Lithium-Rich Giants in Globular Clusters

1.5 MB·

by  Evan N. Kirby

Accepted to ApJ on 2016January14
PreprinttypesetusingLATEXstyleemulateapjv.5/2/11
LITHIUM-RICH GIANTS IN GLOBULAR CLUSTERS*
Evan N. Kirby1, Puragra Guhathakurta2, Andrew J. Zhang3,4, Jerry Hong5, Michelle Guo4,6, Rachel Guo6,
Judith G. Cohen1, Katia Cunha7,8
Accepted to ApJ on 2016 January 14
ABSTRACT
Although red giants deplete lithium on their surfaces, some giants are Li-rich. Intermediate-mass
6
1 asymptoticgiantbranch(AGB)starscangenerateLithroughthe Cameron–Fowlerconveyor,butthe
0 existence of Li-rich, low-mass red giant branch (RGB) stars is puzzling. Globular clusters are the
2 best sites to examine this phenomenon because it is straightforwardto determine membership in the
cluster and to identify the evolutionary state of each star. In 72 hours of Keck/DEIMOS exposures
n
in 25 clusters, we found four Li-richRGB and two Li-richAGB stars. There were 1696 RGB and 125
a
AGB stars with measurements or upper limits consistent with normal abundances of Li. Hence, the
J
frequencyofLi-richnessinglobularclustersis(0.2±0.1)%fortheRGB,(1.6±1.1)%fortheAGB,and
5
(0.3±0.1)% for all giants. Because the Li-rich RGB stars are on the lower RGB, Li self-generation
1
mechanismsproposedto occur atthe luminosity function bump orHe coreflashcannotexplainthese

four lower RGB stars. We propose the following origin for Li enrichment: (1) All luminous giants
]
R experience a brief phase of Li enrichment at the He core flash. (2) All post-RGB stars with binary
companionsonthe lowerRGB willengageinmass transfer. This scenariopredictsthat 0.1%of lower
S
RGB stars will appear Li-rich due to mass transfer from a recently Li-enhanced companion. This
.
h frequency is at the lower end of our confidence interval.
p Subject headings: stars: abundances — stars: chemically peculiar — stars: evolution — globular
-
clusters
o
r
t
s 1. INTRODUCTION nificantly below the primordial value, A(Li) = 2.72
a
(Coc et al. 2012). Although the factor of 2–4 drop
[ Lithium was created in the Big Bang at a concen-
in Li abundance from the primordial value to the
tration of about 0.5 parts per billion (Coc et al. 2012).
2 Spite plateau has been attributed to atomic diffusion
Since then, many of the Universe’s Li nuclei have been
v and turbulent transport below the convection zone on
destroyed in nuclear burning because Li is susceptible
5 the main sequence (Richard et al. 2005; Mel´endez et al.
to proton capture at relatively low temperatures (T &
1 2010)andconvectiveovershootonthepre-mainsequence
2.5×106 K). Li burning occurs in the centers of stars,
3 (Fu et al. 2015), models of rotationally induced mix-
1 buttheir surfacesarecoolenoughtopreserveLi. There- ing (e.g., Pinsonneault et al. 1989) offer an explanation
0 fore, Li is observable only in stars with outer envelopes
with less fine tuning. Pinsonneault et al. (1992, 1999,
. that haveneverbeen fully mixed downto high tempera-
1 2002) showed that calibrating mixing parameters to the
tures.
0 Sun also explains Li depletion in other stars, includ-
The atmospheres of most old, metal-poor stars on
6 ing the mean and dispersion of the Li abundance on
the main sequence display the same amount of Li
1 the Spite plateau. In addition, the rotation models
(Spite & Spite 1982). This value, A(Li) ∼ 2.2, is
: also explain the behavior of other light elements, like
v called the Spite plateau.9 However, the plateau is sig-
Be and B (de la Reza et al. 1997; Deliyannis et al. 1998;
i
X Boesgaardet al. 2005).
r *ThedatapresentedhereinwereobtainedattheW.M.Keck In metal-rich stars, mixing more efficiently de-
a Observatory,whichisoperatedasascientificpartnershipamong pletes surface lithium than in metal-poor stars (e.g.,
the California Institute of Technology, the University of Cali- Mel´endez et al. 2014; Tucci Maia et al. 2015). Further-
forniaandtheNational AeronauticsandSpaceAdministration.
more,novae cangenerate Li for metal-rich, PopulationI
The Observatory was made possible by the generous financial
stars(Romano et al.1999;Tajitsu et al.2015;Izzo et al.
supportoftheW.M.KeckFoundation.
1CaliforniaInstituteofTechnology, 1200E.CaliforniaBlvd., 2015). As a result, the Spite plateau breaks down at
MC249-17,Pasadena,CA91125,USA [Fe/H] & −1.2 (e.g., Chen et al. 2001). The constancy
2UCO/Lick Observatory and Department of Astronomyand
of Li on the Spite plateau makes Li anomalies in metal-
Astrophysics, University of California, 1156 High St., Santa
Cruz,CA95064,USA poor stars readily apparent. For example, some carbon-
3TheHarkerSchool,500SaratogaAve.,SanJose,CA95129, rich stars show deficiencies in Li that can be explained
USA by mass transfer from a binary, Li-depleted companion
4Stanford University, 450 Serra Mall, Stanford, CA 94305,
(Masseron et al. 2012).
USA
5PaloAltoHighSchool,50EmbarcaderoRd.,PaloAlto,CA, However, it is more difficult to explain stars that
94301, USA are anomalous for being enhanced in Li. This is es-
6Irvington High School, 41800 Blacow Rd., Fremont, CA pecially true for giant stars. Stars at the main se-
94538, USA
7Observato´rioNacional,S˜aoCrist´ova˜oRiodeJaneiro,Brazil quence turn-off experience a rapid drop in Li abun-
8UniversityofArizona,Tucson,AZ85719,USA
9A(Li)=12+logn(Li)/n(H),wheren(Li)isthenumberdensity ofLiatomsandn(H)isthenumberdensityofHatoms.
2 Kirby et al.
dance(Pilachowski et al.1993;Ryan & Deliyannis1995; calledthis non-standardphenomenonextra deep mixing
Lind et al. 2009b). As a low-mass star evolves on combined with “cool bottom processing.” The mecha-
to the red giant branch (RGB), its surface convec- nism for the mixing remains elusive.
tion zone deepens enough to dredge up material that Nonetheless, Li-rich red giants do exist. Kraft et al.
has been processed through nuclear fusion, including (1999) discovereda luminous red giant with A(Li)=3.0
Li burning. Although those regions are no longer hot in the globular cluster (GC) M3. The star is unremark-
enough to burn Li, they were once hot enough to do able except for having over 1000 times more Li than it
so. Hence, the dredge-up brings up Li-depleted ma- should have, based on its position on the RGB. Other
terial while Li on the surface is subducted into the GCs with Li-rich giants include M5 (in a post-AGB
star. The dredge-up dilutes the surface Li abundance Cepheid, Carney et al. 1998), NGC 362 (Smith et al.
to 5–10% of its original value. Models of dilution 1999; D’Orazi et al. 2015b), and M68 (Ruchti et al.
causedby the dredge-up(Deliyannis et al.1990)explain 2011). Kumar & Reddy (2009) and Kumar et al. (2011)
the surface abundance of Li as a function of the sub- found over a dozen Li-rich field K giants around solar
giant’s increasing luminosity or decreasing temperature. metallicity. They also found tentative evidence for clus-
When the star reaches a luminosity of M ∼ 0, the teringofLi-richgiantsaroundtheredclump,orhorizon-
V
hydrogen-burning shell expands beyond the molecular tal branch (HB), where stars burn helium in their cores
weightboundaryestablishedbythefirstdredge-up(Iben after the He core flash at the tip of the RGB. The idea
1968). “Extra” mixing—possibly thermohaline mixing thattheHecoreflashcouldactivatetheCameron–Fowler
(Charbonnel & Zahn2007;Charbonnel & Lagarde2010; conveyor was bolstered by Silva Aguirre et al.’s (2014)
Denissenkov 2010; Wachlin et al. 2011; Angelou et al. discoveryofaLi-richHBstarwhoseHecoreburningwas
2012; Lattanzio et al. 2015)—beyond the canonical stel- confirmedbyasteroseismologicalmeasurementsfromthe
lar model changes the surface composition for stars at Kepler spacecraft (Gilliland et al. 2010). Monaco et al.
the RGB bump. This mixing rapidly destroys any Li (2014) also discovered a Li-rich, HB star in the open
remaining in the red giant’s atmosphere. cluster Trumpler 5, and Anthony-Twaroget al. (2013)
Nonetheless, some giants are Li-rich (see found a Li-rich giantin the open cluster NGC 6819 that
Wallerstein & Conti 1969). Cameron (1955) and is too faint to be on the HB or AGB. This star is par-
Cameron & Fowler (1971) suggested a mechanism (the ticularly interesting for showing asteroseismic anomalies
“Cameron–Fowler conveyor”) for producing excess Li that could indicate rotationally induced mixing, which
in the atmospheres of giant stars. The central nuclear in turn could generate Li (e.g., Denissenkov 2012). In-
processes for the conveyor comprise the pp-II chain of deed, Carlberg et al. (2015) found that the star is rotat-
hydrogen burning. ing rapidly for a red giant, but puzzlingly, they did not
find any additional evidence for deep mixing.
Metal-rich stars can have a complicated evolution
p+p → d+e++ν (1)
e of Li, as illustrated by the ∼ 1.5 dex scatter in
d+p → 3He+γ (2) Li abundance—even at fixed effective temperature—in
3He+4He → 7Be+γ (3) Delgado Mena et al.’s (2015) survey of lithium in open
clusters. Surveys for Li enhancement among metal-poor
7Be+e− → 7Li+νe (4) starscanbeeasiertointerpret. InspiredbyKraft et al.’s
7Li+p → 24He (5) (1999) discovery of a Li-rich giant in a metal-poor GC,
Pilachowskiet al. (2000) surveyed 261 giants in four
Reaction 3 is very active (compared to the pp-I chain) metal-poorGCs,buttheyfoundnoLi-richgiants. There-
at temperatures around 2×107 K. Li destruction, reac- fore, the frequency of Li-rich red giants in GCs is less
tion 5, is very efficient at T & 2.5×106 K. Hence, 7Li than 0.4%. D’Orazi et al. (2014, 2015a) also surveyed
will be destroyed as soon as it is created in reaction 4 red giants in GCs and found one Li-rich giant out of
unless 7Be canbe broughtto coolertemperaturesbefore about 350 giants, corresponding to a Li-rich frequency
it captures an electron. Although the half-life for reac- of (0.3±0.3)%. Ruchti et al. (2011) searchedfor Li-rich
tion 4 is 53 days under terrestrial conditions, Cameron giants in the Milky Way (MW) halo in the Radial Ve-
(1955) theorized that the scarcity of bound K-shell elec- locity Experiment (RAVE, Steinmetz et al. 2006). They
trons available for reaction 4 at T > 106 K, where 7Be found eight Li-rich giants out of 700 metal-poor field
is almostentirely ionized, extends the half-life to 50–100 giants. They also found one Li-rich giant in the GC
years. M68. Dom´ınguez et al. (2004) and Kirby et al. (2012)
The mixing that accompanies thermal pulses in alsofound15Li-richgiantsinMWdwarfsatellite galax-
intermediate-massstarsonthesecond-ascentasymptotic ies. However, the MW field and dwarf galaxies are not
giant branch (AGB) is deep enough to reach the pp-II amenable to easily distinguishing between the AGB and
burningzone. Asaresult,theCameron–Fowlerconveyor upper RGB. In fact, many of the Li-rich giants discov-
is a plausible explanation for Li-rich AGB stars in the eredbyRuchti et al.(2011)andKirby et al.(2012)could
mass range 4–7 M⊙ (Sackmann & Boothroyd 1992). In be AGB stars.
fact, Li-rich AGB stars are not uncommon (Plez et al. Explanations for Li-rich RGB stars fall into three
1993; Smith et al. 1995). However, the convective en- categories: engulfment of a substellar companion,
velopes of first-ascent RGB stars and less massive AGB self-generation, and mass transfer. In the engulfment
stars are not deep enough to activate the conveyor. Any scenario (e.g., Siess & Livio 1999; Denissenkov & Weiss
excessLiinRGBstarsmustbearesultofprocessesout- 2000; Melo et al. 2005; Villaver & Livio 2009;
side of “standard” stellar evolution of single stars with Adam´ow et al. 2012), a red giant expands into the
ordinary rotation rates. Sackmann & Boothroyd (1999) orbit of a rocky planet, a hot Jupiter, or a compan-
Li-Rich Giants 3
ion brown dwarf. The destroyed companion could RGB or AGB. Section 2 describes our observations,and
potentially enrich the star with Li and other volatile Section 3 details the measurement of Li and other spec-
elements that concentrate in planets (Carlberg et al. troscopic properties. In Section 4, we define what it
2013). Even if the engulfed companion does not donate means to be “Li-rich” and quantify the statistics of Li-
Li to its host star, it would provide angular momentum. richgiantsin GCs. We addressthe possible originsof Li
The resulting increase in rotation rate could itself enhancement in Section 5, and we summarize our con-
inspire deep mixing that activates the Cameron–Fowler clusions in Section 6.
conveyor (Denissenkov & Herwig 2004).
2. SPECTROSCOPIC OBSERVATIONS
In the self-generation scenario, stars can experience
deep mixing events that dredge Li to the stellar sur- Weobserved25GCswithKeck/DEIMOS(Faber et al.
face, where it is observable. Rotationally induced mix- 2003)overeightyears. Table1liststheclustersandtheir
ing is one example. Indeed, some Li-rich giants are coordinates. Someoftheseslitmaskswereobservedwith
rapid rotators (Drake et al. 2002; Guillout et al. 2009; the purpose of validating a method to measure metal-
Carlberg et al. 2010), but others are not (Ruchti et al. licities and α element abundances from DEIMOS spec-
2011). Other possible causes are mixing at the RGB tra. Kirby et al. (2008, 2010) previously published these
luminosity function bump (Charbonnel & Balachandran observations. Most of the remaining slitmasks were de-
2000) or deep mixing inspired by He core flashes at signed expressly to search for Li-rich red giants.
the tip of the RGB or on the HB (Kumar et al. 2011;
2.1. Source Catalogs
Silva Aguirre et al. 2014; Monaco et al. 2014). For ex-
ample,D’Orazi et al.(2015b)foundaLi-richgiantinthe We used custom slitmasks designed to observe giant
GC NGC 362thatmaybe either atthe RGB bump(hy- starsintheclusters. Inordertodesigntheslitmasks,we
drogen shell burning) or on the red clump (helium core used photometric catalogs from various sources.
burning). On the other hand, Anthony-Twarog et al.’s Our primary source of photometry was P.B. Stetson’s
(2013)Li-richgiantinNGC6819isonecounter-example databaseofphotometricstandardfields. Wedownloaded
below the RGB bump. The chemical analysis of that some of these from Stetson’s public web page, but he
starbyCarlberg et al.(2015)foundnoevidencefordeep provided some of these catalogs to us privately (see
mixinginanyelementotherthanLi. Furthermore,most Kirby et al. 2010). Several of these clusters were also
deep mixing scenarios predict that the Li-rich giants previously published (Stetson 1994, 2000). These cata-
would cluster at a specific evolutionary phase (luminos- logs were made with DAOPHOT (Stetson 1987, 2011),
ity). However, Lebzelter et al. (2012) found no luminos- whichmodelsthepointspreadfunctions(PSFs)ofstars.
ity clustering of Li-rich red giants. This approach performs better than aperture photome-
Finally, stars can alter their surface compositions try in crowded fields, like GCs.
through binary mass transfer. AGB stars are known Some of Stetson’s clusters had dense sampling over
to generate carbon and neutron-capture elements, like a field comparable in size to a DEIMOS slitmask. In
barium (Busso et al. 1995). Hence, binary companions these cases, we relied on his photometry alone. The cat-
to AGB stars or former AGB stars can be enhanced alogsforotherclusters sampledonly tens orhundredsof
in those elements (McClure et al. 1980). Intermediate- stars for the purposes of providing a photometric cal-
mass AGB stars can also dredge up Li in the Cameron– ibration field rather than a science catalog. In these
Fowler conveyor. Even less massive AGB stars might be cases,wesupplementedStetson’sphotometrywithother
able to generate Li with the help of thermohaline mix- sources. Table 1 lists the source catalogs for each clus-
ing (Cantiello & Langer 2010). If the star transferred ter. Notable sources include the Sloan Digital Sky Sur-
mass to a companion during a phase of Li dredge-up, vey (SDSS, Abazajian et al. 2009) and An et al. (2008).
then that companionwouldbe enhancedin Li. This is a Because the primary SDSS catalog uses aperture pho-
possible explanationfora Li-richturn-offstarin the GC tometry, An et al. (2008) re-reduced the photometry of
NGC6397(Koch et al.2011;Pasquini et al.2014). That select GCs with DAOPHOT.
starwill remainenhancedin Liuntilthe firstdredge-up. Allofthecatalogsusedhavecoverageinatleasttwoof
Assuming that the dredge-up dilutes a fixed percentage the three filters B, V, and I. We correctedthe observed
of Li for all stars of similar mass and composition, then magnitudes for extinction according to the dust maps of
the star would still appear Li-enhanced relative to other Schlegel et al. (1998).
post-dredge-up stars in the cluster.
2.2. Target Selection
GCsarethebestenvironmentstostudylow-massstel-
lar evolution. The common distance to all the member We designedthe slitmasks with a minimum slit length
stars makes it easy to determine stellar luminosity. The of4′′ andseparationbetweenslitsof0′.′35. Thesechoices
commonage andsmallabundance dispersionin mostel- allowedjustenoughseparationbetweenstarsto(1)avoid
ements implies a similar evolution for all stars. To first overlappingspectra and(2)permit skysubtractionfrom
order, a GC is a snapshot of stellar evolution over a se- the empty portions of the slits. However, these restric-
quence of stellar masses at fixed age and mostly fixed tions also forced us to choose among the many stars in
metallicity. With reasonably attainable photometric un- the dense GCs. Although several hundred GC giants
certainty,theAGBandRGBcanbedistinguishedwitha might have been visible in a single DEIMOS pointing,
color–magnitude diagram (CMD) except for the bright- theslitmaskwouldallowonlyabout150targetsatmost.
estgiants,where the AGB nearlymergeswith the RGB. Wedevelopedtargetselectionstrategiestopickoutlikely
We exploitedthe controlledstellarpopulations ofGCs giant members of the GCs.
to study the phenomenon of Li-richgiants. We searched Because the 75 slitmasks were designed for different
for Li-rich giants and classified them photometrically as projects over many years, the target selection strategy
4 Kirby et al.
Table 1
GlobularClustersObserved
GC RA(J2000) Dec(J2000) SourceCatalogs
NGC288 005245 −263457 Stetson; Bellazzinietal.(2001)
Pal2 044605 +312253 Stetson
NGC1904(M79) 052411 −243128 Stetson; Rosenbergetal.(2000)
NGC2419 073808 +385256 Stetson (2000)
NGC4590(M68) 123927 −264438 Stetson; Walker(1994)
NGC5024(M53) 131255 +181005 Stetson; Anetal.(2008)
NGC5053 131627 +174200 Stetson; Anetal.(2008)
NGC5272(M3) 134211 +282238 Stetson (2000)
NGC5634 142937 −055835 Stetson; Bellazzinietal.(2002)
NGC5897 151724 −210036 Stetson; Testaetal.(2001)
NGC5904(M5) 151833 +020451 Stetson (2000);Anetal.(2008)
Pal14 161100 +145727 Sahaetal.(2005)
NGC6205(M13) 164141 +362735 Stetson
NGC6229 164658 +473139 SDSS
NGC6341(M92) 171707 +430809 Stetson (2000);Anetal.(2008)
NGC6656(M22) 183623 −235417 Stetson; Peterson&Cudworth(1994)
NGC6779(M56) 191635 +301100 Hatzidimitriouetal.(2004)
NGC6838(M71) 195346 +184645 Stetson
NGC6864(M75) 200604 −215516 Kravtsovetal.(2007)
NGC7006 210129 +161114 Stetson (2000);Anetal.(2008)
NGC7078(M15) 212958 +121001 Stetson (1994,2000)
NGC7089(M2) 213327 −004923 Stetson (2000);Anetal.(2008)
NGC7099(M30) 214022 −231047 Stetson; Sandquistetal.(1999)
Pal13 230644 +124619 Stetson
NGC7492 230826 −153641 Stetson
References. — Cluster coordinates are from the compilation of
Harris (1996, updated 2010) and references therein. “Stetson” refers to
photometry by P.B. Stetson. Most of the photometry is available at
http://www2.cadc-ccda.hia-iha.nrc-cnrc.gc.ca/community/STETSON/standards/,
but Stetson provided some of it directly to us. “SDSS” refers to photometry from the
SloanDigitalSkySurvey(Abazajianetal.2009).
was not uniform. Although most masks were designed GCs are excellent laboratories to study stellar evolu-
for giants, some included main sequence stars. In gen- tion because they are nearly single-age populations of
eral,selectionalongthe RGB wasperformedby defining nearly uniform metallicity.10 For these reasons,GCs are
selection regions in the CMD. In some cases, where the the best stellar populations for distinguishing between
RGB was well-defined and distinct from the foreground, theAGBandtheRGB. Thisdistinctionhelpsdetermine
we drew an irregular polygon around the RGB and se- how evolutionary phase plays a role in Li-richness.
lected stars inside of it. In other cases, we drew an old Althoughmodelisochronescouldbeusedforthistask,
(∼12Gyr)isochronecorrespondingto the metallicity of wefoundthatsmallimperfectionsinthemodelsresulted
the cluster (Harris 1996, updated 2010). We used both inmisidentificationathighstellarluminosities,wherethe
Victoria–Regina (VandenBerg et al. 2006) and Yonsei– AGB asymptotically approaches the RGB. Instead, we
Yale (Demarque et al. 2004) isochrone models. The se- identifiedthe AGB“byeye.” We drewaselectionregion
lection region was defined within a color range (typi- around the AGB for each GC. AGB stars are shown in
cally0.1mag)aroundtheisochrone. Formostslitmasks, blue in Figure 1. RGB stars are shown in red.
brighter stars were given higher priority for selection.
The target selection favored first-ascent RGB stars 2.4. Observations
rather than helium-burning stars on the HB or AGB.
Table 2 lists the observing log, including the slitmask
The HB was particularly disfavored because the spec-
name,the numberoftargetsonthe slitmask,the date of
tra of hot, blue stars do not readily lend themselves to
observation,the airmassandseeingatthe time ofobser-
themeasurementofradialvelocityandmetallicity,which
vation, the number of exposures, and the total exposure
wasthe originalintentformanyofthe slitmasks. There-
fore, this data set is not ideal to search for Li-richness 10 ThereisextensiveobservationalevidencethatGCsarechem-
on the HB. However, it is suitable for quantifying the icallycomplex(e.g.,Grattonetal.2004,2012). Inparticular,pri-
frequency of Li-rich giants on the RGB or upper AGB. mordial intracluster variation in certain elements, such as O, Na,
Mg,andAl,indicatesthatclusterstarsweredifferentiallyenhanced
Figure1showstheextinction-andreddening-corrected
withtheproductsofhigh-temperaturehydrogenburning. Liisnot
CMDs for all of the GCs in our sample. M53 and immune to the primordial variation, as exhibited by weak Li–Na
NGC 7492 are shown with (B − V)0 color, and all of and Li–Al anti-correlations observed in some clusters (Lindetal.
the other GCs are shown with (V −I) . Stars that we 2009b; Monacoetal. 2012; D’Orazietal. 2015a). Unfortunately,
0
we cannot distinguish between first and later generation stars in
identified asmembers (Section 3.3)areshownas colored
our sample because we cannot observe Na in our spectra. How-
points or black, five-pointed stars. ever,GCsarestillsimpleenough forourpurposes. Specifically, it
isstraightforwardtodistinguishtheAGBfromtheRGB,andthe
2.3. Separation of RGB and AGB heavyelements,likeFe,areinvariantwithineachoftheclustersin
oursample.
Li-Rich Giants 5
NGC 288 Pal 2 M79 NGC 2419 M68
−2 −2 −2 −2 −2
0 0 0 0 0
0
MV,
RGB
2 2 2 2 2
AGB
Li−rich
non−member
not observed
4 4 4 4 4
0.6 1.0 1.4 0.6 1.0 1.4 0.6 1.0 1.4 0.6 1.0 1.4 0.6 1.0 1.4
(V − I) (V − I) (V − I) (V − I) (V − I)
0 0 0 0 0
M53 NGC 5053 M3 NGC 5634 NGC 5897
−2 −2 −2 −2 −2
0 0 0 0 0
0
MV,
2 2 2 2 2
4 4 4 4 4
0.6 1.0 1.4 0.6 1.0 1.4 0.6 1.0 1.4 0.6 1.0 1.4 0.6 1.0 1.4
(B − V) (V − I) (V − I) (V − I) (V − I)
0 0 0 0 0
M5 Pal 14 M13 NGC 6229 M92
−2 −2 −2 −2 −2
0 0 0 0 0
0
MV,
2 2 2 2 2
4 4 4 4 4
0.6 1.0 1.4 0.6 1.0 1.4 0.6 1.0 1.4 0.6 1.0 1.4 0.6 1.0 1.4
(V − I) (V − I) (V − I) (V − I) (V − I)
0 0 0 0 0
Figure 1. Color–magnitude diagramsforall25GCs observedwithDEIMOS. Thepanel forNGC2419 includes afigurelegend. Li-rich
stars are shown as black, five-pointed stars. The hollow star indicates the Li-rich giant IV–101 in M3 (Kraftetal. 1999), which is not
partof our sample. Spectroscopically confirmed, Li-normalmembers areshown as red(RGB) and blue(AGB) points. Non-members are
shownasblackcrosses. GraypointsshowstarsthatwedidnotobservewithDEIMOS. WedistinguishedbetweenRGBandAGBstarsby
drawingselectionregionsintheCMDs. Figure5showsdetailofthegrayboxesaroundtheLi-richstars.
6 Kirby et al.
M22 M56 M71 M75 NGC 7006
−2 −2 −2 −2 −2
0 0 0 0 0
0
MV,
2 2 2 2 2
4 4 4 4 4
0.6 1.0 1.4 0.6 1.0 1.4 0.6 1.0 1.4 0.6 1.0 1.4 0.6 1.0 1.4
(V − I) (V − I) (V − I) (V − I) (V − I)
0 0 0 0 0
M15 M2 M30 Pal 13 NGC 7492
−2 −2 −2 −2 −2
0 0 0 0 0
0
MV,
2 2 2 2 2
4 4 4 4 4
0.6 1.0 1.4 0.6 1.0 1.4 0.6 1.0 1.4 0.6 1.0 1.4 0.6 1.0 1.4
(V − I) (V − I) (V − I) (V − I) (B − V)
0 0 0 0 0
Figure 1. — continued —
Li-Rich Giants 7
time. The number of targets is the number of science 1.1
slitlets in the mask (excluding alignment boxes). It is 1.0
not the number of stars in the final sample. In addition 0.9
M68 Stet−M68−S232
to member giants,the slitmasks included main sequence 0.8 T = 4462 K log g = 0.98
stars as well as non-members. 0.7 [Fefef/H] = −2.38 M = −2.01
All slitmasks were observed with the 1200G grating, A(Li) = +3.17 ± 0.1V0,0 AGB
which has a groove spacing of 1200 mm−1 and a blaze 1.04
wavelengthof7760˚A. Theslitwidthsweretypically0′.′7. 1.00

The resulting resolution was 1.2 ˚A, which corresponds 0.96 M68 Stet−M68−S534
lteongathr.esEolavcihngpipxoewl eenrcoofmRpas≈ses6500.303a˚At,tshuechbltahzaetwaarvees-- 0.92 AT[Fe(fefL /=Hi) 5]= 4= 8+ −822 .K4.41 4 ± l o 0Mg. 1gV5, 0= = 3 +.125.27 RGB
olution element spans 3.6 pixels. Slitmasks with the let- 1.05

ter “l” were observedat a central wavelengthof 7500 ˚A. 1.00

O7a8nt0dh0ehr˚Aisg.lhiTtemrh-eaosrOkdsGerw55leir0gehootrbdfsereror-mbvleodccokanittnaagmcfieilnntaetrtriabnllgowctakhveeedlessnpegecctohtnrdoa-f. ed flux 000...899505 ANT[Fe(GfefL /C=Hi) 5]5= 30= 6+5 −7322 .KN7.32 50 0± l o5 0Mg3. −1gVS4, 0=7 = 92 +. 7 1 3 . 31 RGB
We usedDEIMOS’sflexurecompensationsystem,which z
i
provideswavelengthstability ofabout0.03˚Aduring the mal 1.00
observation of one slitmask. Afternoon calibrations in-
r
cludedexposuresofaquartzlampforflatfieldingandan no NGC 5897 Tes01−WF4−703
exposureofNe,Kr,Ar,andXearclampsforwavelength 0.96 T[Fefef /=H 4] 7=7 −31 K.9 9 l o Mg g = = 1 −.700.62
calibration. A(Li) = +1.50 ± 0.1V1,0 AGB
We reduced the DEIMOS spectra with the spec2d 1.05

IDL data reduction pipeline developed by the DEEP2 1.00

team (Cooper et al. 2012; Newman et al. 2013). The
0.95 M30 132
pipeline excises the 2-D spectrum for each slitlet. The
T = 5640 K log g = 3.54
2-D spectrum is flat-fielded and wavelength-calibrated. 0.90 [Fefef/H] = −2.43 M = +3.04
Thewavelengthcalibrationfromthearclampsisrefined 0.85 A(Li) = +2.66 ± 0.1V4,0 RGB
with night sky emission lines. All of the exposures are
1.00
combined,andcosmicraysareremoved. Finally,the1-D

0.96
spectrumisextractedwithoptimalextraction. Thesoft- M30 7229
ware tracks the variance spectrum at every step. The 0.92 T = 5509 K log g = 3.28
result is a flat-fielded, wavelength-calibrated, 1-D spec- A[Fe(fefL/Hi) ]= = + −22.8.372 ± 0M.1V3,0 = +2.49 RGB
trum ofthe targetalongwith anestimate ofthe errorin
each pixel. 6700 6705 6710 6715
Figure 2 shows the spectra of the six giants that we rest wavelength (Å)
determinedtobeLi-richmembersoftheirrespectiveGCs
(see Section 4). Only a small spectral region around the Figure 2. DEIMOS spectra (black) of the six Li-rich giants
around the Liiλ6707 absorption line. Best-fit synthetic spectra
Liiλ6707 line is shown.
are shown in red. The pink lines show synthetic spectra with no
Li. Each panel gives the star’shostcluster, thestar’sname, tem-
3. SPECTROSCOPIC MEASUREMENTS perature, gravity, metallicity, luminosity, and NLTE-corrected Li
abundance.
We measured four important parameters from each
spectrum: radial velocity, vhelio; effective temperature, 3.2. Atmospheric Parameters
T ; metallicity, [Fe/H]; and Li abundance, A(Li).
eff We measured T and [Fe/H] in the same manner as
eff
Kirby et al. (2008, 2010). This section summarizes the
3.1. Radial Velocities procedure. First, we shifted the spectrum into the rest
We measured v in the same manner as frame, removed telluric absorption by dividing by the
helio
Simon & Geha (2007). First, we measured v , the ve- spectrum of a hot star, and divided by the continuum,
obs
locity requiredto shift the spectrum into the rest frame. approximated by a spline with a breakpoint spacing of
Todoso,wecross-correlatedeachspectrumwith16tem- 100 ˚A. Next, we searched for the best-fitting synthetic
plate spectra observed with DEIMOS, kindly provided spectrum among a large grid of spectra computed with
by Simon & Geha. Because of imperfect centering in MOOG(Sneden 1973) and ATLAS9 model atmospheres
the slitlet, a star can have an apparent radial velocity (Kurucz 1993; Sbordone 2005).
with respect to the telluric absorption lines. To coun- WeestimatedinitialguessesatTeff andsurfacegravity,
teractthis error,we cross-correlatedeachspectrum with logg, by comparing the star’s color and magnitude to
a template spectrum of a hot star, which is dominated model isochrones shifted by the distance modulus of its
by telluric absorption. The resulting velocity is vcenter, respectiveGC. Insearchingthe grid, Teff wasallowedto
the velocity required to shift the spectrum into the geo- vary in a range around the photometrically determined
centric frame. We also computed v , the correction value, but logg was fixed at the photometric value. On
corr
required to shift from the geocentric to the heliocen- the other hand, no restrictions were imposed on [Fe/H].
tric frame. The final heliocentric velocity of the star is Wemade“firstdraft”measurementsofTeff and[Fe/H]
v =v +v +v . by minimizing χ2 between the observed and synthetic
helio obs center corr
8 Kirby et al.
spectra. We refined these measurements by using the
best-fit synthetic spectrum to improve the continuum Table 3
LineList
determination. We repeatedthis iterative continuumre-
finement until T and [Fe/H] changed by a negligible
eff Wavelength (˚A) Species EP(eV) log(gf)
amount between iterations. The values of T and logg
eff
··· ··· ··· ···
atthe endofthe lastiterationwereregardedasthe final
6707.752 Sci 4.049 −2.672
measurements.
6707.7561 7Lii 0.000 −0.4283
6707.7682 7Lii 0.000 −0.2062
3.3. Membership 6707.771 Cai 5.796 −4.015
6707.799 CN 1.206 −1.967
Weconsideredonlystarsthataremembersofoursam- 6707.9066 7Lii 0.000 −1.5086
ple of GCs for the purposes of this project. Our mea- 6707.9080 7Lii 0.000 −0.8069
surements of atmospheric parameters are valid only for 6707.9187 7Lii 0.000 −0.8069
member stars because we used model isochrones to esti- 6707.9196 6Lii 0.000 −0.4789
mate T and logg. The measurements are not valid for 6707.9200 7Lii 0.000 −0.8069
eff ··· ··· ··· ···
mon-member stars at unknown distances.
First,weremovedduplicatespectra. Whereastarwas
observed multiple times on different slitmasks, we kept References. — Lithium lines are from
Hobbsetal. (1999). Other lines are from
the measurement with the lowest estimate of error on
Kirbyetal.(2008),whichiscompilationofatomic
[Fe/H],whichisessentiallyaS/Ncriterion. We removed linesfromVALD(Kupkaetal.1999)andmolecu-
437 duplicate spectra. larlinesfromKurucz(1993).
Second, we eliminated any stars that were obviously Note. —Wavelengths areinair. (Thistableis
non-members or non-giants based on their positions in availableinitsentiretyinamachine-readableform
theCMD. Althoughtheslitmasksweredesignedtoavoid in the online journal. A portion is shown here for
guidance regardingitsformandcontent.)
non-members, some obvious non-members were placed
on the slitmask merely to fill it with targets. We drew a
generous CMD selection region around the stellar locus Other lines are from Kirby et al.’s (2008) compilation
and flagged stars outside of the region as non-members. fromVALD(forneutralandionizedatoms,Kupka et al.
Figure 1 shows some of these non-members as crosses. 1999)andKurucz(formolecules,1993). TheLilinesare
We alsoeliminated non-giantstars by imposing a cuton separated by isotope (6Li and 7Li).
surface gravity: logg <3.6. We prepared the spectrum by performing a local con-
Third, we restricted the member list on the basis of tinuum correction around Liiλ6707. We used MOOG
radialvelocity. Weestimatedthecluster’smeanvelocity, and Kirby’s (2011) grid of ATLAS9 model atmospheres
hvhelioi, and velocity dispersion, σv, by calculating the to compute a synthetic spectrum devoid of Li. The at-
meanvelocityofallstarswithin40kms−1 ofthemedian mospheric parameters (T , logg, [Fe/H]) were tailored
eff
velocity. We compiled a list of all stars that satisfied to each star following the procedure in Section 3.2. The
|v −hv i|<2.58σ (99%ofallstarsina Gaussian microturbulent velocity (ξ) was calculated based on a
helio helio v
velocity distribution). From this list, we re-computed calibration between ξ and logg (Kirby et al. 2009). We
hv i and σ . The member list includes only those divided the observed spectrum by this model. We fit
helio v
stars that have |v −hv i|−δv <3σ , where δv is a straight line with variable slope and intercept to the
helio helio v
the uncertainty on the radial velocity. In other words, residual in the wavelength range 6697–6717 ˚A, but ex-
any star whose 1σ velocity error bar overlapped the 3σv cluding the Lidoublet (6705.7–6709.9˚A). This linearfit
membership cut was allowed as a member. Although comprised the local continuum correction, by which we
the different criteria for stars used in the computation divided the observed spectrum.
of σv versus the member list may seem capricious, we WemeasuredA(Li)intheobservedspectrumbymini-
found from examining the velocity histograms that this mizingχ2betweenthecontinuum-refined,observedspec-
procedure reliably identified stars in the GC’s velocity
trumanda modelspectrum. The only freeparameterin
peak. thefitwasA(Li). Weminimizedχ2 withtheLevenberg–
Finally, we restricted the member list on the ba-
MarquardtIDL code MPFIT (Markwardt2012). This re-
sis of [Fe/H]. The procedure was nearly identical to
quiredcomputing manyspectralsyntheseswithMOOG,
the velocity membership criterion. The mean metallic-
whichwe didinthe samemanneras forthe Li-freespec-
ity, h[Fe/H]i, and metallicity dispersion, σ([Fe/H]), were
trum described in the previous paragraph.
computedfromallstarsinthecluster. Then,thesevalues We set the 7Li/6Li isotopic ratio to 30. Although
were re-computed from a more restricted list: |[Fe/H]− Liiλ6707 spectra modeled with 3D, NLTE model at-
h[Fe/H]i| < 2.58σ([Fe/H]) and [Fe/H] < −0.5. With mospheres show no detectable 6Li (Lind et al. 2013),
these refined values, the final membership list was those
a 7Li/6Li ratio of ∼ 30—while not an accurate rep-
stars with |[Fe/H]−h[Fe/H]i|−δ[Fe/H] < 3σ([Fe/H]),
resentation of the atmospheric composition—gives the
where δ[Fe/H] is the uncertainty on [Fe/H].
best-fitting line shape in a 1D, LTE spectral synthesis
(Smith et al. 1998), such as ours. Our results are nearly
3.4. Li Abundance
insensitive to this choice because the resolution of our
We measured Li abundances by spectral synthesis of spectra is smaller than the isotopic splitting of the Li
the Liiλ6707 doublet. We compiled a line list (Table 3) doublet.
of absorption lines in the region 6697–6717 ˚A. The Li Wetookthe1σerroronA(Li)tobethevaluebywhich
absorption lines come from Hobbs et al.’s (1999) list. A(Li)neededtochangeinordertoraiseχ2 by1fromthe
Li-Rich Giants 9
minimum χ2. For spectra with S/N > 300 pixel−1, this the firstdredge-up). As aresult, astar withA(Li)=1.1
estimate of the error could be even smaller than 0.01, and M =−2 would be Li-rich.
V
which is unrealistically low because it does not account Lind et al. (2009b) conducted the definitive study of
for systematic error, such as imperfections in the spec- Li in GC stars. They measured A(Li) for hundreds of
tral model. We imposed a minimum error of 0.1 dex by starsinthe metal-poorGCNGC 6397fromR=14,000,
adding 0.1 dex in quadrature with the statistical error. high-S/N VLT/FLAMES spectroscopy. Figure 3 shows
MostofthestarshadnodetectableLi. Forthesestars, their measurements in red. The main sequence stars at
theχ2contourflattenedtoaconstantvalueatlowA(Li). M > +3.3 have a constant A(Li) = 2.3. The first
V
We computed 2σ upper limits as the value of A(Li) cor- dredge-up begins at M = +3.3 and depletes A(Li) to
V
responding to anincrease in χ2 of 4 abovethe minimum 1.1. The Liremainsbriefly untoucheduntil the luminos-
χ2. We found this value using a truncated Newton min- ity function bump at MV = 0.0, which further depletes
imization method.11 Li to an undetectable level.
We examinedthe spectrumofeveryLidoublettocon- Our DEIMOS spectra have lower spectral resolution
firm that the measurement of A(Li) or its upper limit than Lind et al.’s FLAMES spectra. Consequently, we
is valid. We plotted the best-fitting synthetic spectrum did not detect Li in the majority of our stars. Those
over the continuum-corrected observed spectrum. If the stars with detections also have larger A(Li) uncertain-
fit appeared to fail, then we removedthe spectrum from ties than the FLAMES measurements. The DEIMOS
oursample. Commonreasonsfor failure included single- detections tend to be for stars with largerA(Li) at fixed
pixelnoisespikes(possiblyduetocosmicrays)andbadly MV than the FLAMES detections. That is why most
placed continuum measurements due to spectral arti- of our detections of Li trace the upper envelope of the
facts. We also flagged every spectrum with a convinc- NGC6397data. We havedetectedLionlyinthosestars
ing detection of Li. Although we technically measured with upward fluctuations in A(Li) due to intrinsic vari-
A(Li) for every spectrum, we present upper limits for ation in the cluster or, more likely, random noise in the
those spectra with unconvincing detections. DEIMOS spectra.
Lind et al. (2009a) computed corrections to A(Li) to We also quantified what it means to be Li-normal by
counteractdeficienciesfromtheassumptionoflocalther- coadding DEIMOS spectra of RGB stars in six bins of
modynamic equilibrium (LTE) in computing synthetic MV. The least luminous bin was MV > +2, and the
spectra. The non-LTE (NLTE) correction depends on most luminous bin was MV < −2. The other four bins
the LTE lithium abundance and stellar parameters, like were 1 mag wide in the range +2 > MV > −2. In each
T and logg. Lind et al. provided convenient tables to bin, we coadded all RGB spectra (excluding the AGB)
eff
compute NLTE corrections for most cool stars. All of thatdonotsatisfytheLi-richcriterion(Equation6). We
the values of A(Li), including upper limits, in the text, did not include spectra that we identified in Section 3.4
figures, and tables in this paper have these NLTE cor- to be problematic. The spectra were interpolatedonto a
rectionsapplied. Welinearlyextrapolatedthecorrection commonwavelengtharrayandcoaddedwithinversevari-
for stars with stellar parameters outside of the range of ance weighting. We also averaged MV and atmospheric
Lind et al. (2009a)’s tables. parametersin eachbin, weighting by the medianinverse
Table 4 gives Li measurements or 2σ upper limits for variance within 10 ˚A of Liiλ6707. We measured A(Li),
our sample. The table also identifies whether the star treatingthecoaddedspectrumasasinglespectrumwith
fellintheRGBorAGBselectionwindow. Non-members a single T , logg, and [Fe/H], which were fixed at the
eff
andstarsthatwereremovedfromthesampleuponvisual weighted averagevalues for all the spectra in the bin.
inspection are not shown in the table. The table gives Figure 4 shows the coadded spectra. Most absorption
the six Li-rich giants first, followed by all other stars in lines become stronger with increasing luminosity (note
order of right ascension. The photometric magnitudes the increasing y-axis range) because T decreases with
eff
and colors are corrected for extinction and reddening. increasing luminosity on the RGB. However, Liiλ6707
becomes weaker because Li is depleted with decreasing
4. LI ENHANCEMENT Teff. Figure 3 compares our coaddition measurements of
hA(Li)i (green) to individual stars in NGC 6397 (red).
In this section, we quantify the number of Li-rich gi-
ExceptforthehM i=−0.4bin, thegreenpointslie in
antsinoursample. Todoso,weestablishaquantitative V,0
themidstoftheredpoints. The binwithhM i=−0.4
definition for “Li-richness.” We also separate the statis- V,0
tics on Li-richness by stellar evolutionarystate (RGB or shows a likely spurious absorption feature at 6702 ˚A,
AGB). which pushes up the continuum. Therefore, our mea-
surement of A(Li) might be slightly low in this bin.
Thehigh-qualityNGC 6397dataalongwithourcoad-
4.1. Defining “Li-Rich”
ded DEIMOS data define a clear trend of A(Li) with
Inordertodefine“Li-rich,”weexaminewhatitmeans M . We drew a boundary in Figure 3 along the upper
V
to be “Li-normal.” The definition should depend on the envelope of our measurements. The following equation
star’s luminosity because surface Li is progressively de-
defines the boundary:
pletedasthestarascendstheRGB. Forexample,agiant
withA(Li)=1.1andM =+1wouldnotbeLi-rich,but
V
slutamrisnwositithyMfuVnc<tio0nbebguimnpap(ihnasceonotfrLaistdewsittrhucdtiioluntiaotnthaet A(Li)=21..6500+ 3.00−.3M3V +2.7>MMVV ≥>+−20..72 (6)
1.76+0.79M M ≤−0.2
V V
11 TNMIN, an IDL code by C. Markwardt 
(http://cow.physics.wisc.edu/~craigm/idl/idl.html). Six Li detections fall above the boundary. Although
10 Kirby et al.
4
3
2
)
Li
(
A 1
0
RGB, DEIMOS
AGB, DEIMOS
RGB, coadded DEIMOS
−1
RGB, NGC 6397 (Lind et al. 2009)
4 2 0 −2
M
V,0
Figure 3. NLTE-correctedLiabundancesversusabsolutemagnitude. OurDEIMOSdetectionsofLiareshownasblack(RGB)andblue
(AGB)points. Upperlimitsareshowningray(RGB)andfadedblue(AGB). Forcomparison,high-resolutionspectroscopicmeasurements
of Li in the GC NGC 6397 (Lindetal. 2009b) and DEIMOS spectra of red giants coadded in bins of MV,0 are shown in red and green,
respectively. Thebluecurve(Equation6)separatesLi-richfromLi-normalstars.
theexactplacementoftheboundaryissomewhatsubjec- Li-rich giants appear in M68, NGC 5053, M3,
tive, Figure 3 shows that there is little ambiguity about NGC 5897, and M30. S232, the more luminous giant in
which stars are Li-rich. The assignment of Li-rich and M68 was previously discovered by Ruchti et al. (2011).
Li-normal could be questioned only for the faintest Li- IV–101, the M3 giant discovered by Kraft et al. (1999),
rich star, M30 132. A more rigorous analysis might use is not in our sample. These clusters do not appear re-
multiple levels of Li-richness, such as “Li-normal,” “Li- markable in any way other than hosting Li-rich giants.
rich”,and“superLi-rich,”orevenacontinuouslydefined Table 5 shows that these clusters have typical luminosi-
“Li-richness” variable. For simplicity, we retain our bi- ties and metallicities.
nary (yes/no) definition, accepting that the Li-richness Two GCs host two Li-rich giants each. M68 has one
of M30 132 is ambiguous. Li-rich RGB star and one Li-rich AGB star, and M30
has two Li-rich RGB stars. We conducted 106 random
4.2. Li-Rich Frequency draws of stars from our sample in order to test for the
significance of this apparent clustering of Li-rich giants.
Stars with Li detections above the boundary are con-
We drew at least two RGB stars from the same GC in
sidered Li-rich. Stars with Li detections or upper limits
32% of the trials, and we drew one RGB and one AGB
below the boundary are Li-normal. Upper limits above
star from the same GC in 46% of the trials. At least
the boundary do not indicate whether the star is Li-rich
two stars of any type were drawn from each of two or
or Li-normal. We calculated the frequency of Li-rich
more GCs in 11% of the trials. Therefore, clustering of
stars as the number of Li-rich stars divided by the to-
Li-richgiantscannotbe ruledout,butthe significanceis
tal number of detections and “useful” upper limits. If
marginal.
weweretoraisetheboundaryforLi-richness,fewerstars
The fraction of Li-rich giants across all GCs in our
wouldbeconsideredLi-rich,andmoreupperlimitswould
sample is (0.3±0.1)%, notably less than the commonly
be considered useful, both of which would decrease the
quoted 1%. The statistics do not include IV–101 in M3
Li-rich frequency.
because it was not included in our sample. We obtained
Table 5 shows the Li-rich frequency for each GC
a longslit spectrum of this star and confirmed its Li en-
in our sample and for the combined sample of all 25
hancement, but we did so because it was pre-selected to
GCs. The “Members” column shows stars that passed
be Li-rich. In order to avoid biasing our results, Table 5
the membership criteria, regardless of their Li abun-
includes only stars that were included in our random
dances. “Li-rich” shows stars with A(Li) that exceed
sample.
the boundary set by Equation 6. “Li-normal” includes
stars with detections or upper limits below the bound-
ary. The “Li-Rich Frequency” is (Li-Rich)/(Li-Rich+ 4.3. Stellar Evolutionary State
Li-Normal). The error bars on the frequencies are Pois- It is useful to identify the stellar evolutionary phase
sonian: (Li-Rich)/(Li-Rich+Li-Normal). of the Li-rich giants in order to determine whether they
p