Astronomy & Astrophysics manuscript no. lebz0164 February 2, 2008 (DOI: will be inserted by hand later) Long period variables in the globular cluster 47 Tuc: radial velocity variations 1 2 3 3 4 T. Lebzelter , P.R. Wood , K.H. Hinkle , R.R. Joyce , and F.C. Fekel 1 Institutefor Astronomy (IfA),University of Vienna, Tu¨rkenschanzstrasse 17, A-1180 Vienna,Austria 5 e-mail: [email protected] 0 2 Research School for Astronomy & Astrophysics, Mount Stromlo Observatory,Weston, ACT 2611, Australia 0 3 National Optical Astronomy Observatory⋆, PO Box 26732, Tucson, AZ 85726, USA 2 4 Center of Excellence in Information Systems, Tennessee StateUniversity,Nashville, TN 37203-3401, USA n a Received ; accepted J 7 Abstract. Wepresent near infrared velocity curvesfor 12 long period variables (LPVs) in theglobular cluster 47 1 Tuc (NGC 104). New light curves are also presented for these variables. Results are compared with the period- 1 luminosity sequences occupied by the LPVs in the LMC. Sequence C variables (fundamental mode pulsators) v have larger velocity amplitudes than sequence B variables (first overtone pulsators). We show that, at similar 6 luminosities, highermass loss rates areassociated with larger pulsation amplitudes.Onevariable(V18) does not 3 fitthenormalperiodluminositysequencesandithasanunusuallylargeamountofcircumstellardust,suggesting 3 that it has recently undergone a thermal pulse on the AGB. Finally, we report the discovery of three new long 1 period variable stars in thecluster core, all previously found to havea large infrared excess. 0 5 Key words.stars: late-type– stars: AGBand post-AGB– stars: evolution 0 / h p 1. Introduction minosities,amplitudesandmasslossrateswiththeoretical - o calculationsis complicated.Awayaroundthese problems Inrecentyears,therehavebeensignificantadvancesinour r istoobservepulsatingredgiantstarsinstarclusterswhere t understanding of pulsation in long period variable stars s the giantshaveacommon(initial)mass,compositionand a (LPVs). Particularly important is the discovery of multi- distance. We can then readily compare theoretical and : v ple period-luminosityrelationsforAGB starsinthe LMC observed period-luminosity relations, and we can see how i and the interpretation of all but one of these relations in X the mass loss rate depends on pulsation properties and terms of radial pulsation in low order modes (Wood et quantities such as mass, luminosity and metallicity. r a al.1999).Anotheradvanceisthe realizationoftheimpor- The driving of mass lossby pulsationis a complicated tance ofpulsationinthe productionofmasslossfromred process.Levitationofthestellaratmospherecausedbythe giants(e.g.Wood1979;Bowen& Willson1991;Ho¨fneret pulsationleadstothe formationabovethe photosphereof al.1996). acoolanddenseenvironmentwheredustgrainsformand One problem with the observational study of pulsa- grow efficiently. Radiation pressure on dust grains then tionandmasslossinredgiantsisthatnearlyallthe stars combines with momentum in shock waves to drive the studied so far are field stars. This means that we have no massloss(e.g.Wood1979;Bowen&Willson1991;Ho¨fner way of directly determining the initial or current stellar et al.1996). It is found that larger pulsation velocity am- mass. Any relations that are found for the field stars will plitudes and higher luminosities drive a higher mass loss be broadened by the range in mass (age) and metallicity rate.One of the main purposes of this study is to observe existingamongthestars.ForGalacticstars,butnotLMC the pulsation velocity amplitudes in a goodsample of red stars,we have the additionalproblemthat the luminosity variables in a star cluster so that a comparison of mass (distance) is generally poorly known. As a result of these loss rate with pulsation velocity amplitude (and luminos- uncertainties, precise comparison of observed periods, lu- ity) can be made. Send offprint requests to: T. Lebzelter The pulsation velocities ofLPVs are best measuredin ⋆ operated by the Association of Universities for Research the near infrared since optical absorption lines measure in Astronomy, Inc. under cooperative agreement with the velocities in the outer layersonly. These optical velocities National ScienceFoundation are nearly always directed inward relative to the stellar 2 T. Lebzelter et al.: Long period variables in theglobular cluster 47 Tuc: radial velocity variations center-of-mass (e.g. Wood 1979). The near-infrared lines Light curves of the LPVs V1 to V81 were measured revealthe deeper, largeramplitude pulsation. Monitoring by Arp et al.(1963). Severaladditional variables were de- of the velocity variations in field variables using infrared tectedbyLloyd-Evans&Menzies(1973).Afurtherstudy, (1.6 µm) CO lines has been carriedout by Hinkle (1978), including the variables V3 to V7, V11(=W12), V13 and Hinkleetal.(1982;1997)andLebzelteretal.(2000).These V18,waspresentedbyFox(1982).The lightvariabilityof observations revealedvelocity curves with amplitudes be- these stars will be further discussed in Sect.4.1. tween 3 and 30km/s. The velocity curves of Miras are sawtooth-shaped and of large amplitude, with phases of 3. Measurements of mass loss in 47 Tuc linedoublingaroundmaximumlight.Thetypicalvelocity curves of semiregular variables (SRVs) are of smaller am- Therehavebeenmanyattempts to measurethe massloss plitude than in Miras and they are rather sinusoidal (i.e. from AGB stars in globular clusters. Mass loss at the continuous with no line doubling), although they can re- end of the red giant branch (RGB, also called the first flectthepartlyirregularbehaviourseeninthelightcurves. giant branch or FGB) phase has already been proposed A summary is given in Lebzelter & Hinkle (2002). to explain the observed gaps in the horizontal branch of Theglobularcluster47Tuc(NGC104)waschosenfor globular clusters (Soker et al.2001; Schr¨oder & Sedlmayr this study since it contains the richest known collection 2001).Cohen(1976)proposedtheexistenceofcircumstel- of long periodvariables (LPVs).Properties ofthis cluster lar shells (and hence mass loss) around first giant branch from the literature are summarized in Sect.2. Along its stars in globular clusters to explain emission components giant branch, four Miras and more than 10 semiregular in the Hα line. Bates et al.(1990) summarized different and irregular variables have been detected (Sawyer-Hogg studies on Hα lines of globular cluster stars and found a 1973; Lloyd-Evans1974). Severalinfrared searches for in- further indication for circumstellar shells from the NaD dications of mass loss from giant branch stars have also line profiles. Lyons et al.(1996) discussed mass motions been reported (see Sect.3). in the atmospheres of 63 red giants from five different globular clusters with the help of Hα and NaD lines. Core shifts of these lines indicate mass flow. The lower luminosity limit for outflow from Hα and NaD lines is 2. 47 Tuc and its variables log(L/L⊙)=2.5and2.9,respectively.Onlypartofthered 47 Tuc is a prototype “metal-rich” globular cluster as giantsshowedanoutflow.Standarddeviationsinthefinal well as being one of the closest. Recently, Gratton et al. masses ofwhite dwarfsin globularclusters ofthe orderof (2003) presented a very detailed study on the fundamen- 0.1mag may indicate a stochastic nature of the mass loss tal parameters of 47 Tuc. They derive a metallicity of (e.g.see the discussion by Alves et al.2000). [Fe/H]= −0.66±0.04 on the Carretta & Gratton (1997) Not only is mass loss known to be essential in stel- scale, which is in agreement with the findings of several lar evolution but the mass loss from red giants in glob- other groups (see references in Gratton et al.). A metal- ular clusters should result in intracluster gas and dust licity based on the Zinn & West (1984) scale is given by (Evans et al.2003). Searches for the intracluster gas have e.g. Briley et al. (1995) with [Fe/H]=−0.76. beenlimitedintheirresults.Thelowmetallicitynodoubt plays arole inreducing silicate emission,making dustde- An accurate distance to the cluster is still a matter tection difficult (Frogel & Elias 1988; Helling et al.2002). of some debate. Values in the literature for the distance However,findingsfromAGBstarsintheLargeMagellanic modulus of 47Tuc scatter by about ±0.2 mag (for an Cloud (LMC) confirm that low metallicity stars reach overview see Gratton et al.2003). Gratton et al. (2003) high mass loss rates (e.g.Wood et al.1992; van Loon et find a distance modulus of (m−M)V=13.50±0.08 from al.1999). main sequence fitting. All studies agree that there is very little reddeningtowards47Tuc(E(B−V)=0.024±0.05). Afewrecentpapersprovidemoredetailedinformation on circumstellar shells around AGB stars in 47 Tuc. One Based on their estimates for the distance and metallic- of the most definitive results is by Ramdani & Jorissen ity Gratton et al. (2003) derived an age of 47 Tuc of 11.2±1.1Gyr. Hesser et al.(1987) derived a turnoff mass (2001) who used the ISO satellite to measure the mid- infraredemissionaround6AGBvariablesin47Tuc.Their of0.9M⊙,butusedanageestimateof13.5Gyr.However, findings, however, highlight the difficulty of correlating withtheagefromGrattonetal.andthemorerecenttheo- pulsational properties with mass loss. Half of the vari- reticalisochronesfromBertellietal.(1994)weagaincome ables, namely V5, V7 and V15, show no or only marginal to a turnoff mass between 0.86 and 0.9M⊙. 12µm-excesswhiletheotherthreestars,V3,V11andV18, A large number of photometric measurements in do have a detectable excess flux at 12µm. The excess of the blue, visual and near- and mid-infrared (Lee 1977; V11israthersmall.V3isaMiraandwouldbeexpectedto Frogel 1983; Montegriffo et al.1995; Origlia et al.1997; show an excess. But the largest infrared excess was mea- Ramdani&Jorissen2001)establishaveryaccuratecolor- sured for V18, an irregular AGB variable of only modest magnitude diagram of this cluster showing a well defined AGB. Luminosities and surface temperatures have been 1 Throughout this paper we use the nomenclature from the derived for the AGB stars (e.g. Whitelock 1986). catalogueofglobularclustervariablesbyClementetal.(2001). T. Lebzelter et al.: Long period variables in theglobular cluster 47 Tuc: radial velocity variations 3 luminosity. Origlia et al. (2002) suggested that the mass 1,0 loss from red giant stars is episodic. They found an IR 0,95 excess for V8 and no excess for V21 (see also Sect.5.3). cTdfieVixenur8ctdcceeOu.acsimAnstvneeddsVdsrtiuaceV-bbalbllly1ts,aait1nrGsoaodnnbmimlstylpaieefatootFlteilrnaoirenairdrtafitnizrlecaaaailernt.etxeiti(foidor1snana9tersl8xwe.o8cdffa()oe2,strseh0sbxf10euoci2n2eutp)snVrtAshedi3Gnisen,edsiBnnVeiVcc1aaveV51uta1oitaranhrinnagtodaodbrbtlsVVeVhsdse1e14ni38di,cen.wexVnAi4aoo6sn7st-f, 000,,,468 OH/CO blend Fe CO blendFe CO 4-1 CO 6-3Fe CO 6-3 CO 4-1 CO 6-3CO 6-3CO 6-3 CO 6-3Fe Fe CO 4-1CO blendFe 000,,,889050 tence of circumstellar material around these stars. Glass 0,75 & Feast (1973) reported an L−band excess in V3. Frogel OH & Elias (1988) detected a 10µm excess in the four clus- 0,2 0,70 ter variables V1, V2, V3 and V4. They calculated total 16280 16290 16300 16310 16320 16330 -20-100 1020 km/s wavelength [A] mass loss rates for these objects ranging between 4.6 and 12.3x10−6M⊙yr−1. A 10µm excess was also detected in Fig.1. Left panel: Example spectrum of the star V11. V1 and V3 by Origlia et al. (1997). However their esti- Several prominent features are identified. Right panel: mated dust mass loss rates are about ten times less than Example of line doubling (CO 4-1, V3). those of Frogel & Elias (1988). To determine proper phases for the velocity curve, 4. Observations and results photometric measurements of all variables in our sam- We selected 12 long period variables in the globular clus- ple except V5 (see below) were done as well. Data in ter 47 Tuc for which a period or at least some informa- the blue and red MACHO filters were obtained with the tion on the variability type was given in the catalogue of 50inch telescope at Mount Stromlo (also destroyed in the variable stars in globular clusters published by Clement January 2003 fires). V and I measurements were ob- et al. (2001). These stars are moderately bright infrared tained with ANDICAM at the YALO telescope (before sources, with K∼6-8. Table 1 summarizes the properties the MACHO dataset), with a few additional measure- of the stars. mentsfromANDICAMattheCTIO1.3mtelescope(after Time series of infrared spectra in the H band were the MACHO dataset). The different data sets have been obtained in 2001 and 2002 with the 74inch telescope at combined using transforms based on about 40 nonvari- Mount Stromlo Observatory, Australia. The NICMASS able cluster stars ranging in V −I between 0.9 and 2.1. detector, already successfully used for a preceding pro- The MACHO blue filter has a mean wavelength close to gram at Kitt Peak (Joyce et al. 1998), was used at the that of the V filter so MACHO blue magnitudes can be Coud´e focus of the telescope. The standard infrared ob- reliably transformed to V. Long time series are therefore servation technique was used. Spectra of each star were availableintheV filter.Noobviousphaseshiftrelativeto obtained at two different slit positions to allow sky sub- the other filters was noticed, but a conclusive result can- traction. Resolution was set to R=37000. We achieved not be drawn from our data. From the V light curves we a S/N ratio of 30 or better. An example spectrum is determined the time of the light maxima, phase zero. shown in the left part of Fig.1 with an identification of some relevant spectral features. The spectral range cov- 4.1. Light curves ered a number of second overtone CO lines, some OH lines, and a few metallic lines. When 47Tuc was visible Light curves for most of the variables are shown in the at Mount Stromlo spectra were obtained approximately upper panels of Figs.2 to 13. Periods derived from our onceamonth.The observingprogramhadanunexpected measurementswereinmostcasesingoodagreementwith end when Mount Stromlo Observatory was destroyed by the values from the literature. In our data V4 shows a a bush fire in early 2003. Nevertheless time coverage and main period of ∼170d in agreement with the value of sampling of the 47Tuc AGB variables, combined with a 165d from Arp et al.(1963). Fox (1982) found a period few additional spectra taken in the same wavelength re- of 82 days for V4 and suggested that this star switched gion with the PHOENIX spectrograph at Gemini South, between two modes. This type of multimode behaviour is sufficient for the investigation presented here. has subsequently been found to be common in the light The bright stars α Cet and δ Oph have been used as curves of LPVs (e.g. Wood et al.1999). For V5, we un- primaryvelocitystandards(Udryetal.1999).Velocitiesof fortunately do not have sufficient usable data as this star the variableswere determined by a crosscorrelationtech- was located in a bad area of the CCD. The Sawyer-Hogg nique, using the IRAF task fxcor. Typical velocity uncer- value (1973) is 60 days, but different values can be found tainties, determined from multiple observations of some in the literature, ranging between 29 and 70days (Fox stars in the same or consecutive nights, were found to be 1982). From our velocity data we rather favour a period ∼0.4 kms−1. of50days.The shortperiods of52and100daysreported 4 T. Lebzelter et al.: Long period variables in theglobular cluster 47 Tuc: radial velocity variations Table 1. Properties of LPVs in 47Tuc. Table 2. Data on the AGB variables in 47Tuc. Name α (2000) δ (2000) J K P [d] Name velocity amplitude variability comments V1 00 24 12.4 −72 06 39 7.45 6.21 221 [km/s] type V2 00 24 18.4 −72 07 59 7.52 6.29 203 V1 20 Mira-like line doubling V3 00 25 15.9 −72 03 54 7.49 6.27 192 V2 23 Mira-like line doubling V4 00 24 00.3 −72 07 26 7.87 6.69 165,82 V3 22 Mira-like line doubling V5 00 25 03.7 −72 09 31 8.65 7.47 50 V4 18 Mira-like double-period V6 00 24 25.5 −72 06 30 8.54 7.43 48 V5 8 regular V7 00 25 20.6 −72 06 40 8.18 6.97 52 V6 7 semiregular V8 00 24 08.3 −72 03 54 7.94 6.70 155 V7 4 semiregular V11 00 25 09.0 −72 02 17 7.91 6.71 52,160? V8 16 Mira-like V13 00 22 58.3 −72 06 56 8.79 7.70 40 V11 4 irregular V18 00 25 09.2 −72 02 39 8.59 7.47 83? V13 12 long period double-period V21 00 23 50.1 −72 05 50 8.07 6.78 76 V18 5 irregular Notes: For V18 and V21 both period determinations are from V21 7 long period this study. For V1 we suggest a period of 221 days from our data instead of 212 days listed in the literature. For V5 we favour a period of 50 days (instead of 60 days listed in the Clement et al. catalogue). For V11 see text. All other periods solarneighborhood(Lebzelter &Hinkle 2001).The veloc- are from the catalogue of Clement et al. (2001). The J and ity amplitude of V1, V2 and V3 are all similar to nearby K magnitudes are averages of the maximum and minimum Miras, too. The other two stars show a similar shape but observed values from the following sources: the 2MASS Point a clearly smaller amplitude. While we confirm the Mira- SourceCatalog; Fox(1982);Menzies &Whitelock(1985);and likenatureofthesefivevariables(e.g.notedbyWhitelock Frogel et al. (1988). All magnitudes have been converted to 1986), we note that the smaller velocity amplitude of the the AAO system using conversions in Allen & Cragg (1983) lattertwostarsisalsoaccompaniedbyasmallerlightam- and Carpenter (2004). plitude. The second group of variables, consisting of V5, V6 and V7 is comparable to local semiregular variables for V11 could not be confirmed in our data. Instead, the (Lebzelter & Hinkle 2001). Their amplitudes are much star shows some long period variation lasting more than smaller than in the first group. V5 shows a very regu- 200days(Fig.10).ComparedtothedatapresentedbyFox larlightchange,whilethe othertwoobjectsareobviously (1982) the amplitude ofthe variationis smaller.This star not strictly periodic, in agreement with the semiregular appearstobeanothermultimodepulsator.ForV13,there nature of their light curves. The third group consists of isahintofthecatalogued∼40dperiodinourlightcurve, V11, V13, V18, and V21. In all four of these stars, there but a period of ∼90d may also be present. Additionally, isnoobviouscorrelationbetweenthelightchangeandthe our light curve suggests a long period (Fig.11) about 10 velocity variations. In the case of V13, it cannot be ruled timestheshortperiod.Thisstarappearstobeyetanother out that variations occur on a time scale similar to the multimode pulsator. shortoreventhelongperiod(seeabove).ForV11ourob- V18 was reported variable both in the catalogue of servationsdonotallowaclearpictureofthe lightchange, Sawyer-Hogg(1973)andbyLloydEvans(1974),whileFox so a correlation is not possible. On the other hand, V18 (1982) found no variability in this star. We found photo- and V21 do show a rather well defined light change, but metric variability with an amplitude of about 0.2 mag in velocity variability occurs on a different time scale. The this object (Fig.12). Our short light curve would suggest reason for that is not clear and longer spectroscopic and a period of about 83 days.For V21, no period determina- photometric time series will be needed to understand this tion existed in the literature. Our data set (see Fig. 13) phenomenon. Table 2 summarizes the results on the ve- shows a period of 76 days. locity variations in the 47Tuc AGB variables. The table gives the total velocity amplitude, the characteristics of thevelocitycurveandcommentsontheoccurrenceofline 4.2. Velocity curves doubling. Velocity curves were determined for all 12 stars of our sample. All stars show at least some velocity variability 5. Discussion aboveourdetectionthreshold.Linedoublingwasdetected in three stars of our sample (Tab.2). An example of line 5.1. Pulsation and Stellar Evolution doubling is shown in the right part of Fig.1. The velocity curves we derived are shown in the lower Observations from the Magellanic Clouds indicate that partsofFigs.2to13togetherwiththecorrespondinglight AGB variables are found on four distinct period- change.Thevelocitycurvescanberoughlyseparatedinto luminosity relations (e.g. Wood 2000). Three of these se- threegroups:V1,V2,V3,V4,andV8showvelocitycurves quences can be interpreted in terms of stars pulsating in verysimilarinshapetothosetypicalofMirasfoundinthe thefundamentalmodeandthefirst,secondandthirdover- T. Lebzelter et al.: L ong period variables in theglobular cluster 47 Tuc: radial velocity variations 5 10 10 12 12 ag] g] m ma V [ 14 V [ 14 165 16 -15 0 -20 m/s] -5 m/s] -25 k ocity [ -10 city [k -30 vel elo -35 v -15 -40 -20 -45 2452200 2452400 2452600 2452200 2452400 2452600 JD JD Fig.2. Light(upper panel)andvelocity variations(lower Fig.4. Same as Fig.2 for V3. panel) for V1. Filled symbols in the lower panel indicate individualvelocitymeasurements.Forabetterillustration of the velocity change data are repeated shifted by an 2 tone modes . The fourth period-luminosity relation still integralnumber of periods forwardand backwardin time lacks a definite interpretation (Wood et al.2004). For the (open symbols). The period listed in Table 1 is used. The MilkyWayAGBvariablestheabsenceofreliabledistances typical error bar for the velocity data is indicated. limits the establishment of such a period-luminosity rela- tion(e.g.Bedding&Zijlstra1998).Inthecaseofboththe 11 LMCandtheMilkyWay,thediversityofthestarsinage, andtherefore initial mass,complicates the interpretation. 12 The AGBs of globular clusters are characterized by a g] a much more homogeneous set of stellar parameters. The m 13 richness of 47Tuc in AGB variables allows us to use this V [ advantage for a discussion of stellar pulsation in relation 14 to luminosity, i.e. evolutionary stage. Period-luminosity 1150 diagrams for globular clusters have been constructed by several authors (e.g.Feast et al.2002). Typically, only a relation for the large amplitude, long period variables 5 (’Miras’) was derived. This relation is in agreement with the sequence C found from LMC data (Wood 2000). In 0 s] Fig.14 we placed the AGB variables of 47Tuc studied in m/ this paper into a logP-K-diagram.The approximate loca- k -5 y [ tion of the P-K-relationsB (first overtone)and C (funda- cit mental mode) determined in the LMC (Wood 2000) are o -10 el indicated by lines. Relations were transformed from the v LMC to 47Tuc distance using a LMC distance module -15 of 18.515 (Clementini et al.2003). Stars with more than one well established period in the literature or from our -20 own light curve data are plotted twice and are marked 2452200 2452400 2452600 accordingly in Table 2. JD 2 SequenceAisattributedtobothsecondandthirdovertone Fig.3. Same as Fig.2 for V2. mode pulsators. 6 T. Lebzelter et al.: L ong period variables in theglobular cluster 47 Tuc: radial velocit y variations 11 11.5 g] g] ma 12 ma V [ V [ 12.0 13 10 -20 -22 m/s] 5 m/s] k city [k 0 ocity [ -24 o el el v v -26 -5 -28 2452200 2452400 2452600 2452200 2452400 2452600 JD JD Fig.7. Light (upper panel) and velocity (lower panel) Fig.5. Same as Fig.2 for V4. variationsforV6.Atypicalerrorbarforthevelocitydata is indicated. -18 determined,asthere is asignificantscatter(probablydue s] -20 to the largerscatter in stellar parametersand maybe also m/ the extension of the LMC) around the average locations k y [ used here. We therefore do not want to overinterpret our cit findings on that point. o -22 el There is a clear bipartition in our logP-K-diagram v (Fig.14): At lower luminosities only sequence B is pop- ulated (see below for the case of V18), no variables are knownin47TucthatwouldfallonthelowerpartoflogP- -24 K-relation C. It is very likely that the stars V5, V6, V7, 2452500 2452550 2452600 2452650 and V13, which we find in this lower luminosity part are JD probably in an earlier evolutionary state than the stars Fig.6. Velocity change of V5. Only part of the time of at higher luminosity (see below). In a first approach we monitoring is shown. Symbols as in the lower panel of assume that the luminosity of the stars in Fig.14 corre- Fig.2. No parallel light curve data exist for this star. A sponds to the evolutionary status. typical error bar for the velocity data is indicated. At the highest luminosities only sequence C is occu- pied, and the stars there are the Miras V1, V2, and V3. BelowthesestarsonsequenceClieV4andV8.Bothstars Itcanbeseenthatthevariablesnicelyfollowthesetwo arelocatedclosetothetipoftheRGB.Theyhavesimilar relations. The good fit of the P-L-relations to the 47Tuc velocityamplitudeandKbrightness.V4appearstoswitch data after scaling from the LMC is indeed remarkable. modes between sequence C (P = 165d) and sequence B Obviously, the physical mechanism that determines the (P =82d).Fromourdatawehaveto concludethatV4is location of at least these two sequences in the logP-K- currently only on sequence C. V11 having approximately diagram is the same for LMC and 47Tuc. Furthermore, the same K luminosity as V4 is another star suspected of thedistancemoduliweusedinthispaperseemtobevery showing mode switching (see Sect. 4.1) located clearly off well determined at least concerning the relative distance sequenceBtowardsshorterperiods(i.e.nearsequenceA). betweenLMCand47Tuc.However,ithastobenotedthat The period used in the plot is taken from Fox (1982). It thelocationofthesequencesintheLMCisnotaccurately is interesting that both stars with possible mode switch- T. Lebzelter et al.: Long period variables in theglobular cluster 47 Tuc: radial velocit y variations 7 11.4 12.0 11.6 g] ag] ma 11.8 m V [ V [ 12.2 12.0 -2 -26 m/s] -4 m/s] k city [k -6 ocity [ elo vel v -28 -8 2452200 2452400 2452600 2452200 2452400 2452600 JD JD Fig.8. Same as Fig.7 for V7. Fig.10. Same as Fig.7 for V11. 11.8 11 g] ag] 12.0 a 12 m V [m V [ 12.2 12.4 13 -10 -15 -12 -20 s] -14 s] m/ m/ k y [k -25 city [ -16 ocit elo -18 el v v -30 -20 -22 -35 2452200 2452400 2452600 2452200 2452400 2452600 JD JD Fig.11. Same as Fig.7 for V13. Fig.9. Same as Fig.2 for V8. (2000). The nature of this period-luminosity relation is ing arefound atsimilarK luminosity,namely closeto the not yet understood (Wood et al. 2004). RGB tip. Symbol size in our P-K-diagram is linearly propor- For V13 we found indications of a second period, and tional to the measured velocity amplitude. In the case of the star is therefore shown twice in Fig.14. The long pe- V4 we stress again that the shorter period, marked with riod of V13 is located close to sequence D from Wood an open symbol in Fig.14, is clearly not dominating the 8 T. Lebzelter et al.: L ong period variables in theglobular cluster 47 Tuc: radial veloc ity variations 11.8 6 g] a V1 m 11.9 V3 V [ V2 12.0 V11 V4 V8 V4 -8 V21 RGB tip s] K 7 V7 m/ k y [ cit -10 o el v V6 V5 V18 V13 V13 -12 2452200 2452400 2452600 B C JD 8 Fig.12. Same as Fig.7 for V18. 1.6 1.8 2.0 2.2 2.4 2.6 log P 12.0 Fig.14. logP vs. K diagram for the long period vari- 12.2 ables in 47Tuc. Symbol size denotes the velocity ampli- tude (ranging between 4 and 23kms−1). Lines indicate g] 12.4 the approximate location of sequences B and C found by a m Wood(2000)forLMClongperiodvariablesshiftedtothe V [ 12.6 distance of 47Tuc. The dotted horizontal line marks the tip of the RGB according to Ferraro et al.(2000). Stars 12.8 with possible multiple periods are shown twice. -12 rently monoperiodic. It can be seen that the large ampli- -14 tude variablesareallfound alongsequenceC.Our results m/s] indicate further that the velocity amplitude is increasing k -16 along sequence C. Velocity amplitudes along sequence B y [ aremuchsmalleranddonotshowasteadyincreaseofve- cit locity amplitude with luminosity. The large amplitude of o el -18 V13seems to be associatedwiththe longperiod(∼400d) v variation rather than the 40d variation of sequence B. -20 These results lead us to the following probable evolu- tionaryscenario:Stars evolveupthe AGB fromlowlumi- 2452200 2452400 2452600 nosities. They pulsate first in the first overtone (sequence B)then,atintermediateluminositiesclosetotheRGBtip, JD they go through an interval of mode switching back and Fig.13. Same as Fig.7 for V21. forth between first overtone and fundamental mode and finally, at the highest luminosities, they remain pulsating in the fundamental mode. There is clearly an increase in light or velocity variations at the moment. The symbol both the light and velocity amplitude associatedwith the size of the V4 data point on sequence B may therefore switchfromfirstovertonetofundamental.Thelargeveloc- be misleading as we have no indications that the veloc- ity amplitudes found at the tip of sequence C are similar ity amplitude would be the same if the 82 days period to what is expected from models for fundamental mode would dominate. According to our data the star is cur- pulsation (e.g.Scholz & Wood 2000). T. Lebzelter et al.: Long period variables in theglobular cluster 47 Tuc: radial velocity variations 9 TheroleofV18inthisscenarioisnotclear.Asnotedin tip are clearly AGB stars, namely V1, V2, V3. V4 and Section3,thisstarhasoneofthelargestinfraredexcesses V8 are both fundamental mode pulsators slightly above among the 47Tuc variables. In Fig.14 we find the star in the RGB tip and thus they canbe countedas AGB stars, betweensequencesBandCataratherlowluminosity.Its too. For the other stars of our sample, we cannot decide velocity amplitude is rather small. However, locating this ontheir AGB naturefromluminosity alone.Colourinfor- star on the period axis is difficult as the star has shown mation may be a good indicator for a separation,but our periodic, nonperiodic and constant phases in the past. photometry is not homogeneous enough to decide on this rather small effect as a mean colour over the light cycle Comparing our results with evolutionary models re- would be required. quires some adaption of this scenario: Due to luminosity Can variability properties allow us to distinguish be- variationsduringaThermalPulsethecurrentlocationofa tween AGB and RGB stars? Using artificial luminosity starinthelogP-K-diagrammaynotcorrespondtoitsevo- functions, Alves et al.(1998) and Wood et al.(1999) ar- lutionary status along the AGB. However, as was shown gued that most pulsating stars below the RGB tip are e.g. by Boothroyd & Sackmann (1988) the probability to in fact AGB stars. However, using data more sensitive to observethestaratluminosityvalueshighlydeviatingfrom smaller amplitude variables, Ita et al.(2002) and Kiss & the mean (and for the star’s evolutionary status typical) Bedding (2003) have shown that a substantial fraction of value is comparatively low. This fact is even more ex- variables below the RGB tip should be RGB stars. pressedin stars of 1M⊙ or less where the interpulse time According to the logP-K-diagram for LMC variables is very large compared to the duration of the pulse itself given by Kiss & Bedding (2003, see their Fig.4), most of (Vassiliadis&Wood1993).FromthemodelsofVassiliadis the variables detected below the tip of the RGB are pul- & Wood (1993, see their Fig.13) for a star of 1M⊙ and satinginthesecondorthirdovertonemode.Furthermore, Z=0.004 we would expect to see about 1/6 of our sample Kiss & Bedding split their logP-K-diagraminto six parts in a phase strongly deviating from the mean global pa- according to the light amplitude in I. Most of the vari- rameters of the star, in our case these would be 2 stars ables below the tip of the RGB show amplitudes between (if our sample consists of AGB stars only, see Sect. 5.2). 0.01 (their detection limit) and 0.14mag. The luminos- Most probably, these stars are expected to be found at a ity function for these stars indicates a substantial frac- lower luminosity and a shorter period. It is therefore un- tion are on the RGB. However, for larger light ampli- likely that one of the stars on sequence C belongs to this tudes,theRGBcomponentisnolongerdetectable.Kiss& group. Following these considerations we have to add to Bedding also give colour information showing that by far our scenario the possible occurrence of loops in the logP- thelargestfractionofstarsbelowtheRGBtipwithampli- K-diagramduringtheevolutionuptheAGB.V18withits tudes >0.14magisfoundonthe blueside ofthe redgiant large IR excess and odd position in the logP-K-diagram branch,whichagainfavoursattributing these starsto the may be an example of such a star. However,it is unlikely AGB. Thus larger light amplitudes seem an indication of that e.g. the mode switch of V4 is an indication of such a AGB status. loop because of the time scale of the mode switch. Arp et Kiss & Bedding (2003) used I band data, whereas we al. (1963) gave a period of 165 days from data obtained have V band light curves. Fox (1982) presents V and I in 1955/1956.Our light curve also favours this period. In light curves for several of our sample stars. For the stars between these two measurements are the light curve data oflowerluminositylikeV5orV6,theI amplitudeisabout presented by Fox (1982) giving the shorter period. The a factor of 2 less than the V amplitude. The V band am- star therefore switched its pulsation mode back and forth plitudesofthestarsofoursampleareall≥0.2mag,i.e.we within about 50 years. This seems to be hardly compati- estimate an I amplitude of ≥0.1mag. ble withthe longertimescalesexpectedfromevolutionary V21, V5 and V6 all show light amplitudes of several models. The mode switches seem to be concentrated in tenth of a magnitude. Such amplitude values favour an the luminosity range close to the RGB tip. AGB nature for these stars as well. V7 shows a rather small light amplitude, and for V11 the parameters of the 5.2. AGB or RGB stars? light change are not clear. Therefore, we cannot decide if these two objects are RGB or AGB stars. Interestingly, Throughoutthis paper wehaveassumedthatallthe vari- both objects also have smaller velocity amplitudes. ablesdiscussedhereareonthe AGB.Ascanbeseenfrom AnotherunclearcaseisV13.Thetotalamplitudemea- Fig.14 a significant part of our sample is located below suredby us is ≈0.7mag.But this amplitude is dominated thetipofthe redgiantbranch(RGB).TheRGBtipplot- bythelongperiodvariation,theshortperiodvariationhas ted in Fig.14 is taken from the recent study of Ferraro amuchsmalleramplitude.Theamplitudesusedinthedis- et al.(2000). They give a value of KTRGB=6.75±0.2mag. cussionbyKiss&BeddingarederivedfromFourieranaly- Thisresultisingoodagreementwiththetheoreticalvalue sis ofthe individual frequency components.Therefore the expected for the metallicity of 47Tuc (Salaris & Cassisi amplitude of the short period is the relevant one in our 1998). All stars are above the low luminosity limit for argumentation. As a result V13 may well be a RGB star. thermally pulsing stars, so that they could belong to ei- Our approach is based on statistical arguments and ther the RGB orthe AGB.The starswellabovethe RGB not on a complete understanding of RGB pulsation. In 10 T. Lebzelter et al.: Long period variables in theglobular cluster 47 Tuc: radial velocity variations particular,wenotethatthe LMCstarsofKiss&Bedding Jorissen(2001)laterfoundaninfraredexcessofV18from (2003)arelikelytobemoremassivethanthe47Tucstars ISOobservations,whileV11hasonlyaverysmallinfrared and their amplitude behaviour may be quite different to excess.Weconcludethat,atsimilarluminosity,starswith thatof47Tucstars.Wethereforecannotexcludethepos- higher pulsation amplitudes show higher mass loss rates. sibility that some of the variables below the RGB tip are V5, V6 and V18 form a second group of variables at ontheRGB,butformanyofthemalocationontheAGB similar luminosity. Unlike the group of stars around the seems to be more likely. RGB tip, these stars all have very similar velocity ampli- tudes. Neither V5 nor V6 has an infrared excess reported inthe literature.ButV18hasanoutstandinginfraredex- 5.3. Comparison with mass loss cess according to Ramdani & Jorissen (2001). As noted above, the high infrared excess of V18 at a relatively low Using the velocity amplitude we can separate the logP- luminositysuggeststhatthis staris currentlyinthe lumi- K-diagram (Fig.14) of 47Tuc into three to four regions. nosity minimum following a thermal pulse on the AGB. At highest luminosity we find the Mira variables (V1, V2 V13, the star with the long secondary period, shows andV3)withthelargestvelocityamplitudes.Goingdown no indication of circumstellar dust. It appears that these sequence C, we find the second group (V4 and V8) with long periods on sequence D are not directly related to a intermediateamplitudes.Thethirdgroupconsistsofstars mass loss phenomenon (compare also Wood et al.2004). with low velocity amplitude on sequence B (and possibly Origliaetal.(2002)detectedfivefurtherstarsnearerto A). Finally, V13 may form a special case with its large thecoreof47Tucthatshowconsiderableinfraredexcesses amplitude due to a long secondary period. butthatarenotknowntobevariablestar.Wecheckedour Mass loss data available in the literature are limited photometric monitoring data coveringtwo months in late to measurements of circumstellar dust. As there may be 2003 for variable stars at the positions given by Origlia a dust-free mass loss as well, e.g. during the RGB phase, et al. Four of the five stars had variable counterparts on we cannot rule out that stars with indications for no or our V frames. The fifth object is the one with the lowest only small amounts of circumstellar dust still have a con- mass loss rate in the list of Origlia et al. All four ob- siderablemasslossrate.However,itisverylikelythatthe jects varied during the two months by several tenths of a existence of dust enhances the mass loss rate (e.g.Ho¨fner magnitude. More detailed results on these stars including etal.1996),soweexpectthattheinfraredexcessisatleast period determination will be presented in a forthcoming ameasurefortherelativemasslossratewithinthecluster. paper (Lebzelter & Wood in prep.). As a caveat we have to mention the lack of infrared pho- Summarizing our results regarding the correlation of tometry beyond 12µm for the stars of our sample. This mass loss and velocity amplitude: Stars with significant certainly limits the reliability ofestimates forthe amount mass loss are all pulsating, a fact already noted by other of dust surrounding a star. authors(e.g.Ramdani&Jorissen2001).Starsonsequence Note that the circumstellar material currently around C combine large velocity amplitudes with comparably eachstarmayoriginatefromatimewhenthestar’spulsa- highmass lossrates.Onsequence B allstarshavelow ve- tion was different. Another factor to keep in mind is that locity amplitudes and most of them also have no or very itis wellknown(e.g.Olofssonetal.2000)that,atleastin modestmassloss.Forstarsofsimilarluminosity,increased somecases,masslossinAGBstarsmaybeepisodicrather pulsational amplitude seems to significantly increase the than continuous. This has also been noted in the context mass loss rate. One star (V18) currently showing small of globular clusters by Origlia et al. (2002). pulsation amplitude has considerable amounts of circum- AccordingtoReimers’(1975)law,wewouldexpectthe stellar dust, presumably from a past interval of higher mass loss rate to increase with luminosity. Therefore it is amplitude pulsation. We speculated that V18 is currently difficult to separate the effects of luminosity and pulsa- doing a loop in the logP-K-diagram as a consequence of tion(velocityamplitude)onthe masslossrate:highmass a recent or ongoing thermal pulse. Its circumstellar dust loss rates are found at high luminosities, where we also may then originate not only from a phase of higher am- find large velocity amplitudes. We can, however, try to plitude pulsation but also from a time the star was more compare mass loss rate and velocity amplitude for some luminous. stars of similar luminosity. Such a group of stars is found around the RGB tip (V8, V4, V11 and V21). Origlia et Acknowledgements. TL has been supported by the Austrian al. (2002) measure K−[12] values for V8 and V21. V8 Academy of Science (APART programme). PRW has been partially supported by a grant from the Australian Research shows a clear dust excess while V21 does not. Comparing Council. This research at Tennessee State Universitywas par- the velocity amplitudes for these two stars shows that V8 tiallyfundedbyNASAgrantNCC5-511andNSFgrantHRD- alsohasasignificantlylargervelocityamplitudethanV21. 9706268. This publication makes use of data products from V4 is also a star with an infrared dust excess (Frogel & the Two Micron All Sky Survey, which is a joint project of Elias 1988). It has a similar velocity amplitude as V8. the University of Massachusetts and the Infrared Processing Gillettetal.(1988)reportaninfraredexcessforV11,but andAnalysisCenter/CaliforniaInstituteofTechnology,funded a separationbetween V11 and the nearby variable V18 is by the National Aeronautics and Space Administration and very difficult on the IRAS images they used. Ramdani & the National Science Foundation. Partly based on obser-