ebook img

Carbon in Red Giants in Globular Clusters and Dwarf Spheroidal Galaxies PDF

1.5 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Carbon in Red Giants in Globular Clusters and Dwarf Spheroidal Galaxies

Accepted to ApJ on 2015Jan27 PreprinttypesetusingLATEXstyleemulateapjv.5/2/11 CARBON IN RED GIANTS IN GLOBULAR CLUSTERS AND DWARF SPHEROIDAL GALAXIES* Evan N. Kirby1, Michelle Guo2,3, Andrew J. Zhang4, Michelle Deng5, Judith G. Cohen1, Puragra Guhathakurta6, Matthew D. Shetrone7, Young Sun Lee8, Luca Rizzi9 Accepted to ApJ on 2015 Jan 27 ABSTRACT 5 We present carbon abundances of red giants in Milky Way globular clusters and dwarf spheroidal 1 galaxies (dSphs). Our sample includes measurements of carbon abundances for 154 giants in the 0 clusters NGC 2419, M68, and M15 and 398 giants in the dSphs Sculptor, Fornax, Ursa Minor, and 2 Draco. ThissampledoublesthenumberofdSphstarswithmeasurementsof[C/Fe]. The[C/Fe]ratio b in the clusters decreases with increasing luminosity above log(L/L⊙) ≃ 1.6, which can be explained e by deep mixing in evolved giants. The same decrease is observed in dSphs, but the initial [C/Fe] of F the dSph giants is not uniform. Stars in dSphs at lower metallicities have larger [C/Fe] ratios. We 5 hypothesize that [C/Fe] (corrected to the initial carbon abundance) declines with increasing [Fe/H] 2 due to the metallicity dependence of the carbon yield of asymptotic giant branch stars and due to the increasing importance of Type Ia supernovae at higher metallicities. We also identified 11 very ] carbon-rich giants (8 previously known) in three dSphs. However, our selection biases preclude a A detailed comparison to the carbon-enhanced fraction of the Milky Way stellar halo. Nonetheless, the G stars with [C/Fe] < +1 in dSphs follow a different [C/Fe] track with [Fe/H] than the halo stars. Specifically, [C/Fe] in dSphs begins to decline at lower [Fe/H] than in the halo. The difference in the . h metallicity of the [C/Fe] “knee” adds to the evidence from [α/Fe] distributions that the progenitors p of the halo had a shorter timescale for chemical enrichment than the surviving dSphs. - Subjectheadings: galaxies: dwarf—LocalGroup—galaxies: abundances—stars: evolution—stars: o r abundances t s a 1. INTRODUCTION element in the photosphere of a star relative to the rest [ of the star. Nuclear burning at the base of a convec- Variationsofelementalabundanceswithinastellarsys- 2 tive zone that reachesthe photosphere canalso decrease temareevidenceforphysicalprocessesbothinternaland v or increase the observable abundances of light elements, external to individual stars. Chemical evolution on a 8 likeLi(Lind et al.2011),C,N,O,andevens-processel- galactic scale leads to a dispersion in metallicity. As 0 ements,likeBaandRb(Liang et al.2000;Karakas et al. progressive generations of supernovae enrich the inter- 9 2012). This process is called astration when nuclear stellar medium of a galaxy, stars that form later have 6 burning leads to the destruction of an element. higher metallicity than their older predecessors (Tinsley 0 Carbon exhibits abundance variations within a stellar . 1980). Onthe otherhand, variationsin certainlightele- 1 ments can arise from internal stellar evolution. Gravita- populationdue to astrationand galactic chemicalevolu- 0 tion. As a star ascends the red giant branch (RGB), tionalsettling andradiativelevitation(Behr et al.1999) 5 it encounters episodes of mixing that bring material deplete and enhance, respectively, the abundance of an 1 processed through the CNO cycle to the photosphere. : Within 2–3 magnitudes of the tip of the RGB, C and v *ThedatapresentedhereinwereobtainedattheW.M.Keck O are depleted and N is enhanced relative to stars on i X Observatory,whichisoperatedasascientificpartnershipamong the main sequence or the lower RGB. Proposed mech- the California Institute of Technology, the University of Cali- anisms for mixing deep CNO-processed material into r forniaandtheNational AeronauticsandSpaceAdministration. a The Observatory was made possible by the generous financial the upper atmosphere include Rayleigh-Taylor instabil- supportoftheW.M.KeckFoundation. ity (Eggleton et al. 2006) or, more likely, thermohaline 1CaliforniaInstituteofTechnology, 1200E.CaliforniaBlvd., mixing (Charbonnel & Zahn 2007). MC249-17,Pasadena,CA91125,USA 2Irvington High School, 41800 Blacow Rd., Fremont, CA Globularclusters(GCs)areexcellentsitestostudythe 94538, USA evolution of carbon. It is relatively easy to test whether 3Stanford University, 450 Serra Mall, Stanford, CA 94305, variationsincarbonabundanceswithinaclusterarepri- USA 4TheHarkerSchool,500SaratogaAve.,SanJose,CA95129, mordialormodifiedbystellarevolutionaryprocessesdur- USA ingthelifetimeoftheclusterbecauseGCsareessentially 5Harvard University, Massachusetts Hall, Cambridge, MA single-age, mono-metallic populations at a uniform dis- 02138, USA tance. Itisnotnecessarytocontrolforageormetallicity. 6UCO/Lick Observatory and Department of Astronomyand Carbonshouldnotbeastratedinmetal-poorstarsonthe Astrophysics, University of California, 1156 High St., Santa Cruz,CA95064,USA main sequence and subgiant branch, which do not have 7McDonaldObservatory,UniversityofTexasatAustin,HC75 convective envelopes deep enough to dredge up CNO- Box1337-MCD,FortDavis,TX79734, USA processed material. However, GCs do show primordial 8Chungnam National University, Department of Astronomy abundance variations. Cohen (1999) and Briley et al. andSpaceScience,Daejeon305-764,RepublicofKorea 9Keck Observatory, 65-1120 Mamalahoa Hwy, Kamuela, HI (2002) found variations in [C/Fe] on the main sequences 96743, USA ofM71andM13,respectively. Becausethephotospheres 2 Kirby et al. of main sequence stars are not affected by convection bon abundances in 18 stars in Draco and Ursa Minor that can reach CNO-processed material, these carbon from the G band of CH, and Sku´lado´ttir et al. (2015) variations must be primordial. Regardless, astration on measured C abundances for about 80 Sculptor giants the upper RGB causes the most severe variations (e.g., from a near-infrared CN band. Recently, ultra-faint Suntzeff 1981; Smith et al. 1996; Smith & Briley 2005, galaxies have received a lot of attention, including car- 2006). Martell et al. (2008c) showedthat the rate of de- bon abundance measurements in Segue 1 (Norris et al. pletion depends on metallicity, as predicted by thermo- 2010; Frebel et al. 2014), Ursa Major II and Coma haline mixing (Charbonnel & Zahn 2007). Berenices (Frebel et al. 2010b), Boo¨tes I (Norris et al. Stars can also become very carbon-rich. “Carbon 2010; Lai et al. 2011), and Boo¨tes II (Koch & Rich stars”(seethereviewbyWallerstein & Knapp1998)dis- 2014). Interestingly, all of these measurements are con- play extremely strong C , CH, and/or CN bands, and sistent with GC abundances on the upper RGB except 2 they can be significantly redder than the RGB in op- one star in Sculptor (Sku´lado´ttir et al. 2015) and four tical colors (e.g., Bond & Neff 1969; Mould & Aaronson stars in Segue 1, including its three most metal-poor 1979). Carbon can form in helium-burning asymptotic stars (Norris et al. 2010). The tendency of carbon-rich giant branch (AGB) stars through the triple-α process. stars to appear among the metal-poor population led It can be brought to the surface of an AGB star in the Shetrone et al.(2013)tomeasurecarbonabundancesfor third dredge-up (Iben 1975). This process is most ef- 35 stars in the metal-poor dSph Draco and Frebel et al. ficient in stars around 2–2.5 M⊙ (Mouhcine & Lan¸con (2010a)andStarkenburg et al.(2013)toexplicitlytarget 2003), which explains why most ancient GCs—where themostmetal-poorstarsinSculptor. Still, Segue1and the highest-mass stars are less than 1 M⊙—lack C stars Sculptorcontaintheonlyexamplesofdwarfgalaxystars (Cohen et al.1978). Cstarsalsorequireatleast0.3Gyr with quantified high carbon abundances. (As discussed to form (Mouhcine 2002), which makes them good indi- above, C stars still exist in dSphs.) cators of intermediate-age populations. The MW halo containsa largenumber of C-richstars, Dwarf and RGB stars can acquire carbon through bi- and their frequency increases with decreasing metallic- nary mass transfer with a C star (see Lucatello et al. ity. However, the exact fraction of carbon-enhanced, 2005;Cohen et al.2006;Starkenburg et al.2014). C-rich metal-poor (CEMP) stars as a function of metallicity RGBstarsarerarelyseentohaveaphotometriccompan- is contentious (Cohen et al. 2005; Marsteller et al. 2005; ion because the AGB donor has almost always evolved Lucatello et al. 2006; Frebel et al. 2006; Carollo et al. into a faint white dwarf long ago. Instead, binarity of 2012; Yong et al. 2013; Lee et al. 2013; Placco et al. C-rich RGB stars is observed spectroscopically through 2014),partlybecausethe searchforlow-metallicitystars variability in radial velocity (McClure & Woodsworth isfraughtwithselectionbiases(e.g.,Scho¨rck et al.2009). 1990). The recipient of carbon through mass transfer Very roughly, the fraction of stars with [Fe/H] < −2.5 can exist in any evolutionary stage, such as the main that are carbon-enhanced is about 20%. sequence (Dahn et al. 1977). The MW halo is expected to be at least partly com- Aaronson & Mould (1980) and Aaronson et al. (1982) posed of the disrupted dSphs from past accretion events used the existence of carbon stars in dwarf spheroidal (Searle & Zinn 1978). One apparent challenge to this galaxies (dSphs) to argue for the presence of ideawasthediscrepancybetweenthe[α/Fe]ratioofhalo intermediate-age populations. These stars were selected and dSph stars at [Fe/H] & −2 (Shetrone et al. 1998a, for spectroscopy from their infrared colors, and their 2001,2003;Venn et al.2004). However,Robertson et al. spectra showed very strong C bands. However, car- (2005) and Font et al. (2006) showed that the discrep- 2 bon enhancements need not be so large that they af- ancy is in fact an expected outcome of hierarchical as- fect broadband colors. For example, Norris et al. (2010) sembly. The inner halo, where most of the observed discoveredaC-rich([C/Fe]=2.3)redgiantin the ultra- halo stars lie, is composed of more massive satellites faint galaxy Segue 1. This star (Segue 1–7) could only that do not resemble the present-day, surviving dSphs. havebeenfoundspectroscopicallybecauseitsbroadband At low enough metallicities, most elemental abundance colors are consistent with carbon-normalRGB stars. ratios between dwarf galaxies and the MW halo agree The statistics of carbon abundances in dwarf galaxies (e.g., Frebel et al. 2010b), though the neutron-capture arelesscompletethanforGCsorthe MilkyWay(MW). elements are deficient in ultra-faint dwarf galaxies com- Although Aaronson & Mould (1980) and their contem- pared to the halo, even at very low metallicities (e.g., poraries discovered carbon stars in dwarf galaxies, they Koch et al. 2013). didnotquantifythecarbonenhancement. Kinman et al. Thefactthatsofewcarbon-richstarshavebeenfound (1981) published in conference proceedings the first car- indSphs,evenatlowmetallicities,mightbetroublesome bon abundances from low-resolution spectroscopy in a for the theory of the hierarchical assembly of the MW dSph, specifically Draco. Smith (1984) followed up with halo. The fraction of CEMP stars in Draco, Ursa Mi- CH and CN indices in Draco, and Bell (1985) pub- nor,andSculptorisaround10%orless(Cohen & Huang lished more low-resolution [C/H] abundances in Draco 2009,2010;Starkenburg et al.2013;Shetrone et al.2013; and Ursa Minor. Shetrone et al. (1998b) later measured Sku´lado´ttir et al. 2015). However, this number does not low-resolution[C/Fe] abundances for two stars in Sculp- include C-rich stars for which carbon abundances were tor,findingonetobemildlycarbon-rich. Fulbright et al. not quantified. (2004) measured the first carbon abundance from high- Inthis work,we addto the body ofcarbonabundance resolution spectroscopy in a dSph, specifically Draco. measurements in dSphs. We measured carbon abun- Abia et al. (2008) measured the 12C/13C isotopic ratio dances of 398 red giants (8 previously known) in the and the C/O ratio—but not [C/H]—in two carbonstars dSphsSculptor,Fornax,UrsaMinor,andDraco. Wealso in Carina. Cohen & Huang (2009, 2010) measured car- identified 11 giants that are too carbon-rich to quantify Carbon in dSphs 3 their abundances. The stars observed here are a subset 1.0 of the stars observed by Kirby et al. (2009, 2010), who 0.6 m30e0a0surreeddgMiangt,sSiin, eCigah,tTdi,Spahnsd. Fe abundances in about 0.2 lNToeGgff C=g =423 4041.979 0 KN2419−S1166 [[FCe/F/He]] == −−S11/N..90 88= ±±4 300 ..Å1115−1 1.0 0.6 2. SPECTROSCOPIC OBSERVATIONS 0.2 lMToegf6f 8=g S=4t8 e12t.−07M 8K68−S199 [[FCe/F/He]] ==S −−/N20.. 44=53 1 ±±1 300 ..Å1114−1 We obtained Keck/DEIMOS (Faber et al. 2003) spec- 1.0 tra of the carbon G band in red giants in MW GCs and 0.6 dSphs. The GCs are NGC 2419, NGC 4590 (M68), and NGC7078(M15). ThedSphsareSculptor,Fornax,Ursa x 0.2 lMToegf1f 5=g 5=574 320.7928 4K [[FCe/F/He]] == −+S20/N..56 00= ±±4 700 ..Å1124−1 MKiirnboyr,eatnadl.D(2r0a0c9o,.2W01e0u),sewdhtohoebstaamineeddSrepdhdselritDmEaIsMksOaSs zed flu 01..60 spectraformeasuringMg,Si,Ca,Ti,andFeabundances. ormali 0.2 lSToecgfuf =lgp t=4o4 r13 .1320 6K03443 [[FCe/F/He]] ==S −−/N11.. 80=13 1 ±±3 000 ..Å1114−1 Were-observed3of5slitmasks(198of376memberstars) n 1.0 Fornax 7 6382 inSculptor,1of5slitmasks(125of675members)inFor- 0.6 nanadx,33ooff84sslliittmmaasskkss((81781ofof212298mmemembebresr)s)ininUrDsraaMcoi.nor, 0.2 lToegff =g =41 01.46 9K [[FCe/F/He]] == −−S11/N..10 33= ±±8 900 ..Å1211−1 1.0 0.6 2.1. Target selection 0.2 lUToergffs a=g M=47 i2n5.o00r 5K Bel60099 [[FCe/F/He]] == −−S10/N..76 89= ±±3 400 ..Å1235−1 We selected stars for the DEIMOS slitmasks from 1.0 a variety of photometric catalogs, which provided as- 0.6 trometry, magnitudes, and colors of the red giants that we observed. We used Stetson’s (2000) catalog for all 0.2 lDToergffa =cgo =4 68 142.6146 8K87 [[FCe/F/He]] == −−S20/N..55 45= ±±8 700 ..Å1125−1 three GCs. We supplemented Stetson’s list of M68 stars 4260 4280 4300 4320 4340 rest wavelength (Å) with photometry from Alcaino et al. (1990) and Walker (1994). We used Johnson V and Cousins I magnitudes Figure 1. Portions of DEIMOS spectra around the G band for for all GC stars. We corrected the magnitudes for red- onestarineachoftheGCsanddSphsobserved. Theblacklineis theobservedspectrum,andtheredlineisthebest-fittingsynthetic deningandextinctionbasedonE(B−V)values of0.08, spectrum. Thefitwasperformedinthegreenshadedregion. The 0.05,and0.10forNGC2419,M68,andM15,respectively pale blue lines show synthetic spectra with [C/Fe]±0.2 different (Harris 1996, updated 201011). fromthebest-fittingvalue. Kirby et al. (2009) described the target selection for Sculptor, and Kirby et al. (2010) detailed the target se- lection for Fornax, Ursa Minor, and Draco. For Sculp- tor, we used Westfall et al.’s (2006) photometric catalog wide. This configuration gives an approximate wave- in Washington M and T magnitudes. We used sources length range of 4000–7400 ˚A at a resolution of 2.1 ˚A 2 from Stetson et al.’s (1998) catalog in Cousins B and FWHM (R ∼ 2100 at 4300 ˚A). The exact wavelength Johnson V magnitudes for Fornax. For Ursa Minor, we range of each spectrum depends on the location of the usedBellazzini et al.’s(2002)VI catalog,andforDraco, slitontheslitmask. Thestartingandendingwavelengths we used S´egall et al.’s (2007) gi catalog in the CFHT of the spectra vary by ∼500 ˚A across the slitmask. MegaCam system. Weobtainedinternalflatfieldandarclampexposures. We placed stars with colors and magnitudes appropri- We exposed a quartz lamp three times for 12 s and six ate for red giants on the slitmasks. For the GCs, we times for 45 s, which is appropriate for the red and blue drewapolygonaroundthe RGB inthe color–magnitude halvesoftheCCDmosaic,respectively. Wealsoobtained diagram (CMD). Stars within the polygon were in- twosetsofarclampexposures. First,weobtainedsimul- cluded in the slitmask. Brighter stars were preferred taneousspectraofNe,Ar,Kr,andXelampsfor1s. This overfainterstarsinthe caseswhereslitmask designcon- exposure is suitable to calibrate the wavelengthsolution straints forced a choice among multiple red giants. For for the red halves of the spectra. Then, we obtained the dSphs, we selected stars from the CMD with the a single exposure for several lamps. We controlled the guidance of theoretical isochrones (see Kirby et al. 2010 amount of flux from each lamp by turning off lamps in for additional details). The Sculptor catalog addition- themiddleoftheexposure. ThelampswereHg(1s),Cd ally has DDO21 magnitudes, which we used to discrimi- (5 s), Kr (6 s), Ar (12 s), and Zn (12 s). This exposure nate between red giants and foreground dwarf stars (see is appropriate for the blue halves of the spectra. Kirby et al. 2009). All slitmasks were also observedwith the 1200Ggrat- ing, which has a groove spacing of 1200 mm−1 and a 2.2. Observations blazewavelengthof7760˚A. Thegratingwascenteredat Table 1 lists the observationdates and exposure times 7800˚A. Table2ofKirby et al.(2010)listsalloftheseob- ofeachslitmask. We usedthe 900ZDgrating with a rul- servations except for the n2419b and n4590a slitmasks. ingof900linesmm−1 andablazewavelengthof5500˚A. We observed n2419b with the 1200G grating and this The grating was tilted such that the blaze wavelength 7800 ˚A setting on 2012 March 19 for 1800 s, and we fell on the center of the CCD mosaic. Slits were 0.7′′ observed n4590a on 2011 June 2 for 2400 s. 11 http://www.physics.mcmaster.ca/$\sim$harris/mwgc.dat 2.3. Reductions 4 Kirby et al. Table 1 DEIMOSObservations System Slitmask #targets Date Airmass Seeing Individual Exposures TotalExposureTime (′′) (s) (s) NGC2419 n2419b 112 2012Mar19 1.1 0.8 3×900 2700 M68 n4590a 96 2014 Feb 2 1.6 0.8 1200+937 2137 M15 7078d 164 2011 Jul 29 1.0 1.1 3×600 1800 7078e 167 2011 Jul 30 1.0 0.9 3×900 2700 Sculptor bscl1 86 2011 Jul 31 1.7 0.8 4×1200+900 5700 bscl2 106 2011Aug6 1.8 0.7 2×1200+2×840 4080 bscl6 91 2011Aug4 1.7 0.8 3×1200+1260 4860 Fornax bfor6 169 2011Aug5 1.8 1.3 2×780 6360 2011Aug6 1.8 0.8 1200 2011Aug7 1.9 0.8 3×1200 UrsaMinor bumi1 125 2011 Jul 29 1.5 0.7 4×1200+600 5400 bumi2 134 2011 Jul 31 1.7 0.8 4×1200 4800 bumi3 137 2011Aug4 1.8 0.6 4×1200 4800 Draco bdra1 135 2011 Jul 30 1.4 1.2 5×1200 6000 bdra2 144 2011Aug7 1.3 0.7 4×1200 4800 bdra3 129 2011Aug5 1.3 1.0 5×1200 6000 We reduced the raw data with the spec2d pipeline12 (Cooper et al. 2012) developed by the DEEP2 survey 1.0 team (Newman et al. 2013). This software is optimized 0.6 for extended galaxies observed with the 1200G grating. M15 WmoeremsauditeambloedfoerrabtleuerervDisEioInMsOtoSsthpeecptripa.elFinierstt,owmeamkeodit- 0.2 4lI(o0V7 g= 8− 6(1 LI63)/0L.9 =O •5) 0=.8 16.91 [C/Fe] = +0.88 ± 0.10 ified the software to accept separate flat lamp exposures for the red and blue CCDs. Second, we revised the arc 1.0 lamplinelistsforHg,Cd,Kr,Ar,andZnbetween4158˚A 0.6 icanhnadinn5gd4eivd9i6dtu˚Ahae.l WtDoEleeIreMalinmOceSinsaaftroecrdlaaliumnteposmetxahtpaitocswuidereecsn.otuifiTldchanitriodotn, sweoeef zed flux 0.2 UBlI(o0V erg= sl− 6a (1 L 0IM5)0/0L.39 i=O n•16) o 1=r.4 39.24 C star awrecaklienresblsuoetlhinaets.thFeopuirpthel,inweewcohualndgeindctluhdeewsaoymtehaotf tthhee normali 1.0 two-dimensional spectrum is rectified. The curvature of 0.6 thefocalplaneandtheslitpositionanglecanintroducea D67r0ac0o92 tcioltrrienspthonedsspatotiaalddiiargeocntiaolnlisnuecrhatthhaerttahsainngalesiwngavleelreonwgtohf 0.2 l(io0g g =− ( 1Li6)/0.L 3=O •7 )1 =.3 26.56 C star pixels. Previously,thistiltwasremovedbyshiftingCCD 1.0 columnsbyintegervalues. Wechangedthecodetoallow non-integer shifts, which slightly increased the effective 0.6 Fornax spectralresolutionandsignal-to-noiseratio(S/N). Fifth, 91907 we implemented a correction for atmospheric dispersion 0.2 l(RoB0g =− ( L1R/7)L0.9 O •=)5 =2 .22.254 [C/Fe] = −1.14 ± 1.99 along the slit, which is important for our bluer spectra. 7800 7900 8000 8100 8200 8300 8400 Becauselightat4000˚Acanbedisplacedalongtheslitby rest wavelength (Å) a few pixels from light at 7400 ˚A, we curved the extrac- Figure 2. The top panel shows a moderately carbon-rich AGB tion window to match the shape of the atmospheric dis- starinM15. TheCNbandheadisat∼7850˚A,andCNabsorption persion. Finally, we adopted the optimizations for point isvisibleredwardofthebandhead. ThenexttwopanelsshowCN sources developed by Simon & Geha (2007). absorption in two of the 11 carbon stars in the DEIMOS sample. The results of the reductions were wavelength- The bottom panel shows a carbon-normal RGB star for compari- son. calibrated, sky-subtracted, one-dimensional spectra. Cosmic rays were removed before the one-dimensional spectra were extracted. The spec2d code also calculated 2.4. Carbon Stars and saved variance (error) spectra, which we used to es- timaterandommeasurementuncertaintyon[C/Fe](Sec- We found 11 examples of carbon stars in our sample. tion 3.4.1). The red spectra of these stars are dominated by very Figure1showsoneexamplespectrumfromeachstellar strong CN absorption, which precluded measurement of system in Table 1. The figure shows spectra at a variety atmospheric parameters like [Fe/H]. The G bands are of S/N. Each panel lists the S/N, which is calculated as also strong enough that we would not have found our the median absolute deviation of the observed spectrum measurements of carbon abundances reliable due to sat- (black) from the best-fitting synthetic spectrum (red). uration. Figure 2 shows examples of the near-infrared spectra of some of these stars. We discuss carbon stars 12 http://www2.keck.hawaii.edu/inst/deimos/pipeline.html further in Sections 5.1 and 6. Carbon in dSphs 5 3. SPECTROSCOPIC MEASUREMENTS Table 2 Before we could measure carbon abundances from the LineList spectra,weneededtomeasurethestars’radialvelocities (vr) and atmospheric parameters. Wavelength (˚A)a Species EP(eV) loggf 4100.005 Mnii 8.129 −1.184 3.1. Radial Velocity Measurements 4100.013 NH 2.202 −3.421 4100.014 NH 2.387 −4.433 We previously measured v for all of the spectra r 4100.021 NH 2.335 −4.289 from our red DEIMOS observations (Kirby et al. 2009, 4100.022 13CH 0.642 −4.345 2010). However, the wavelength calibration for the blue 4100.048 NH 2.824 −2.929 DEIMOS spectra is not as precise as the red spectra 4100.051 NH 2.335 −4.321 because of the relative sparsity of blue arc lamp lines. 4100.055 NH 2.599 −2.932 4100.084 NH 2.202 −4.982 Rather than apply the known vr from the red spectra 4100.088 Cri 4.535 −1.241 to the blue spectra, we measured vr directly from the ··· ··· ··· ··· G band in the blue spectra. Although v is not as r,blue precise as vr,red (for example, for measuring galactic ve- References. — Atomic data come from locity dispersions), v is more suited to shifting the VALD (Piskunovetal. 1995; Kupkaetal. 1999) r,blue G band into the rest frame. and NIST (Kramidaetal. 2014). CH data come from SCAN (Jorgensen etal. 1996). NH and CN We measured v by cross-correlating the observed r,blue datacomefromKurucz(1992). spectrumwithasyntheticspectrum. Thesyntheticspec- Note. — (This table is available inits entirety trum had the atmospheric parameters previously mea- in a machine-readable form in the online journal. suredfromtheredspectra(seeSection3.2)and[C/Fe]= Aportionisshownhereforguidanceregardingits 0.0. The spectra were synthesized according to the pro- formandcontent.) ceduredescribedinSection3.4. Weperformedthecross- a Wavelength inair. correlationintherange4200–4400˚A,andwesetv tobe r the velocity corresponding to the cross-correlationpeak. of the Sun and Arcturus. Then, we computed MOOG For the remaining analysis,we shifted the spectruminto synthetic spectra using ATLAS9 model atmospheres. the rest frame using this measurement of v . r We also evaluated the membership of each star by 3.3.1. Line List its radial velocity. Kirby et al. (2010, their Section 2.6) The G band has a high density of CH molecular ab- gavethedetailsofthemembershipcriteria. Insummary, sorption lines. At the low resolution of our DEIMOS weeliminated starsmorethanthree standarddeviations spectra, individual CH lines are indistinguishable from away from the mean radial velocity. individual atomic and other molecular lines. Therefore, we neededto model the absorptiondue to many species, 3.2. Atmospheric Parameters notjustCH. Wecompiledalistofatomicandmolecular The measurement of carbon abundances (Section 3.4) transitions in the range 4100–4500˚A. requires knowledge of effective temperatures (T ), sur- We startedby including allthe neutralandsinglyion- eff facegravities(logg),metallicities([Fe/H]),andalphaen- ized atomic transitions listed in the Vienna Atomic Line hancements([α/Fe]). Kirby et al.(2009,2010)measured Database13 (VALD, Piskunov et al. 1995; Kupka et al. these parameters for most of the spectra presented here. 1999) in our spectral range of interest. We discarded They obtainedpreliminaryestimates ofT and logg by transitions with excitation potentials (EP) greater than eff comparing the stars’ colors and magnitudes to theoret- 10 eV or with oscillator strengths loggf <−5. ical isochrones. They refined T and measured [Fe/H] We then compiled a separate atomic line list from the eff and [α/Fe] by minimizing χ2 between the observed red National Institute of Standards and Technology (NIST) DEIMOS spectra and a large grid of synthetic spectra database14 (Kramida et al. 2014). This database is (Kirby 2011). very incomplete, but it generally has more trustworthy, Weadoptedwithoutmodificationthepreviouslydeter- laboratory-measuredatomictransitionprobabilitiesthan mined atmospheric parameters. However, the DEIMOS other databases. We replaced entries in the VALD line slitmasks n2419b and n4590a were not included in list with NIST entries where available. the previous catalogs of atmospheric parameters. We Next, we added molecular transitions. We used the measured these parameters in the same manner as CN and NH line lists of Kurucz (1992)15 and the Kirby et al. (2010). SCAN line list16 for CH from Jorgensenet al. (1996). (Masseron et al. 2014 published a new CH line list long 3.3. Grid of Synthetic Spectra after we computed the spectral grid and shortly before we submitted this article.) We preservedthe isotopic in- Wemeasured[C/Fe]bycomparingobservedspectraof formationforeachCHtransitionsothatwecouldchange theGbandtoagridofsyntheticspectraspanning4100– 12C/13C in the spectral syntheses. 4500 ˚A. We computed this grid in a similar fashion to No line list is perfectly accurate or perfectly com- thewayinwhichKirby, Guhathakurta, & Sneden(2008) plete. Wetestedourownlinelistagainstobserved,high- computed their synthetic spectral grid spanning 6300– 9100 ˚A in order to measure [Fe/H] and [α/Fe] from red 13http://vald.astro.univie.ac.at/$\sim$vald/php/vald.php DEIMOS spectra. First, we compiled a line list from 14 http://www.nist.gov/pml/data/asd.cfm publiclyavailablecatalogsofatomicandmoleculardata. 15 http://kurucz.harvard.edu/ We refined the line list by comparing it to the spectra 16 http://www.astro.ku.dk/$\sim$uffegj/ 6 Kirby et al. 1. Continuum division: MOOG generates synthetic Table 3 spectra normalized to the continuum. We divided GridofSyntheticSpectra eachobservedspectrum by a continuumfit so that it could be compared to the synthetic spectra. We Parameter Start End Step fit the spectrum with a spline with a breakpoint λ 4100˚A 4500˚A 0.02˚A spacing of 200 ˚A. Pixels with fluxes that were Teff 3500K 5600K 100K greater than 5 standard deviations above or 0.1 5600K 6400K 200K standard deviations below the spline were itera- logg 0.0 4.0 0.5 [Fe/H] −4.0 0.0 0.1 tively excluded from the fit so that stellar absorp- [α/Fe] −0.8 +1.2 0.1 tion was not included in the continuum determi- [C/Fe] −2.4 +1.0 0.2 nation. The spectrum was divided by the spline. +1.0 +3.0 0.4 This continuum was refined in step 3. +3.0 +3.5 0.5 2. Preliminary [C/Fe] measurement: Wesearchedthe resolutionspectraoftheSunandArcturus(Hinkle et al. spectral grid for the best-fitting value of [C/Fe]. 2000). We computedsynthetic spectraforbothstarsus- The synthetic spectra were smoothed through a ing MOOG17 (Sneden 1973) with an updated treatment Gaussian kernel (2.1 ˚A FWHM) to match the ob- of electron scattering (Sobeck et al. 2011). For the Sun, served resolution. We fixed Teff, logg, [Fe/H], and we used an ATLAS9 (Kurucz 1993) model atmosphere [α/Fe]atthevaluespreviouslydeterminedfromthe with T = 5798 K and logg = 4.44. For Arcturus, we red DEIMOS spectra (see Section 3.2). We used eff used Peterson et al.’s (1993) model atmosphere. Where the Levenberg–Marquardt optimization code MP- atomic lines did not match the observed spectra, we FIT (Markwardt 2012) to find the value of [C/Fe] changedoscillatorstrengths. Wealsointroducedahand- that minimized χ2 in the spectral range 4260– fulofartificialFe ilinestomimicobservedlinesthatdid 4325 ˚A (green region in Figures 1 and 3), where not appear in our original line list. We avoided chang- the spectrum is most sensitive to changes in car- ing molecular line strengths. Table 2 gives the final line bon abundance. listthatweusedinourcomputationofsyntheticspectra (Section 3.3.2). 3. Continuumrefinement: Thecontinuumdetermina- tioninstep1doesnottakeintoaccounttheknown 3.3.2. Spectral Synthesis variationinfluxasafunctionofwavelength. Were- finedthecontinuumbydividingtheobservedspec- We used MOOG with our existing grid of ATLAS9 trum by the best-fitting synthetic spectrum deter- model atmospheres (Kirby 2011) to compute synthetic mined in step 2. We fit that quotientwith a spline spectra at a range of atmospheric parameters. Table 3 with a breakpoint spacing of 150 ˚A. This spacing shows the parameters of the spectral grid: the lower is slightly finer than in step 1 for a slightly more and upper ranges and step sizes of wavelength, effective detailed continuum determination. We iteratively temperature, surfacegravity,metallicity, alpha enhance- discarded pixels that were above or below the fit ment, and carbon abundance. We used 256 CPUs to by more than one standard deviation. compute the 4,835,376 spectra. Although [C/Fe] varied in the spectral syntheses, the input model atmospheres 4. Refined [C/Fe] measurement: We repeated steps 2 were all computed assuming [C/Fe]=0. and 3 until [C/Fe] changed by less than 0.001 be- As stars ascend the RGB, they dredge up products tween iterations. of the CNO cycle. We examine this phenomenon in Section 4, but many others have investigated it be- In some cases, the spectra allowed only measurements fore. In particular, Keller et al. (2001) showed how the of upper limits. We computed χ2 contours for each star 12C/13C isotope ratio varies as a function of evolution- at several values of [C/Fe] around the minimum χ2. For ary state. The following relation, which we deduced stars whose χ2 contours did not rise by at least 1 on fromKeller et al.’sFigure4,approximateshowtheratio both sides of the minimum, we calculated a 2σ upper changes with decreasing surface gravity: limit. The upper limit was the value of [C/Fe] at which χ2 was 4 above the minimum χ2. 12C/13C=50 if logg >2.7 Figure1showsmodelspectraofthebest-fittingGband 12C/13C=63 logg−120 if 2.0<logg ≤2.7 spectruminred. Alsoshowninlightbluearemodelspec- (1) 12C/13C=6 if logg ≤2.0 tra deviant from the best fit at [C/Fe]±0.2. Figure 3 shows two pairs of spectra of stars—one pair in Sculp- torandone pair inUrsa Minor—withsimilarluminosity We used Equation 1 in computing the grid of synthetic andmetallicitybutdifferentcarbonabundances. Thefig- spectra. ure demonstrates that the stars measured to have larger [C/Fe] have noticeably stronger G bands. 3.4. Carbon Abundance Measurements We also computed the S (CH) index (Martell et al. 2 Wemeasuredcarbonabundancesbycomparingtheob- 2008a). This index is defined to have minimal sensitiv- served spectra to our new grid. This process involved ity to nitrogen abundance. It is the ratio of flux in the several steps. G band to flux in two side bands. The blue side band starts at 4212 ˚A. We measured S (CH) only in spec- 2 17 http://www.as.utexas.edu/$\sim$chris/moog.html tra with minimum rest wavelengths less than 4212 ˚A. Carbon in dSphs 7 0.3 Sculptor NGC 2419 1.0 0.2 MM6185 Sculptor Fornax 0.1 Ursa Minor e] Draco 0.6 C/F -0.0 ∆[ -0.1 -0.2 ed flux 0.2 1l[[oFC0ge1/ F/0(HeL0]]/6 L==9O • )−− =11.. 23303.1 ±±9 00..1151 1l[[oFC0ge0/ F/5(HeL1]]/0 L==3O • )−− =01.. 72347.2 ±±6 00..1151 -0.3 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 ormaliz 1.0 Ursa Mi nor Figure 4. Theresponseof[C/Fe∆][αto/Fec]hanging[α/Fe]inthestellar n atmosphere. The x-axis shows the difference between the [α/Fe] values used in our fiducial measurements of [C/Fe] and [α/Fe] = 0.0. 0.6 bution with a standard deviation equal to spec2d’s esti- mate of the flux error in that pixel. We repeated steps Pal157 Bel60017 0.2 l[oFge /(HL]/ L=O • )− =2. 027.5 ±6 0.11 l[oFge /(HL]/ L=O • )− =2. 027.6 ±3 0.11 2–4 above with the noise-added spectrum. [C/Fe] = −0.86 ± 0.15 [C/Fe] = +0.31 ± 0.29 We repeated this process for a total of 1000 noise re- 4260 4280 4300 4320 4340 alizations. We took the base uncertainty on [C/Fe] as rest wavelength (Å) the standard deviation of all 1000 trials. Although this Figure 3. Gbandspectraofpairsofstarswithsimilarluminos- procedure accounts for random uncertainty, it does not ity,similarmetallicity,anddifferingcarbonabundanceinSculptor includeallsourcesofsystematicerror. Weadded0.1dex (top) and UrsaMinor (bottom). The stellar parameters listedon as an estimate of systematic error in quadrature with theleft(right)ofeachpanelcorrespondtothesolidblack(dashed the Monte Carlo uncertaintyfor the finaluncertainty on blue) spectrum. As in Figure 1, only the green shaded region is usedinthemeasurementof[C/Fe]. [C/Fe]. Thissystematicuncertaintyisapproximatelythe value that is required to account for the rms of the dis- Martell et al. (2008a) defined the band to be measured tributionofourmeasurementsof[C/Fe]versusliterature in units of ADU per pixel. However, the throughput of measurements (Figure 5, discussed in Section 3.4.3). DEIMOSinthevicinityoftheGbanddiminishesrapidly We computed errors on S (CH) from the same noise- 2 toward the blue. Consequently, we measured S2(CH) in added spectra. We did not add any systematic error to continuum-corrected flux rather than the original ADU these measurements. per pixel. Table 4 gives the [C/Fe] and S (CH) measurements 3.4.2. Effect of Nitrogen and Oxygen 2 for our GC and dSph sample. The last column iden- The measurement of carbon could be affected by the tifies the 11 carbon stars for which we were unable abundancesofotherelements. First,otherelementsthat to measure [Fe/H] or [C/Fe]. The table also gives form diatoms with carbon affect the molecular equilib- photometric information, including the absolute lumi- rium, changing the amount of carbon available to form nosity for each star relative to the solar luminosity. CH. Second, some elements contribute to the opacity in We used bolometric corrections based on theoretical thestellaratmosphere. Changesintheirabundancescan isochrones (Demarque et al. 2004; Girardi et al. 2004) affect the structure of the atmosphere and the strength and previously measured distance moduli to compute oftheGband. Themostimportantelementstoconsider the luminosities: 19.83 for NGC 2419 (Harris et al. fortheseeffectsarenitrogenandoxygenbecausetheyare 1997), 15.21 for M68 (McClure et al. 1987), 15.08 abundant,andtheyformdiatomswithcarbon. Unfortu- for M15 (Durrell & Harris 1993), 19.67 for Sculptor nately,wecanmeasureneitherelementdirectlyfromour (Pietrzyn´ski et al. 2008), 20.72 for Fornax (Rizzi et al. spectra. However, we can quantify their possible effects 2007),19.18forUrsaMinor(Mighell & Burke1999),and on our carbon abundance measurements. 19.84 for Draco (Bellazzini et al. 2002). The table also We explored the effects of molecular equilibrium be- includes T , logg, [α/Fe], and S/N, calculated as the eff tween CH and CN by making syntheses with MOOG medianabsolutedeviationoftheobservedspectrumfrom with varying abundances of N. We took three stars in the best-fitting spectrum in the range 4250–4350˚A. M15 ([Fe/H] ≈ −2.4) as test cases: 13701, which is be- low the RGB luminosity function bump; 43026, which 3.4.1. Estimation of Uncertainties is at the top of the RGB; and 47866, which is carbon- Although MPFIT provides a covariance matrix for enhanced (see Section 4). We also tested two stars in eachminimization,thisestimateofuncertaintyon[C/Fe] Fornax to explore more metal-rich ([Fe/H] ≈ −1) stars: does not accountfor uncertainty in the placementof the 52752 at log(L/L⊙) = 2.0 and 40828 at the top of the continuum. Instead of using the covariance matrix, we RGB. Wesynthesizedspectrawith[N/Fe]valuesof−2.0, calculated Monte Carlo uncertainties on [C/Fe] by re- −1.0,−0.5,−0.2,0.0,+0.2,+0.5,+1.0,and+2.0. Inno sampling the observed spectrum. case did the flux of any pixel in the G band change by After we determined the best-fitting carbon abun- more than 1%. dance, we added noise to the spectrum after it was Ourcarbonabundance measurements alreadyaccount continuum-normalizedaccordingtostep1inSection3.4. for varying elemental abundances that might affect the The new flux in each noise-added pixel was the original structureofthestellaratmospheres. Wepreviouslymea- flux plus a value drawn from a Gaussian random distri- sured four α elements (Mg, Si, Ca, and Ti) from red 8 Kirby et al. DEIMOS spectra (Kirby et al. 2010). The stellar atmo- The origin of the discrepancies might be spectral res- spheres we used to measure carbon abundances reflect olution or the details of the carbon abundance measure- those [α/Fe] ratios. VandenBerg et al. (2012) showed ments. At first glance, the spectral resolution is suspect that at temperatures found in the atmospheres of cool because our low-resolution measurements agree most giants, Mg and Si are the most important metals for de- withanotherlow-resolutionstudy(Shetrone et al.2013). termining the opacity. As a result, changing [α/Fe] (in- However, our measurements disagree with Venn et al.’s cluding [Mg/Fe] and [Si/Fe]) can alter the atmospheric (2012)intheoppositedirectionfromtheothertwohigh- structure and therefore affect the measurements of all resolution studies. Therefore, we investigated the possi- elemental abundances. bilitythatdifferencesinthelinelistcausedtheoffset. We Werecomputedallofthe[C/Fe]measurementsassum- comparedourlinelist(Table2)withJ.Cohen’sline list. ing [α/Fe]=0.0 ratherthan the values previously deter- Between 4274 ˚A and 4232 ˚A, our list contains 2.5 times mined by Kirby et al. (2010). Here, α refers to O, Ne, more CH lines. The higher density of lines makes the Mg, Si, S, Ar, Ca, and Ti. This value of [α/Fe] is used G band stronger. Hence, for a fixed observed G band to compute both the modelatmosphereandthe spectral strength, we would measure a lower carbon abundance synthesis. Hence, the synthesesinclude the effects ofthe with our line list than with Cohen’s line list. Our line varying atmospheric structure and the changing molec- list has a similar density of lines to Shetrone et al.’s list. ular equilibrium between CH and CO. Figure 4 shows Therefore, it is plausible that a different line list is the theresponseof[C/Fe]tothechangein[α/Fe]. Changing main source of disagreement of carbon abundance mea- [α/Fe] by ±0.6 dex—much more than the mean uncer- surements from the G band. tainty in [α/Fe], 0.2 dex—results in a change of approx- imately ±0.1 dex in [C/Fe]. Forcing [α/Fe] to be 0.0 4. ASTRATION OFCARBON ON THE UPPERRGB results in |∆[C/Fe]|<0.1 dex for 94% of the dSph stars ThedestructionofcarbonontheRGBabovethelumi- and|∆[C/Fe]|<0.05dex for 75%ofthe dSph stars. We nosity function bump is well documented (e.g., Suntzeff concludethatuncertaintyin[α/Fe]isnotamajorsource 1981;Carbon et al.1982;Smith & Briley2006). As pre- of uncertainty in our carbon abundance measurements. viousinvestigatorshavedone,wesearchedforsignatures In conclusion, uncertainties in nitrogen, oxygen, and of carbon astration in GCs. GCs are the easiest stel- other α element abundances are minor contributors to lar systems to isolate stellar luminosity as an indepen- the error budget of our [C/Fe] measurements. dentvariablebecause the cluster starshavesimilarages, metallicities, and distances. 3.4.3. Validation Figure 6 shows the trend of carbon enhancement ([C/Fe]) with luminosity. The carbon abundance re- We compared our low-resolution measurements of [C/Fe] with previous high- and low-resolution measure- mains flat until giants reach log(L/L⊙) = 1.6, where- upon[C/Fe]declines,presumablyduetomixingthatbe- mentsof[C/Fe]forthesamestars. Wefoundoverlapping ginsattheluminosityfunctionbump. Wefitalinetothe starswithhigh-resolutionmeasurementsintheliterature combined trend of [C/Fe] with [Fe/H] for all three GCs, for NGC 2419 (Cohen & Kirby 2012), M68 (Venn et al. 2012), Draco (Fulbright et al. 2004; Cohen & Huang fixing the luminosity of the bump at log(L/L⊙) = 1.6. The best fit line, represented in green in Figure 6, is 2009), and Ursa Minor (Cohen & Huang 2010). Shetrone et al. (2013) also published measurements of [C/Fe]inDracobasedonlow-resolutionKeck/LRISspec- [C/Fe]=0.12±0.06 if log(L/L⊙)≤1.6 troscopy. Allofthesemeasurementsarealsobasedonthe [C/Fe]=(1.42±0.06) (2) G band. −(0.82±0.02) log(L/L⊙) if log(L/L⊙)>1.6 The left panel of Figure 5 shows the comparison between [C/H] measured from DEIMOS and [C/H] as published in the aforementioned references. The The amountofcarbonastrationin GCs couldbe com- right panel shows [C/Fe]. We shifted all abundances puted from this empirical trend, but we also considered to our adopted solar abundance scale18: A(C) = theoretical estimates of the amount of carbon depletion. Placco et al.(2014)calculatedcorrectionsbasedonmod- 8.56 (Anders & Grevesse 1989) and A(Fe) = 7.52 els of mixing on the upper RGB. The corrections de- (Sneden et al.1992). Thedeviationsreachupto0.8dex. pendonsurfacegravity,metallicity,anduncorrectedcar- The deviations are also about the same for [C/Fe] as bon abundance.19 We used their models to compute a they are for [C/H], which indicates that differences in correction for each star in our sample. Figure 6 shows atmospheric parameters like T are not the reason for eff in faded colors Placco et al.’s corrections applied to the the discrepancy. On the other hand, the degree of de- GC sample. The faded points have less dependence on viation depends largely on the source of the measure- luminosity than the uncorrected points. Although the ment. For example, our [C/H] measurements are 0.3– empirical linear fit to the data (Equation 2) is a better 0.4 dex above Venn et al.’s [C/H] measurements from fit than Placco et al.’s models, the models are ab initio VLT/FLAMES, but our measurements are on average ∼ 0.4 dex below Cohen & Huang’s, Cohen & Kirby’s, 19 The correction depends slightly on [N/Fe]. We assumed and Fulbright et al.’s measurements from Keck/HIRES. [N/Fe] = 0. Grattonetal. (2000) showed that [N/Fe] climbs Our measurements largely agree with Shetrone et al.’s to about 0.5 at the tip of the RGB in metal-poor field stars. Keck/LRIS measurements. Placcoetal.’s(2014)correctionschangeby<0.05dexforchanges in[N/Fe]of0.5dex. Fortherangeofstellarparametersinoursam- ple, the difference fromthe [N/Fe]=0correction exceeds 0.1dex 18 A(X)=12+log(n(X)/n(H)) wheren(X)isthenumberden- onlyfor[N/Fe]≥+1,aconditionthatveryfew,ifany,ofthestars sityofelementX. inoursamplewouldsatisfy. Carbon in dSphs 9 N GC 2419 (Cohen & Kirby 2012) -0.2 M 68 (Venn et al. 2012) --22..00 D raco (Cohen & Huang 2009) D raco (Fulbright et al. 2004) -0.4 D raco (Shetrone et al. 2013) Ursa Minor (Cohen & Huang 2010) --22..55 -0.6 OS OS M M DEI DEI -0.8 H] --33..00 e] F C/ C/ -1.0 [ [ --33..55 -1.2 -1.4 --44..00 H]other 0.5 Fe]other -01..56 C/ C/ [ [ − − S 0.0 S 0.0 O O M M EI EI D -0.5 D -0.5 ] ] H e C/ C/F [ -4.0 -3.5 -3.0 -2.5 -2.0 [ -1.6 -1.4 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2 [C/H] [C/Fe] other other Figure 5. Comparison between Keck/DEIMOS and literature carbon abundances for stars in the GCs NGC 2419 and M68 and the dSphs Draco and Ursa Minor. All of the literature measurements come from Keck/HIRES except for the studies of Vennetal. (2012, Magellan/MIKE)andShetroneetal.(2001,Keck/LRIS). predictions. They can be applied to data in environ- individually to calculate the dispersion. We then nor- ments that are not as tightly controlled as GCs. We malized the residuals by the measurement uncertainty: use the Placco et al. (2014) correctionsfor dSph stars in ∆ = ([C/Fe] − [C/Fe](trend))/δ[C/Fe]. If there is no Section 5. intrinsic scatter in [C/Fe] beyond the measurement Martell et al. (2008c) found that the rate of carbon uncertainty, then the standard deviation of ∆ should be depletion as giants evolve depends on metallicity, which 1. Instead,wefound1.6forNGC2419,1.5forM68,and is consistent with the thermohaline mixing mechanism 2.1 for M15. (We excluded star 47866, discussed below, for the change in elemental composition with luminos- from M15.) This measurement of dispersion demands ity (Charbonnel & Zahn 2007; Charbonnel & Lagarde that we estimated measurement uncertainties accu- 2010). Unfortunately, we can not test for metallicity rately. If we underestimated the uncertainties, then the dependence because the GCs we observed have similar standard deviation of ∆ will appear erroneously larger metallicities. According to Harris (1996), the metal- than 1. Taken at face value, we have found evidence licities are [Fe/H] = −2.15 (NGC 2419, Suntzeff et al. for intrinsic scatter in [C/Fe]. We see no evidence for 1988), −2.23 (M68, Lee et al. 2005), and −2.37 (M15, bimodality in [C/Fe], but the scatter is not much larger Takeda et al. 2009). Martell et al. found carbon deple- than the uncertainties. Nonetheless, CN bands do show tion rates of d[C/Fe]/dlog(L/L⊙)=−0.7 to −1.5 above bimodality in some GCs (Smith & Briley 2005, 2006; theluminosityfunctionbumpforthemetallicityrangeof Martell et al. 2008b), reflecting that primordial C and thesethreeGCs. Ourvalueof−0.82±0.02(Equation2) N abundances in clusters were bimodal, independent falls in this range. of the present-day astration on the upper RGB. This GCs have primordial abundance variations bimodality can even be found on the main sequence of (Gratton et al. 2004), including carbon (e.g., Cohen some clusters (e.g., 47 Tucanae, Harbeck et al. 2003). 1999). The dispersion in [C/Fe] at fixed luminosity can Cohen et al. (2011), Cohen & Kirby (2012), and give a sense of the differential enrichment in carbon Mucciarelli et al. (2012) discovered that NGC 2419 has between the first and subsequent generations of star an extremely unusual chemical composition. The Mg formation. We calculated this dispersion by finding the abundances vary from star to star by a factor of 25, residuals between [C/Fe] and the kinked linear trend. and the variation is anticorrelated with abundances of The trend is similar to Equation 2, but we fit each GC other elements, including K and Ca. In a sample of 13 10 Kirby et al. Note the one carbon-rich star in M15, which is also 1.0 NGC 2419 visible in Figure 6. Star 47866, a confirmed radial ve- 0.5 locity member of M15, stands out as lone orange point ([C/Fe] = +0.88± 0.10) in Figure 7 in a sea of dark 0.0 green points ([C/Fe]∼ −0.2). The carbon abundance is -0.5 highenoughthatthenear-infraredspectrumshowssome CN absorption(Figure 2). This star was also previously -1.0 Ca enhanced identified by J. Cohen to be carbon-richfrom an unpub- -1.5 C a normal lished Keck/HIRES spectrum. That spectrum shows a 1.0 M68 weakC2 bandheadat5163˚Ainadditiontoaverystrong Gband. Alsonotethatthisstarliesslightlybluewardof 0.5 the main RGB locus. The CMD position indicates that e] 0.0 star 47866 is an AGB star. F Carbon-enhanced stars in GCs are rare. Cohen et al. C/-0.5 (1978)andMouhcine (2002)explainthatCstarsappear [ after severalhundredmillion years,long after the end of -1.0 starformationinaGC. NoneoftheGCstarswouldhave -1.5 been pre-enriched by carbon from an AGB star. Addi- 1.0 tionally,thelowest-massAGBstarthatproducescopious M15 amountsofcarbonisabout2M⊙. Suchastarwouldlive 0.5 forabout1Gyr. Hence,allcarbon-producingAGBstars 0.0 inGCsshouldhavediedlongago. Nonetheless,thereare a few examples of C-rich stars in GCs. The GC ω Cen- -0.5 tauri hosts at least seven known C stars (Harding 1962; -1.0 Dickens 1972; Bond 1975; Cowley & Crampton 1985; van Loon et al. 2007). However, ω Centauri is known to -1.5 have an unusually long star formation history (SFH) for aGC (Hilker et al.2004;Villanova et al.2007). Regard- 1.0 1.5 2.0 2.5 3.0 3.5 less, C stars can be found in even more ordinary GCs, log (L/L ) like M14 (Cˆot´e et al. 1997), Lynga 7 (Matsunaga 2006; O • Feast et al. 2013), and NGC 6426 (Sharina et al. 2012). Figure 6. Thedecreaseincarbonenhancementwithredgiantlu- C stars can be formed as the result of mass trans- minosityintwoglobularclusters. ThegiantsinNGC2419aresep- fer from an AGB star. In most cases, the AGB star arated into calcium-enhanced (red triangles) and normal-calcium has now evolved into a white dwarf. There is pho- groups (red circles). The green lines illustrate of the destruction tometric (B¨ohm-Vitense et al. 2000) and spectroscopic of carbon as photospheric material is processed by the CNO cy- cle on the upper RGB (Equation 2). The faded points show the (Lucatello et al. 2005; Starkenburg et al. 2014) evidence [C/Fe]measurements correctedforthedestructionofcarbonwith thatmostorallC stars—especiallythose enhancedins- increasingluminosityaccordingtothecalculationsofPlaccoetal. process elements—have binary companions. Binarity is (2014). often diagnosed through radial velocity variability over multiple measurements. Although we observed 47866 red giants with high-resolution spectra, Cohen & Kirby twice, the observationswere separatedby only 24 hours. (2012) did not find any correlation between C and any We found no evidence for binarity from this very short of the elements with an intracluster dispersion. We time baseline. An alternative possibility is that the ex- looked in our larger sample for such a correlation by isting C-rich AGB star formed in a stellar merger, the separating the NGC 2419 stars into Ca-enhanced and resultofwhichwouldbemassiveenoughtogeneratecar- Ca-normal. We estimated Ca enhancement by measur- bon in its AGB phase (e.g., Feast et al. 2013). Stellar ing the equivalent widths (EWs) of the near-infraredCa mergers are a plausible origin for blue stragglers in GCs triplet. WeconvertedtheseEWstometallicityestimates (Mateo et al. 1990; Sills et al. 2013), so it is conceivable usingStarkenburg et al.’s(2010)formula. Thestarswith thatthiscarbon-richstarisalsotheproductofamerger. [Fe/H] < −2.2 (Ca-normal) are represented by blue CaT circles in Figure 6. Stars with [Fe/H] ≥ −2.2 (Ca- CaT 5. CARBON IN DWARF SPHEROIDAL GALAXIES enhanced) are red triangles. There is no noticeable dif- ference between the two populations. Hence, we confirm Unlike GCs, dSphs contain stars with a range of ages thatcarbondidnotparticipateintheunusualnucleosyn- and metallicities. The more complex stellar populations thesis that affected heavier elements in NGC 2419. complicate the evolution of carbon. First, dSph stars Figure7showsanalternatewaytodisplaythedestruc- need not start their lives with the same C abundance. tion of carbonas red giants ascendthe RGB. The CMD GC starswereformedfromwell-mixedgaspollutedonly color-codesthestars,withcoolercolorscorrespondingto byTypeIIsupernovae(SNe)andAGBstarsinalimited lower [C/Fe]. In general, the plotting color is cooler at massrange. Figure6andEquation2showthatallofthe the top of the RGB compared to less luminous stars. In pre-bumpstarswithinaGChaveroughlythesamevalue M68 and M15, which are the GCs observed to sufficient of [C/Fe] ∼ +0.12. On the other hand, the progenitors depth, most of the stars below the luminosity function ofdSphstarscanhavearangeofmetallicities. Theycan bump(I ∼14)arecoloredgreen. Thiscolorcorresponds also form from inhomogeneously mixed gas polluted by 0 to the constant pre-bump value of [C/Fe]=+0.12. SNe of various types and AGB stars of various masses

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.