1 Astronomy (DOI:willbeinsertedbyhandlater) & (cid:13)c ESO2008 Astrophysics NLTE determination of the sodium abundance in a homogeneous sample of extremely metal-poor stars S.M.Andrievsky1,2,M.Spite1,S.A.Korotin2,F.Spite1,P.Bonifacio1,3,R.Cayrel1, V.Hill1,andP.Franc¸ois1 1 GEPI,CNRSUMR8111,ObservatoiredeParis-Meudon,F-92125MeudonCedex,France 7 2 DepartmentofAstronomyandAstronomicalObservatory,OdessaNationalUniversity,ShevchenkoPark,65014Odessa,Ukraine 0 3 CIFISTMarieCurieExcellenceTeam 0 2 n ABSTRACT a J Context.Abundanceratiosinextremelymetal-poor(EMP)starsareagoodindicationofthechemicalcompositionofthegasintheearliest 9 phasesoftheGalaxyevolution.IthadbeenfoundfromanLTEanalysisthatatlowmetallicity,andincontrastwithmostoftheotherelements, 2 thescatterof[Na/Fe]versus[Fe/H]wassurprisinglylargeandthat,ingiants,[Na/Fe]decreasedwithmetallicity. v Aims.Sinceitiswellknownthattheformationofsodiumlinesisverysensitivetonon-LTEeffects,tofirmlyestablishthebehaviourofthe 9 sodiumabundanceintheearlyGalaxy,wehaveusedhighqualityobservationsofasampleofEMPstarsobtainedwithUVESattheVLT,and 9 wehavetakenintoaccountthenon-LTElineformationofsodium. 1 Methods.The profiles of the two resonant sodium D lines (only these sodium lines are detectable in the spectra of EMP stars) have been 1 computedinasampleof54EMPgiantsandturn-offstars(33ofthemwith[Fe/H]<−3.0)withamodifiedversionofthecodeMULTI,and 0 comparedtotheobservedspectra. 7 Results.Withthesenewdeterminationsintherange−4<[Fe/H]<−2.5,both[Na/Fe]and[Na/Mg]arealmostconstantwithalowscatter. 0 In the turn-off stars and ”unmixed” giants (located in the low RGB): [Na/Fe]=−0.21±0.13 or [Na/Mg]=−0.45±0.16. These values / h are in good agreement with the recent determinations of [Na/Fe] and [Na/Mg] in nearby metal-poor stars. Moreover we confirm that all p the sodium-rich stars are ”mixed” stars (i.e., giants located after the bump, which have undergone an extra mixing). None of the turn-off - starsissodium-rich.Asaconsequenceitisprobablethatthesodiumenhancementobservedinsomemixedgiantsistheresultofadeepmixing. o r t s a Keywords. Line:Formation–Line:Profiles–Stars:Abundances–Stars:Mixing–Stars:Supernovae–Galaxyevolution : v i X 1. Introduction Such abundance ratios as [α/Fe], [Na, Al/Fe], and others can r bederivedfromhigh-resolutionspectraandthencomparedto a Duringthelastdecadetheinterestintheextremelymetal-poor theSNeIIyieldprediction.InparticularinCayreletal.(2004) (EMP)galacticstarshassignificantlyincreased.Thesestars,in andinBonifacioetal.(2006),ahomogeneoussampleofEMP fact,arereservoirsoftheprimordialGalacticmaterialthatwas stars has been studied: 35 giants and 18 turn-off stars (39 of only slightly polluted by nuclides producedin SNe II (by the themwith[Fe/H]≤−3).IthasbeenshownthatintheseEMP firstgenerationofmassivestars).Thus,elementalabundances stars the scatter of the abundanc e ratios at metallicity below providedbymetal-pooroldstars,andespeciallyabundancera- –3.0 dex is generally very small and comparable to the mea- tios, are of the highest importance for the testing of the SN surementerrors.Buttherearesomeexceptions,andoneofthe theoreticalmodels,theiryields,andfinallyforconstrainingthe most striking cases is [Na/Fe], which varies from star to star initialmassfunctionslopeintheearlyGalaxyandunderstand- byafactorclosetoten.IntheseEMPstars,onlytheresonance ingitsformationhistoryandearlyevolution. linesofNa(D1andD2)arevisible,andtheselinesareknown Withtheadventoflargeaperturetelescopes,thepreciseel- to be very sensitive to NLTE effects (e.g., Mashonkina et al. emental abundances in these faint objects became available. 1993;Baumu¨lleretal.1998). Sendoffprintrequeststo:M.Spite In Cayrel et al. (2004) and Spite et al. (2005) the sodium e-mail:[email protected] abundance has been calculated under the LTE hypothesis. 2 Andrievskyetal.:NLTEdeterminationoftheSodiumabundance ing power of 43000 in the region of the D lines and a typi- cal S/N ratio per pixel of ∼ 200; they have been reduced us- ing the UVES context (Ballester et al. 2000) within MIDAS. Stellar parameters (effective temperature, surface gravity, mi- croturbulentvelocity,and metallicity) for each star have been takenfromCayreletal.(2004)andBonifacioetal.(2006). 3. Methodofanalysis Toderivethesodiumabundanceintheprogramstars,weper- formanNLTEanalysisofthetworesonantsodiumD1andD2 lines(onlytheselinesareseeninthespectraoftheEMPstars) Fig.1.[Na/Fe]vs.[Fe/H]foranhomogeneoussampleofEMP usingamodifiedversionoftheMULTIcode(Carlsson1986). stars(Cayreletal.2004;Bonifacioetal.2006):dwarfs(black These modifications are described in Korotin et al. (1999ab). dots), low RGB stars (grey dots), and upper RGB stars (open To more adequately calculate the continuous opacity taking circles). The abundance of Na is deduced from the D lines into account the absorption produced by the great number of throughanLTEanalysis.Asamean,theratio[Na/Fe]islower spectral lines, the additional opacity sources from ATLAS9 in turn-off stars (black dots) than in RGB stars and is also (Kurucz 1992) are included. Simultaneous solution of the ra- lowerinlowerRGBstars(greydots)thaninupperRGBstars diative transfer and statistical equilibrium equations are ob- (opencircles).Thisgraphsuggestsavariationofthecomputed tainedusingtheapproximationofacompletefrequencyredis- sodiumabundancewiththegravity,whichcouldbetheconse- tributionforallthesodiumlines.Atmospheremodelsofappro- quenceofanNLTEeffect. priatemetallicityareinterpolatedinKurucz’sgridofstellarat- mospheremodels(microturbulentvelocity2kms−1,α=1.25). ThismodifiedcodehasalreadybeenusedforNLTEabundance determinationofcarbon,oxygen,andsodiuminstarsofdiffer- Later,asaveryfirstapproximation,auniformcorrectionLTE- enttypes(see,e.g.,Korotinetal.1999ab;Korotin&Mishenina NLTE of –0.5 dex, extrapolated in the table of Baumu¨ller 1999;Andrievskyetal.2001;Andrievskyetal.2002,andref- et al. (1998) for log g = 2, was applied to all the giants. erencestherein). The scatter of the relation [Na/Fe] versus [Fe/H] in giants We employ the modified model of the sodium atom (first was then very large, and [Na/Fe] was decreasing with the consideredbySakhibullin1987)thatconsistsof27energylev- metallicity in contrast to the turn-off stars (Spite et al., 2005) elsofNaIatomsinadditiontothegroundleveloftheNaIIion. where the ratio [Na/Fe] was almost constant in the interval The fine splitting has been taken into accountonly for the 3p −3.5<[Fe/H]<−2.5. level. This enables one to calculate the sodium doublet more In Fig. 1 we show the relation [Na/Fe] vs. [Fe/H] for our precisely. The radiative transitions between the first 20 levels sampleofdwarfsandgiantswithoutanyNLTEcorrection.The of Na I and the ground level of Na II are considered, while lower RGB stars (below the bump, see Spite et al. 2006) and transitions between the other levels are used for the particle theupperRGBstars(abovethebump)havebeenplottedwith number conservation. Linearization includes 46 b−b and 20 differentsymbols.Inthe mean,thedwarfshavealowervalue b − f transitions. Radiative rates for 34 transitions are fixed. of [Na/Fe] than the giants, and the lower RGB stars a lower Photoionization rates were taken from TOPBASE. Collisions value of [Na/Fe] than the upper RGB stars. This variation of withelectrons,aswellaswithhydrogenatoms,havebeenin- [Na/Fe]withthegravityofthestarsuggeststhat,asexpected, cluded. non-LTE effects are very pronounced on the Na D lines, and Figure2showsthebehaviourofthedeparturecoefficients that as a consequence it is important to carefully take them of Na I for dwarfs and giants of different metallicities. It is into account. In this paper we present new determinations of clearlyseenthatwithintheregionoftheline-formationinthe thesodiumabundanceinthesampleofEMPstarsofCayrelet atmosphere, the lower levels of the Na I atom are overpopu- al.(2004)andBonifacioetal.(2006)throughacarefulNLTE lated, leading to a larger opacity in D1,2 lines compared to analysis,andwediscussthenewtrendsobtained. the pure LTE case, and therefore to larger equivalent widths of these lines. This means that NLTE abundance corrections willbenegative. 2. Thestarsample To find the relative-to-solar sodium abundance, we com- The spectraused here havebeenpresentedin detailin Cayrel puted NLTE sodium abundance in the Sun. This was done etal.(2004)andBonifacioetal.(2006).Theobservationswere withthefollowinglines:4496.05,4982.81,5148.84,5682.63, performedwith the ESO VLT andits highresolutionspectro- 5688.20, 5889.95, 5895.92, 6154.22, 6160.75, 8183.25, graphUVES (Dekkeret al. 2000).The spectra have a resolv- 8194.82,and 12679.14Å. Their profiles were extracted from Andrievskyetal.:NLTEdeterminationoftheSodiumabundance 3 Log bi00..12 3s Teff = 6000 K [ FLeo/gH g] == -4 4 000...246 3s Teff = 6000 K [ FLeo/gH ]g = = - 4 2 RELATIVE FLUX00001.....67890 00001.....67890 4p 3p 3p 0.5 0.5 0.0 0.0 4p 0.4 BD+17 3248 0.4 5s 3d 0.3 0.3 4s -0.2 5s3d 4s 0.2 5889.5 5890.0 5890.5 λ(A) 0.2 5895.5 5896.0 5896.5 λ(A) -0.1 -0.4 E FLUX01..90 01..90 -0.2 V -4 -3 -2 -1Log τ50000 -4 -3 -2 -1Log τ05000 RELATI00..78 00..78 0.6 0.6 bi0.4 Teff = 5200 K Log g = 1.5 0.6 Teff = 5200 K Log g = 1.5 0.5 BS16477-003 0.5 Log 0.2 3s [Fe/H] = - 4 0.4 3s [Fe/H] = - 2 0.4 5889.5 5890.0 5890.5 λ(A) 0.4 5895.5 5896.0 5896.5 λ(A) 4p 0.2 4p 3p X1.0 1.0 5s U -00..20 33dp 4s --000...420 5s 4s3d RELATIVE FL000...789 000...789 0.6 CS29499-60 0.6 -0.6 -0.4 0.5 0.5 5889.5 5890.0 5890.5 λ(A) 5895.5 5896.0 5896.5 λ(A) -4 -3 -2 -1 0 -4 -3 -2 -1 0 Log τ5000 Log τ5000 X1.0 1.0 U E FL0.9 0.9 Fig.2.Departurecoefficientsinatmospheresofdwarfsandgi- ATIV0.8 0.8 antsofthedifferentmetallicities. EL R0.7 0.7 BS17570-63 0.6 0.6 UX1.0 1.0 0.5 5889.5 5890.0 5890.5 λ(A) 0.5 5895.5 5896.0 5896.5 λ(A) E FL0.9 0.9 TIV0.8 0.8 Fig.4. Best fit between observed (dots) and calculated A REL0.7 0.7 (solid line) profiles of the Na D1 (5889.95Å) and D2 0.6 0.6 (5895.92Å)lines,forsomeprogramstars. 0.5 BS16477-003 0.5 0.4 5889.5 5890.0 5890.5 λ(A) 0.4 5895.5 5896.0 5896.5 λ(A) Fig.3. The observed spectra of the D1 and D2 lines (dots) Fig. 3). Figure 4 shows the best fitting between observedand in BS 16477-003 are compared to the theoretical profiles calculated profilesfor some EMP giantsand dwarfs. In Table computed for the best-fit Na abundance +0.10 and – 0.10 1theparametersofthemodels(Teff,loggandvt)whichhave (solidlines). been used and the new sodium abundancein logarithmǫ(Na) (forǫ(H)=12)aregivenforeachstar.Thestellarratios[Na/H] havebeencomputedassumingǫ(Na) =6.25(seeSect.3). ⊙ the Kurucz et al. (1984) solar flux spectrum. Similarly to Three stars in our sample are very peculiar carbon- the stellar spectra analysis, to derive the NLTE sodium abun- rich stars: CS 22892-52 (Sneden et al. 1996, 2000, 2003), dance in the Sun, we used the solar atmosphere model from CS 22949-37 (Norris et al. 2001; Depagne et al. 2002), and the Kurucz’s grid and the solar microturbulent velocity rec- CS 29528-41 (Sivarani et al. 2006). The models we used for ommended by Maltby et al. (1986). Our NLTE solar sodium theNLTEcomputationsdonottakethecarbonenhancementin abundance is ǫ(Na)⊙ =6.25±0.04, in agreement with the these stars into accountand thus the sodium abundancecom- valuefoundbyBaumu¨lleretal.(1998):ǫ(Na)⊙ =6.30±0.03, putedheremustbeconsideredasverypreliminary.Thesestars and Mashonkina et al. (2000): ǫ(Na)⊙ = 6.20. The value arenottakenintoaccountinthediscussion. ǫ(Na)⊙ =6.25 will be used later, as a reference to determine InFig.5wepresentanestimateofthedifferenceinsodium [Na/H]. abundanceobtainedwhenLTEorNLTEprofilesoftheDlines areusedalongtheRed GiantBranchoftheHRdiagram.The computationshavebeendoneforasetoftemperaturesbetween 4. Results 6500K(turn-offstars)to4500K(upperRGBstars)andmetal- New sodiumabundanceshavebeendeducedfromNLTEpro- licities from−2.5to −4.0(Table2). Foreach temperaturethe files of the D lines for our sample of EMP stars. Typical un- gravity has been deduced from the isochrones of Kim et al. certaintyinthederivedsodiumabundanceisabout±0.05(see (2002) for 14 Gyr, [α/Fe]=+0.6, [Fe/H] = −3.7and − 2.7 4 Andrievskyetal.:NLTEdeterminationoftheSodiumabundance Table 1. Model parameters and NLTE sodium abundance in our sample of stars. The three stars marked with an asterisk are carbon-richstars.An”m”inthelastcolumndenotesthemixedgiantsfollowingSpiteetal.(2005). Star T , K log g v,kms−1 [Fe/H] ǫ(Na) [Na/H] [Na/Fe] Rem eff t Turnoffstars BS16023–046 6360 4.5 1.4 –2.97 3.33 –2.92 +0.05 BS16968–061 6040 3.8 1.5 –3.05 2.80 –3.45 –0.40 BS17570–063 6240 4.8 0.5 –2.92 3.12 –3.13 –0.21 CS22177–009 6260 4.5 1.2 –3.10 2.95 –3.30 –0.20 CS22888–031 6150 5.0 0.5 –3.28 2.70 –3.55 –0.25 CS22948–093 6360 4.3 1.2 –3.43 2.44 –3.81 –0.51 CS22953–037 6360 4.3 1.4 –2.89 2.90 –3.35 –0.46 CS22965–054 6090 3.8 1.4 –3.04 3.10 –3.15 –0.11 CS22966–011 6200 4.8 1.1 –3.07 2.95 –3.30 –0.23 CS29499–060 6320 4.0 1.5 –2.70 3.23 –3.02 –0.32 CS29506–007 6270 4.0 1.7 –2.91 3.15 –3.10 –0.19 CS29506–090 6300 4.3 1.4 –2.83 3.40 –2.85 –0.02 CS29518–020 6240 4.5 1.7 –2.77 3.20 –3.05 –0.25 CS29518–043 6430 4.3 1.3 –3.24 2.75 –3.50 –0.30 CS29527–015 6240 4.0 1.6 –3.55 2.50 –3.75 –0.20 CS29528–041* 6170 4.0 1.3 –3.06 3.85 –2.40 +0.66 CS30301–024 6330 4.0 1.6 –2.75 3.30 –2.95 –0.20 CS30339–069 6240 4.0 1.3 –3.08 3.00 –3.25 –0.17 CS31061–032 6410 4.3 1.4 –2.58 3.50 –2.75 –0.17 Giants HD2796 4950 1.5 2.1 –2.47 3.65 –2.60 –0.13 m HD122563 4600 1.1 2.0 –2.82 3.20 –3.05 –0.23 m HD186478 4700 1.3 2.0 –2.59 3.45 –2.80 –0.21 m BD+17:3248 5250 1.4 1.5 –2.07 4.44 –1.81 +0.26 m BD–18:5550 4750 1.4 1.8 –3.06 2.90 –3.35 –0.29 CD–38:245 4800 1.5 2.2 –4.19 2.15 –4.10 +0.09 m BS16467–062 5200 2.5 1.6 –3.77 2.38 –3.87 –0.10 BS16477–003 4900 1.7 1.8 –3.36 2.80 –3.45 –0.09 BS17569–049 4700 1.2 1.9 –2.88 3.25 –3.00 –0.12 m CS22169–035 4700 1.2 2.2 –3.04 3.30 –2.95 +0.09 m CS22172–002 4800 1.3 2.2 –3.86 2.15 –4.10 –0.24 CS22186–025 4900 1.5 2.0 –3.00 3.15 –3.10 –0.10 m CS22189–009 4900 1.7 1.9 –3.49 2.47 –3.78 –0.29 CS22873–055 4550 0.7 2.2 –2.99 3.40 –2.85 +0.14 m CS22873–166 4550 0.9 2.1 –2.97 3.20 –3.05 –0.08 m CS22878–101 4800 1.3 2.0 –3.25 2.95 –3.30 –0.05 m CS22885–096 5050 2.6 1.8 –3.78 2.50 –3.75 +0.03 CS22891–209 4700 1.0 2.1 –3.29 3.10 –3.15 +0.14 m CS22892–052* 4850 1.6 1.9 –3.03 3.00 –3.25 –0.22 CS22896–154 5250 2.7 1.2 –2.69 3.47 –2.78 –0.09 CS22897–008 4900 1.7 2.0 –3.41 2.69 –3.56 –0.15 CS22948–066 5100 1.8 2.0 –3.14 2.97 –3.28 –0.14 m CS22949–037* 4900 1.5 1.8 –3.97 3.85 –2.40 +1.57 m CS22952–015 4800 1.3 2.1 –3.43 3.33 –2.92 +0.51 m CS22953–003 5100 2.3 1.7 –2.84 3.20 –3.05 –0.21 CS22956–050 4900 1.7 1.8 –3.33 4.30 –1.95 -- CS22966–057 5300 2.2 1.4 –2.62 3.60 –2.65 –0.03 CS22968–014 4850 1.7 1.9 –3.56 2.35 –3.90 –0.34 CS29491–053 4700 1.3 2.0 –3.04 3.05 –3.20 –0.16 m CS29495–041 4800 1.5 1.8 –2.82 3.22 –3.03 –0.21 CS29502–042 5100 2.5 1.5 –3.19 2.63 –3.62 –0.43 CS29516–024 4650 1.2 1.7 –3.06 3.05 –3.20 –0.14 CS29518–051 5200 2.6 1.4 –2.69 3.40 –2.85 –0.16 m CS30325–094 4950 2.0 1.5 –3.30 2.84 –3.41 –0.11 CS31082–001 4825 1.5 1.8 –2.91 3.30 –2.95 –0.04 Andrievskyetal.:NLTEdeterminationoftheSodiumabundance 5 Table 2. Abundance corrections (ǫ(NLTE)−ǫ(LTE)) for the D1andD2lines. [M/H] -2.5 -3.0 -3.5 -4.0 Teff=4500logg=0.8 EW corr EW corr EW corr EW corr D1 182-0.48 138-0.63 107-0.44 81 -0.24 D2 161-0.54 122-0.52 93 -0.32 66 -0.14 Teff=4600logg=1.0 D1 167-0.62 - - - - - - D2 149-0.60 - - - - - - Teff=4750logg=1.4 D1 151-0.65 124-0.53 100-0.34 74 -0.13 D2 135-0.58 110-0.41 85 -0.20 57 -0.08 Teff=5000logg=2.0 D1 142-0.59 115-0.44 91 -0.24 63 -0.08 D2 126-0.52 100-0.31 75 -0.14 45 -0.06 Teff=5250logg=2.7 D1 137-0.56 108-0.39 82 -0.19 52 -0.06 D2 119-0.47 92 -0.28 64 -0.12 34 -0.06 Teff=5550logg=3.3 D1 122-0.49 95 -0.32 67 -0.14 38 -0.06 Fig.5.IndividualcorrectionapplyingtoanLTEanalysisofthe D2 105-0.38 79 -0.22 49 -0.09 23 -0.06 NaD1andD2linesalongtheRGB, formetal-poorstarswith −4.0<[Fe/H]<−2.5andthehypothesisthat[Na/Fe]=0. Teff=6000logg=3.7 D1 101-0.39 75 -0.22 45 -0.10 21 -0.06 D2 85 -0.28 57 -0.13 29 -0.08 12 -0.06 Teff=6350logg=4.1 withtheclassicalformula: D1 88 -0.34 61 -0.18 32 -0.09 13 -0.06 D2 72 -0.23 43 -0.12 19 -0.08 7 -0.06 logL/L =logM/M +4logT /T −logg/g ⊙ ⊙ eff eff,⊙ ⊙ 5. Discussion For this set of models, the LTE profiles have been com- 5.1.Mixingeffectsingiantstars putedwiththehypothesisthat[Na/H]=[Fe/H](i.e.,[Na/Fe]= 0),andthenfittedbyNLTEprofilesthatenabledthederivation InFig.7wepresenttheNLTEvaluesof[Na/Fe]foroursample of the appropriate NLTE corrections. A first cause of uncer- ofstarsasafunctionof[Fe/H](thecarbon-richstarshavebeen taintyisthattheshapesofLTEandNLTEprofilesareslightly omitted).ThesymbolsarethesameasinFig1.Dwarfsandgi- different,butthe correctionmainlydependson the equivalent antsnowlieinthesamelocus,andthetrendofthe[Na/Fe]ra- widthofthe sodiumlines,as canbeseen in Fig.6. Asa con- tio,intherange−4.0<[Fe/H]<−2.5,isflat.Themeanvalue sequence,thecorrectionsgivenin Table2 canbeusedonlyif of[Na/Fe]iscloseto–0.2. theequivalentwidthofthelinesissimilarto thatgiveninthe A few stars appear to be ”Na-rich” with [Na/Fe]>+0.1. table. If there is a strong anomaly of the sodium abundance Allthesestarsbelong,followingSpiteetal.(2006),totheclass in the star, it is better to deduce the correction from the Fig. ofthe”mixedstars”.Theypresentthecharacteristicsofamix- 6 (or to directly compute the non-LTE profile of the line as ing with the H-burningshell, low abundanceof carboncorre- it has been done for the stars of our sample). Takeda et al. lated with a high abundance of nitrogen, and a low value of (2003)havemadeafirstattempttocomputethenon-LTEcor- the 12C/13C ratio close to the equilibriumvalue.However,all rection of the sodium abundance in a large metallicity range themixedstarsarenotNa-rich.Letusremarkthatallthestars (−4.0<[Fe/H]<+0.4). We havecheckedthatinourdomain foundherewith [Na/Fe]>+0.1were alreadysuspected from ofmetallicity([Fe/H]<−2.5)thereisarathergoodagreement LTEcomputations,tobeNa-rich.Allofthem,butBD17:3248, withtheircomputations. havealsobeenfoundtobealuminumrich,andfromtheirpo- 6 Andrievskyetal.:NLTEdeterminationoftheSodiumabundance Fig.7. [Na/Fe] vs. [Fe/H] for the sample of EMP stars. The symbolsarethesameasinFig.1.Thesodiumabundancehas been computed in NLTE with a modified version of the pro- gram of Carlsson. The trend of the [Na/Fe] ratio is flat in the region −4.0<[Fe/H]<−2.5, and there is a good agreement betweentheabundancesofthegiantsandturn-offstars.Allthe starswith[Na/Fe]>0.1havebeenfoundtobeextra-mixed,in Spiteetal.(2005). LTEeffects.InFig.8wecomparetheratio[Na/Fe]inourEMP Fig.6.Individualcorrectionsplottedasafunctionoftheequiv- starsandinthesampleofGehrenetal.(2006);unfortunately, alentwidthsofthelinesforthedifferentmodelsgiveninTable we haveno star in commonto checktheinfluenceof system- 2. The solid line represents the corrections for models with aticeffects,buttheyhaveonestarinourdomainofmetallicity [Fe/H]=-2.5, the dashed lines models with [Fe/H]=-3.0, the (G64-12 with [Fe/H]=–3.1), and this star falls exactly in the dotted-dashed lines models with [Fe/H]=-3.5, and the dotted middleofoursampleinFig.8. linesmodelswith[Fe/H]=-4.0.Forasameequivalentwidthof Sodium has only one (stable) isotope, 23Na, which is the D lines the correction little depends on the parameters of mainly produced in the carbon burning process operating in- themodels sidemassivestars.AsnotedbyWoosley&Weaver(1995)the resulting amount of sodium nuclei in the SN ejecta is sensi- tivetotheexcessofneutrons,thereforeitsproductionissome- sition in the HR diagram they could be AGB stars (see also whatmetaldependent,butintheearlyGalaxy(verylowmetal Johnson(2002),andBond(1980)forBD17:3248).Noneofthe content)thiselementproductionshouldbethesameasforthe turn-off stars or ”unmixed”giants are sodium rich. As a con- primary elements. There were several attempts to incorporate sequence the sodium rich stars are suspected to have suffered theobservedsodiumabundanceinthediscandhalostarsinto a mixing between the atmosphere and the H-burning shell, a Galacticevolutionarymodels.Onecanmentionheretheworks mixing deep enough to bring the products of the Ne-Na cy- of Timmes et al. (1995), Samland (1998), and Goswami & cletothesurface.WhenonlyEMPdwarfsandunmixedgiants Prantzos(2000).Alltheabove-mentionedmodelsarebasedon are taken into account, we find that [Na/Fe]=−0.21±0.13 the Woosley & Weaver (1995) yields from massive stars, but or[Na/Mg]=−0.45±0.16;thestar-to-starvariationbecomes use different initial-mass functions, star formation rates, and smallandcomparabletowhatisobtainedfortheotherelements othermodelspecifications,whichcanexplaintheslightdiffer- (seeCayreletal.2004). encesinthepredictions. InFig.8ourobservationsandthoseofGehrenetal.(2006) 5.2.Comparisonwiththenearbymetal-poorstars, arecomparedtothepredictionsofTimmesetal.(1995).These authors predict a local minimum around [Fe/H] = –1.5 with andthepredictionofthemodelsofgalactic subsequent increasing of [Na/Fe] as the metallicity decreases chemicalevolution to[Fe/H]=–3.Thepredictionsfortwodifferentvaluesofthe RecentlyGehrenetal.(2006)havecomputedthesodiumabun- ironyieldsfrommassivestars(whichdependonthemasscutin danceinasampleof55nearbymetal-poorstars,mostofthem themassivesupernovae)arerepresentedinthefigure.Between belongingtothethinorthickGalacticdiscandsomeofthemto modelsT(A)andT(B),the massofironejectedbySN IIdif- thehalo.Theyhaveverycarefullytakenintoaccountthenon- fers by a factor of two. At a metallicity higher than [Fe/H]= Andrievskyetal.:NLTEdeterminationoftheSodiumabundance 7 Fig.8.[Na/Fe]vs.[Fe/H]forthesampleofEMPstars(Cayrel Fig.9. [Na/Mg] vs. [Fe/H] for the sample of EMP stars and et al. 2004; Spite et al. 2005 ), dwarfs (black dots), unmixed the nearby metal-poor stars studied by Gehren et al. (2006). giants(greydots), andthe nearbymetal-poorstars studied by Symbols are the same as in Fig. 8. The dashed line repre- Gehrenet al. (2006) (open triangles).The dashed lines repre- sents the prediction of the model of Timmes et al. (1995). sentthepredictionsoftwomodelsofTimmesetal.(1995)that (The [Na/Mg] ratio is independent of the mass cut at vari- differbythequantityofironejectedbymassivestarsandthus ance with [Na/Fe] and the predictions are reduced to a sin- bythepositionofthemasscut. gle curve). It can be seen in the figure that in the interval −4<[Fe/H]<−2.8 [Na/Mg] is constant. However, a slight increase of [Na/Mg] when the metallicity decreases (from [Na/Mg]= –0.5 at [Fe/H]=–3. to [Na/Mg]= –0.3 at [Fe/H]=– –3, there is a rather good agreement with the predictions of 4.)cannotbeexcluded. themodelT(A),whichsupposesthesmallestvalueoftheiron yieldsfrommassivestars,andsuchanagreementisgenerally foundfortheotherelements(seeCayreletal.2004). Incontrastwith[Na/Fe],theratio[Na/Mg]isindependent the distribution of EMP stars in the [Na/Mg] or [Al/Mg] ver- ofthemasscut.InFig.9wepresent[Na/Mg]asafunctionof sus [Mg/H] plane. The observed trends and the scatter of the [Fe/H]foroursampleofEMPstarsandthesampleofGehren observations are mainly based on the work of McWilliam et et al. (2006). (We neglected the NLTE effects on the magne- al. (1995). This sample contains several mixed giants which sium lines that are not supposed to be very important). The explain the very large scatter of [Na/Mg] and [Al/Mg] they agreementisrathergoodfor[Fe/H]>−3. observed at low metallicity. Such a scatter, which is an im- Timmes et al. (1995) do not predict the behaviour of portant point in the definition of the model of Tsujimoto et [Na/Fe]or[Na/Mg]for[Fe/H]<−3.Fromourmeasurements, al. (2002), is not observed in the unmixed stars of our sam- [Na/Fe]and[Na/Mg]becomeflat:therelativeabundancesare ple(dwarfsand”unmixed”giants)andthusdoesnotprobably independent of the metallicity. However, a slight increase of reflectascatterinthepristinematerial.LetusnotethatCohen [Na/Mg]whenthemetallicitydecreases(from[Na/Mg]=–0.5 et al. (2004), who studied a sample of turnoff stars in the in- at[Fe/H]=–3.to[Na/Mg]=–0.3at[Fe/H]=–4.)cannotbeex- terval−3.6<[Fe/H]<−2.2,alsofoundaverysmallscatterof cluded. [Al/Mg].Theprecise determinationof[Al/Mg]in oursample ofstarsfromnon-LTEsyntheticprofilesisinprogress. Samland (1998) has considered the chemical evolution of many elements and found that the expected distribution of [Na/Fe] vs. [Fe/H] in the extreme metallicity region from 6. Conclusion [Fe/H] = –4 to –2.5 is almost independent of [Fe/H], and the mean [Na/Fe] value equals approximately –0.2. This was NLTE abundances of sodium have been derived for not supported by observations available at that time, but is a sample of extremely metal-poor dwarfs and giants us- in good agreement with our observations. Qualitatively the ing high-resolution spectra obtained with VLT. Both sub- sameconclusionwasreachedbyGoswami&Prantzos(2000). samples of dwarfs and unmixed giants now have the According to their model, [Na/Fe] must be almost constant same mean [Na/Fe] or [Na/Mg] values, within the standard within this metallicity interval and equal approximately –0.5, errors: [Na/Fe]=−0.21±0.13 or [Na/Mg]=−0.45±0.16. butthisissignificantlylowerthanwhatisgivenbyourcalcu- However,aslightincreaseof[Na/Mg]atlowmetallicity,(from lations. [Na/Mg]= –0.5 at [Fe/H]=–3. to [Na/Mg]= –0.3 [Fe/H]=–4.) RecentlyTsujimotoetal.(2002)havealsoinvestigatedthe cannotbeexcluded. chemicalevolution of sodium and aluminum duringthe early •Inthedomain−4<[Fe/H]<−2.5,thegiantsofoursam- epochsoftheGalaxy.Theyhaveusedamodelthatreproduces pledonotexhibitanysystematicdependenceof[Na/Fe]upon 8 Andrievskyetal.:NLTEdeterminationoftheSodiumabundance [Fe/H], contrary to the preliminary findings of Cayrel et al. GoswamiA.,PrantzosN.,2000,A&A359,191 (2004). This effect was a consequence of a coarse treatment JohnsonJ.A.,2002,ApJ139,219 of the non-LTEeffects: in Cayrel et al. (2004) the LTE abun- KimY.-C.,DemarqueP.,YiS.K.,AlexanderD.R.,2002, ApJS143, danceshadbeenuniformelycorrectedby–0.5dex.Infact,the 499 KorotinS.A.,AndrievskyS.M.,LuckR.E.,1999a,A&A351,168 non-LTEcorrectionstrongly dependson the equivalentwidth Korotin S.A., Andrievsky S.M., Kostynchuk L.Yu., 1999b, Ap&SS ofthesodiumlinesandthusisgenerallysmalleratlowermetal- 260,531 licity. (When the lines are smaller they formdeeperin the at- KorotinS.A.,MisheninaT.V.,1999,ARep43,533 mosphere where, in particular, collisions are more important KuruczR.,1992,RMxAA23,45 andthusnon-LTEcorrectionsaresmaller.) KuruczR.L.,FurenlidI.,BraultJ.,TestermanL.,1984,NationalSolar •Amongthegiantstars,the”mixed”giants(starsthathave Observatory,Atlas,Sunspot,NewMexico undergone an extra-mixing between the atmosphere and the MaltbyP.,AvrettE.H.,CarlssonM.,Kjeldseth-MoeO.,KuruczR.L., H-burningshell following Spite et al. 2005), often presentan LoeserR.,1986,ApJ306,284 overabundanceofNa.Itissuggestedthatinthesestars,mate- MashonkinaL.I.,SakhibullinN.A.,ShimanskijV.V.,1993,ARep37, rialprocessedinNe-Nacycleinthehothydrogenburningshell 192 hasbeenbroughttothesurface.FromtheirpositionintheHR MashonkinaL.I.,ShimanskiiV.V.,SakhibullinN.A.,2000,ARep44, 790 diagram,someofthesestarscouldbeAGBstars. McWilliam A., Preston G.W., Sneden C., Searle L., 1995, AJ 109, • Qualitatively, our observational result on sodium distri- 2757 butionin themetallicityregion[Fe/H]from–4to–2.5agrees Mishenina T.V., Kovtyukh V.V., Korotin S.A., Soubiran C., 2003, withpredictionsofmodelselaboratedbySamland(1998)and ARep47,422 Goswami & Prantzos (2000). Both models produce the al- NorrisJ.E.,RyanS.G.,BeersT.C.,2001,ApJ561,1034 most constant [Na/Fe] value in the region of extreme metal- SamlandM.,1998,ApJ496,155 poorstars.Inparticular,theSamland’smodelgivesanaverage SivaraniT.,BeersT.C.,BonifacioP.,MolaroP.,etal.2006,(submitted [Na/Fe] value that approximately equals –0.2, and this value toA&A) appearstobeinquantitaviveagreementwithourestimateofthe SnedenC.,McWilliamA.,PrestonG.W.,etal.,1996,ApJ467,819 mean relative-to-iron sodium abundance in the early Galaxy. SnedenC.,CowanJ.J.,IvansI.I.,etal.,2000,ApJ533,L139 ThemodelofTimmesetal.(1995)alsoagreeswithourobser- SnedenC.,CowanJ.J.,LawlerJ.E.,IvansI.I.,BurlesS.,BeersT.C., PrimasF.,HillV.,etal.,2003,ApJ591,936 vationsaround[Fe/H]=–3.Inqualitativeagreement,themodel SpiteM.,CayrelR.,PlezB.,HillV.,F.Spite,E.Depagneetal.,2005, of Goswami & Prantzos predicts a slightly too low relative A&A430,655 sodiumabundanceforextremelymetal-poorstars. Spite M., Bonifacio P., Cayrel R., Hill V., Francois P., Spite F. et al., 2005 in the Proceedings of the IAU symposium 228 Acknowledgements. SMA kindly acknowledges the C.N.R.S./INSU ”FromlithiumtoUranium:ElementalTracersofEarlyChemical forfinancialsupportandtheParis-MeudonObservatoireforitshospi- Evolution”,V.Hill,P.Franc¸ois,&F.Primaseds.,p.185 talityduringhisproductivestayinMeudon. SpiteM.,CayrelR.,HillV.,SpiteF.,FrancoisP.,PlezB.,Bonifacio P., Molaro P., Depagne E. et al., 2006, A&A (in press), astro- ph(06050561) References Takeda Y., Zhao G., Takada-Hidai M., Chen Y.-Q., Saito Y. Zhang Andrievsky S.M., Kovtyukh V.V., Korotin S.A., Spite M., Spite F., H.W.,2003,ChJAA3,316 2001,A&A367,605 TimmesF.X.,WoosleyS.E.,WeaverT.A.,1995,ApJS98,617 AndrievskyS.M.,EgorovaI.A.,KorotinS.A.,BurnageR.,2002,A&A TsujimotoT.,ShigeyamaT.,YoshiY.,2002,ApJ565,1011 389,519 WoosleyS.E.,WeaverT.A.,1995,ApJS101,181 BallesterP.,ModiglianiA.,BoitiquinO.,CristianiS.,HanuschikR.et al.,2000,ESOMessenger101,31 Baumu¨llerD.,ButlerK.,GehrenT.,1998,A&A338,637 BondH.E.1980ApJS44517 CarlssonM.,1986,UppsalaObs.Rep.33 CayrelR.,DepagneE.,SpiteM.,HillV.,SpiteF.,FrancoisP.,PlezB., BeersT.,PrimasF.,AndersenJ.,BarbuyB.,BonifacioP.,Molaro P.,Nordstro´mB.,2004,A&A416,1117 CohenJ.,ChristliebN.,McWilliamA.,SchectmanS.,ThompsonI.et al.,2004,ApJ612,1107 Dekker H., D’Odorico S., Kaufer A., et al. 2000 in Optical and IR TelescopesInstrumentationandDetectors,edsI.Masanori&A.F. MorwoodProc.SPIE4008,534 DepagneE.,HillV.,SpiteM.,SpiteF.,PlezB.,etal.,2002,A&A390, 187 GehrenT.,ShiJ.R.,ZhangH.W.,ZhaoG.,KornA.J.,2006,A&A451,