive to non-LTE effects that affect line formation (for,e.g., Pb; Mashonkina, Ryabtsev, Frebel 2012), since stars42009955 and 61005163 have identical stellar parametersand metallicities.An unmistakable correlation emerges when these dif-ferences are plotted as a function of the s -process contri- bution to the solar system abundance of each element, asshown in the bottom panel of Figure 17. Elements withminimal s -process contribution ( <
10 per cent; Eu, Tb) totheir solar abundances show no difference between the r -only and r + s groups. Elements with major s -process con-tributions ( >
80 per cent; Ba, Ce, Pb) show the largest dif-ferences. Other n -capture elements fall between these twoextremes. The lone exception, Hf, only deviates from themean trend by < σ . We regard this as compelling evidencethat s -process nucleosynthesis is responsible for the differ-ences in the abundance patterns between stars 42009955 and61005163.To summarize, we find that 25 of the 26 stars observedin NGC 5824 share a common n -capture element abundancepattern consistent with a nucleosynthetic origin in some c (cid:13) , 1–26 bundances in NGC 5824 Figure 15.
Selections of the MIKE spectra of stars 42009955 (thin red line) and 61005163 (bold blue line) around various lines of n -capture elements. The few differences in the top panel that do not correspond to n -capture element lines are due to differences in theCH abundance. form of r -process. One of these 26 stars shows substantialamounts of s -process material, as well. We have found only one star with an unusual abundancepattern in NGC 5824, which raises the possibility that thecomposition of the atmosphere of this star does not reflect its natal composition. In this section, we consider, and dismiss,the possibilities that star 61005163 received its s -processenhancement by self-enrichment or by mass-transfer froma companion star that passed through the AGB phase ofevolution. As a reminder, in this context “self-enrichment”refers to the mechanism by which the composition of thesurface layers of a star are changed through internal nucle-osynthesis and/or mixing episodes, not polluters from onestellar generation enriching other stars in a later generation. c (cid:13) , 1–26 Roederer et al.
Figure 16.
Logarithmic abundances of the n -capture elementsin NGC 5824 as derived from the MIKE spectra. In the top panel,blue squares mark star 61005163, and the red circles mark star42009955. The gray line marks the n -capture element abundancepattern observed in the metal-poor r -process-enhanced standardstar BD+17 3248 (Cowan et al. 2002, 2005; Sneden et al. 2009;Roederer et al. 2009, 2010a, 2012a). This pattern is normalizedto the Eu abundance. In the bottom panel, the squares mark theoffset between the two stars in NGC 5824, characterized as “ r + s minus r -only.” The dashed line represents a difference of zero. Star 61005163 ( M bol = − s -process material throughself-enrichment. The latest update to the Full-networkRepository of Updated Isotopic Tables & Yields (FRUITY,version 4; Cristallo et al. 2015) database presents physicalparameters and s -process yields for AGB stars with initialmasses from 1.3 to 6.0 M ⊙ . The models for low-metallicitystars have − < M bol < − s -process nucle-osynthesis occurs. Several stars in our sample are more lu-minous than star 61005163, and none shows evidence of s -process enhancement. We therefore discard the possibilitythat star 61005163 is an intrinsic s -process-enriched AGBstar.Ba enhancement is always accompanied by C enhance-ment in low-metallicity field stars whenever the Ba enhance-ment does not originate from r -process nucleosynthesis (e.g.,Aoki et al. 2007, Sneden et al. 2008, Allen et al. 2012). Suchstars are classified as carbon-enhanced metal-poor starswith s -process enhancement, or CEMP- s (Ryan et al. 2005;Beers & Christlieb 2005). Frequently, these stars are foundin binary (or multiple) star systems, as revealed by radial ve-locity variations. They are presumed to have acquired theircarbon and s -process enhancement from a more massivecompanion that evolved through the AGB phase of evolu-tion and transferred the C and s -process material to thelonger-lived star, which we now observe.Star 61005163 does not exhibit radial velocity variationsin our two observations spaced ≈ & d) orbital period.Star 61005163 is not C-enhanced, and it probably was Figure 17.
Abundance enhancement due to the s -process inglobular clusters M2 (Yong et al. 2014b), M4 (Ivans et al. 1999,2001; Yong et al. 2008a,b), M22 (Roederer et al. 2011), NGC 5286(Marino et al. 2015), and NGC 5824. In M2, M22, NGC 5286,and NGC 5824, the abundance differences are computed in thesense of “ r + s minus r -only.” In M4, the difference is computed inthe sense of “M4 minus M5.” The s -process solar system fractionsare taken from the models of Bisterzo et al. (2011). Only elementswith Z >
56 are shown, since processes other than the r - and s -process may contribute to the lighter n -capture elements.c (cid:13) , 1–26 bundances in NGC 5824 not C-enhanced before evolving up the RGB (Section 10.1).It is also not C-enhanced when evaluated by its C/O ratio,which is ≪
1, even if the LTE O abundance is overestimatedby 1 dex. Both models and observations support this view.All of the low-metallicity models of Cristallo et al. (2015)end the TP-AGB phase with C/O >
1. The compilation ofMasseron et al. (2010) lists 11 CEMP- s stars with [Fe/H] > −
3, and all have C/O > s stars are usually several dex higher thanthe solar ratios and significantly higher than the ratios ob-served in star 61005163 in NGC 5824 (e.g., Van Eck et al.2001; Sneden, Preston, & Cowan 2003; Aoki et al. 2008;Allen et al. 2012). The lack of radial velocity variations, lackof C enhancement, and relatively low [ n -capture/Fe] ratioscollectively support our assertion that star 61005163 is notrelated to the stars in the CEMP- s class.Searches for Ba-enhanced stars in normal globular clus-ters have found that the fraction of such stars (5 out of 1205,or 0.4 per cent; D’Orazi et al. 2010) is lower than in the field( > n -capture abundance pattern in one of them, Lee 4710in cluster 47 Tuc, is consistent with s -process enrichmentfrom an AGB companion with an initial mass of ∼ ⊙ (Cordero et al.) No such comparison was performed for theother star, in globular cluster M75. One simple explanationfor the lack of Ba-enhanced stars could be the dense stel-lar environments of globular clusters, which would disruptmost wide binary systems and produce a lower binary frac-tion than in the field (cf., e.g., Cˆote et al. 1996; Mayor et al.1996; Milone et al. 2012).One star out of 26 (4 per cent) analyzed by us showsan anomalous abundance pattern. Finding additional starswith abundance patterns like star 61005163 in NGC 5824would greatly alleviate the concerns that the compositionof this particular star does not reflect its natal composition.At first glance, this percentage is much lower than the per-centage of spectroscopically-confirmed members of the r + s groups in other clusters: M2 (40 per cent, 10 stars total),M22 (40 per cent, 35 stars total), and NGC 5286 (43 percent, 7 stars total). However, it is important to recognizethat those samples are highly biased in favor of stars in the r + s groups. More representative percentages can be esti-mated from photometric analyses of the multiple subgiantor red giant sequences (which correspond to the r -only and r + s groups), for example. The populations associated withthe r + s groups in M2, M22, and NGC 5286 comprise only3 per cent, 40 per cent, and 14 per cent of stars (Milone et al.2015b; Piotto et al. 2012; Marino et al. 2015). NGC 5824 isnot an outlier with respect to other complex clusters in thisregard. It would be of great interest to obtain new ultravioletbroadband photometry of NGC 5824 with the Hubble SpaceTelescope (cf. Piotto et al. 2015) to search for the presenceof multiple sequences in this cluster. s -process Material Stars in most low-metallicity globular clusters only showenrichment patterns like that found in the r -only group inNGC 5824. Three other low-metallicity clusters host groups Figure 18.
Comparison of the [ hs / ls ] and [Pb/ hs ] indices in fiveglobular clusters and AGB model predictions from the FRUITYdatabase (Cristallo et al. 2015). In M4, the s -process residual iscomputed using the abundance pattern in the (physically unre-lated) globular cluster M5 as the r -only group. In NGC 5824,the difference is computed in the sense of “star 61005163 minus42009955.” We recompute the [ hs / ls ] and [Pb/ hs ] indices usingour definitions and the yields of the FRUITY database. of stars resembling the distinct r -only and r + s patterns: M2(Yong et al. 2014b), M22 (Marino et al. 2009, 2011; Roed-erer et al. 2011), and NGC 5286 (Marino et al. 2015). Allstars in another low-metallicity cluster, M4, resemble the r + s pattern (Ivans et al. 1999; Yong et al. 2008a,b), thusdistinguishing this cluster from other low-metallicity ones.In this section we examine the n -capture elements in M2,M4, M22, and NGC 5286, along with NGC 5824, to attemptto constrain the mass range of AGB stars that may have pol-luted the cluster ISM from which star 61005163 formed.Figure 17 illustrates the differences between the r + s and r -only groups in these five clusters. In all cases, ∆ log ǫ is calculated as “ r + s minus r -only” (or “ s -rich minus s -poor” in other nomenclature; e.g., Marino et al. 2011).We use the globular cluster M5, whose n -capture elementsoriginated mainly in some form of r -process nucleosynthe-sis (Ivans et al. 2001; Yong et al. 2008a,b; Lai et al. 2011),as a foil for M4 to perform the subtraction “M4 minusM5.” Qualitatively, the abundance patterns in these fivephysically-unrelated clusters are remarkably similar, whichcould indicate that they share similar chemical enrichmenthistories.We compute a few indices from well-measured abun-dance ratios to serve as indicators of the s -process nucleosyn-thesis patterns. We define [ ls /Fe] = ([Y/Fe]+[Zr/Fe])/2, c (cid:13) , 1–26 Roederer et al. [ hs /Fe] = ([La/Fe]+[Ce/Fe]+[Nd/Fe])/3, [ hs / ls ] =[ hs /Fe] − [ ls /Fe], and [Pb/ hs ] = [Pb/Fe] − [ hs /Fe]. Thesedefinitions are constructed so that all ratios are measuredin each of the five clusters, and thus they differ slightlyfrom some versions of [ ls /Fe] and [ hs /Fe] in the literature.These indices are computed from the s -process residual ineach cluster, not the derived ratios in the r + s groups. Theindices [ hs / ls ] and [Pb/ hs ], unlike [X/Fe] ratios, are insen-sitive to the details of how material was acquired by thestar observed today. For NGC 5824, [ hs / ls ] = +0.84 ± hs ] = +0.23 ± s -process peak (Ba, La, Ce, Pr, Nd)are significantly overproduced relative to the elements at thefirst s -process peak (Sr, Y, Zr) and mildly underproducedrelative to one of the elements at the third s -process peak(Pb).The AGB models in the FRUITY database are com-puted at several metallicities that bracket the metallicitiesof the clusters of interest. The differing neutron fluxes, am-bient physical conditions, and subsequent AGB evolutionaffect the s -process yields to produce a substantial depen-dence on the mass of the AGB progenitor. We overlay thesepredictions on the cluster values shown in Figure 18. A fewpoints are notable. First, the [ hs / ls ] index in NGC 5824 ishigher by ≈ hs / ls ] indices in the other four clusters span a range of ≈ hs / ls ] index in NGC 5824, and the modelswould predict AGB masses ranging from ≈ ⊙ for theother four clusters. Second, the [Pb/ hs ] index in NGC 5824is in very good agreement with the other three clusters wherePb has been detected. The models consistently imply AGBmasses ≈ ⊙ based on the [Pb/ hs ] index. The modelspredict AGB masses of ≈ ⊙ produced the s -process ma-terial in M4; however, for all other clusters, no single massmodel can self-consistently account for the observed indices.In principle, the timescale for star formation and chem-ical enrichment in each cluster matches the lifetime ofthe lowest-mass stars known to contribute to the met-als in the present-day stars. A 6 M ⊙ star has a life-time of ≈
60 Myr, while a 3 M ⊙ star has a lifetime of ≈
300 Myr. Roederer et al. (2011) examined tables of AGBnucleosynthesis yields and concluded that AGB stars with
M > ⊙ were responsible for producing the s -process ma-terial in M22, implying a formation timescale of ≈
300 Myror less. Straniero, Cristallo, & Piersanti (2014) re-examinedthe M22 abundances derived by Roederer et al. using a self-consistent approach to the AGB models, integrating theyields of all stars above a certain mass and weighting bythe initial mass function. They confirmed the suspicionsof Roederer et al., deriving a timescale of 144 ±
49 Myrfor M22, requiring the yields from AGB stars with masses > ⊙ or so. Shingles et al. (2014) adopted a differentset of models and found a minimum enrichment timescaleof ≈ ±
60 Myr, with slightly lower-mass AGB starscontributing. The difference between the two results canbe attributed to the prescription for the formation of the C pocket in AGB stars with 3 < M/ M ⊙ < ∼
300 Myr) by the split subgiant branch in M22(Marino et al. 2012).The weak s -process, which operates in massive ( ∼
25 M ⊙ ), rapidly-rotating, low-metallicity stars, is an-other proposed mechanism for producing s -process mate-rial in the early universe (e.g., The, El Eid, & Meyer 2000;Pignatari et al. 2008). These rapidly-evolving stars leave lit-tle time for the s -process to flow to the highest-mass nu-clei. Using Figure 1 of Frischknecht, Hirschi, & Thielemann(2012), we estimate that [ hs / ls ] can take a wide range of val-ues, but [Pb/ hs ] < − hs ]index is slightly super-solar in the four clusters shown inFigure 18, so this is not a likely candidate for the s -processmaterial found in these clusters.Detailed comparisons with AGB models, like the workof Straniero et al. (2014) or Shingles et al. (2014), are be-yond the scope of the present study. The similarity of the[Pb/ hs ] ratios between NGC 5824 and M22 argues for a rel-atively short (few hundred Myr or less) enrichment timescalefor NGC 5824. We have no reason to discount the measure-ments that form the [ hs / ls ] index in NGC 5824, but its valueis quite different from that found in the other four clustersand predicted using low-metallicity AGB models. We rec-ommend the issue be revisited should more stars in the r + s group in NGC 5824 be found. Th Nuclear Chronometer inNGC 5824
We derive an upper limit from the non-detection of the Th ii line at 4019.13 ˚A in the MIKE spectrum of star 42009955.The Th isotope is radioactive, and it can only be pro-duced by r -process nucleosynthesis. We calculate an age ofthe r -process material in NGC 5824 by comparing the cur-rent ratio of Th and a stable element produced by the r -process (Eu) with their initial production ratio (Table 9,Roederer et al. 2009). The current ratio, log ǫ (Th/Eu) < − ± ǫ (Th/Eu) ratio to exclude the presence of a rare, poorly-understood aspect of r -process nucleosynthesis, the so-called“actinide boost” (Hill et al. 2002; Schatz et al. 2002). Thisphenomenon is characterized by enhanced Th abundances,and it is found in a handful of low-metallicity field stars.If NGC 5824 is 13 Gyr old, the radioactive decay from Thproduced in an actinide boost would yield a current ratioof log ǫ (Th/Eu) − ± ǫ (Th/Eu) ratios, in sharper tension withour observed limit.Roederer & Thompson (2015) summarized the Thabundances or upper limits reported for stars in six clusters.To this list we add NGC 5824 (and M75, Kacharov et al.2013, which was inadvertently omitted). None of these clus-ters, including NGC 5824, show evidence of an actinideboost, hinting that it may not occur or its signature is di-luted beyond recognition in globular cluster environments.This observation may help in the hunt to identify or excludecandidate sites for the actinide boost phenomenon. c (cid:13) , 1–26 bundances in NGC 5824
11 SUMMARY
We have examined the first sets of high-resolution spectro-scopic observations obtained for the luminous, metal-poorglobular cluster NGC 5824. Fifty stars have been observedusing the M2FS spectrograph at Magellan, and 26 of themhave S/N sufficient to perform a detailed abundance analy-sis of 20 species of 17 elements. Two stars were re-observedusing the MIKE spectrograph to achieve broader wavelengthcoverage.We derive h [Fe/H] i = − ± ± ω Cen, M3, M13, M15,NGC 2419, NGC 2808, NGC 4833, and NGC 6752. Our lim-ited data on O, Na, and Al do not permit us to examine thedetails of the light-element correlations and anti-correlationsproduced by p -capture reactions. However, these elementsvary with each other, and Mg, in the usual manner in thetwo stars observed with MIKE.The n -capture abundance patterns in 25 of the 26 starsobserved with M2FS are effectively identical and consistentwith r -process nucleosynthesis patterns found in field starswith similar stellar parameters and metallicities. One starshows a significant enhancement of s -process material, aswell. We consider, and dismiss, the possibilities that thisstar self-enriched or obtained the s -process material from acompanion that passed through the AGB phase of evolu-tion. The s -process pattern resembles that found in groupsof stars in the other low-metallicity clusters M2, M22, andNGC 5286. Mounting evidence suggests that intermediate-mass AGB stars ( ≈ ⊙ ) were responsible for producingthe s -process material during the time of star formation inthese clusters. The percentage of s -process-enhanced stars inNGC 5824 is within the range of that found in other clustersin this peculiar class.We close by noting that several other low-metallicityclusters show evidence for internal dispersion among the n -capture elements. M2, M22, and NGC 5286 have beendiscussed extensively in Section 10.5. Heavy element dis-persion in clusters ω Cen (Smith et al. 2000; D’Orazi et al.2011), M15 (e.g., Sneden et al. 1997; Otsuki et al. 2006;Worley et al. 2013), and NGC 1851 (Yong & Grundahl2008; Villanova, Geisler, & Piotto 2010; Carretta et al.2011) has been confirmed by multiple investigators. Com-pelling evidence for variations in M19 has been pre-sented recently by Johnson et al. (2015b). Other clusterswith more subtle variations or less secure evidence in-clude 47 Tuc (Cordero et al. 2015), M75 (Kacharov et al.2013), M80 (Carretta et al. 2015), NGC 362 (Carretta et al.2013a), NGC 4372 (San Roman et al. 2015), and NGC 5897(Koch & McWilliam 2014). We have not included these clus-ters in the discussion here due to their enormous complex- ity, small sample sizes, or small numbers of n -capture el-ements studied. Spectroscopic followup observations of the n -capture elements in these clusters would be greatly wel-comed. ACKNOWLEDGMENTS
I.U.R. thanks G. Da Costa for thoughtful, encouraging dis-cussions throughout the course of this study; D. Kelson andE. Villanueva for their assistance installing and maintainingthe CarPy MIKE reduction pipeline; A. Koch for check-ing the list of clusters with possible n -capture variations;E. Olszewski, I. Thompson, and M. Walker for helping tomake M2FS a reality; and V. Placco for sharing insights onCEMP star binaries. We appreciate our referee’s thoughtfulcomments that have substantially improved this manuscript.This research has made use of NASA’s Astrophysics DataSystem Bibliographic Services, the arXiv preprint serveroperated by Cornell University, the SIMBAD and VizieRdatabases hosted by the Strasbourg Astronomical Data Cen-ter, the Atomic Spectra Database (Kramida et al. 2014)hosted by the National Institute of Standards and Technol-ogy, and the r suite of software (R Core Team 2014). iraf is distributed by the National Optical Astronomy Observa-tories, which are operated by the Association of Universitiesfor Research in Astronomy, Inc., under cooperative agree-ment with the National Science Foundation. M.M. and J.I.B.gratefully acknowledge support from the U.S. National Sci-ence Foundation to develop M2FS (AST-0923160). REFERENCES
Aldenius M., Lundberg H., Blackwell-Whitehead R. 2009,A&A, 502, 989Allen D. M., Ryan S. G., Rossi S., Beers T. C., TsangaridesS. A. 2012, A&A, 548, A34Alonso A., Arribas S., Mart´ınez-Roger C. 1999, A&AS, 140,261Aoki W., Beers T. C., Christlieb N., Norris J. E., RyanS. G., Tsangarides S. 2007, ApJ, 655, 492Aoki W., et al. 2008, ApJ, 678, 1351Aoki W., Suda T., Beers T. C., Honda S. 2015, AJ, 149, 39Asplund M. 2005, ARA&A, 43, 481Asplund M., Grevesse N., Sauval A. J., Scott P. 2009,ARA&A, 47, 481Bailey III J. I., Mateo M. L., Bagish A., Crane J., SlaterC. T. 2012, Proc. SPIE, 8446, 5G-1Bedell M., Mel´endez J., Bean J. L., Ram´ırez I., Leite P.,Asplund M. 2014, ApJ, 795, 23Beers T. C., Christlieb N. 2005, ARA&A, 43, 531Bellazzini M., et al. 2008, AJ, 136, 1147Bernstein R., Shectman S. A., Gunnels S. M., MochnackiS., Athey A. E. 2003, Proc. SPIE, 4841, 1694Bessell M. S. 1979, PASP, 91, 589Bi´emont ´E., Garnir H. P., Palmeri P., Li Z. S., Svanberg S.2000, MNRAS, 312, 116Bi´emont ´E., et al. 2011, MNRAS, 414, 3350Bisterzo S., Gallino R., Straniero O., Cristallo S., K¨appelerF. 2011, MNRAS, 418, 284 c (cid:13) , 1–26 Roederer et al.
Bohlin R. C., Hill J. K., Jenkins E. B., Savage B. D., SnowJr., T. P., Spitzer Jr., L., York D. G. 1983, ApJS, 51, 277Booth A. J., Blackwell D. E., Petford A. D., Shallis M. J.1984, MNRAS, 208, 147Brocato E., Buonanno R., Malakhova Y., Piersimoni A. M.1996, A&A, 311, 778Carballo-Bello J. A., Sollima A., Mart´ınez-Delgado D.,Pila-D´ıez B., Leaman R., Fliri J., Mu˜noz R. R., Corral-Santana J. M. 2014, MNRAS, 445, 2971Cardelli J. A., Clayton G. C., Mathis J. S. 1989, ApJ, 345,245Carretta E., 2014, ApJL, 795, L28Carretta E., Bragaglia A., Gratton R. G., Lucatello S., Mo-many Y. 2007, A&A, 464, 927Carretta E., et al. 2009a, A&A, 505, 117Carretta E., Bragaglia A., Gratton R., Lucatello S. 2009b,A&A, 505, 139Carretta E., Bragaglia A., Gratton R. G., Recio-Blanco A.,Lucatello S., D’Orazi V., Cassisi S. 2010a, A&A, 516, A55Carretta E., et al. 2010b, A&A, 520, A95Carretta E., et al. 2010c, ApJL, 722, L1Carretta E., Lucatello S., Gratton R. G., Bragaglia A.,D’Orazi V. 2011, A&A, 533, A69Carretta E., et al. 2013a, A&A, 557, A138Carretta E., Gratton R. G., Bragaglia A., D’Orazi V., Lu-catello S., Sollima A., Sneden C. 2013b, ApJ, 769, 40Carretta E., et al. 2014, A&A, 564, A60Carretta E., et al. 2015, A&A, 578, A116Casagrande L., Ram´ırez I., Mel´endez J., Bessell M., As-plund M. 2010, A&A, 512, A54Casagrande L., et al. 2014, MNRAS, 439, 2060Castelli F., Kurucz R. L. 2003, Proc. IAU Symp. No 210,Modelling of Stellar Atmospheres, N. Piskunov et al., eds.,A20 (arXiv:0405087)Cohen J. G. 2011, ApJL, 740, L38Cohen J. G., Kirby E. N. 2012, ApJ, 760, 86Cohen J. G., Kirby E. N., Simon J. D., Geha M. 2010, ApJ,725, 288Cordero M. J., Hansen C. J., Johnson C. I., PilachowskiC. A. 2015, ApJL, 808, L10Correnti M., Bellazzini M., Dalessandro E., Mucciarelli A.,Monaco L., Catelan M. 2011, MNRAS, 417, 2411Cˆote P., Pryor C., McClure R. D., Fletcher J. M., HesserJ. E. 1996, AJ, 112, 574Cowan J. J., et al. 2002, ApJ, 572, 861Cowan J. J., et al. 2005, ApJ, 627, 238Cox A. N. 2000, Allen’s Astrophysical Quantities, Springer-Verlag, New York, p. 340–341Cristallo S., Straniero O., Piersanti L., Gobrecht D. 2015,ApJS, in press (arXiv:1507.07338)Da Costa G. S., Held E. V., Saviane I., Guillieuszik M.2009, ApJ, 705, 1481Da Costa G. S., Held E. V., Saviane I. 2014, MNRAS, 438,3507Den Hartog E. A., Lawler J. E., Sneden C., Cowan J. J.2003, ApJS, 148, 543Den Hartog E. A., Lawler J. E., Sneden C., Cowan J. J.2006, ApJS, 167, 292Den Hartog E. A., Lawler J. E., Sobeck J. S., Sneden C.,Cowan J. J. 2011, ApJS, 194, 35Di Criscienzo M., et al. 2011, MNRAS, 414, 3381D’Orazi V., Gratton R., Lucatello S., Carretta E., Bra- gaglia A., Marino A. F. 2010, ApJL, 719, L213D’Orazi V., Gratton R. G., Pancino E., Bragaglia A., Car-retta E., Lucatello S., Sneden C. 2011, A&A, 534, A29Dubath P., Meylan G., Mayor M. 1997, A&A, 324, 505Fabbian D., Asplund M., Barklem P. S., Carlsson M., Kisel-man D. 2009, A&A, 500, 1221Frebel A., Collet R., Eriksson K., Christlieb N., Aoki W.2008, ApJ, 684, 588Frischknecht U., Hirschi R., Thielemann F.-K. 2012, A&A,538, L2Gallino R., Arlandini C., Busso M., Lugaro M., TravaglioC., Straniero O., Chieffi A., Limongi M. 1998, ApJ, 497,388Gonzalez G., Wallerstein G. 1998, AJ, 116, 765Gratton R. G., Sneden C., Carretta E., Bragaglia A. 2000,A&A, 354, 169Gratton R. G., et al. 2001, A&A, 369, 87Gratton R., Sneden C., Carretta E. 2004, ARA&A, 42, 385Gratton R. G., Carretta E., Bragaglia A., Lucatello S.,D’Orazi V. 2010, A&A, 517, A81Grillmair C. J., Freeman K. C., Irwin M., Quinn P. J. 1995,AJ, 109, 2553Harris W. E. 1996, AJ, 112, 1487Hill V., et al. 2002, A&A, 387, 560Ibata R. A., Gilmore G., Irwin M. J. 1995, MNRAS, 277,781Ivans I. I., Sneden C., Kraft R. P., Suntzeff N. B., SmithV. V., Langer G. E., Fulbright J. P. 1999, AJ, 118, 1273Ivans I. I., Kraft R. P., Sneden C., Smith G. H., Rich R. M.,Shetrone M. 2001, AJ, 122, 1438Ivans I. I., Simmerer J., Sneden C., Lawler J. E., CowanJ. J., Gallino R., Bisterzo S. 2006, ApJ, 645, 613Ivarsson S., Litz´en U., Wahlgren G. M. 2001, Phys. Scr.,64, 455Johnson J. A., Bolte M. 2002, ApJ, 579, 616Johnson C. I., Pilachowski C. A. 2010, ApJ, 722, 1373Johnson C. I., Kraft R. P., Pilachowski C. A., Sneden C.,Ivans I. I., Benman G. 2005, PASP, 117, 1308Johnson C. I., et al. 2015a, AJ, 149, 71Johnson C. I., Rich R. M., Pilachowski C. A., Caldwell N.,Mateo M., Bailey III J. I., Crane J. D. 2015, AJ, 150, 63Kacharov N., Koch A., McWilliam A. 2013, A&A, 554, A81Kelson D. D. 2003, PASP, 115, 688King I. R., et al. 2012, AJ, 144, 5Koch A., McWilliam A. 2008, AJ, 135, 1551Koch A., McWilliam A. 2011, AJ, 142, 63Koch A., McWilliam A. 2014, A&A, 565, A23Kraft R. P., Sneden C., Smith G. H., Shetrone M. D.,Langer G. E., Pilachowski C. A. 1997, AJ, 113, 279Kramida A., Ralchenko Y., Reader J., and the NIST ASDTeam. 2014, NIST Atomic Spectra Database (v. 5.2), on-line, URL: http://physics.nist.gov/asdKunder A., et al. 2014, A&A, 572, A30Kurucz R. L., Bell B. 1995, Cambridge, MA: SmithsonianAstrophysical Observatory, CD-ROM 23Lai D. K., Smith G. H., Bolte M., Johnson J. A., LucatelloS., Kraft R. P., Sneden C. 2011, AJ, 141, 62Langer G. E., Fischer D., Sneden C., Bolte M. 1998, AJ,115, 685Lawler J. E., Dakin J. T. 1989, J. Opt. Soc. Am. B OpticalPhysics, 6, 1457Lawler J. E., Bonvallet G., Sneden C. 2001, ApJ, 556, 452 c (cid:13) , 1–26 bundances in NGC 5824 c (cid:13) , 1–26 Roederer et al.
Gullieuszik M., Barbuy B., Ortolani S. 2012, A&A, 450,A27Sbordone L., Salaris M., Weiss A., Cassisi S. 2011, A&A,534, A9Schatz H., Toenjes R., Pfeiffer B., Beers T. C., Cowan J. J.,Hill V., Kratz K.-L. 2002, ApJ, 579, 626Shetrone M. D. 1996, AJ, 112, 1571Shimanskaya N. N., Mashonkina L. I., Sakhibullin N. A.2000, Astron. Rep., 44, 530Shingles L. J., Karakas A. I., Hircshi R., Fishlock C. K.,Yong D., Da Costa G. S., Marino A. F. 2014, ApJ, 795,34Simmerer J., Ivans I. I., Filler D., Francois P., CharbonnelC., Monier R., James G. 2013, ApJL, 764, L7Smith V. V., Lambert D. L., Nissen P. E. 1998, ApJ, 506,405Smith V. V., Suntzeff N. B., Cunha K., Gallino R., BussoM., Lambert D. L., Straniero O. 2000, AJ, 119, 1239Sneden C. A. 1973, Ph.D. Thesis, Univ. of Texas at AustinSneden C., Kraft R. P., Shetrone M. D., Smith G. H.,Langer G. E., Prosser C. F. 1997, AJ, 114, 1964Sneden C., Pilachowski C. A., Kraft R. P. 2000, AJ, 120,1351Sneden C., et al. 2003, ApJ, 591, 936Sneden C., Preston G. W., Cowan J. J. 2003, ApJ, 592,504Sneden C., Kraft R. P., Guhathakurta P., Peterson R. C.,Fulbright J. P. 2004, AJ, 127, 2162.Sneden C., Cowan J. J., Gallino R. 2008, ARA&A, 46, 241Sneden C., Lawler J. E., Cowan J. J., Ivans I. I., Den Har-tog E. A. 2009, ApJS, 182, 80Sobeck J. S., Lawler J. E., Sneden C. 2007, ApJ, 667, 1267Sobeck J. S., et al. 2011, AJ, 141, 175Straniero O., Cristallo S., Piersanti L. 2014, ApJ, 785, 77Takeda Y., Zhao G., Chen Y.-Q., Qiu H.-M., Takada-HidaiM. 2002, PASJ, 54, 275The L.-S., El Eid M. F., Meyer B. S. 2000, ApJ, 533, 998Truran J. E., Cowan J. J., Pilachowski C. A., Sneden C.2002, PASP, 114, 1293Van Eck S., Goriely S., Jorissen A., Plez B. 2001, Nature,412, 793Ventura P., D’Antona F., Di Criscienzo M., Carini R.,D’Ercole A., Vesperini E. 2012, ApJL, 761, L30Villanova S., Geisler D., Piotto G. 2010, ApJL, 722, L18Wickliffe M. E., Lawler J. E., Nave G. 2000, J. Quant.Spectrosc. Rad. Trans., 66, 363Wood M. P., Lawler J. E., Sneden C., Cowan J. J. 2013,ApJS, 208, 27Wood M. P., Lawler J. E., Sneden C., Cowan J. J. 2014a,ApJS, 211, 20Wood M. P., Lawler J. E., Den Hartog E. A., Sneden C.,Cowan J. J. 2014b, ApJS, 214, 18Worley C. C., Hill V., Sobeck J., Carretta E. 2013, A&A,553, A47Yong D., Grundahl F. 2008, ApJ, 672, L29Yong D., Grundahl F., Lambert D. L., Nissen P. E.,Shetrone M. D. 2003, A&A, 402, 985Yong D., Grundahl F., Nissen P. E., Jensen H. R., LambertD. L. 2005, A&A, 438, 875Yong D., Lambert D. L., Paulson D. B., Carney B. W.2008a, ApJ, 673, 854Yong D., Karakas A. I., Lambert D. L., Chieffi A., Limongi M. 2008b, ApJ, 689, 1031Yong D., et al. 2013, MNRAS, 434, 3542Yong D., et al. 2014a, MNRAS, 439, 2638Yong D., et al. 2014b, MNRAS, 441, 3396Zinn R. 1980, ApJ, 241, 602 c (cid:13)000