A homogeneous analysis of globular clusters from the APOGEE survey with the BACCHUS code. I. The Northern clusters
T. Masseron, D.A.García-Hernández, Sz. Mészáros, O. Zamora, F. Dell'Agli, C. Allende Prieto, B. Edvardsson, M. Shetrone, B. Plez, J. G. Fernández-Trincado, K. Cunha, H. Jönsson, D. Geisler, T. C. Beers, R. E. Cohen
aa r X i v : . [ a s t r o - ph . S R ] D ec Astronomy & Astrophysicsmanuscript no. NdCeClusters c (cid:13)
ESO 2018December 24, 2018
A homogeneous analysis of globular clusters from the APOGEEsurvey with the BACCHUS code. I. The Northern clusters
T. Masseron , , D.A.García-Hernández , , Sz. Mészáros , , O. Zamora , , F. Dell’Agli , , C. Allende Prieto , , B.Edvardsson , M. Shetrone , B. Plez , J. G. Fernández-Trincado , , , K. Cunha , , H, Jönsson , D. Geisler , , ,T. C. Beers , and R. E. Cohen Instituto de Astrofísica de Canarias, E-38205 La Laguna, Tenerife, Spain Departamento de Astrofísica, Universidad de La Laguna, E-38206 La Laguna, Tenerife, Spaine-mail: [email protected] ELTE Eötvös Loránd University, Gothard Astrophysical Observatory, Szombathely, Hungary Premium Postdoctoral Fellow of the Hungarian Academy of Sciences Theoretical Astrophysics, Department of Physics and Astronomy, Uppsala University, Box 516, SE-751 20 Uppsala, Sweden McDonald Observatory, University of Texas at Austin, Fort Davis, TX 79734, USA Laboratoire Univers et Particules de Montpellier, Université de Montpellier, CNRS, 34095, Montpellier Cedex 05, France Instituto de Astronomía y Ciencias Planetarias, Universidad de Atacama, Copayapu 485, Copiapó, Chile Departamento de Astronomía, Casilla 160-C, Universidad de Concepción, Concepción, Chile Institut Utinam, CNRS UMR 6213, Université Bourgogne-Franche-Comté, OSU THETA Franche-Comté, Observatoire de Be-sançon, BP 1615, 25010 Besançon Cedex, France Observatório Nacional / MCTI, Rua Gen. José Cristino, 77, 20921-400 Rio de Janeiro, Brazil Steward Observatory, University of Arizona Tucson 85719, USA Lund Observatory, Department of Astronomy and Theoretical Physics, Lund University, Box 43, SE-22100 Lund, Sweden Instituto de Investigación Multidisciplinario en Ciencia y Tecnología, Universidad de La Serena. Avenida Raúl Bitrán S / N, LaSerena, Chile Departamento de Física y Astronomía, Facultad de Ciencias, Universidad de La Serena. Av. Juan Cisternas 1200, La Serena, Chile Department of Physics and JINA Center for the Evolution of the Elements, University of Notre Dame, Notre Dame, IN 46556,USA Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21210, USAReceived ; accepted
ABSTRACT
Aims.
We aim at providing abundances of a large set of light and neutron-capture elements homogeneously analyzed and covering awide range of metallicity to constrain globular cluster (GC) formation and evolution models.
Methods.
We analyze a large sample of 885 GCs giants from the APOGEE survey. We used the Cannon results to separate the red giantbranch and the asymptotic giant branch stars, not only allowing for a refinement of surface gravity from isochrones, but also providingan independent H-band spectroscopic method to distinguish stellar evolutionary status in clusters. We then use the BACCHUS codeto derive metallicity, microturbulence, macroturbulence and many light-element abundances as well as the neutron-capture elementsNd and Ce for the first time from the APOGEE GCs data.
Results.
Our independent analysis helped us to diagnose issues regarding the standard analysis of the APOGEE DR14 for low-metallicity GC stars. Furthermore, while we confirm most of the known correlations and anti-correlation trends (Na-O, Mg-Al,C-N), we discover that some stars within our most metal-poor clusters show an extreme Mg depletion and some Si enhancementbut at the same time show some relative Al depletion, displaying a turnover in the Mg-Al diagram. These stars suggest that Alhas been partially depleted in their progenitors by very hot proton-capture nucleosynthetic processes. Furthermore, we attempted toquantitatively correlate the spread of Al abundances with the global properties of GCs. We find an anti-correlation of the Al spreadagainst clusters metallicity and luminosity, but the data do not allow to find clear evidence of a dependence of N against metallicity inthe more metal-poor clusters.
Conclusions.
Large and homogeneously analyzed samples from on-going spectroscopic surveys unveil unseen chemical details formany clusters, including a turnover in the Mg-Al anti-correlation, thus yielding new constrains for GCs formation / evolution models. Key words. stars: abundances – globular clusters: general –
1. Introduction
The existence of multiple populations in Globular Clusters(GC) can be unambiguously observed in appropriate colour-magnitude diagrams (e.g. Milone et al. 2017, and referencetherein) and the variations in colours are associated to abun-dances variations (see e.g. Monelli et al. 2013; Mészáros et al. 2018). The colour indices sensitive to multiple populations havesuch a sensitivity because their band-pass includes the spectralfeatures that are changing, for example C
UBI or the Milone etal. "magic trio" with WFC3 UVIS. Some broadband colours(V-I in most cases) are mostly insensitive to such (C,N,O etc.)variations, but most colours indices are sensitive to metal-
Article number, page 1 of 17 & Aproofs: manuscript no. NdCeClusters licity in various ways. Indeed, it is known for a long timethat some elemental abundances vary from star to star withinthe clusters. The most observed elements showing abundancevariations are C, N, O, Mg, Na and Al (Smith 1987; Kraft1994; Gratton et al. 2004, 2012, see past reviews and referencestherein). Si has been more recently revealed to vary in some clus-ters (e.g. Yong et al. 2005; Carretta et al. 2009b). Ca is anotherelement that has been revealed to show spread in some clusters(Marino et al. 2009). Finally, K has been observed to vary in onlytwo clusters: NGC 2808 and NGC 2419 (Cohen & Kirby 2012;Mucciarelli et al. 2015a). Regarding neutron capture elements,a few clusters show significant dispersion (Marino et al. 2009;Sobeck et al. 2011; Roederer & Sneden 2011; Shingles et al.2014, including M 22, M 15, M 92 and M4). Last but notleast, some colour indices may suggest some He enhancement(Milone et al. 2015). Correlation or anti-correlation betweenthose elements have been observed and provide hints to de-cipher their origins. While there is a consensus on a hot H-burning nucleosynthesis source, a broad range of polluters havebeen proposed to explain those chemical trends: fast rotatingmassive stars, massive asymptotic giant branch (AGB) stars,intermediate-mass binaries and supermassive stars (see discus-sion of the various models by Renzini et al. 2015; Charbonnel2016).Nevertheless, all those models try to establish a universalscenario for the formation of GCs. One way to distinguishthose scenarios may come from confronting the various abun-dance trends and spread against global properties of the clus-ters, as already observationally attempted by Carretta et al.(2009b) and Milone et al. (2017) and theoretically predicted byDell’Agli et al. (2018) and Szécsi & Wünsch (2018). To do this,large samples of homogeneously analyzed stars are required toallow to compare trends or (anti-)correlations against clusterproperties and thus draw an overall picture of GC formation.But, as emphasized by Bastian & Lardo (2017), to date, therehave only been a few stars in a handful of GCs that have beenfully characterized in terms of their chemistry.In fact, the largest published spectroscopic homogeneousanalysis of GCs stars to date is from Carretta et al. (2009c) whogathered nearly 1400 stellar spectra from the VLT / GIRAFFEspectrograph, but they study only Na and O because of relativelylimited resolution. In parallel, Carretta et al. (2009b) measuredseveral elements over 17 GCs, but the sample is only of 202stars. Given the potential scientific impact of such studies, largespectroscopic surveys have now dedicated generous amount oftelescope time for the observation of GCs. Pancino et al. (2017)makes use of the Gaia-ESO survey (Gilmore et al. 2012) andstudied 510 stars over 9 clusters, but limited their conclusionsto Mg and Al elements. Mészáros et al. (2015) presented an in-dependent analysis of 428 Northern cluster stars using spec-tra published in the 10th data release of the SDSS III-ApachePoint Observatory Galactic Evolution Experiment (APOGEE)(Ahn et al. 2014). However, this analysis su ff ered from largerthan expected uncertainties in the fundamental C and N abun-dances. Nowadays, the fourteenth data release (DR14) of theSDSS IV / APOGEE2 survey (Gunn et al. 2006; Zasowski et al.2017; Majewski et al. 2017; Blanton et al. 2017) contains one ofthe largest samples of GCs stars with high enough resolution andsignal-to-noise ratio to allow the determination of abundancesfor many key elemental abundances for GCs studies. However,the extreme elemental abundances of GC stars are still uncer-tain in DR14 standard analysis as demonstrated by Jönsson et al.(2018). Given the larger than expected errors in the derivation of Cand N abundances by Mészáros et al. (2015) and the APOGEEStellar Parameters and Chemical Abundances Pipeline (ASP-CAP) (Holtzman et al. 2018), we independently revisit here theanalysis of APOGEE spectra from 10 Northern GCs and nearlydouble the sample compared to Mészáros et al. (2015)’s work(now 885) as more stars have been observed by APOGEEsince then. The current analysis includes measurement of C,N, O, Mg, Al, Si, K and Ca as well as the neutron-capture el-ements Ce and Nd for the first time in GCs since those ele-ments have been shown to be measurable in the APOGEE spec-tra (Hasselquist et al. 2016; Cunha et al. 2017).
2. Data
The general steps of target selection were the same as used byMészáros et al. (2015). After selecting radial velocity members,we cut stars that were beyond the tidal radius given by Harris(2010), and stars that had an o ff set larger than ± / APOGEE2 survey DR14(Abolfathi et al. 2018). The spectra have been reduced and com-bined following Holtzman et al. (2015, 2018), but the spectralnormalization has been done within our own code.
3. Analysis
We use the Brussels Automatic Code for Characterizing HighaccUracy Spectra (BACCHUS) (Masseron et al. 2016) to deter-mine metallicity, microturbulence and macroturbulence / v sin i aswell as abundances (C, N, O, Mg, Al, Na, Ca, Si, K, Nd andCe). The code runs on the fly the spectral synthesis code Tur-bospectrum (Plez 2012) and rely on a very large grid of MARCSmodel atmospheres (Gustafsson et al. 2008), extended over var-ious C values (-1.0 < [C / Fe] < + e ff , logg, metallicity). Note that although the gridalso contains a range of [ α / Fe] ( α being O, Ne, Mg, Si, S, Ar,Ca, and Ti), it does not contain models with variation in O, Mgand Si, but fixed Ne, Ar, Ca and Ti. Therefore we decided tolimit the grid such that [ α / Fe] =+ ff ective temperatures and surfacegravities (see Sec.3.1), the code determines the macroturbulencebased on a selection of Si I lines (simultaneously to adjust the Siabundance). Then, metallicity is determined from Fe I lines withmacro- and microturbulence parameters fixed. Because of theimportance of molecular lines in the H band, we successively de-termine the O, C and N abundances that dominate the molecularline strengths. Finally, we derived the other element abundances. Article number, page 2 of 17. Masseron et al.: BACCHUS and the 10 GCs
Teff2.521.510.5 5000 45002.521.510.5 5000 4500 5000 4500 5000 4500 5000 4500
Fig. 1. E ff ective temperatures and surface gravities derived by the Cannon (DR14) for the sample stars. Open magenta squares are RHB / eAGBstars and green full squares are RGB stars as determined from photometry except for M 107 and M 71, for which the separation has been donefrom spectroscopic parameters only. The figure illustrates the potentiality of separating RGB and RHB / eAGB solely from the Cannon results. Teff2.521.510.5 5000 45002.521.510.5 5000 4500 5000 4500 5000 4500 5000 4500
Fig. 2. E ff ective temperatures and surface gravities derived by ASPCAP (DR14) for the sample stars. Open magenta squares are RHB / eAGB starsand green full squares are RGB stars as determined from photometry except for M 107 and M 71, for which the separation has been done fromspectroscopic parameters only. In comparison to Fig. 1, this figures illustrates that the Cannon has more precise results when it comes to separat-ing the RGB / RHB / eAGB, but as shown in Jönsson et al. (2018), the ASPCAP parameters are more accurate when comparing to independentlyanalyzed local disk stars. Article number, page 3 of 17 & Aproofs: manuscript no. NdCeClusters
Fig. 3.
Microturbulence velocity values obtained for the sample stars bythe ASPCAP pipeline (upper panel) and by our preliminary run with theBACCHUS code (lower panel). Red points indicate stars with [Fe / H] < -1.5 and black points stars with [Fe / H] > -1.5. The dashed line showthe relation derived in the optical by the Gaia-ESO survey assumingT e ff = / H] = -1.5. Both works show a good correspondencewith the optical data for the metal-rich stars, but a larger dispersion isobtained for low-metallicity stars. The overall process was iterated twice to ensure self-consistencyand ran over a couple of days for the whole sample on our HT-Condor sytem.
Jönsson et al. (2018) illustrated that the ASPCAP DR14 analy-sis leads to an excellent precision in abundances for disk starsand first generation GC stars, lending support for the asso-ciated values of e ff ective temperature and gravity. Neverthe-less, Jönsson et al. (2018) has also demonstrated that AS < PCAPDR14 analysis do not provide satisfying results for second gen-eration GC stars. Moreover, as detailed in Sec. 3.4, we foundthat the temperatures provided by ASPCAP for GC stars intro-duce a bias in the oxygen abundances. Our BACCHUS analysisdoes not determine T e ff and logg, so we derive those parame-ters similarly to Mészáros et al. (2015), i.e. temperatures are in-ferred from photometry and surface gravities from isochrones.We computed the e ff ective temperatures by using the J-Ks re-lation from González Hernández & Bonifacio (2009), while thereddening values were selected from the Harris catalog (Harris2010) and kept as constant for each cluster.In order to assign correctly the surface gravities withisochrones, a distinction between early asymptotic giant branch(eAGB) or red horizontal branch (RHB), and red giant branch(RGB) stars must be made. This has been done by usingcolour-colour diagrams based on the most accurate ground-based photometry currently available using the same method asGarcía-Hernández et al. (2015). However, such photometric dataare only available for the most metal-poor clusters, and not forthe two most metal-rich clusters in our sample, M 71 and M 107.For these two clusters, we chose to use the Cannon (Ness et al.2015) analysis of DR14 parameters.While the accuracy of this machine learning algorithm is lim-ited to that of its training set, it has the advantage of exploitingall of the information that may be present in the stellar spec-tra, thus improving the precision of the stellar parameters de-termination. We show in Fig. 1 that the parameters derived bythe Cannon allow us to disentangle e ffi ciently the RGB fromthe RHB / eAGB stars. In comparison, we show in Fig. 2 a simi-lar diagram but with the ASPCAP DR14 calibrated parameters.Although the ASPCAP results also allow to separate the twogiant branches, the Cannon results show it more clearly. Con-sequently, we have separated empirically the RHB / eAGB andthe RGB stars in M 107 and M 71 using the Cannon results.This provides a new method for separating solely from H-bandspectroscopy the evolutionary status of giants in clusters, or inmono-age populations in general. In contrast, we found that thee ff ective temperatures and the surface gravities provided by theCannon are much less accurate than the ASPCAP values, thusthe latter are preferred for parameter determinations of globularclusters. We note that a few stars in M 5, M 13, M 3 and M 15have been apparently misclassified by the Cannon and / or ASP-CAP. However, it is not clear at this stage of this study whether itis due to an error in the Cannon determination of the parameters,or due to an error in the photometric classification. More workwould be needed to clarify the situation of those few outliers andit is beyond the scope of this paper.Once the evolutionary status is assigned, the surface gravitiesare determined by using the set of isochrones from the Padovagroup (Bertelli et al. 2008; Marigo et al. 2017). We use 12 bil-lion years isochrones for all GCs, except for M 107 and M 71for which we use 10 billion years. Regarding metallicities, we Article number, page 4 of 17. Masseron et al.: BACCHUS and the 10 GCs
Fig. 4.
Convolution values obtained for the sample stars against sur-face gravities (upper panel) and metallicity (lower panel). Red squaresare spectra with a low mean fiber number ( < > / s (R ∼ adapt the isochrones to the metallicities of the Harris (2010) cat-alog. Microturbulence velocities have been fixed to the relation ob-tained in the optical by the Gaia-ESO survey (such as describedby Smiljanic et al. 2014). However, in order to check the valid-ity of this relation in the H-band infrared spectra of APOGEE,we let the microturbulence parameter free in a preliminary run.The code determines the microturbulence by cancelling the trendof abundances against equivalent widths of a set of Fe I lines.Fig. 3 shows the results obtained with BACCHUS and comparesthem to the values derived in APOGEE DR14 and to the relationof the Gaia-ESO survey. All the relations agree well and showin particular an expected dependence with surface gravity. Nev-ertheless, the dispersion of the microturbulence values increasesignificantly for the most metal-poor stars ([Fe / H] < -1.5). Thisis due to the fact that the lines used for the microturbulence rela-tion become very weak and very sensitive to random uncertain- ties such as noise or continuum placement, and thus increasingthe uncertainties of the microturbulence relation. Because ourstudy contains a large number of very low-metallicity stars, wedecided to fix the microturbulence velocity to the optical relationfrom Gaia-ESO in order to reduce the impact of the growing er-ror at low metallicity on the abundances.Concerning the macroturbulence value, the BACCHUS codedoes not derive it directly, but rather derives the value of thespectral convolution necessary to match the observed line pro-file after the thermal, radiative, collisional and microturbulencebroadening have been taken into account in the synthesis. Thisconvolution parameter approximately represents the quadraticsum of instrumental resolution, projected velocity of the star( v sin i ) and macroturbulence. To determine it, the BACCHUScode iterates over a series of selected lines until it finds conver-gence between abundances measured from the few points aroundthe core of the line (being sensitive to the convolution value) andfrom the equivalent width of the lines (insensitive to the convo-lution). For this particular analysis, we use a Gaussian kernel forthe convolution and probe it over a set of clean Si I lines (thoselines being strong enough in all our metal-poor stars in contrastto Fe I lines, see next Subsection).In Fig. 4, we show the resulting convolution values derivedfrom our procedure against log g and metallicity. We highlightedthe average fiber number of each target in the figure. It is remark-able that we obtained higher convolution values on average forstars with a larger fiber number than the ones with a lower fibernumber. This is fully consistent with the fact that the APOGEEspectrograph has a variable resolution depending on the fiber(Nidever et al. 2015). Although the BACCHUS code is makingsome approximations regarding the macroturbulence and instru-ment line profile broadening, the recovery of the fiber impact onthe spectra demonstrates the ability of the code to account forthe various source of line broadening parameters when derivingthe stars abundances.In the same Figure, we compare our convolution values to theadopted value for macroturbulence from DR14 APOGEE datarelease (Holtzman et al. 2018). Holtzman et al. (2018) founda dependence only in metallicity in DR14. Our results tendto rather show a dependence in log g , consistently with pre-vious findings from Hekker & Meléndez (2007); Gray (2008),but we do not find a dependence on metallicity. We remindthe reader that the macroturbulence velocity relation derived inHoltzman et al. (2018) is based on the whole APOGEE sample,which is dominated by more metal-rich stars than present in ourGC sample and that may be responsible for some of the disagree-ment. In any case, we stress that if the macroturbulence is reallyoverestimated at low metallicity, we expect that all abundancesand parameters derived from the ASPCAP pipeline to be signif-icantly a ff ected for all stars with metallicities below -2.0. Once the macroturbulence, microturbulence, T e ff and logg havebeen fixed, the metallicity is derived from a selection of threeFe I lines. Fig. 5 shows the median metallicity obtained for eachcluster and compared to various literature values. The agreementbetween all the studies is good overall. While there is evidencefor metallicity variations within some peculiar clusters (e.g. ω Cen, NGC 6934 Norris & Da Costa 1995; Marino et al. 2018),the relatively low star-to-star metallicity dispersion as shownin Fig. 5 is consistent with the conclusions of Carretta et al.(2009a), claiming that most of the clusters can be consid-ered mono-metallic regarding the Fe abundance. However, we
Article number, page 5 of 17 & Aproofs: manuscript no. NdCeClusters
Fig. 5.
Median metallicities obtained by the present work, APOGEEDR14 and Mészáros et al. (2015) for the 10 same GCs stars. Errorbarsindicate the star-by-star metallicity rms. Green points are literature val-ues (Carretta et al. 2009a).
Fig. 6. Di ff erence in abundances obtained by BACCHUS usingAPOGEE T e ff and logg with those obtained using our parameters. Thepoints are colour-coded by temperature and the points are ordered bycluster metallicity for each element. The squares represent the meanrandom line-by-line scatter uncertainties of each element. Systematicuncertainties are comparable to random uncertainties except for C, N,O, Ce and Nd. find a systematic o ff set of ∼ ff ects in optical Fe I lines in M 2 stars a ff ect the metal-licity. In the H-band we also strongly suspect 3D-NLTE e ff ects,notably because 1D-LTE synthesis of a very high resolution ofArcturus spectrum does not provide a satisfying fit for Fe I lineprofiles. A more extended discussion will be given in a forthcom-ing paper. It is for this reason that we carefully select only threeFe I lines which are as insensitive as possible to NLTE e ff ects todetermine the metallicities.In any case, such an o ff set in metallicity does not a ff ect ourdiscussion, because we mostly focus on the relative trends orranges of abundance-abundance diagrams, cluster by cluster. -101-101-101-101-101-101-101-101-101-101-101-101-101-101-101-101 4000 4500 5000-101 4000 4500 5000 4000 4500 5000M154000 4500 5000-101 4000 4500 5000 4000 4500 5000M924000 4500 5000-101 4000 4500 5000 4000 4500 5000M534000 4500 5000-101 4000 4500 5000 4000 4500 5000N54664000 4500 5000-101 4000 4500 5000 4000 4500 5000M24000 4500 5000-101 4000 4500 5000 4000 4500 5000M134000 4500 5000-101 4000 4500 5000 4000 4500 5000M34000 4500 5000-101 4000 4500 5000 4000 4500 5000M5 Fig. 7.
The C, N and O abundances as a function of e ff ective tempera-ture for the same stars as a function of e ff ective temperature for threestudies (this work, first column, uncalibrated APOGEE DR14, secondcolumn, and Mészáros et al. 2015, third column). Blue triangles indi-cates upper limits. Our results show that C, N and O abundances havemostly upper limits for T e ff > e ff > Abundances have been derived by the code on a line-by-line ba-sis. The lines used for abundance determination are displayedin Tab. 1. Note that the BACCHUS code automatically adjuststhe window / mask on a star-by-star and line-by-line basis (seethe manual for more details Masseron et al. 2016). Therefore weprovide only central wavelengths in Table 1. Among the severalabundance indicators that the code o ff ers (equivalent width, coreline intensity or χ ) and their respective flags, we selected theabundances corresponding to the minimum χ , but still use theother methods to reject any suspicious line. The final abundanceis the mean of the abundances of the non-flagged lines. Systematic errors have been evaluated by comparing the abun-dances obtained using the uncalibrated e ff ective temperature andsurface gravities from APOGEE DR14 and running the BAC-CHUS code following the same procedure as described above.The corresponding results are reported for each star of the sam-ple in Fig. 6 and compared to the mean random uncertainties Article number, page 6 of 17. Masseron et al.: BACCHUS and the 10 GCs -0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5 4000 4500 5000-0.500.5 4000 4500 5000 4000 4500 5000M154000 4500 5000-0.500.5 4000 4500 5000 4000 4500 5000M924000 4500 5000-0.500.5 4000 4500 5000 4000 4500 5000M534000 4500 5000-0.500.5 4000 4500 5000 4000 4500 5000N54664000 4500 5000-0.500.5 4000 4500 5000 4000 4500 5000M24000 4500 5000-0.500.5 4000 4500 5000 4000 4500 5000M134000 4500 5000-0.500.5 4000 4500 5000 4000 4500 5000M34000 4500 5000-0.500.5 4000 4500 5000 4000 4500 5000M5
Fig. 8.
Comparison Mg, Al, Si and Ca abundances for the same stars asa function of e ff ective temperature for three studies (this work, first col-umn, uncalibrated APOGEE DR14, second column, and Mészáros et al.2015, , third column). Blue triangles are upper limits. To clear up thediagrams with intrinsic clusters abundances variations, only stars of thefirst population ([Al / Fe] < (the latter being derived from the line-by-line abundance disper-sion). In this figure, systematic uncertainties are comparable torandom uncertainties for most of the elements. However, theyappear significantly larger for C, N, O, Nd and Ce. This is due tothe high sensitivity of their abundances to e ff ective temperatureand / or surface gravity. Indeed, the APOGEE DR14 parameterstend to show systematic parameter di ff erences with optical spec-troscopy, as highlighted by Jönsson et al. (2018), a ff ecting theabundance determination of the most sensitive elements. Never-theless, in Fig. 6 we illustrate the imàct of systematic errors onthe abundances of Na, Mg, K, Si, Ca and Fe as well as on C,N, and O, by using another set of e ff ective temperature or sur-face gravity. These e ff ects are relatively small compared to theintrinsic variations observed in GCs (see Sec. 4), and thus do nota ff ect our conclusions regarding those elements. Determining upper limits of abundances can be particularly use-ful when lines become too weak to be accurately measured. Ac-tually, many lines used for abundance determination become par-ticularly weak at the metallicity of many GC spectra. The BAC-CHUS code is able to determine for each line an upper limit to -0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5-0.500.5 4000 4500 5000-0.500.5 4000 4500 5000M134000 4500 5000-0.500.5 4000 4500 5000M34000 4500 5000-0.500.5 4000 4500 5000M54000 4500 5000-0.500.5 4000 4500 5000M1074000 4500 5000-0.500.5 4000 4500 5000M71
Fig. 9.
Comparison Na, K and Ce abundances as a function of e ff ec-tive temperature for two studies (this work, first column, uncalibratedAPOGEE DR14, second column). Blue triangles are upper limits. Themore metal-poor clusters are not shown because those clusters containsmostly upper limits values. abundances by comparing the variance of the observed spectrumto the behaviour of the line strength in the synthetic spectra. Theupper limit corresponds to the abundance where the line inten-sity is comparable to the variance, and is consequently in BAC-CHUS sensitive to stellar parameters, as well as spectral reso-lution or signal-to-noise ratio. An illustration of the importanceof flagging and determining upper limits is given in Fig. 7. Inthis figure, we can observe that most of the upper limits for C,N and O elements are set for stars with e ff ective temperaturesabove 4600 K (100 K higher than what Mészáros et al. (2015)have used). This naturally occurs because the strength of themolecular lines used for those elements quickly weakens as tem-perature increases. However, while a flagging system is clearlymissing in the ASPCAP pipeline (García Pérez et al. 2016), andit is unclear to which extent the pipeline can actually measurethose elements. Furthermore, to avoid any bias in the interpreta-tion and discussion of our results, we will not show C, N and Oabundances for stars with T e ff > Figures 7, 8 and 9 show the resulting abundances as a functionof e ff ective temperature for three distinct analysis of the sameAPOGEE spectra: this work, the uncalibrated ASPCAP DR14results, and those from Mészáros et al. (2015). For clarity, we donot show all the sample in these diagrams, but rather select the Article number, page 7 of 17 & Aproofs: manuscript no. NdCeClusters
Fig. 10.
O abundances as a function of e ff ective temperature (upperpanel) and against Al abundances for three of our clusters (bottom pan-els) determined by the BACCHUS code with the APOGEE DR14 un-calibrated T e ff and logg (black dots) and by the ASPCAP pipeline (un-calibrated values, red triangles). The oxygen trend is slightly positive,at the opposite of Fig. 7. Moreover, almost no O dispersion nor anti-correlation is observed, which cast some doubts on the stellar parame-ters. most relevant stars or clusters to highlight any residual trends oro ff sets.Apart from the stars with upper limits, the abundances we deriveagree fairly well with previous studies. Nevertheless, in Fig. 7we notice that the N values derived by the ASPCAP pipeline forthe DR14 data do not reach values higher than 1.0. This is issueis due to the limits of the model grid, that will be enlarged for thenext 16th data release of SDSS. Nevertheless, we note that suchextreme values in N and C are only expected in specific casessuch as GCs stars (e.g. Mészáros et al. 2015) or peculiar N-rich stars (Fernández-Trincado et al. 2016; Schiavon et al. 2017;Fernández-Trincado et al. 2017, 2018). Mészáros et al. (2015)have extended the limit of [N / Fe] up to + / Fe] ≥ -0.75, which made the abundances of the most C-poorand N-rich stars unreliable.Furthermore, none of the measured elements seem to showa temperature dependence consistent with the results in litera-ture, except for C, N and O. While the C trend is probably re-lated to expected changes in the evolution of giants, N shouldrather increases with decreasing temperatures while O is notexpected to remain constant with temperature. Moreover, boththe Mészáros et al. (2015) and APOGEE DR14 results show aslight trend in [O / Fe] as a function of T e ff , but our is significantlystronger. Several hypothesis for such a trend can be invoked: element wavelength (Å air )C (CO) 15578.0 15775.5 15978.7 16185.5 16397.2 16614.016836.0 17063.0 17448.6 17456.0N (CN) 15119.0 15210.2 15222.0 15228.8 15242.5 15251.815309.0 15317.6 15363.5 15410.5 15447.0 15462.415466.2 15495.0 15514.0 15581.0 15636.5 15659.015825.7 15391.0 15569.0 15778.5 16052.9 16055.516650.0 16704.8 16714.5 16872.0 16909.4O (OH) 15391.0 15569.0 15778.5 16052.9 16055.5 16650.016704.8 16714.5 16872.0 16909.4Na I 16373.9 16388.8Mg I 15740.7 15749.0 15765.7Al I 16719.0 16750.6Si I 15361.2 15376.8 15557.8 15827.2 15833.6 15884.515960.1 16060.0 16094.8 16129.0 16163.7 16170.216215.7 16241.8 16680.8 16828.2K I 15163.0 15168.3Ca I 16136.8 16150.8 16155.2 16157.4 16197.1Fe I 15207.5 15294.6 15662.0Ce II 15277.6 15784.7 15829.8 15958.4 15977.1 16327.316376.4 16595.1 16522.5Nd II 15284.4 15368.1 15912.2 15977.9 16053.6 16262.016303.7 16382.9 16558.2 16634.6 Table 1.
Lines used in this work for abundance determination. – i) our chosen T e ff -color transformation maybe poorly calibrated for metal-poor GCs stars(González Hernández & Bonifacio 2009). Given theextreme sensitivity of O in the H-band as demonstrated byJönsson et al. (2018), the O trend may be a residual fromthis colour-T e ff approximation. However, our temperaturescale is similar to Mészáros et al. (2015), thus should lead toa similar trend which, is not the case – ii) we are using spherical stellar atmospheres, whereasAPOGEE DR14 and Mészáros et al. (2015) used plane par-allel models. The di ff erences basically lead to cooler layer inthe outer layers in the spherical case compared to the planeparallel, where molecular lines form. Consequently, the OHmolecule from which O is measured is stronger in the syn-thesis with spherical models and thus lead to lower O abun-dances. While this may explain the di ff erence between ourstudy and previous analysis, the fact that we are using animproved approximation for radiative transfer and model at-mosphere still does not explain why we obtain a trend. – iii) it has been demonstrated that 3D corrections af-fect the OH lines in the H-band at very low-metallicity(Dobrovolskas et al. 2015), such that O abundances are over-estimated in a 1D analysis. But this e ff ect goes towards theopposite direction to what is observed, where the coolest gi-ants with larger 3D e ff ects should have even lower O thanwhat is shown in Fig. 7 . – iv) last, we use model atmosphere with α elements composi-tion (including O, Mg, and Si), as well as Na and Al, fixed.But in globular clusters, O, Si and Mg are known to vary in-dependently. Moreover, Al and Na are also known to varyin GCs. Those elements are non-negligible electron donors,which could also a ff ect the atmospheric structure. We havetested the impact of changing the initial α elements contentby 0.5 dex, but no significant change has been obtain regard-ing the trend of O versus e ff ective temperature. Article number, page 8 of 17. Masseron et al.: BACCHUS and the 10 GCs
Finally, we have derived O abundances with the BACCHUScode adopting the calibrated temperatures and surface gravitiesin APOGEE DR14. In Fig. 10) we show these abundances as afunction of the APOGEE uncalibrated temperatures and againstAl. With the temperatures provided by APOGEE DR14, we canreproduced very well the APOGEE DR14 results and we do notobserve anymore any significant trend with T e ff as we do withour photometric temperatures. But those measurements do notexhibit any oxygen variations in clusters and the well-establishedAl-O anti-correlation. This is clearly in contradiction with allfindings from previous GCs cluster studies. Knowing that OH isvery temperature sensitive in the H band, we interpret such para-dox by the fact that the ASPCAP is biasing the e ff ective tem-peratures in order to obtain O abundances such that the built-inrelation [O / α ] = ff ect the precision of our abun-dances it does not a ff ect our conclusions. Fig. 11.
Na abundances as a function of O (black points). Althoughthere are only few stars in our sample where Na could be measured, theNa and O abundances follow the expected anti-correlation as measuredby Carretta et al. (2009c) (green crosses) in the optical.
4. Discussion
Final abundances, number of lines used, upper limits and randomerrors are reported in Table 2. For some stars, the Fe I lines weretoo weak to be detected and only upper limits could be obtainedfor the metallicity. For those stars, the literature metallicity val-ues from Harris (2010) have been adopted to build the plots. Allplots in the following section display random errors.
Na is very di ffi cult to measure in the APOGEE spectra of metal-poor stars because the Na I lines are very weak. Actually, thoselines appear only in the coolest stars of the most metal-rich clus-ters (hence mostly M 71 and M 107 and very few in M 5, M 3 andM 13). Given that O is also di ffi cult to measure for the warmeststars, our version of the O-Na anti-correlation diagram (Fig. 11)appears quite unpopulated. Nevertheless, the values are con-sistent with very high resolution optical studies (Carretta et al.2009c) and thus confirm the high quality of the APOGEE dataand of our analysis.However, the H band contains very clear Al I lines that aremore easily measurable than the Na I lines. Therefore, Al rep-resents a more robust abundance indicator in the APOGEE datafor the study of the multiple stellar population phenomenon inGCs, as first presented by Mészáros et al. (2015). Indeed, in Fig.12 we demonstrate that we can observe the Al-O anti-correlationfor the ten clusters separately, except for M 107 and M 71. Giventhat those two latter GCs are the most metal-rich ones of thesample, this may indicate that the temperature conditions in thepolluters of those two clusters are too low to e ffi ciently produceAl, as illustrated in the case of massive AGB stars polluters byVentura et al. (2016) and Dell’Agli et al. (2018). In Fig. 13, we establish the Mg-Al anti-correlation for all clus-ters as already found in literature (except M 107 and M 71 as al-ready noted by Mészáros et al. 2015). This implies that the Mg-Al chain is active in almost all the clusters (including M5) incontrast to the conclusions of Carretta et al. (2009b). An interest-ing aspect shown in these data when scrutinizing those diagrams,is that stars with [Mg / Fe] < / Fe] abundances. We stress that, re-garding any possible analysis bias, we have not been able tofind any dependence on e ff ective temperature nor on evolution-ary status (RGB or eAGB / RHB). Moreover, in Fig. 15 we showthe spectra around the two Al I lines present in the APOGEEspectra. This figure unambiguously demonstrates that the Al Ilines are weaker in the most Mg-depleted stars. Thus, we con-clude that the relatively low Al in such extremely Mg-depletedstars in M 15 and M 92 is real.The Al-Si plane can be seen in Fig. 14. Overall Si is con-stant and consistent with field star value for similar metallici-ties. But three clusters (M15, M 92 and M 13) show a signifi-cant Si enhancement for the Al richest stars. Yong et al. (2005),Carretta et al. (2009b) and Mészáros et al. (2015) interpreted theAl-Si correlation as a signature of Si leakage from the Mg-Al chain. Interestingly, most of the Si enriched stars in M 15and M 92 also seem to correspond to the extreme Mg-depletedand midly enhanced Al stars mentioned above. According toPrantzos et al. (2017) during H-burning processes, Si begins to
Article number, page 9 of 17 & Aproofs: manuscript no. NdCeClusters [O/Fe]-0.500.511.5 -1 -0.5 0 0.5 1-0.500.511.5 -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
Fig. 12.
Al abundances as a function of O for all GCs in our sample. Black dots are measurements and blue triangles are upper limits in O. Theanti-correlation between those elements can be clearly seen for all clusters individually, except M 107 and M 71. [Mg/Fe]-0.500.511.5 -0.5 0 0.5-0.500.511.5 -0.5 0 0.5 -0.5 0 0.5 -0.5 0 0.5 -0.5 0 0.5
Fig. 13.
Al abundances as a function of Mg abundances for all GCs in our sample. The green points show the Si- and Al-enhanced stars fromFig. 14. Magenta points highlight the extreme Mg-depleted stars. These stars also show a relatively lower Al abundance. be produced above 80MK. According to the same authors Al isexpected to be progressively produced up to ∼
80 MK but beginsto be depleted above 80 MK. Therefore, temperatures in pol- luters above 80 MK would explain satisfactorily the existence ofMg- and Al-weak (and Si enhanced) stars in M 15 and M 92.
Article number, page 10 of 17. Masseron et al.: BACCHUS and the 10 GCs [Si/Fe]-0.500.511.5 0 0.2 0.4 0.6 0.8 1-0.500.511.5 0 0.2 0.4 0.6 0.8 10 0.2 0.4 0.6 0.8 10 0.2 0.4 0.6 0.8 10 0.2 0.4 0.6 0.8 1
Fig. 14.
Al abundances as a function of Si abundances for all GCs in our sample. The magenta points highlight the very low Mg and Al-weak starsfrom Fig. 13. The green points show the Si- and Al-enhanced stars. There is generally no correlation and Si is very homogeneous except in M 15,M 92 and possibly M 13.
Fig. 15.
Al I lines in M15 stars.The green lines show the Si-enhancedstars from Fig. 14, and the magenta lines show the Mg- and Al-weakstars from Fig. 13.
Interestingly, M 13 also shows at least as high Al enhance-ments as M 15 and M 92 ([Al / Fe] > / depletion yields of the Al source.Finally, if we now consider the whole sequence formedby the Mg-depleted stars with the more standard Mg-Al anti- correlation, its appearance is similar to a hook, because the mostMg-depleted stars have smaller than expected Al abundances.Although our data is certainly one of the most extensive spec-troscopic studies of those clusters, some data may still miss, andthe di ffi culty of analyzing such low metallicity stars can possiblyresult in biases in the trends. On the other hand, if the most Mg-poor stars do indeed have lower than expected Al abundances,this would imply that standard correlation and anti-correlationpatterns observed in GCs are in fact more complex than previ-ously thought. This is well in line with the recent partitioning ofthe clusters photometric map (the so-called chromosome map)into various sub-populations by Milone et al. (2017) and in par-ticular the discovery of the puzzling extension of the first popu-lation by Lardo et al. (2018). In Fig. 16, the C and N anti-correlation can be observed for mostof the clusters except the most metal-poor ones where CO andCN lines are too weak to be detected. We first wondered aboutthe influence of the intrinsic red giant branch extra-mixing overthe C and N data. Indeed, the existence of the extra-mixing andits C and N signature is clearly seen in the APOGEE field stars(Masseron et al. 2017; Shetrone 2018) and can have a very largeimpact on the C and N yields at very low metallicity. We plot inFig. 17 the C / N ratios over the e ff ective temperature and com-pare to the model expectations using a prescription for the extra-mixing from Lagarde et al. (2012). The model predicts a signif-icant drop of the C / N ratio after the bump luminosity around4700 K. However, it is di ffi cult to evaluate the impact of the Article number, page 11 of 17 & Aproofs: manuscript no. NdCeClusters [C/Fe]00.511.52 -2 -1 000.511.52 -2 -1 0 -2 -1 0 -2 -1 0 -2 -1 0
Fig. 16.
N abundances as a function of C abundances for all GCs in our sample. Black dots are measurements and blue triangles are upper limitsin C. The N-C anti-correlation is clearly observed for the most metal-rich clusters (bottom panels). extra-mixing on the C and N yields by comparing the C and Nabundances before and after the luminosity bump because the[C / N] data show a very large scatter. Indeed, it is very likelythat the extra-mixing signature depends also on the initial abun-dances of C and N, which have been proven to vary greatly inglobular clusters from the observation of unevolved stars (seeGratton et al. 2004, and references therein). Consequently, westress that only field stars represent reliable test-beds for study-ing extra-mixing along the RGB but no GCs stars should be usedto do so, in contrast to Lagarde et al. (2018)’s work.However, extra-mixing is believed to induce CN-cycling, i.e.that extra-mixing is a ff ecting C and N yields but not a ff ecting theC + N + O yields. By looking at the C + N + O yields we can ver-ify if the C + N + O is varying in a cluster, and thus constrain thenucleosynthesis of the cluster polluters independently of extra-mixing e ff ect. We have checked the C + N + O values in our clus-ters. Compared to Mészáros et al. (2015), it appears quite clearnow that no correlation exists between C + N + O and Al for atleast three clusters (M 13, M 3 and M 5). Furthermore, theC + N + O data are consistent with no variations within the errorsand are also consistent with field stars values of similar metal-licities (e.g. Gratton et al. 2000). Therefore, the CNO cycle isalso occurring in the GCs polluters. Moreover, we do not findsignificant enhancement in any of our clusters (including thosewith upper limits) compared to the value of field stars such asit has been claimed by Yong et al. (2009, 2015) in NGC 1851,although disputed by Villanova et al. (2010).
Teff-4-3-2-10 5000 4500 4000-4-3-2-10 5000 4500 4000 5000 4500 4000
Fig. 17. [C / N] ratios as a function of e ff ective temperature for the mostmetal-rich GCs of our sample. RHB / eAGB stars are in magenta, whileRGB stars are in green. The data are compared with a Lagarde et al.(2012) 0.85M ⊙ Z = / N to match approximately the post-firstdredge up C / N data value.
In the previous sections, we have confirmed that Al is (anti-) cor-related with many elements (C, N, O, Na, Mg and Si). Given thecompleteness of the Al measurements in our sample as well asits large variations, we consider this element as the best repre-sentation to evaluate the extent of the cluster’s multiple popula-
Article number, page 12 of 17. Masseron et al.: BACCHUS and the 10 GCs
Fig. 18.
The spread in Al as a function of global cluster parame-ters (metallicity, absolute magnitude and total masses). M 15, M 92and NGC 5466 have probably underestimated spread and have beenhighlighted accordingly. The metallicity and the absolute magnitudesare extracted from Harris (2010). The total masses are computed byMcLaughlin & van der Marel (2005) (black line) and from a compi-lation by Boyles et al. (2011) (red line). There is probably an anti-correlation with metallicty and maybe a correlation with absolute mag-nitude. tions. As already suggested by Carretta et al. (2009b) (but seealso Mészáros et al. 2015; Ventura et al. 2016; Dell’Agli et al.2018), Al spread within each cluster decreases with increasingcluster metallicity. Thanks to our large sample, we can now attempt to compare quantitatively this spread against clustersmetallicities as well as other global properties. In Fig.18 , weplot the Al spread (derived from the deviation from the median)against cluster metallicity, absolute magnitude and mass. We re-mind that M 15 and M 92 probably have lower than expected Alabundance in the most extreme cases. Therefore, the Al spreadmay be underestimated for those clusters. We also remark thatNGC 5466 has a quite low number of stars, thus we also prob-ably under-evaluate the real Al spread. The diagrams presentedFig. 18 suggest an anti-correlation between the Al spread andthe metallicity, as well as possible correlation between the Alspread and absolute magnitude. From a nucleosynthetic point ofview, this potentially implies that the Mg-Al chain reaction isbecoming less important with increasing metallicity and / or clus-ter luminosity. Moreover, although cluster absolute luminosity isknown to be a proxy for cluster mass, we found that the correla-tion of the Al spread with cluster mass is quite uncertain, mostlybecause cluster mass determination is model dependent.Similarly, Carretta et al. (2009c) found a bilinear anti-correlationbetween their Na max ( ∝ Na spread) and clusters luminositiesand metallicities. This is consistent with our results because Aland Na have been demonstrated to be correlated (Carretta et al.2009b). But this is contrasting with Milone et al. (2017) whorather found a correlation between metallicity - as well as clus-ter magnitude - and the width of the RGB. This apparent con-tradiction could be explained by the fact that the RGB width ina colour-magnitude diagram is known to be sensitive to N be-cause of molecular bands (but not to Al which has no molecularbands). Thus, we can deduce that in the Milone et al. (2017)’swork, it is the N spread which is increasing with metallicity.But this phenomenon is di ffi cult to be confirmed by our data (Fig. 19) because N is measured in only a fraction of the stars ofour sample, statistically weakening the possibility of measuringaccurately the N spread. Actually, in all the clusters we couldmeasure N in a large enough number of stars (M 13, M 3, M 5,M 107 and M 71) the N spread seems rather constant ( ∼ Although the exact temperature for the onset of the K produc-tion by H-burning nucleosynthesis may be still debated (120-180 MK Ventura et al. 2012; Iliadis et al. 2016), the tempera-ture must be much higher than the one for the onset of Al orSi production( ∼
80 MK). We certainly observe Si production inM 15 and M 92 (Fig. 14), indicating that in these clusters the pol-luters had reached the highest temperatures. But neither in thoseextreme clusters -nor in the others- is a variation in K observed(Fig.20). We conclude that the K measured in those clusters isconsistent with no K production, thus relatively low-temperaturenucleosynthesis conditions (but still high enough to produce Si),and that K production in GCs is rare, in agreement with the con-clusions of Takeda et al. (2009) and Carretta et al. (2013). Still,it has been clearly established by Cohen & Kirby (2012) andMucciarelli et al. (2015a) that NGC 2808 and NGC 2419 showK enhancement correlated with notably Al. It is remarkable thatthese two clusters have metallicities in the same range as oursample. Therefore, in contrast to Si production (Sec.4.2) or Alspread (Sec.4.4), K production in GCs is not only a functionof metallicity. In any case, the absolute luminosities of thoseK-enhanced clusters are among the largest. This may corrobo-rate the idea that the clusters peculiar chemistry is not only afunction of metallicity but also a function of cluster luminosity(Carretta et al. 2009b; Milone et al. 2017)
Article number, page 13 of 17 & Aproofs: manuscript no. NdCeClusters [N/Fe]-0.500.511.5 0 0.5 1 1.5 2-0.500.511.5 0 0.5 1 1.5 2 0 0.5 1 1.5 2 0 0.5 1 1.5 2 0 0.5 1 1.5 2
Fig. 19.
Al abundances as a function of N abundances for all the clusters sorted by metallicity. Blue triangles are upper limits in N. Al and N arecorrelated, but the Al spread is decreasing with increasing metallicity. [K/Fe]-0.500.511.5 -1 0 1 2-0.500.511.5 -1 0 1 2-1 0 1 2-1 0 1 2-1 0 1 2
Fig. 20.
Al abundances as a function of K abundances for all GCs inour sample. Blue triangles are upper limits in K. While K is di ffi cult tomeasure in very low metallicity GCs, there is no significant spread in theremaining ones. Note that M 13 does not contain measurements becausethe K lines at the cluster doppler shift fall in one of the APOGEE CCDgap where no measurement can be made. Marino et al. (2009) observed a Ca spread in M 22. But Ca is notexpected to be a ff ected by H-burning processes (Prantzos et al.2017). Fig. 21 shows the Ca abundances measured in our 10GCs. It is clear in that figure that no correlation with Al isobserved. The star-to-star scatter is also small enough to show [Ca/Fe]-0.500.511.5 0 0.5-0.500.511.5 0 0.5 0 0.5 0 0.5 0 0.5 Fig. 21.
Al abundances as a function of Ca abundances for all GCs inour sample. Blue triangles are upper limits in Ca. While Ca is di ffi cultto measure in very low metallicity GCs, there is no significant spread inany of the clusters. no enhancement in any of the clusters. Therefore, M 22 re-mains certainly an exceptional cluster regarding Ca enhance-ment, although Mucciarelli et al. (2015b) suspect that NLTE ef-fects could be also responsible for such an observational spread. Article number, page 14 of 17. Masseron et al.: BACCHUS and the 10 GCs
Fig. 22.
Al abundances as a function of Ce abundances for all GCs in oursample. Black dots are measurements and blue triangles are upper limitsin Ce. Except for M 15 (and perhaps M 92), all clusters are consistentwith a homogeneous Ce abundance.
Fig. 23.
Nd abundances as a function of Ce abundances for all GCs inour sample. Although there are few measurements of Nd, they correlatewell with Ce.
In Fig. 22, we confirm that most GCs do not seem to show asignificant spread in Ce nor show any enhancement comparedto field stars value as already remarked by Gratton et al. (2004).In contrast, M 15 shows a significant spread. Nd is di ffi cult tomeasure and we obtain some constraining values only for themost favorable spectra, i.e. in the coolest giants of the mostmetal-rich GCs. We still observe a consistent correlation with Ce(Fig. 23). Unfortunately, with only one neutron-capture elementmeasured in M 15, it is not possible to assign the nucleosyn-thetic origins in those stars, and notably disentangle a r-processand a s-process nucleosynthesis such as discussed in M 22 andM 2 (Roederer & Sneden 2011; Yong et al. 2014). Therefore,our conclusions concerning M 15 are bound to Sobeck et al.(2011), i.e. that the Ce we have measured has probably a purer-process origin. We also agree that the Ce dispersion in M 15is comparable to that of the halo field at the same metallicity,although we have now clearly established that M 15 also shows large Al variation. But we hardly see a correlation with Al, con-firming that polluters in our GCs do not produce Ce and is prob-ably from the GCs primordial formation gas.
5. Conclusions
We confirm that APOGEE spectra provide precise elementalabundances for many elements. However, we show that the latestdata release (DR14) still su ff ers from uncertainties, in particularregarding extreme abundances such as those found in some casesfor C and N, as already pointed out by Jönsson et al. (2018), butmore generally for very low metallicity spectra ([Fe / H] < -1.5) inwhich metallic lines become very weak and are extremely sen-sitive to choice of prescription for parameters like macroturbu-lence and NLTE e ff ects. We emphasize the crucial lack of anupper limits flagging system fro APOGEE. Finally, we demon-strate that the ASPCAP DR14 T e ff are probably biased whenever[O / α ] are not solar.With our independent analysis, we measure almost all theelemental abundances needed for GCs studies, (newly-includedthe neutron-capture elements Ce and Nd in this survey), ex-cept He and Na. Although those latter two elements abundanceswould certainly be very interesting to be measured as well, wedemonstrate that all known light elements anti- / correlation canbe recovered with a high precision. Consequently, we confirmthat H-burning reactions are the main nucleosynthesis processesthat have occurred in all the clusters in this analysis.Moreover, along with literature, we have collected somecorroborating evidences suggesting that cluster luminosity andmetallicity are the two main parameters which are driving thevarious GCs chemical patterns. Unfortunately, our sample ofGCs is such that we can hardly disentangle what is the main fac-tor controlling the amplitude of the pollution between metallicityand cluster magnitude. Indeed, in our sample both are correlated,so that the most metal-rich clusters are also the less luminous,and conversely, the most metal-poor clusters are, on average, thebrightest ones. To validate those main dependencies, it will beinteresting to analyze more peculiar clusters such as NGC 2808,NGC 2419, M 2, M 4 M 22, NGC 1851, ω Cen or any youngmassive stellar clusters planned to be observed by the ongoingSDSS IV / APOGEE-2 survey.Furthermore, we discovered some puzzling stars extremelydepleted in Mg and weakly enhanced in Al that seem to occuronly in our most metal-poor clusters, suggesting that the temper-ature conditions reached in the corresponding polluters are highenough to start burning Al. Finally, in the same Mg-Al plane, wehave observed that the data are forming a hook. Any model thatis attempting to explain the multiple populations in GCs mustnow be able to self-consistently account for such a turnover inthe Mg-Al anti-correlation as seen in these new detailed and ex-tensive observations.
Acknowledgements.
We are very grateful for the fruitful comments from P. Ven-tura and the help from S. Villanova. T.M. acknowledges support from Span-ish Ministry of Economy and Competitiveness (MINECO) under the 2015Severo Ochoa Program SEV-2015-0548. T.M., D.A.G.H., O.Z., and F.D.A.also acknowledge support by the MINECO under grant AYA-2017-88254-P.SzM has been supported by the Premium Postdoctoral Research Program ofthe Hungarian Academy of Sciences, and by the Hungarian NKFI Grants K-119517 of the Hungarian National Research, Development and Innovation Of-fice. H. J. acknowledges support from the Crafoord Foundation, Stiftelsen OlleEngkvist Byggmästare, and Ruth och Nils-Erik Stenbäcks stiftelse. D.G. ac-knowledges support from the Chilean Centro de Excelencia en Astrofísica yTecnologías Afines (CATA) BASAL grant AFB-170002. D.G. also acknowl-edges financial support from the Dirección de Investigación y Desarrollo dela Universidad de La Serena through the Programa de Incentivo a la Inves-tigación de Académicos (PIA-DIDULS). T.C.B. acknowledges partial support
Article number, page 15 of 17 & Aproofs: manuscript no. NdCeClusters from grant PHY 14-30152; Physics Frontier Center / JINA Center for the Evolu-tion of the Elements (JINA-CEE), awarded by the US National Science Foun-dation. This paper made use of the IAC Supercomputing facility HTCondor(http: // research.cs.wisc.edu / htcondor / ), partly financed by the Ministry of Econ-omy and Competitiveness with FEDER funds, code IACA13-3E-2493. Fundingfor the Sloan Digital Sky Survey IV has been provided by the Alfred P. SloanFoundation, the U.S. Department of Energy O ffi / University of Tokyo, the Korean ParticipationGroup, Lawrence Berkeley National Laboratory, Leibniz Institut für AstrophysikPotsdam (AIP), Max-Planck-Institut für Astronomie (MPIA Heidelberg), Max-Planck-Institut für Astrophysik (MPA Garching), Max-Planck-Institut für Ex-traterrestrische Physik (MPE), National Astronomical Observatories of China,New Mexico State University, New York University, University of Notre Dame,Observatório Nacional / MCTI, The Ohio State University, Pennsylvania StateUniversity, Shanghai Astronomical Observatory, United Kingdom ParticipationGroup, Universidad Nacional Autónoma de México, University of Arizona, Uni-versity of colourado Boulder, University of Oxford, University of Portsmouth,University of Utah, University of Virginia, University of Washington, Universityof Wisconsin, Vanderbilt University, and Yale University.
References
Abolfathi, B., Aguado, D. S., Aguilar, G., et al. 2018, ApJS, 235, 42Ahn, C. P., Alexandro ff , R., Allende Prieto, C., et al. 2014, ApJS, 211, 17Bastian, N. & Lardo, C. 2017, ArXiv e-prints [ arXiv:1712.01286 ]Bertelli, G., Girardi, L., Marigo, P., & Nasi, E. 2008, A&A, 484, 815Blanton, M. R., Bershady, M. A., Abolfathi, B., et al. 2017, AJ, 154, 28Boyles, J., Lorimer, D. R., Turk, P. J., et al. 2011, ApJ, 742, 51Brooke, J. S. A., Bernath, P. F., Western, C. M., et al. 2016, J. Quant. Spec. Ra-diat. Transf., 168, 142Carretta, E., Bragaglia, A., Gratton, R., D’Orazi, V., & Lucatello, S. 2009a,A&A, 508, 695Carretta, E., Bragaglia, A., Gratton, R., & Lucatello, S. 2009b, A&A, 505, 139Carretta, E., Bragaglia, A., Gratton, R. G., et al. 2009c, A&A, 505, 117Carretta, E., Gratton, R. G., Bragaglia, A., et al. 2013, ApJ, 769, 40Charbonnel, C. 2016, in EAS Publications Series, Vol. 80, EAS PublicationsSeries, ed. E. Moraux, Y. Lebreton, & C. Charbonnel, 177–226Cohen, J. G. & Kirby, E. N. 2012, ApJ, 760, 86Cunha, K., Smith, V. V., Hasselquist, S., et al. 2017, ApJ, 844, 145D’Antona, F. & Ventura, P. 2007, MNRAS, 379, 1431Dell’Agli, F., García-Hernández, D. A., Ventura, P., et al. 2018, MNRAS, 475,3098Di Criscienzo, M., Ventura, P., D’Antona, F., Dell’Agli, F., & Tailo, M. 2018,MNRAS, 479, 5325Dobrovolskas, V., Kuˇcinskas, A., Bonifacio, P., et al. 2015, A&A, 576, A128Fernández-Trincado, J. G., Robin, A. C., Moreno, E., et al. 2016, ApJ, 833, 132Fernández-Trincado, J. G., Zamora, O., García-Hernández, D. A., et al. 2017,ApJ, 846, L2Fernández-Trincado, J. G., Zamora, O., Souto, D., et al. 2018, ArXiv e-prints[ arXiv:1801.07136 ]García-Hernández, D. A., Mészáros, S., Monelli, M., et al. 2015, ApJ, 815, L4García Pérez, A. E., Allende Prieto, C., Holtzman, J. A., et al. 2016, AJ, 151,144Gilmore, G., Randich, S., Asplund, M., et al. 2012, The Messenger, 147, 25González Hernández, J. I. & Bonifacio, P. 2009, A&A, 497, 497Gordon, I. E., Rothman, L. S., Hill, C., et al. 2017, J. Quant. Spec. Ra-diat. Transf., 203, 3Gratton, R., Sneden, C., & Carretta, E. 2004, ARA&A, 42, 385Gratton, R. G., Carretta, E., & Bragaglia, A. 2012, A&A Rev., 20, 50Gratton, R. G., Sneden, C., Carretta, E., & Bragaglia, A. 2000, A&A, 354, 169Gray, D. F. 2008, The Observation and Analysis of Stellar PhotospheresGunn, J. E., Siegmund, W. A., Mannery, E. J., et al. 2006, AJ, 131, 2332Gustafsson, B., Edvardsson, B., Eriksson, K., et al. 2008, A&A, 486, 951Harris, W. E. 2010, ArXiv e-prints [ arXiv:1012.3224 ]Hasselquist, S., Shetrone, M., Cunha, K., et al. 2016, ApJ, 833, 81Hekker, S. & Meléndez, J. 2007, A&A, 475, 1003Holtzman, J. A., Hasselquist, S., Shetrone, M., et al. 2018, AJ, 156, 125Holtzman, J. A., Shetrone, M., Johnson, J. A., et al. 2015, AJ, 150, 148 Iliadis, C., Karakas, A. I., Prantzos, N., Lattanzio, J. C., & Doherty, C. L. 2016,ApJ, 818, 98Jönsson, H., Allende Prieto, C., Holtzman, J. A., et al. 2018, AJ, 156, 126Kraft, R. P. 1994, PASP, 106, 553Lagarde, N., Decressin, T., Charbonnel, C., et al. 2012, A&A, 543, A108Lagarde, N., Reylé, C., Robin, A. C., et al. 2018, ArXiv e-prints[ arXiv:1806.01868 ]Lardo, C., Salaris, M., Bastian, N., et al. 2018, A&A, 616, A168Li, G., Gordon, I. E., Rothman, L. S., et al. 2015, ApJS, 216, 15Majewski, S. R., Schiavon, R. P., Frinchaboy, P. M., et al. 2017, AJ, 154, 94Marigo, P., Girardi, L., Bressan, A., et al. 2017, ApJ, 835, 77Marino, A. F., Milone, A. P., Piotto, G., et al. 2009, A&A, 505, 1099Marino, A. F., Yong, D., Milone, A. P., et al. 2018, ApJ, 859, 81Masseron, T. & Gilmore, G. 2015, MNRAS, 453, 1855Masseron, T., Lagarde, N., Miglio, A., Elsworth, Y., & Gilmore, G. 2017, MN-RAS, 464, 3021Masseron, T., Merle, T., & Hawkins, K. 2016, BACCHUS: Brussels AutomaticCode for Characterizing High accUracy Spectra, Astrophysics Source CodeLibraryMcLaughlin, D. E. & van der Marel, R. P. 2005, ApJS, 161, 304Mészáros, S., García-Hernández, D. A., Cassisi, S., et al. 2018, MNRAS, 475,1633Mészáros, S., Martell, S. L., Shetrone, M., et al. 2015, AJ, 149, 153Milone, A. P., Marino, A. F., Piotto, G., et al. 2015, ApJ, 808, 51Milone, A. P., Marino, A. F., Renzini, A., et al. 2018, MNRAS, 481, 5098Milone, A. P., Piotto, G., Renzini, A., et al. 2017, MNRAS, 464, 3636Monelli, M., Milone, A. P., Stetson, P. B., et al. 2013, MNRAS, 431, 2126Mucciarelli, A., Bellazzini, M., Merle, T., et al. 2015a, ApJ, 801, 68Mucciarelli, A., Lapenna, E., Massari, D., et al. 2015b, ApJ, 809, 128Ness, M., Hogg, D. W., Rix, H.-W., Ho, A. Y. Q., & Zasowski, G. 2015, ApJ,808, 16Nidever, D. L., Holtzman, J. A., Allende Prieto, C., et al. 2015, AJ, 150, 173Norris, J. E. & Da Costa, G. S. 1995, ApJ, 447, 680Pancino, E., Romano, D., Tang, B., et al. 2017, A&A, 601, A112Plez, B. 2012, Turbospectrum: Code for spectral synthesis, astrophysics SourceCode LibraryPrantzos, N., Charbonnel, C., & Iliadis, C. 2017, A&A, 608, A28Renzini, A., D’Antona, F., Cassisi, S., et al. 2015, MNRAS, 454, 4197Roederer, I. U. & Sneden, C. 2011, AJ, 142, 22Schiavon, R. P., Zamora, O., Carrera, R., et al. 2017, MNRAS, 465, 501Shetrone, M. 2018, in prep.Shetrone, M., Bizyaev, D., Lawler, J. E., et al. 2015, ApJS, 221, 24Shingles, L. J., Karakas, A. I., Hirschi, R., et al. 2014, ApJ, 795, 34Smiljanic, R., Korn, A. J., Bergemann, M., et al. 2014, A&A, 570, A122Smith, G. H. 1987, PASP, 99, 67Sneden, C., Lucatello, S., Ram, R. S., Brooke, J. S. A., & Bernath, P. 2014, ApJS,214, 26Sobeck, J. S., Kraft, R. P., Sneden, C., et al. 2011, AJ, 141, 175Szécsi, D. & Wünsch, R. 2018, ArXiv e-prints [ arXiv:1809.01395 ]Takeda, Y., Kaneko, H., Matsumoto, N., et al. 2009, PASJ, 61, 563Ventura, P., D’Antona, F., Di Criscienzo, M., et al. 2012, ApJ, 761, L30Ventura, P., García-Hernández, D. A., Dell’Agli, F., et al. 2016, ApJ, 831, L17Villanova, S., Geisler, D., & Piotto, G. 2010, ApJ, 722, L18Yong, D., Grundahl, F., D’Antona, F., et al. 2009, ApJ, 695, L62Yong, D., Grundahl, F., Nissen, P. E., Jensen, H. R., & Lambert, D. L. 2005,A&A, 438, 875Yong, D., Grundahl, F., & Norris, J. E. 2015, MNRAS, 446, 3319Yong, D., Roederer, I. U., Grundahl, F., et al. 2014, MNRAS, 441, 3396Zasowski, G., Cohen, R. E., Chojnowski, S. D., et al. 2017, AJ, 154, 198 Article number, page 16 of 17. Masseron et al.: BACCHUS and the 10 GCs star cluster status * Te ff logg [Fe / H] σ [C / Fe] σ / Fe] σ · · · + · · · · · · · · · · · · · · · · · · · · · + < · · · · · · + · · · · · · · · · · · · · · · + < · · · · · · + · · · · · · · · · · · · · · · · · · · · · + < · · · · · · + < · · · < · · · · · · + · · · · · · · · · · · · · · · + · · · · · · · · · · · · · · · · · · · · · + · · · · · · · · · · · · · · · · · · · · · · · · < · · · · · · < · · · · · · < · · · · · · < · · · · · · · · · · · · < -0.057 · · · < · · · · · ·· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · Notes. * eAGB are distinguished from HB stars such that T e ff < Table 2.
Parameters and abundances for the sample stars.