The Chemical Abundances of Stars in the Halo (CASH) Project. II. A Sample of 16 Extremely Metal-poor Stars
Julie K. Hollek, Anna Frebel, Ian U. Roederer, Christopher Sneden, Matthew Shetrone, Timothy C. Beers, Sung-Ju Kang, Christopher Thom
aa r X i v : . [ a s t r o - ph . GA ] A ug Draft version August 12, 2018
Preprint typeset using L A TEX style emulateapj v. 5/14/03
THE CHEMICAL ABUNDANCES OF STARS IN THE HALO (CASH) PROJECT. II.A SAMPLE OF 16 EXTREMELY METAL-POOR STARS
Julie K. Hollek , Anna Frebel , Ian U. Roederer , Christopher Sneden , Matthew Shetrone , Timothy C.Beers , Sung-ju Kang , and Christopher Thom Draft version August 12, 2018
ABSTRACTWe present a comprehensive abundance analysis of 20 elements for 16 new low-metallicity stars fromthe Chemical Abundances of Stars in the Halo (CASH) project. The abundances have been derivedfrom both Hobby-Eberly Telescope High Resolution Spectrograph snapshot spectra (R ∼ , ∼ , − − < − .
7. We find four starsto be carbon-enhanced metal-poor (CEMP) stars, confirming the trend of increasing [C/Fe] abundanceratios with decreasing metallicity. Two of these objects can be classified as CEMP-no stars, addingto the growing number of these objects at [Fe/H] < −
3. We also find four neutron-capture enhancedstars in the sample, one of which has [Eu/Fe] of 0.8 with clear r-process signatures. These pilot samplestars are the most metal-poor ([Fe/H] . − .
0) of the brightest stars included in CASH and are usedto calibrate a newly-developed, automated stellar parameter and abundance determination pipeline.This code will be used for the entire ∼
500 star CASH snapshot sample. We find that the pipelineresults are statistically identical for snapshot spectra when compared to a traditional, manual analysisfrom a high-resolution spectrum.
Subject headings:
Galaxy: halo—methods: data analysis—stars: abundances—stars: atmospheres—stars: Population II INTRODUCTION
The first stars formed from metal-free material in theearly universe and therefore are thought to have beenmassive ( ∼ M ⊙ ; e.g., Bromm et al. 1999). Many ofthese first stars polluted the surrounding local environ-ment with their chemical feedback through core-collapsesupernovae. From this enriched material, subsequentgenerations of stars were born. Due to the presence ofadditional cooling mechanisms, these stars had a rangeof lower masses and thus were longer lived (e.g., Bromm& Loeb 2003).Today, we observe the surviving low-mass stars as themost metal-poor stars in the Galaxy. The atmospheresof these objects contain the chemical signatures of earlysupernova events. By studying these stars, constraintscan be placed on the specific types of nucleosynthetic Based on observations obtained with the Hobby-EberlyTelescope, which is a joint project of the University of Texas atAustin, the Pennsylvania State University, Stanford University,Ludwig-Maximilians-Universit¨at M¨unchen, and Georg-August-Universit¨at G¨ottingen. Based on observations gathered with the 6.5 meter MagellanTelescopes located at Las Campanas Observatory, Chile. Department of Astronomy, University of Texas at Austin,Austin, TX 78712; julie,chris,[email protected] Harvard-Smithsonian Center for Astrophysics, Cambridge,MA 02138; [email protected] Carnegie Observatories, Pasadena, CA 91101;[email protected] McDonald Observatory, University of Texas, Fort Davis, TX78734 Department of Physics and Astronomy, and JINA: JointInstitute for Nuclear Astrophysics, Michigan State University,East Lansing, MI 48824; [email protected] Department of Physics and Astronomy, Iowa State University,Ames, IA 50011; [email protected] Space Telescope Science Institute, Baltimore, MD 21218;[email protected] events responsible for the observed abundances patterns.Efforts to classify metal-poor stars have been basedupon metallicity, [Fe/H] , and chemical composition,[X/Fe], to better understand the diversity of the ob-served abundance patterns (Beers & Christlieb 2005).Stars with [Fe/H] < − . ∼
25 of these EMP stars have [Fe/H] . − .
5. Be-low [Fe/H] ∼ − . < − < − .
0; Christlieb et al. 2004; Frebel et al.2005; Norris et al. 2007).The majority of metal-poor stars ([Fe/H] < −
1) showabundance patterns similar to the Solar System, butscaled down by metallicity, with two main differences:there is an enhancement in the α -elements (e.g., [Mg/Fe])and a depletion in some of the Fe-peak elements(e.g., [Mn/Fe]) compared to the solar abundance ratios.This pattern can be explained with enrichment by pre-vious core-collapse supernovae (e.g., Heger & Woosley2010). The chemical outliers among stars with [Fe/H] < − .
0, which make up perhaps 10%, show great diver-sity in their abundance patterns. Many stars have over-abundances in selected groups of elements, e.g., therapid (r) neutron-capture process elements (Sneden et al.2008) and/or the slow (s) neutron-capture process ele-ments. The frequency of chemically unusual stars in- [A/B] ≡ log( N A /N B ) − log( N A /N B ) ⊙ for N atoms of elementsA, B, e.g., [Fe/H] = − . Hollek et al.creases with decreasing metallicity, with stars often be-longing to multiple chemical outlier groups. Not in-cluded in this estimate of chemically unusual stars arethe so-called carbon-enhanced metal-poor (CEMP) stars(where [C/Fe] > . ∼
15% ofstars with [Fe/H] < − .
0. At the lowest metallicities,the frequency of CEMP stars also increases (Beers &Christlieb 2005; Lucatello et al. 2006; Frebel et al. 2006;Cohen et al. 2006; Carollo et al. 2011). In fact, all three[Fe/H] < − . ∼ , ∼ , ∼ . ∼ , , ∼
40) fill the gapbetween time intensive high- and medium-resolution ob-servations. From such snapshot data, abundances for ∼
15 elements can be derived with moderate uncertain-ties ( ∼ .
25 dex; Barklem et al. 2005). This allows fora more efficient confirmation of EMP stars and chemicaloutliers. The Barklem et al. (2005), Hamburg/ESO R-process Enhanced Star, (hereafter HERES) study itselfdetermined abundances (and upper limits) for a total of ∼
250 stars based on VLT/UVES snapshot spectra.The Chemical Abundances of Stars in the Halo(CASH) project is a dedicated effort that aims to pro-vide abundances for ∼
500 metal-poor stars primarilybased on R ∼ , − ∼ . ∼ − . − .
9. The aim of the present study istwofold: to present abundance analyses for the 16 mostmetal-poor stars included in the CASH project and touse those abundances to calibrate the newly developedstellar parameter and abundance pipeline. We will useit to obtain abundances for the full ∼
500 star CASHsample (Hollek et al. 2012, in prep.).In Section 2 we discuss the spectra in terms of thesample selection, observational information, and data re-duction. In Section 3 we introduce our spectral analysistools, including our linelist, equivalent width measure-ment routines, and model atmosphere analysis code. InSections 4 and 5 we describe acquisition of our stellarparameters for both sets of data and a comparison be-tween the two. In Section 6 we discuss the abundanceanalysis methods for each element we measure, includ-ing the error analysis and comparison of our results tothose in the literature. Section 7 includes a summaryof our abundance results and discussion of the implica- tions of our derived abundances. In Section 8 we list oursummary. OBSERVATIONS
Sample Selection
The stars of the pilot study were chosen from the Ham-burg/ESO Bright Metal-Poor Sample (BMPS; Frebelet al. 2006) of the Hamburg/ESO objective-prism plateSurvey (HES; Christlieb et al. 2001). These stars(B < < − . ∼ − .
0; however, these spectra require further processingin order to derive accurate stellar parameters and abun-dances. The snapshot-derived results for these stars willbe included in a future paper, though we present theabundances derived from the high-resolution MIKE spec-tra here.
Spectroscopy
The snapshot spectra for the CASH project were ob-tained using the fiber-fed HRS on the HET at McDonaldObservatory. The CASH spectra were obtained with a2 ′′ fiber yielding R ∼ , × − ∼
65, with a median S/N value for the pilotsample of 70; see Table 1. There is substantially lowerS/N at the blue ends of the spectra, given the combina-tion of the somewhat poor blue response of the HRS andthe lack of blue flux for many of the objects observed inCASH, especially the cool giants.High-resolution spectra for 21 stars were obtained us-ing the MIKE instrument (Bernstein et al. 2003) on theMagellan-Clay Telescope at Las Campanas Observatory.We used the 0 ′′ .7 slit with 2 × ∼ ,
000 in the blue and28,000 in the red with average S/N ∼
85 at 5200 ˚A. MIKEASH II. A Sample of 16 Extremely Metal-poor Stars 3
Fig. 1.—
MIKE (left) and HRS (left) spectra of three stars with derived spectroscopic T eff ∼ spectra have nearly full optical wavelength coverage from ∼ Data Reduction
The HRS spectra were reduced using the IDL pipelineREDUCE (Piskunov & Valenti 2002), which performsstandard echelle reduction techniques (trimming, biassubtraction, flat fielding, order tracing, extraction). Thedata were wavelength calibrated using ThAr lamp expo-sures taken before or after every observation. Compar-isons have been made between a by-hand IRAF reduc-tion and the REDUCE reduction of medium-S/N HRSdata. Both yield comparable S/N across the spectrum,and the measured equivalent widths for 121 different linesdiffer between the two different reductions by 3 ±
8m ˚A,which is statistically insignificant (Roederer et al. 2008).In addition, earlier tests of REDUCE versus IRAF haveshown that the optimized extraction in REDUCE forhigh S/N spectra yields an extracted spectrum that isless noisy than that of a spectrum extracted in IRAF (seeFigure 8 in Piskunov & Valenti 2002). Standard IRAFroutines were then used to coadd (in the case of multipleobservations) and continuum normalize the individualobservations into a final one-dimensional spectrum. Ra- IRAF is distributed by the National Optical Astronomy Ob-servatories, which is operated by the Association of Universities forResearch in Astronomy, Inc., under cooperative agreement with theNational Science Foundation. dial velocities (RVs) were computed by cross-correlatingthe echelle order containing the Mg b triplet against an-other metal-poor giant observed with the same instru-mental setup. Typical uncertainties were 2 − and then normalized and coadded using thesame method as the HRS spectra. SPECTRAL ANALYSIS
Linelist x The linelist to analyze the HRS spectra was basedon the lines included in Roederer et al. (2010). Onlythose lines which are unblended at the median S/N andresolution of the typical HRS snapshot spectrum wereincluded in our final list.The linelist for the MIKE data is a composite of thelines from Roederer et al. (2010), supplemented with ad-ditional lines from Cayrel et al. (2004), and Aoki et al.(2007b). This linelist includes those lines used for theHRS snapshot spectra analysis. In the instances wherethe same line was included in more than one linelist, themost up to date oscillator strength was used, followingRoederer et al. (2010). We confirmed that all impor-tant lines for our abundance analysis were included in available at http://obs.carnegiescience.edu/Code/python Hollek et al.
Table 1. Observations
Star Telescope UT Date RA Dec t exp
S/N v rad (J2000) (J2000) sec at 5180 ˚A km s − HE 0013 − −
02 41 06 630 65 47.7Magellan 28 Sep 2006 600 34 45.4HE 0013 − −
05 05 52 678 85 − − − − − −
34 06 06 300 88 121.7HE 0324+0152a HET 24 Feb 2008 03 26 53.8 +02 02 28 316 65 107.3Magellan 27 Sep 2006 450 70 106.1HE 0420+0123a HET 05 Jan 2008 04 23 14.4 +01 30 49 207 180 a − · · · − − − −
09 59 36 890 80 a · · · − b
11 18 35.8 −
06 50 46 · · · · · · · · ·
Magellan 03 Jul 2010 1200 121 115.5HE 1311 − −
01 47 16 250 45 125.8Magellan 05 Aug 2010 2736 48 124.7HE 1317 − b
13 19 47.0 −
04 23 10 · · · · · · · · ·
Magellan 03 Jul 2010 487 135 124.7HE 2123 − −
03 16 58 1473 65 − − − −
03 01 17 600 85 a − · · · − − − b
21 50 41.5 −
10 50 58 · · · · · · · · ·
Magellan 06 Aug 2010 300 61 − − −
01 16 16 494 60 − − − −
08 33 28 427 75 − − − −
21 38 07 450 55 − − − −
38 245 Magellan 27 Jul 2009 00 46 36.2 −
37 39 33 250 65 47.1CS 22891 −
200 Magellan 05 Aug 2010 20 19 22.0 −
61 30 15 900 52 137.7CS 22873 −
166 Magellan 27 Jul 2009 19 35 19.1 −
61 42 24 120 53 − −
18 5550 Magellan 27 Jul 2009 19 58 49.7 −
18 12 11 87 157 − a S/N for combined HET spectra b Sky contamination in spectrum, excluded from this analysis this linelist by plotting the position of each line againsta high-resolution MIKE spectrum of a star with a highermetallicity than that of any star in the pilot sample.Compared to the sample stars, this star displays manymore absorption lines. This also allowed us to visually in-spect for features that were not present in both an EMPstar and star that is still considered to be metal-poor,but with substantially higher (1 dex) [Fe/H].
Line Measurements
In each spectrum, we measured equivalent widths ofunblended lines of various elements. Table 2 lists theelement and ionization state, equivalent widths, wave-length, excitation potential, and oscillator strengths ofeach measured line in the MIKE spectra. These equiva-lent widths were used to determine stellar parametersand abundances for ∼
10 elements. The equivalentwidths in the HRS spectra were measured using an IDLroutine. Here we briefly summarize the features impor-tant to this work. The routine works to automatically fita Voigt profile to each line. The user can then manuallyadjust the continuum level, the number of spectral points over which the line is fit, and the line center, among otherfeatures. Given the limitations in resolution and S/N,some lines used in the analysis for HRS spectra departfrom the linear portion of the curve of growth, thus linefits using the Voigt profile, rather than simply a Gaussianfit, are preferred.The equivalent widths in the high-resolution spectrawere measured with an ESO/Midas routine which au-tomatically fits Gaussian profiles to each line. The usercan calculate a fit to the continuum level by selecting linefree continuum regions. This code takes into account anypossible non-zero slope of the continuum.We chose a different equivalent width measurementroutine because the higher S/N of the MIKE spectraallowed us to detect deviations from zero in the slopeof the continuum that may arise from small-scale varia-tion in imperfectly normalized spectra or nearby stronglines. Thus, it was helpful to be able to make a linearfit when determining the continuum. Additionally, thelarger wavelength range allowed for more lines to be mea-sured, thereby enabling us to exclude lines near the flatpart of the curve of growth. Table 2 lists the equivalentASH II. A Sample of 16 Extremely Metal-poor Stars 5
Table 2. Equivalent Widths
Star Ion Wavelength XP log gf EW˚A m ˚AHE 0015+0048 12.0 3986.75 4.35 − − − − − − − − − Note . — Table 2 is published in its entirety in the electronicedition. A portion is shown here for guidance regarding its formand content. We list the ionization state of each element where .0indicates a neutral species and .1 indicates a singly-ionized species. widths for all our stars in the pilot sample.Figure 2 shows our measured equivalent widths fromthe MIKE spectrum of BD −
18 5550 plotted against theequivalent widths measured for the same star by Cayrelet al. (2004). We find a mean difference of − σ = 2 . −
18 5550 because it is not observable from McDon-ald Observatory. We find a mean difference of 2.5 m˚Awith σ = 3 . Analysis Techniques
The large number of stars in the full ∼
500 star CASHsnapshot sample calls for automation of the analysis.Stellar parameters and elemental abundances from thesnapshot HRS spectra were determined using our newlydeveloped spectroscopic stellar parameter and abun-dance analysis pipeline, Cashcode. The pipeline is writ-ten around the existing platform of the local thermody-namic equilibrium (LTE) stellar line analysis and spec-trum synthesis code MOOG (the latest version, 2010;Sneden 1973). The most recent version of MOOG ac-counts for the fact that Rayleigh scattering becomes animportant source of continuum opacity at short wave-lengths, blueward of 4500 ˚A. This is important for oursample stars, as the effect is more pronounced in coolgiants (Sobeck et al. 2011).We compared the results of four representative stars,two from the pilot sample and two standard stars, us-ing the newest version of MOOG and an older versionthat did not distinctly deal with Rayleigh scattering inthe calculation of the continuum opacities. We find thatthe spectroscopic effective temperatures and microturbu-lences are lower, thus the derived [Fe/H] abundances arelower in the version that deals with Rayleigh scattering
Fig. 2.—
Comparison of the equivalent widths measured fromthe MIKE and Cayrel et al. (2004) spectra of BD −
18 5550 . Inthe upper panel, residuals of the equivalent widths (Cayrel − MIKE)are plotted against wavelength. In the bottom panel, the measuredequivalent widths from Cayrel and MIKE are plotted against eachother. by ∼ . . ∼ .
05 dex. These abundances were de-termined using lines down to ∼ STELLAR PARAMETERS
The first step in our abundance analysis is to deter-mine the atmospheric parameters of each star. We ac-complished this in two ways: with the Cashcode pipelineand the traditional, manual way.In order to test the robustness of the snapshot abun-dances one must answer two questions: first, for a givenset of stellar parameters, with what precision can abun-dances be determined from snapshot spectra? Secondly,does the pipeline give reasonable stellar parameters (ef-fective temperature, surface gravity, metallicity, and mi-croturbulent velocity) for snapshot spectra? The firstquestion has been partially answered by the HERESstudy, which finds that the uncertainties are ∼ . HRS Snapshot Data
Hollek et al.
Table 3. Stellar Parameter Comparison
Scattering Treatment Included No Scattering TreatmentStar T eff log g ξ [Fe/H] T eff log g ξ [Fe/H][K] [km/s] [K] [km/s]HD 122563 4450 0.50 2.30 − − −
200 4500 0.45 2.60 − − − − − − − Table 4. Abundance Comparison for Different Treatments of Scat-tering In MOOG
CS 22891 −
200 HD 122653 HE 0015+0048 HE 0432 − scat [X/Fe] non [X/Fe] scat [X/Fe] non [X/Fe] scat [X/Fe] non [X/Fe] scat [X/Fe] non [Mg/Fe] 0.53 0.66 0.54 0.41 0.65 0.69 0.50 0.52[Ca/Fe] 0.68 0.72 0.39 0.36 0.44 0.43 0.39 0.35[Cr/Fe] − − − − − − − − − − − The effective temperature of a star is spectroscopicallydetermined by minimizing the trend of the relation be-tween the abundance and excitation potential of the linesfrom which the abundance is derived. The microturbu-lent velocity is determined by doing the same for theabundance and reduced equivalent width. The surfacegravity is determined from the balance of two ionizationstages of the same element (e.g., Fe I and Fe II). TheHRS snapshot spectra have few detectable Fe II lines,thus we use Ti I and Ti II lines in addition to Fe I andFe II lines in the pipeline to more robustly determine thestellar parameters, in particular the gravity, from thesespectra, with Fe weighted twice compared to Ti. Of-ten, only the ionization balance of Fe is considered. Themetallicity used in the model atmosphere, in this case[M/H], is an average of the abundances from individuallines of Fe I, Fe II, Ti I, and Ti II.The pipeline works iteratively. The first process isto determine the stellar parameters from the equivalentwidth measurements of Ti I, Ti II, Fe I, and Fe II. An ap-proximate initial guess to the effective temperature, sur-face gravity, metallicity, and microturbulent velocity areinput as well as constraints on the parameters over whichthe code iterates. Generally for the HRS snapshot data,we require that the trend between abundance and excita-tion potential is < | . | dex/eV, the reduced equivalentwidth and abundance trend is < | . | dex/log(m˚A), thesurface gravity criterion ∆ ion < | . | dex, and the dif-ference between the metallicity of the model atmosphereand the calculated metallicity is < | . | dex. As a resultof the ionization balance constraints, the abundance dif-ference between Ti I and Ti II is allowed to be no greaterthan 0.3 dex if the Fe I abundance equals the Fe II abun-dance or the abundance difference between Fe I and FeII is allowed to be no greater than 0.2 dex if the Ti Iabundance equals the Ti II abundance. This informa-tion is used to construct an initial Kurucz stellar atmo-sphere with α -enhancement (Castelli & Kurucz 2004) tobegin the stellar parameter determination, in which thepipeline iterates until the various constraints that deter- mine each stellar parameter fall within the user-definedthresholds specified in the beginning.We used the pipeline to derive abundances for eachline from its equivalent width. We used synthetic spectrato determine abundances for particular lines (e.g., Ba II λ High-Resolution MIKE Spectra
The high-resolution MIKE data were analyzed manu-ally. The large wavelength coverage allowed us to per-form a more in-depth analysis with smaller uncertain-ties for a later comparison to the pipeline analysis ofthe corresponding snapshot spectra. Spectroscopic stel-lar parameters for all but two of the MIKE spectra weredetermined from equivalent width measurements of FeI and Fe II lines. The resonance lines of Fe I were ex-cluded in this analysis, as they often are near the flatportion of the curve of growth. In essence, all of thesteps in the Cashcode pipeline were performed, but eachstep was executed manually. This allowed us to comparethe pipeline results with those of a manual analysis toconfirm that the pipeline reproduced the results derivedfrom the high-resolution data.It is also possible to determine effective temperaturesphotometrically, using calibrations between colors andtemperature for given color and metallicity ranges. Wehave chosen to adopt spectroscopic temperatures forthese stars, as well as for the entire CASH sample, in or-der to present a homogeneous set of atmospheric param-eters and, therefore, the resulting chemical abundances.However, we determined photometric temperatures tocheck the accuracy and systematic uncertainties of ourmethod. Accurate long baseline colors (i.e., V − K) for theentire ∼
500 star sample do not exist; however, we used2MASS photometry for the pilot sample and standardstars in order to determine photometric temperatures inorder to compare them to the adopted spectroscopic tem-peratures.ASH II. A Sample of 16 Extremely Metal-poor Stars 7
Table 5. Stellar Parameters
MIKE HRSStar T eff log g ξ [Fe/H] T eff log g ξ [Fe/H][K] [km/s] [K] [km/s]HE 0013 − − − − − − − − − − · · · · · · · · · · · · HE 0324+0152a 4775 1.20 1.80 − − − − − − − − − − a HE 1311 − − − − − − a HE 2123 − − − − − − − − − a HE 2238 − − − − − − − − · · · · · · · · · · · · CS 22891 −
200 4500 0.45 2.60 − · · · · · · · · · · · · HD 122563 4450 0.50 2.30 − − −
18 5550 4600 0.80 1.70 − · · · · · · · · · · · · CD −
38 245 4650 0.95 2.15 − · · · · · · · · · · · · CS 22873 −
166 4375 0.20 2.80 − · · · · · · · · · · · · a Sky contamination in this spectrum
We determined reddening corrections for our stars us-ing the Schlegel et al. (1998) dust maps. We dereddenedthe J − K 2MASS color and then, according to equa-tions 1a-1c in Ram´ırez & Mel´endez (2004), transformed Jand K 2MASS photometry into the TCS system in orderto use the Alonso et al. (1999) calibration to determinephotometric temperatures. We determined the formallinear relation between the (J − K) photometric temper-atures from the sample stars and the spectroscopic tem-peratures. Within the uncertainty of the fit, there ex-ists an offset between the spectroscopic and photometrictemperatures. Thus, we adopted the mean difference be-tween the spectroscopic and photometric temperaturesto be T spec =T ( J − K ) −
225 K. The uncertainties asso-ciated with the spectroscopic and photometric tempera-tures are 160 and 140 K, respectively. See the next sub-section for further details. Figure 3 shows the spectro-scopic temperature plotted against the (J − K) photomet-ric temperature for the sample stars. Plotted are linesthat show a 1:1 agreement, the adopted offset, and theleast squares fit.Two stars in the sample, HE 1116 − − − − Fig. 3.—
Comparison of the spectroscopic temperatures to thosederived from (J − K) 2MASS photometry using the Alonso et al.(1999) calibration. The thin black line is the 1:1 comparison, thedotted blue line is the linear least squares fit to the data, and thethick pink line shows the adopted offset applied to HE 1116 − − to determine the remaining parameters, surface grav-ity, microturbulence, and metallicity spectroscopically,ignoring the nonzero slope of − .
065 and − .
090 dex/eVfor HE 1116 − − Fig. 4.—
HR diagram of the pilot sample and standard stars.For the pilot sample, open symbols represent the stellar parametersderived from the HRS spectra and the solid symbols represent stel-lar parameters derived from the MIKE spectra. For the standardstars, open stars represent stellar parameters from the literatureand asterisks represent stellar parameters derived from the MIKEspectra. Overplotted are the Yale-Yonsei isochrones (Green et al.1984; Kim et al. 2002) for 12 Gyr, at [Fe/H] = − − − the excitation potential. Typically, this slope is < . − . − .
5, and − .
0, as wellas a Cassisi et al. (2004) horizontal-branch mass track.For the standard stars, we show the stellar parametersderived from the MIKE spectra, as well as literature val-ues taken from Cayrel et al. (2004) for BD −
18 5550,CD −
38 245, and CS 22873 − − −
38 245,the corresponding MIKE spectrum shows very few ab-sorption lines, thus we included this star only in the cal-ibration of the stellar parameter offset.
Uncertainties
Each star has two or three measurements of tem-perature: spectroscopic temperatures derived from theMIKE and HRS spectra (when available) and a photo-metric temperature based on JHK 2MASS colors usingthe Alonso et al. (1999) calibration.We determined random uncertainties in the spectro-scopic temperatures based upon the uncertainty in theslope determined by the Fe I line abundances. For arepresentative star, we varied the temperature until theresultant Fe abundance was one standard deviation awayfrom the original derived abundance. We found this value to be ∼
125 K; we adopt this as our random uncertainty.We determined random uncertainties in the photometrictemperatures based upon the uncertainties given for the2MASS colors. We found this uncertainty be ∼
140 K.We also compared the derived spectroscopic and pho-tometric temperatures. We remind the reader that theoffset between the two temperature scales is ∼
225 K,where ∆ T eff = T P hoteff − T Speceff . Given this offset, we de-rive a systematic uncertainty for our spectroscopic tem-peratures of ∼
160 K. This is of the same order as therandom uncertainties.We obtained the random uncertainty in the surfacegravity by allowing the Fe I and Fe II values to vary untilthey no longer agree within the uncertainty of Fe I, whichis ∼ .
25 dex. Since all the pilot sample stars are on thegiant branch, uncertainties in effective temperature atthe ∼
150 K level lead to changes in the surface gravitiesof ∼ . σ logg uncertainty.We calculated the standard error of the mean Fe Iabundance, obtained from individual Fe I line abun-dances and found it to be ∼ .
01 dex. This is rathersmall, as the calculation does not take continuum place-ment uncertainties into account. With that in mind, weadopt the scatter of the individual Fe line abundances asa more conservative random [Fe/H] uncertainty, which is ∼ .
12 dex, although it varies slightly by star.In order to ascertain the uncertainty in the microturbu-lence, we determined the maximum change to the micro-turbulence values which would still yield the same [Fe/H]value within the uncertainties. This leads to σ ξ ∼ . ROBUSTNESS OF THE STELLAR PARAMETER ANDABUNDANCE PIPELINE
Stellar Parameters
One of the main purposes of this paper is to test thestellar parameter and abundance pipeline that will beused for the larger CASH sample. The first test is tocompare the manually derived stellar parameter resultswith those determined with the Cashcode pipeline. Thiscan be done in three ways: i) comparing the stellar pa-rameters and abundances of a manual analysis of a givenspectrum with each Cashcode result; ii) comparing theresults derived from snapshot spectra taken of standardstars with those of well determined literature values ofthe stellar parameters; and iii) comparing the stellar pa-rameter results of a snapshot spectrum with those of acorresponding high-resolution spectrum.We first tested the accuracy of the pipeline and the pre-cision of our iteration criteria by comparing the resultsderived from a representative manual analysis of a high-resolution MIKE spectrum to those derived using thepipeline. We find ∆T eff =10 K, ∆log g = 0.0, ∆ ξ = 0.05,and ∆[Fe/H]= 0.02 dex. We find that the [X/Fe] values(where X is a given element) agree to within ∼ .
05 dex.By comparing the Cashcode results of standard starsto the literature values, we can test how accurately thesnapshot data are able to reproduce the results of inde-pendent studies derived from high-resolution, high-S/Ndata using traditional manual methods. We evaluated anabsolute consistency between the Cashcode results andthe literature. See Subsection 6.7 for details on the abun-ASH II. A Sample of 16 Extremely Metal-poor Stars 9dance differences. A caveat to this comparison is thatstandard stars are usually bright targets, such that theirsnapshot spectra have much higher S/N than the me-dian value for a snapshot spectrum. Generally, Cashcodeproduces abundance results with smaller uncertaintiesfor higher-S/N, higher-resolution data because in thesecases the line abundance scatter is decreased, thereforethe user-defined parameter fitting criteria can be tight-ened.However, the purpose of this study is to test thepipeline for the median S/N snapshot star. Unfortu-nately, we did not observe a low S/N snapshot spectrumof the standard stars, which precludes the best possi-ble comparison. To resolve the issue, we instead turn tohigh-resolution data for testing the pipeline for a medianS/N snapshot data. We compared the HRS pipeline-derived results to those of the manual, high-resolutionanalysis for all stars. We find agreement in the stellarparameters to be within ∆ T eff ±
55 K, ∆ log g ± ± .
15 dex, and ∆ ξ ± .
21 dex.
Chemical Abundances
For each star with an HRS spectrum, we comparedthe derived abundances with the high-resolution MIKEspectrum. For the 12 stars in common, we found theoffset between the two to have a standard error, or stan-dard deviation of the mean, of 0.07 dex. The average∆[X/Fe] over all elements was 0.09 dex, thus the abun-dances derived from the HRS data with Cashcode can re-produce a manual analysis to within 1.5 σ . Over all the el-ements, there is no statistically significant difference be-tween the HRS/Cashcode and the MIKE/manual analy-sis. For individual elements, some discrepancies do exist.We found that the largest discrepancies between the HRSand MIKE spectra arose for those elements whose linesoccurred in regions of low S/N (e.g., the ∆[X/Fe] for SrII is larger than that of Ba II, as the only Sr II line inthe HRS occurs at 4215 ˚A, while there are Ba II lines atlonger wavelengths).For a given element, there is a discrepancy in the mea-sured abundances derived from HRS and MIKE spectra.These range from 0.08 dex, in the case of Ti, which has9 lines across the HRS spectrum, to 0.36 dex, in thecase of Sr, which has one feature at 4215 ˚A in the HRSspectrum. This value is dependent upon the locationin wavelength and number of lines per element; bluerfeatures from lower S/N spectral regions have larger dis-crepancies. Figure 5 illustrates the ∆[X/Fe] for each el-ement that was measured in both the HRS and MIKEspectrum of a particular star. We also compared the fullset of [X/Fe] CASH and MIKE derived abundances tothose of the Cayrel et al. (2004) study. Figure 6 showsthe comparison between the three data sets. Manganeseabundances are sometimes largely discrepant; however,this is likely because there are only four Mn lines avail-able to detection in the HRS spectra and often only oneof these lines was detected. It has been previously noted(Cayrel et al. 2004; Roederer et al. 2010) that the lineselection for Mn is critical in understanding the derivedabundances and the Cayrel et al. (2004) study includedadditional lines not available in the HRS spectra. ABUNDANCE ANALYSIS
Table 6. Abundances
Elem log ǫ (X) [X/Fe] σ n [X/Fe]dex MIKE dex HRSHE 0015-0048Mg I 5.18 0.66 0.18 6 0.62Ca I 3.71 0.45 0.14 15 0.44Sc II 0.14 0.07 0.14 8 0.06Ti I 1.98 0.11 0.18 18 0.11Ti II 2.14 0.27 0.15 39 0.06Cr I 2.40 − .
16 0.20 14 − − .
29 0.14 4 − · · · Ni I 3.13 − .
01 0.16 8 · · ·
Zn I 1.58 0.10 0.14 1 0.68Sr II − − − . − − − . Note . — Table 6 is published in its entirety in the electronicedition. A portion is shown here for guidance regarding its content.
We present chemical abundances and upper limits for13 elements derived from the HRS spectra and 18 ele-ments from the MIKE spectra in Table 6. We now de-scribe the details of our abundance analysis. For eachelement, we discuss the method of abundance determi-nation, the relevant spectral features, the number of starsin which this element was measured, and any differencesin the analysis between the high-resolution and snapshotspectra.
Light Elements: Li, C, Al, Si
Lithium abundances were determined from both theHRS and MIKE spectra through synthesis of the Li I dou-blet at 6707 ˚A. Both spectra cover this feature. Lithiumis most easily detected in main-sequence stars, but asa star ascends the RGB, its Li abundance becomes de-pleted. Thus, we only expect to detect the Li featurein our warmest giants. We detected the doublet in 5 of12 stars stars with HRS spectra and 7 of 20 stars MIKEspectra.Carbon abundances were determined from the CH G-band feature at 4313 ˚A. The S/N near the G-band headregion in the HRS snapshot spectra is low, such that wehave to perform a manual synthesis of the G-band headin order to determine the C abundances. At low S/N,Cashcode does not yield reliable results mostly due touncertainties in continuum placement. For synthesis weassumed an [O/Fe] abundance ratio of 0.0. When the[O/Fe] ratio reaches ∼
1, the C abundances determinedfrom molecular features is affected, though none of ourstars indicate such a high O abundance. We detectedthis feature in all stars in both sets of spectra. The twoavailable O indicators in our spectra are the forbiddenline at 6300 ˚A and the O triplet. At the metallicity ofour sample, we are unable to detect the forbidden lineexcept for cases of extreme O enhancement. The tripletlines are known to have NLTE effects and thus we didnot measure an O abundance.Sodium abundances were not included in this analysisas we have not implemented a routine in the pipeline todiscern the stellar absorption lines from the interstellaremission.Aluminum abundances were determined using equiv-0 Hollek et al.
Fig. 5.—
Abundance differences, ∆[X/Fe], defined as [X/Fe]
MIKE - [X/Fe]
HRS , shown as a function of atomic number for each star thatwas observed with both HRS and MIKE. The black solid line represents zero offset, the green dashed line represents the 0.25 dex randomerror derived in the HERES study, and for comparison, the blue dotted line represents the calculated spread for each star. alent width measurements of the λ λ λ λ λ λ λ δ line, thus we adopt alocal continuum in the wing of the H δ line. We presentspectral synthesis derived abundances from the MIKEspectra using the λ α -Elements: Mg, Ca, Ti ASH II. A Sample of 16 Extremely Metal-poor Stars 11
Fig. 6.— [X/Fe] abundance ratios vs. [Fe/H] for each of the elements measured using Cashcode in the HRS snapshot spectra (bluetriangles) compared with the MIKE abundances (red squares) and the Cayrel et al. (2004) abundances (black circles). The black dottedline represents the solar abundance ratio. The HRS Ni abundances shown are preliminary.
Abundances for Mg, Ca, and Ti were determined for allstars in the sample from both the HRS and MIKE spectrafrom equivalent width analysis. Only 4 unblended Mg Ilines are available in the HRS snapshot spectra: λ λ λ λ ∼
12 detectable lines in theMIKE spectra, although in these spectra we exclude lineson the flat part of the curve of growth. This is determined on a line-by-line basis for each star.There are four available Ca I lines in the snapshot spec-tra: λ λ λ λ ∼
30 Ca I lines in the MIKE spectra. Though the CaI λ Fig. 7.— [X/Fe] abundance ratios vs. [Fe/H] for each of the elements up to Zn manually measured in the MIKE spectra (red squares)compared with the Cayrel et al. (2004) abundances (black squares). The black dotted line represents the solar abundance ratio.
Titanium abundances were determined from 6 Ti I and9 Ti II lines in the snapshot spectra and ∼
30 Ti I linesand ∼
60 Ti II in the high-resolution spectra.
Fe-Peak Elements: Sc, Cr, Mn, Co, Ni, Zn
All abundances were determined from equivalent widthmeasurements except for Zn because it has only two weaklines. There are 4 detectable Sc II lines in the snapshotdata: λ λ λ λ ∼ λ λ λ λ λ ∼
20 Cr I lines and 4 Cr II lines in the MIKE spectra: λ λ λ λ ∼ +0 .
35 dex larger thanCr I. We detected the Cr I lines in 19 MIKE spectraand 12 HRS spectra. In all plots, we adopt the [Cr I/Fe]values as [Cr/Fe].We measured 4 Mn I lines in the HRS spectra: λ λ λ λ Fig. 8.— [X/Fe] abundance ratios vs. [Fe/H] for six neutron-capture elements measured in the MIKE spectra (red squares) comparedwith the Cayrel et al. (2004) abundances (black circles). The black dotted line represents the solar abundance ratio. spectra and 11 HRS spectra.In the MIKE data we measured ∼
30 Ni I lines, butonly the λ λ λ ∼ . λ λ Neutron-Capture Elements
We measured abundances for six neutron-capture ele-ments via spectral synthesis. The Sr II λ λ λ λ λ λ λ λ λ λ λ λ λ λ λ λ Non-LTE Effects
Chemical abundances are generally derived under theassumption of one dimensional (1D) model atmospheresin LTE, but non-LTE effects may alter our derived val-ues. The non-LTE effects in the elements Mg, Sr, andBa have been studied in the stars of the Cayrel et al.(2004) sample by Andrievsky et al. (2010), Andrievskyet al. (2011), and Andrievsky et al. (2009), respectively.Based upon their reanalysis of the Cayrel et al. (2004)sample, the non-LTE corrections would be ∼ .
15 for Mgand Ba in our sample. In the case of Sr, the non-LTEabundances vary, and can be larger and smaller than theLTE abundances even for stars of similar temperatureand gravity, so it is difficult to say what this effect wouldbe in our stars. See also Asplund (2005) for a compre-hensive review of non-LTE effects on stellar abundancesfor a range of elements. Such effects are very sensitiveto stellar parameters and individual lines from which theabundances are derived. In the absence of “full grid”non-LTE correction calculations, we are not able to ap-ply such corrections to our sample; however, we remindthe reader that these corrections have implications for in-terpretations of Galactic chemical evolution models andshould be taken into account for such investigations. Toillustrate the magnitude of non-LTE effects, we discussAl and Mn here, as both have some of the most severenon-LTE corrections. Baumueller & Gehren (1997) present a non-LTE-corrected [Al/H] abundance analysis for four sets of starsfor the λ λ λ λ λ < − . λ − .
00 to 0.00 dex. For a given set of atmosphericparameters, the [Al/H]
NLTE correction increases with de-creasing [Fe/H]. The most evolved star (T eff = 5500 K ,log g = 3.5) in their analysis has a non-LTE correction of0.65 dex. All the stars of the pilot sample are on the giantbranch and generally are more metal poor than the Bau-mueller & Gehren (1997) models, which indicates thatour stars would have a larger non-LTE correction for the λ λ eff =5000 K,log g=4, [Fe/H]= −
3) has an average [Mn/Fe] NLTE cor-rection of 0.42 dex for the lines that we include in ourMn linelist ( λ λ λ λ λ λ λ . Uncertainties
To determine the random uncertainty of our abun-dances, we calculated the scatter of the individual lineabundances for each ionization state of each elementmeasured. For any abundance determined from equiv-alent width measurements of less than 10 lines, we de-termined an appropriate small sample adjustment for the σ (Keeping 1962). In the case of any abundance uncer-tainty that was calculated to be less than the uncertaintyin the Fe I lines, we conservatively adopted the value fromFe I for that particular star. Typically the Fe I uncer-tainty is ∼ .
12 dex.For those lines with abundances determined via spec-tral synthesis, we determined abundance uncertaintiesbased upon the uncertainties associated with equivalentwidth measurements. Continuum placement is the great-est source of uncertainty, along with the S/N of the regioncontaining the particular line. Most of these abundanceswere determined from only two lines, thus we calculatedthe uncertainties for small samples. For those elementswith only one line, we adopt the uncertainties determinedfor the Fe I abundance.To obtain the systematic uncertainties in the abun-dances, we redetermined abundances by individuallyvarying the stellar parameters by their adopted uncer-ASH II. A Sample of 16 Extremely Metal-poor Stars 15
Table 7. Example Systematic Abundance Uncertainties forHE 0015 − Elem ∆T eff ∆ log g ∆ ξ +150 K +0.5 dex +0.3 km/sCH 0.35 − − − − − − − − − − − − − − − − − − − − tainties. We chose a nominal value of 150 K for the effec-tive temperature uncertainty, as this value is similar tothe random and systematic uncertainties. Table 7 showsthese results. We find that the effective temperature con-tributes most to the abundance uncertainty. The uncer-tainty in the surface gravity is somewhat less significantfor most species. For elements with particularly stronglines, especially those whose abundances are determinedwith spectral synthesis, the microturbulence can be animportant source of error. Standard Stars
We compared our stellar parameters and [X/Fe]abundances for our four standard stars (HD 122563,BD −
18 5550, CS 22873 − − −
38 245against McWilliam et al. (1995). The McWilliam et al.(1995) study differs from the other two and this study asthose spectra had comparatively low S/N ( ∼ ∼ − −
38 245) which overlap with our study.That work utilized model atmospheres from Kurucz(1993) with MOOG. We use an updated version of thiscode, which does explicitly deal with Rayleigh scatter-ing as a continuum opacity source; see paragraph 1 ofSubsection 3.3. The effective temperatures for this studywere derived from photometry, with the microturbulencedetermined from Fe I lines and the surface gravity deter-mined from the ionization balance of Fe I and Fe II lines.We find that our temperatures are lower by ∼
170 K.Our derived surface gravities are also lower, as a resultof the lower temperatures. We also find a ∼ − . Table 8. Literature Values for Stellar Parameters
Study T eff log g [Fe/H] ξ [K] [km/s]HD 122563This study 4450 0.50 − − · · · · · · · · · · · · Cayrel 4600 1.10 − −
18 5550This study 4600 0.80 − · · · · · · · · · · · · McWilliam 4790 1.15 − − −
38 245This study 4560 0.95 − · · · · · · · · · · · · McWilliam 4730 1.80 − − − − · · · · · · · · · · · · McWilliam 4480 0.80 − − − − · · · · · · · · · · · · McWilliam 4700 1.00 − · · · · · · · · · · · · Note . — References: Fulbright (2000), McWilliam et al. (1995),Cayrel et al. (2004) and how the most recent version of MOOG explicitlydeals with the Rayleigh scattering opacity (see Table 3).Both lead to lower temperatures, surface gravities, andabundances.Despite these absolute offsets, the derived abun-dance ratios have only small offsets, with the aver-age offset between the two studies in h ∆[X/Fe] i forall three stars is 0.04 ± Standard − [X/Fe] MIKE . For all three stars, Al andSi had the largest offsets, where McWilliam et al. (1995)derive systematically higher abundances. The Al abun-dances for McWilliam et al. (1995) may generally behigh, as they were found to be higher (∆[Al/Fe] ∼ . δ line pro-duces larger uncertainties in our Si measurement, evenwhen the feature is taken into account in the syntheticspectra. These reasons would produce a larger scatterin the measurements of Al and Si lines, compared to theother elements, and may explain the derived abundancediscrepancy. When these two elements are removed fromconsideration the h ∆[X/Fe] i becomes − . ± .
13 dex.The only star which overlaps with the Fulbright (2000)study is HD 122563, though this particular study waschosen because the stellar parameters were determinedspectroscopically, in the same manner as the stellar6 Hollek et al.parameters determined from the MIKE spectra. Ful-bright (2000) used Kurucz (1993) model atmosphereswith MOOG as well. We found that the effective tem-peratures, surface gravities, and microturbulence valuesagree within the uncertainties associated with both stud-ies, while ∆[Fe/H] is − .
17 dex. Similar to the effectsseen in comparison with the McWilliam et al. (1995)study, this is due to the fact that the version of MOOGused in the Fulbright (2000) study did not explicitly han-dle the calculation of scattering from pure absorption interms of the continuum opacity.The derived abundance ratios are in good agreement,with h ∆[X/Fe] i = 0 . ± .
09 dex. The largest discrep-ancy lies with the [Mg/Fe] ratio, with ∆[Mg/Fe] = 0.20dex, which was derived with a different set of lines anddifferent oscillator strengths between this study and theFulbright (2000) study.The Cayrel et al. (2004) study has four stars in com-mon with ours (including CD −
38 245). This studyuses OSMARCS atmospheric models with the LTE syn-thetic spectrum code “turbospectrum”, which does ex-plicitly handle the calculation of the scattering contribu-tion with regard to the continuum opacity. The effectivetemperatures were determined via (V − K) colors, whichleads to higher temperatures ( ∼
200 K) and surface grav-ities ( ∼ . −
18 5550 throughCashcode and found that our derived stellar parameters(4560 K, 0.6 dex, − .
22, and 1.7 km/s in effective tem-perature, surface gravity, [Fe/H], and microturbulence,respectively) were in agreement (∆[Fe/H]= 0 .
02 dex).In comparing the [X/Fe] values between our analysis ofthe Cayrel et al. (2004) equivalent widths and theirs, wefind that our spectroscopically derived stellar parametersalso lead to higher [Mg/Fe] values by ∼ .
25 dex due tothe gravity sensitive nature of the Mg lines.The derived abundance ratios also agree well, with h ∆[X/Fe] i = − . ± .
14 dex. The largest sources ofdiscrepancy for all stars are Al and Si, in addition toMg. In CS 22873 − ∼
20. Additionally, the lines arevery strong, making continuum placement a large sourceof uncertainty, as minor adjustments to the continuumlevel result in large changes in the abundance. ABUNDANCE RESULTS AND DISCUSSION
Table 6 lists abundance results derived from the MIKEspectra. Figures 6 and 7 include the [X/Fe] abundanceratios derived from the MIKE spectra plotted against[Fe/H] for all stars in the sample. These abundance ra-tios are overplotted against the Cayrel et al. (2004) abun-dances as a point of comparison. The table also includesthe abundances derived from the HRS spectra, thoughwe do not discuss these further. These abundances willlater be included in the full ∼
500 star CASH sample.For each element, Table 9 lists the parameters of a leastsquares linear trend versus metallicity, the abundancescatter, and the average [X/Fe] value, if applicable.
Light Elements
We detected Li in six of our stars: HE 0324+0152a,HE 0420+0123a, HE 1311 − − Table 9. Summary of Abundance Trends
Elem h [X/Fe] i slope vs [Fe/H] σ n stars CH 0.02 a − . b · · · · · · · · · − · · · · · · · · · · · · · · · − · · · · · · · · · − − · · · − · · · − · · · − · · · a Average [C/Fe] value calculated for non-CEMP stars. b Slope calculated for log(L/L ⊙ ) vs [C/Fe]. See Figure 11. HE 2138 − − −
18 5550. These detections were seenin both the snapshot and high-resolution spectra forHE 0420+0123a, HE 1311 − − − −
18 5550 and HE 2302 − σ upper limit on the equivalent widthwas calculated following Bohlin et al. (1983) and Frebelet al. (2007b). Figure 10 shows the Li abundance as afunction of metallicity and effective temperature, alongwith the HR diagram of the stars with detected Li.The Spite plateau (Spite & Spite 1982) is an observa-tional discovery that describes a constant Li abundancein low-metallicity stars near the main-sequence turn-off.Metal-poor stars near or on the main sequence have notyet burned their surface Li; it is therefore thought thatthe stars that populate the Spite plateau can be usedto infer details about the primordial Li abundance. Asstars evolve off the main sequence and up the red giantbranch (RGB), their convection zone deepens and theatmospheric Li abundance is depleted through burningand convective dredge up. Any enhancement in the Liabundance (e.g., Roederer et al. 2008) is likely from someform of Li synthesis that occurs during the course of stel-lar evolution, perhaps due to the Cameron-Fowler mech-anism (Cameron & Fowler 1971). All of the stars in thepilot sample are on the giant branch and, thus, are ex-pected to have depleted Li abundances. Due to the evolu-tionary status of the pilot sample, we cannot comment onthe nature of the Spite plateau. Accurate Li abundancesrequire great care in the effective temperature determi-nation, as the Li abundance is extremely temperaturesensitive. Most Li abundance studies (e.g., Ryan et al.1996; Asplund et al. 2006; Garc´ıa P´erez & Primas 2006;Bonifacio et al. 2007; Mel´endez et al. 2010) employ longbaseline (e.g., V − K) photometric effective temperatures.Keeping the different temperature scales in mind, we findthat our abundances qualitatively fall along an extrapo-lation of the Ryan & Deliyannis (1998) Li dilution curve.ASH II. A Sample of 16 Extremely Metal-poor Stars 17
Fig. 9.— Li λ Fig. 10.—
MIKE Li abundances (open squares) and upper lim-its (arrows) plotted against [Fe/H] along with Li abundances fromAsplund et al. (2006), Garc´ıa P´erez & Primas (2006), and Boni-facio et al. (2007) (bottom), and effective temperature with theexpected Li dilution curve from Ryan & Deliyannis (1998) (topleft) overlayed. Effective temperatures and surface gravities areplotted, along with a [Fe/H]= −
3, 12 Gyr Yale-Yonsei isochrone(Green et al. 1984; Kim et al. 2002) for each star plotted in thebottom panel (top right).
Fig. 11.— [C/Fe] abundance ratios plotted against [Fe/H] (left)and luminosity along with the CEMP defining line, which changesover the course of the stellar lifetime (Aoki et al. 2007a) (right),along with the calculated Cayrel et al. (2004) C abundances. The[C/Fe] abundances clearly decline as stars ascend the giant branch.
Fig. 12.—
HRS spectra of a CEMP star (HE 0015+0048a)and a non-CEMP (HE 1317 − ∼ ∼ − . Figure 11 shows a plot of our derived [C/Fe] abun-dances against [Fe/H] and luminosity. The Cayrel et al.(2004) abundances are also shown for reference. Theabundance offset as a function of luminosity between thetwo samples is due to the different temperature scalesused, as also quantified in Table 7. Generally, we finda large spread in the [C/Fe] abundance ratios, from ∼ − .
80 to ∼ . σ = 0.41 dex. In Fig-ure 12, we show spectra for two stars of similar tem-perature ( ∼ ∼ . ∼ . ∼ . ⊙ ) < ∼ . ≥ ⊙ ) > ∼ . > (3.0 − log(L/L ⊙ )) are considered CEMP stars (see theright panel of Figure 11). This definition takes into ac-count the decrease in the surface C abundances as a func-tion of evolutionary status. For our sample, the definitionwould indicate that 4 stars are significantly enhanced intheir [C/Fe] ratios compared to the rest of the sample.This is still true when the effects on the abundances dueto the use of different temperature scales is taken intoaccount. In order to study the initial stellar abundances,corrections, up to ∼ ∼ ∼ α -element ratios. The first and third scenarios,however, are not mutually exclusive. Future modelingof the nucleosynthetic yields of massive Population IIIstars will facilitate a better understanding of early car-bon production. Given the size and metallicity rangeof the pilot sample, unfortunately nothing can be saidabout the CEMP frequency, though we will be able toevaluate this using the full CASH sample.Finally, fine-structure lines of C and O are thought toplay a role in the transition from Population III to Pop-ulation II stars through the cooling of gas clouds of theearly universe (Bromm & Loeb 2003). This hypothesiscan be tested by comparing the abundances of C and Oof metal-poor stars to an abundance transition discrim-inant as described in Frebel et al. (2007a). Stars with[Fe/H] < − . did not play a role. α -Elements The α -elements are created both in core-collapse su-pernova and through stellar nucleosynthesis in high-mass stars. While the dominant isotope of Ti is not technicallyan α -element, it shows a similar abundance pattern tothe α -elements and thus is included here (e.g., Woosley& Weaver 1995). Magnesium and Ca, as well as theother α -elements, have been shown to be overabun-dant, compared to the Solar System [ α /Fe] ratio, atlow metallicities in field halo stars at the ∼ . α -elements compared toFe. Later generations of supernovae, specifically Type Ia,produce more Fe, driving down the [ α /Fe] ratio to whatwe see today in the Sun and similar young, metal-richstars. This downturn in the [ α /Fe] ratio occurs at [Fe/H] ∼ − .
5. As our sample does not reach [Fe/H] > − .
0, wedo not expect to see this downturn in the [ α /Fe] abun-dance ratios.As seen in other halo stars, the [Mg/Fe] ratio is en-hanced relative to the solar ratios in all our sample starsat 0.56 dex. This is also seen in the Cayrel et al. (2004)stars as well, though there is an offset of ∼ .
25 dexbetween these two samples, with ours having the largervalue. This is due to the differences in the effective tem-perature scales chosen. The Cayrel et al. (2004) studyused (higher) photometric temperatures. To demon-strate this, we took the Cayrel et al. (2004) equivalentwidths for BD −
18 5550 and ran them through the Cash-code pipeline. We obtained a temperature different by150 K, resulting in an offset of 0.5 dex in the surface grav-ity. In most elemental ratios these effects cancel, but Mg,especially the triplet, is gravity sensitive, resulting in a ∼ .
25 dex offset in [Mg/Fe] between the two studies.The [Ca/Fe] ratio is also enhanced in our sample starsrelative to the solar ratio at 0.42 dex. The [Ti/Fe] ra-tio is found to be enhanced relative to the solar ratio inall but two stars, HE 2148 − − − − α /Fe] value is ∼ . α /Fe]values. Fe-Peak Elements
The [Sc/Fe] ratio for the stars in the sample is generallyclustered around the solar abundance ratio. The [Cr/Fe]and [Mn/Fe] abundance ratios for all our sample starsare found to be deficient relative to the solar abundanceby − .
23 and − .
49 dex, respectively. We remind thereader that the [Cr/Fe] ratios are based upon only the CrI abundances, as there is a ∼ .
35 dex offset between Cr Iand Cr II derived abundances. The [Co/Fe], [Ni/Fe], and[Zn/Fe] ratios in the sample stars are generally enhancedrelative to the solar abundance ratios by 0.42, 0.05, and0.25 dex, respectively.The Fe-peak elements are created in various late burn-ing stages (see Woosley & Weaver 1995), as well as insupernovae. Our Fe-peak abundance trends follow thoseof other halo star samples and generally indicate a succes-sive increase of these elements over time (e.g., McWilliam1997). We will use the Fe-peak elemental abundances ofASH II. A Sample of 16 Extremely Metal-poor Stars 19the large CASH sample to put constraints on the nucle-osynthesis yields of the progenitor stars. For example,the measured Zn abundances, which we can measure inthe HRS spectra, are sensitive to the explosion energy ofsupernovae (Nomoto et al. 2006; Heger & Woosley 2010).
Neutron-Capture Enhanced Stars
The study of neutron-capture elements allows for test-ing of different sites of nucleosynthesis, beyond protonand α -capture. See Sneden et al. (2008) for a comprehen-sive overview of the neutron-capture stellar abundances.Neutron capture occurs mainly in two locations: inthe envelopes of highly evolved asymptotic giant branch(AGB) stars (s-process) and in some sort of explosiveevent, likely a core-collapse supernova (r-process). Thecontributions of each process to the total elemental abun-dance of the neutron-capture elements in a given starcan be determined by evaluating the collective neutron-capture abundance patterns. For example, in the SolarSystem abundances, the s-process contributes ∼
80% ofthe Ba abundance, with a ∼
20% contribution from ther-process, whereas Eu is made almost entirely from ther-process (Sneden et al. 2008); however, these ratios maydiffer in the early universe. Unfortunately, there are notenough neutron-capture elements detectable in the HRSsnapshot spectra to determine whether the abundancesof neutron-capture elements in a given star have an s- orr-process origin.We measured Sr and Ba for all stars in the MIKE sam-ple. We also detected Y and Zr in many of the stars aswell, though the S/N near the Y lines often precludeus from measuring it, and the Zr feature is often tooweak. We measured La in only four pilot sample stars:HE 2238 − − − − −
166 is depleted in [Eu/Fe] relative to the so-lar abundance ratio. According to Christlieb et al. (2004)HE 0420+0123a is a mildly r-process enhanced (r-I) stardue to its [Eu/Fe] ratio (0.79) and Ba/Eu ratio ( − . ǫ (X) abundances relative to the Baabundance and plotted them against a Solar System r-process abundance pattern. We also normalized the SolarSystem s-process abundance pattern to fit the derived Baabundance. We found that the r-process pattern seemedto fit the ratio of La to Eu better than the s-processpattern, as the Eu abundances would have to be an or-der of magnitude lower to match the s-process pattern.Figure 15 shows this analysis. The first neutron-capturepeak elements (Sr, Y, Zr) show a ∼ . Fig. 13.— La λ λ − λ Fig. 14.— Eu λ λ − λ Stars with [Fe/H] < − . < − . − − . Fig. 15.—
Relative log ǫ (X) abundances for all r-process en-hanced stars in the MIKE sample. Abundances are adjusted tothe Ba abundance to fit a scaled Solar System r-process curve (redline). The Solar System s-process curve is also plotted (blue dashedline). ( − . α /Fe] ratios, depletion in some ofthe Fe-peak abundance ratios, and none are neutron-capture enhanced. We detected the λ − eff = 4675 K. Gen-erally, these stars are indistinguishable from the rest ofthe sample, with the following exception.It is noteworthy that all five EEMP stars show en-hancement in their [C/Fe] abundance ratios. This con-firms the trend toward higher [C/Fe] ratios towards lowervalues of [Fe/H], noted first by Rossi et al. (2005) andLucatello et al. (2006). Without consideration for theevolutionary stage of the stars, the average [C/Fe] ratiofor the EEMP stars is 0.43, while the rest of the samplehas an average [C/Fe] ratio of 0.07 dex. However, giventhe high luminosity of the EEMP stars, the [C/Fe] ratioshave been depleted due to the operation of the CN cy-cle. Taking this effect into account, two stars are mildlyC enriched and two are considered CEMP stars follow-ing the definition of Aoki et al. (2007a). The two CEMPstars lack neutron-capture enhancement, which indicatesthat these stars are CEMP-no stars (Beers & Christlieb2005), adding to the growing number of these objects atthe lowest metallicities.We did not detect either the La or Eu lines in any ofthese stars. This is largely due to the fact that it is diffi-cult to detect Eu in EMP stars unless the Eu abundanceis significantly enhanced. Given that we do not see strongLa enhancement along with the carbon enhancement inthe EEMP stars, it is likely that their observed abun-dance patterns were the result of massive stars, ratherthan low mass stars, which produce La in the AGB stageof stellar evolution. Binary Fraction
For twelve of the pilot sample stars, we derivedmultiple-epoch measurements of their radial velocity. Forthree stars, we have additional radial velocity measure-ments from the solar spectrum-contaminated HET HRSspectra. For the remaining two stars, we do not havemultiple measurements. Generally, the radial velocitymeasurements were taken with a baseline of at leastone year and in some cases, three years. The HRSand MIKE radial velocities agree to within 2.5 km/s forthose stars. We find one binary candidate in our sample.HE 0015+0048a is the one exception; the time span be-tween measurements is three years and we find a radialvelocity variation of 8.0 km/s. Future monitoring of thisstar will confirm its binarity. This sample is too small tomake any definitive statement on the binary fraction ofmetal-poor stars. SUMMARY
We have presented the abundances or upper limits of20 elements (Li, C, Mg, Al, Si, Ca, Sc, Ti, Cr, Mn, Fe,Co, Ni, Zn, Sr, Y, Zr, Ba, La, and Eu) for 16 new starsand four standard stars derived from high-resolution,high S/N MIKE spectra via traditional manual analy-sis methods using the MOOG code. We find that, withthe exception of Mg, our abundances match well withthose of other halo stars reported in the literature, e.g.,Cayrel et al. (2004).In the pilot sample we find several distinct chemicalgroupings of stars, indicating different enrichment mech-anisms may apply for each of these groups, though theexact mechanism is still uncertain. We find four newstars with [Fe/H] < − .
6, where the metallicity distribu-tion function severely drops. We find CS 22891 −
200 tohave a lower [Fe/H] value than reported by McWilliamet al. (1995), bringing the total number of [Fe/H] < − . T eff ±
55 K, ∆ logg ± ± .
15 dex, and ∆ ξ ± .
21 km/s.These fall within the expected uncertainties associatedwith snapshot-quality data. We also find that the abun-dances derived from the HRS spectra using the pipelineASH II. A Sample of 16 Extremely Metal-poor Stars 21are in agreement to within 1.5 σ . The Cashcode pipelinewill be employed for the full ∼
500 star snapshot sam-ple. This sample will be used to determine carbon andneutron-capture enhancement frequencies, to better un-derstand supernova nucleosynthesis and early universechemical enrichment processes, and find new astrophysi-cally interesting stars that merit further study. One starhas already been identified as neutron-capture enhancedin our sample, based on its high Eu abundance and an-other separate star has also been singled out. The latterstar is a CEMP star, with r-process and s-process ele-mental abundance enhancements. Both of these will befurther analyzed in a later paper as part of this series.The Hobby-Eberly Telescope (HET) is a joint projectof the University of Texas at Austin, the Pennsyl-vania State University, Stanford University, Ludwig- Maximilians-Universitt Mnchen, and Georg-August-Universitt Gttingen. The HET is named in honor ofits principal benefactors, William P. Hobby and RobertE. Eberly. We are grateful to the Hobby-Eberly staff fortheir assistance in obtaining the data collected for thispaper. We thank John Norris and Norbert Christlieb fortheir valuable contributions to the discovery of the brightmetal-poor stars analyzed here. J.K.H. and T.C.B. ac-knowledge partial support through grants PHY 02-16783and PHY 08-22648: Physics Frontier Center/Joint Insti-tute for Nuclear Astrophysics (JINA). A.F. acknowledgessupport of a Clay Fellowship administered by the Smith-sonian Astrophysical Observatory. I.U.R. is supportedby the Carnegie Institution of Washington through theCarnegie Observatories Fellowship. C.S. is supportedthrough NSF grant AST-0908978.
REFERENCESAlonso, A., Arribas, S., & Mart´ınez-Roger, C. 1999, A&As, 140,261Andrievsky, S. M., Spite, M., Korotin, S. A., Spite, F., Fran¸cois,P., Bonifacio, P., Cayrel, R., & Hill, V. 2009, A&A, 494, 1083Andrievsky, S. M., Spite, M., Korotin, S. A., Spite, F., Bonifacio,P., Cayrel, R., Fran¸cois, P., & Hill, V. 2010, A&A, 509, A88Andrievsky, S. M., Spite, F., Korotin, S. A., Fran¸cois, P., Spite,M., Bonifacio, P., Cayrel, R., & Hill, V. 2011, A&A, 530, A105Aoki, W., Beers, T. C., Christlieb, N., Norris, J. E., Ryan, S. G.,& Tsangarides, S. 2007a, ApJ, 655, 492Aoki, W., Honda, S., Beers, T. C., Takada-Hidai, M., Iwamoto, N.,Tominaga, N., Umeda, H., Nomoto, K., Norris, J. E., & Ryan,S. G. 2007b, ApJ, 660, 747Asplund, M. 2005, ARA&A, 43, 481Asplund, M., Lambert, D. L., Nissen, P. E., Primas, F., & Smith,V. V. 2006, ApJ, 644, 229Barklem, P. S., Christlieb, N., Beers, T. C., Hill, V., Bessell,M. S., Holmberg, J., Marsteller, B., Rossi, S., Zickgraf, F.-J.,& Reimers, D. 2005, A&A, 439, 129Baumueller, D. & Gehren, T. 1997, A&A, 325, 1088Beers, T. C., Preston, G. W., & Shectman, S. A. 1985, AJ, 90, 2089Beers, T. C., Preston, G. W., & Shectman, S. A. 1992, AJ, 103,1987Beers, T. C. & Christlieb, N. 2005, ARA&A, 43, 531Bergemann, M. & Gehren, T. 2008, A&A, 492, 823Bernstein, R., Shectman, S. A., Gunnels, S. M., Mochnacki, S., &Athey, A. E. 2003, in Society of Photo-Optical InstrumentationEngineers (SPIE) Conference Series, ed. M. Iye & A. F. M.Moorwood, Vol. 4841, 1694Bohlin, R. C., Jenkins, E. B., Spitzer, L., York, D. G., Hill, J. K.,Savage, B. D., & Snow, T. P. 1983, ApJS, 51, 277Bonifacio, P., Molaro, P., Sivarani, T., Cayrel, R., Spite, M., Spite,F., Plez, B., Andersen, J., Barbuy, B., Beers, T. C., Depagne, E.,Hill, V., Fran¸cois, P., Nordstr¨om, B., & Primas, F. 2007, A&A,462, 851Bromm, V., Coppi, P. S., & Larson, R. B. 1999, ApJ, 527, L5Bromm, V. & Loeb, A. 2003, Nature, 425, 812Cameron, A. G. W., & Fowler, W. A. 1971, ApJ, 164, 111Carney, B. W., Laird, J. B., Latham, D. W., & Aguilar, L. A. 1996,AJ, 112, 668Carollo, D., Beers, T. C., Bovy, J., Sivarani, T., Norris, J. E.,Freeman, K. C., Aoki, W., & Lee, Y. S. 2011, astro-ph/1103.3067Cassisi, S., Castellani, M., Caputo, F., & Castellani, V. 2004, A&A,426, 641Castelli, F. & Kurucz, R. L. 2004, astro-ph/0405087Cayrel, R., Depagne, E., Spite, M., Hill, V., Spite, F., Fran¸cois,P., Plez, B., Beers, T., Primas, F., Andersen, J., Barbuy, B.,Bonifacio, P., Molaro, P., & Nordstr¨om, B. 2004, A&A, 416,1117Christlieb, N., Beers, T. C., Barklem, P. S., Bessell, M.,Hill,V.,Holmberg, J., Korn, A. J., Marsteller, B.,Mashonkina,L.,Qian, Y.-Z.,Rossi, S.,Wasserburg, G. J.,Zickgraf, F.-J.,Kratz,K.-L., Nordstr¨om, B.,Pfeiffer, B.,Rhee, J.,Ryan, S. G. 2004,A&A, 428, 1027 Christlieb, N., Gustafsson, B., Korn, A. J., Barklem, P. S., Beers,T. C., Bessell, M. S., Karlsson, T., & Mizuno-Wiedner, M. 2004,ApJ, 603, 708Christlieb, N., Wisotzki, L., Reimers, D., Homeier, D., Koester, D.,& Heber, U. 2001, A&A, 366, 898Cohen, J. G., McWilliam, A., Shectman, S., Thompson, I.,Christlieb, N., Melendez, J., Ramirez, S., Swensson, A., &Zickgraf, F.-J. 2006, AJ, 132, 137Frebel, A. 2010, Astronomische Nachrichten, 331, 474Frebel, A., Aoki, W., Christlieb, N., Ando, H., Asplund, M.,Barklem, P. S., Beers, T. C., Eriksson, K., Fechner, C., Fujimoto,M. Y., Honda, S., Kajino, T., Minezaki, T., Nomoto, K., Norris,J. E., Ryan, S. G., Takada-Hidai, M., Tsangarides, S., & Yoshii,Y. 2005, Nature, 434, 871Frebel, A., Christlieb, N., Norris, J. E., Beers, T. C., Bessell,M. S., Rhee, J., Fechner, C., Marsteller, B., Rossi, S., Thom,C., Wisotzki, L., & Reimers, D. 2006, ApJ, 652, 1585Frebel, A., Johnson, J. L., & Bromm, V. 2007a, MNRAS, 380, L40Frebel, A., Norris, J. E., Aoki, W., Honda, S., Bessell, M. S.,Takada-Hidai, M., Beers, T. C., & Christlieb, N. 2007b, ApJ,658, 534Fryer, C. L., Woosley, S. E., & Heger, A. 2001, ApJ, 550, 372Fulbright, J. P. 2000, AJ, 120, 1841Garc´ıa P´erez, A. E. & Primas, F. 2006, A&A, 447, 299Gratton, R. G. & Sneden, C. 1988, A&A, 204, 193Gratton, R. G., Sneden, C., Carretta, E., & Bragaglia, A. 2000,A&A, 354, 169Green, E. M., Demarque, P., & King, C. R. 1984, BAAS, 16, 997Heger, A. & Woosley, S. E. 2010, ApJ, 724, 341Keeping, E. S. 1962, Introduction to Statistical Inference(Princeton,N.J.: Van Nostrand)Kim, Y.-C., Demarque, P., Yi, S. K., & Alexander, D. R. 2002,ApJS, 143, 499Kobayashi, C., Tominaga, N., & Nomoto, K. 2011, ApJ, 730, L14Kurucz, R. L. 1993, Kurucz CD-ROM 13, ATLAS9 StellarAtmosphere Programs and 2 km/s Grid, Cambridge: SAOLi, H. N., Christlieb, N., Sch¨orck, T., Norris, J. E., Bessell, M. S.,Yong, D., Beers, T. C., Lee, Y. S., Frebel, A., & Zhao, G. 2010,A&A, 521, 10Lucatello, S., Beers, T. C., Christlieb, N., Barklem, P. S., Rossi,S., Marsteller, B., Sivarani, T., & Lee, Y. S. 2006, ApJ, 652, L37McWilliam, A. 1997, ARA&A, 35, 503McWilliam, A., Preston, G. W., Sneden, C., & Searle, L. 1995, AJ,109, 2757Mel´endez, J., Casagrande, L., Ram´ırez, I., Asplund, M., &Schuster, W. J. 2010, A&A, 515, 3Meynet, G., Ekstr¨om, S., & Maeder, A. 2006, A&A, 447, 623Nomoto, K., Tominaga, N., Umeda, H., Kobayashi, C., & Maeda,K. 2006, Nuclear Physics A, 777, 424Norris, J. E., Christlieb, N., Korn, A. J., Eriksson, K., Bessell,M. S., Beers, T. C., Wisotzki, L., & Reimers, D. 2007, ApJ, 670,774Norris, J. E., Gilmore, G., Wyse, R. F. G., Yong, D., & Frebel, A.2010a, ApJ, 722, L1042 Hollek et al.