Stellar metallicity of star-forming galaxies at z ~ 3
Veronica Sommariva, Filippo Mannucci, Giovanni Cresci, Roberto Maiolino, Alessandro Marconi, Tohru Nagao, Andrea Baroni, Andrea Grazian
aa r X i v : . [ a s t r o - ph . C O ] D ec Astronomy & Astrophysics manuscript no. Sommariva c (cid:13)
ESO 2018September 17, 2018
Stellar metallicity of star-forming galaxies at z ∼ ⋆ Veronica Sommariva , Filippo Mannucci , , Giovanni Cresci , Roberto Maiolino , Alessandro Marconi ,Tohru Nagao , Andrea Baroni , and Andrea Grazian INAF-Osservatorio Astrofisico di Arcetri, Firenze, Italy ⋆⋆ e-mail: [email protected] Harvard-Smithsonian Center for Astrophysics, 60 Garden street, Cambridge, MA 02138, USA INAF-Osservatorio Astronomico di Roma, Monte Porzio Catone, Italy Dipartimento di Astronomia, Universit`a di Firenze, Firenze Italy Kyoto University, JapanReceived 22-09-11; accepted 09-12-11
ABSTRACT
The stellar metallicity is a direct measure of the amount of metals present in a galaxy, as a large part of the metalslie in its stars. In this paper we investigate new stellar metallicity indicators suitable for high-z galaxies studying thestellar photospheric absorption lines in the rest frame ultraviolet, hence sampling predominantly young hot stars. Wedefined these new indicators based on the equivalent widths (EW) of selected features using theoretical spectra createdwith the evolutionary population synthesis code
Starburts99 . We used them to compute the stellar metallicity for asample of UV-selected galaxies at z > z >
3. We findthat the metallicity of young, hot stars in galaxies at z ∼ Key words. galaxies: evolution – galaxies:high-redshift
1. Introduction
Metallicity is one of the important properties of galaxies,and its study is able to shed light on the details of galaxyevolution. It is, in fact, an integrated property, relatedto the whole past history of the galaxies. In particular,metallicity is sensitive to whole star formation history,and so to evolutionary stage of the galaxy. Moreover, it isaffected by presence of infalls and outflows, i.e. by feedbackprocesses and by the interplay between the forming galaxyand the intergalactic medium (see e.g. Erb et al. 2008,Mannucci et al. 2009, Cresci et al. 2010). As consequence,it has become an important test of galaxy evolution (e.g.Nagamine et al. 2001, Spitoni et al. 2010, Dav`e et al.2011).Local galaxies show a clear correlation between mass andmetallicity (MZR), for which the galaxies with largerstellar mass have higher metallicities, and this correlationappears to hold both in term of gas-phase metallicity (e.g.Tremonti et al. 2004) and stellar metallicity (Gallazzi etal. 2005, Panter et al. 2008).At high-redshift the gas-phase metallicity of the ISM ofstar-forming galaxies has been measured using primarily ⋆ Based on ESO observations, proposals 082.A-0398 and084.A-0367 ⋆⋆ oxygen abundances. The most common techniques todetermine the gas phase metallicity are based eitheron theoretical calibrations (see Kewley & Ellison 2002,and Kewley & Ellison 2008) or on empirical metallicitycalibrations, the so-called “strong line diagnostics”, whichare based on the ratios of collisionally excited forbiddenlines to hydrogen recombination lines. Previous studieshave shown that the mass-gas phase metallicity relationpresents evidence of strong redshift evolution. Amongothers, Savaglio et al. (2005) and Zahid et al. (2011)studied star forming galaxies at redshift z ∼ . z ∼ . z ∼ .
2. Two projects werespecifically designed to extend the investigation of MZRat z >
3: LSD (Lyman-break Stellar population andDynamic) and AMAZE (Assessing the Mass-Abundanceredshift Evolution). With these projects, Maiolino et al.(2008) and Mannucci et al. (2009) showed for the first timethe evolution of the mass-metallicity relation at z > z ∼ so-called “Fundamental Metallicity Relation (FMR)”, i.e.,a tight relation between stellar mass, gas-phase metallicity,and star formation rate (SFR). Local SDSS galaxies showvery small residuals around this relation, of the orderof 0.05dex. Yates et al. (2011) found a similar relation,with some differences due to the metallicity calibrationadopted. According to Mannucci et al. (2010), the FMRdoes not appear to evolve with redshift up to z ∼ z ∼ .
3, where galaxies tend to have lowermetallicities.All the observational studies mentioned in the previousparagraph refer to the gas-phase metallicity, as measuredby emission lines. Gallazzi et al. (2005) presented the localmass-stellar metallicity relation based on ∼ ∼ × M ⊙ the relation flattens out. In addictionthey noted that gas-phase metallicity is best determinedfor star-forming galaxies, whereas stellar metallicity is bestdetermined for early-type galaxies, and found that thestellar metallicity is generally lower than the gas-phasemetallicity (by 0.5 dex). More recently Panter et al. (2008)inferred the stellar metallicity history of SDSS galaxies anddetermined their stellar mass-metallicity relation. Theyused a different approach respect to Gallazzi et al. (2005),but they found similar results. Moreover, consideringonly the younger population of galaxies ( ≤ starbursts , the strongest features in therest-frame UV spectrum of distant galaxies are interstellarand photospheric absorption lines of C, N, O, Si, andFe, produced by hot, young O-B stars (see Shapley et al.2003). One advantage in using these spectra to measurethe metallicity at z ∼ λ λ λ λ λ λ SiIII 1417, CIII λ λ Starburst99 plus theirnon-LTE model atmosphere code WM-basic, supportedthe conclusions that these lines are useful metallicityindicators, and suggested a new indicator at λ λ λ . < z < .
40 in the FORS deep field. They investigatedthe evolution of the EWs and the metallicity with theredshift, and found that the abundances of heavy elementsis increasing between z ∼ . z ∼ .
5. Halliday etal. (2008), instead, computed the stellar metallicity usingthe λ z ∼
2. Comparing their results with thegas phase metallicity of galaxies with similar mass, theyfound that their stellar metallicities were lower by a factorof ∼ . dex . Quider et al. (2009) computed the stellarmetallicity for the gravitationally lensed galaxies CosmicHorseshoe at z = 2 .
38 by using the λ z = 3 .
07. Because of the presence a strong sky linesin the region of the stellar metallicity indicators, they donot give the exact value of the stellar metallicity, but justan indication by comparing the observed spectra with themodel. Finally, Dessauges-Zavadsky et al. (2010) studiedthe stellar metallicty for the lensed galaxy 8 o’clock arc at z = 2 .
73. Using the λ λ z ∼ z > z >
3, followedby the conclusions.
2. Stellar metallicity from absorption lines
When studying high-redshift galaxies, is useful to have alarge number of calibrated features over a large wavelengthrange to increase the constraints on metallicity and avoidthe effects of atmospheric absorption bands and bright skyemission lines.A feature can be used as a metallicity indicator if thefollowing conditions are satisfied: 1) the EW is deep enoughto be measured in high redshift galaxies; 2) it varies signif-icantly with the metallicity; and 3) it does not dependscritically on age or IMF.The aim of this work is to update the old metallicitycalibrations and to enlarge the number of indices over allthe range of UV spectral region.We want measure the stellar metallicities of high redshiftgalaxies by comparing the observed photospheric lines tothe results of the population synthesis code
Starburts99 (Leitherer et al. 1999). The original version of this codeallowed the creation of synthetic UV spectra with a varietyof ages and IMFs, and used the stellar evolution models tofollow the stellar population over the time. The first releaseincluded only few different metallicities because empiricalstellar libraries were available only for Milky Way O-typestars observed with the International Ultraviolet Explorer(IUE) satellite. The galactic B stars were later included tothat library by de Mello et al. (2000). In order to considerthe sub-solar heavy elements abundences, a new improve-ment was obtained in 2001 with the inclusion of a libraryof O-type spectra obtained from HST STIS observations ofstars in Large and Small Magellanic Clouds (Leitherer etal. 2001). Rix et al. (2004) followed an other approach byutilizing theoretical library spectra instead than empiricalones. Their purpose was synthesized the photospheric ab-sorption lines seen in the spectra of star forming galaxiestaking into account on the non-LTE model atmosphere, thestellar wind and their effects in the spectral synthesis of hotstars. Therefore they replaced the empirical library with agrid of theoretical spectra generated with the hydrodynam-ics code WM-basic. The major difference between this two
Table 1.
List of the metallicity indicators with the corre-sponding elements, and regions where the equivalent widthsare integrated. For an additional indicator at λ Indicator ID element λ rangeF1370 OV, FeV 1360 − a F1425 CIII, FeV, SiIII 1415 − a F1460 NiII 1450 − b F1501 SV 1496 − b F1978 FeIII 1935 − c Notes. ( a ) Leitherer et al. 2001 ( b ) this work ( c ) Rix et al.2004. approaches is that the empirical spectra include prominentUV interstellar absorption lines not present in the theo-retical spectra, which are purely stellar. Nevertheless, theyfound good agreement between the empirical and the the-oretical spectra.Because Rix et al. (2004) in principle focus on pho-tospheric lines, processes such as shock emission that af-fect only the high-ionization wind lines were not includedin their models. These limitations have been addressed tosome extent in the latest generation of the WM-basic code(Leitherer et al. 2010), which is used in this work. The latestgeneration of the WM-basic code is optimized to computethe strong P Cygni type lines originating in the wind of thehot stars. This is a great advantage respect use only thefaint photospheric lines, because the stellar-wind featuresare stronger, therefore more easy to detect in low S/N spec-trum.
Starburst99 with the WM-basic library allows the cre-ation of simulated spectra depending on a number of freeparameters related to star formation history, IMF, age,metallicity, supernova and black hole cut-off, stellar atmo-spheres and microturbolence. We generated galaxy modelspectra using the Padova tracks, with thermally pulsingAGB stars added, for five values of metallicities, 0.02, 0.4,0.2, 1.0 and 2.5 Z ⊙ , and assuming continuous star forma-tion histories, results in Fig. 1 We considered five differentstellar initial mass function (IMF). The reference model isa classical Salpeter power law with exponent α = 2 .
35 andupper mass limit M up = 100 M ⊙ . We also considered IMFswith α = 1 .
85 and α = 2 .
85 between 1 M ⊙ and 100 M ⊙ ,and with α = 2 .
35 and mass limit M up = 60 M ⊙ . Finally,we compute models using the Kroupa IMF. The UV spectrum of star forming galaxies is dominated bystrong photospheric absorption features that are sensitiveto metallicity.Leitherer et al. (2001) investigated the existence ofsome blended photospheric lines whose strengths dependon metallicities only. They found that the two blends oflines near λ λ λ λ λ λ SiIII 1417, CIII λ λ λ ∼ z ∼ Fig. 1.
Variation of the line indices with stellar population age and metallicity. In each plot, the EW is shown in leftvertical axis, while the right axis shows the average fractional depth below the continuum. For each feature we draw themodels constructed with
Starburts99 at five value of metallicity as shown in different colors. For each metallicity, thecolor regions represent the error due to the dependence on the IMF assumed. In this way we can highlight the dependenceof the models and robustness of the indices from the IMF and stellar population age assumed.Here we use the spectra of the new version of
Starburst99 to update the calibrations of the featuresmentioned above and find additional new useful features.Using the new library, our measurements confirm thatthe indicators proposed in the previous studies, (F1370,F1425, and F1978), are stable after ∼ M yr from theonset of star formation, and increase monotonically withthe metallicity with mild dependence on other parameters(see Fig. 1). Previous works have successfully used theseindices (Halliday et al. 2008 and Quider et al. 2009).The predicted EWs derived using the last version of the
Starburst99 code are similar from the previously ones usedfor the 1425 index (0.11˚A at solar metallicity). Howeverfor the 1978 index the difference is higher, especially athigh metallicity (1.95˚A at solar metallicity). Moreover, theF1978 index is quite sensitive to the assumed IMF (seeFig. 3), making it more uncertain among the metallicityindicator.
The first region that we investigate is between 1496 and1506 ˚A . We choose this region because the SV λ λ Starburts99 atmetallicity 0.02, 0.4, 0.2, 1, and 2.5 Z ⊙ . The strengths ofthe absorption features discussed above is clearly a strongand monotonic function of metallicity. z ∼ Fig. 2.
Smoothed and normalized reference models obtained with
Starburts99 at metallicity 0.02, 0.4, 0.2, 1 and 2.5 Z ⊙ .In red are highlighted the metallicity indicators discussed in the text to show their dependence from metallicity. For an accurate determination of metallicity, we interpolatethe relation between the equivalent widths of the indicatorsas a function of
Log ( Z/Z ⊙ ) discussed above.We fitted a second-order polynomial to measured the val-ues of the equivalent widths. The results of the fitting ispresented in Fig. 3: the triangles represent the referencemodel (Salpeter IMF with α = 2 .
35 and age=50 Myr), theblack line is the quadratic fit of these models, and the greyregion represents the uncertainty on the calibration due tothe different IMF assumed, as shown in Fig. 1.We use a second-order fit expressed as: log ( Z/Z ⊙ ) = α + βEW + γEW (1)The coefficients for the five indices are in Table 2. λ In addition to the two new indicators found, we also studiedthe SiII λ Table 2.
Coefficient of the equation
Log ( Z/Z ⊙ ) = α + βEW + γEW for each index considering the real contin-uum. Index α β γ
F1370 -2.501 1.403 -0.1700F1425 -2.003 1.203 -0.1521F1460 -2.023 1.251 -0.1631F1501 -2.152 2.324 -0.5329F1978 -2.051 0.388 -0.0122
Table 3.
Coefficient of the equation
Log ( Z/Z ⊙ ) = α + βEW + γEW for each index, considering the pseudo con-tinuum. Index α β γ
F1370 -2.897 2.346 -0.4208F1425 -2.138 2.116 -0.4424F1460 -2.183 2.336 -0.5212F1501 -2.543 4.800 -1.968F1978 -2.774 1.070 -0.0936
Unfortunately, the SiII is often detected also in emis-sion: see e.g. the very high signal to noise ratio spectrumpresented by Shapley et al. (2003). This emission appearsto be very weak, and in low resolution spectra it may diffi- z ∼ Fig. 3.
EW-Z relations for the five indices considered. The triangles represent equivalent widths measured from thereference models at age 50 Myr, while the solid line is the quadratic fit given by equations in Table 2 for the realcontinuum (left panels) and Table 3 for the “pseudo-continuum” (right panels). The grey region represents the error dueto the dependence of the models on the IMF assumed.cult to recognize it. In addiction, this is a fine line structure,(Pettini et al. 2004): although this feature is normally pho-tospheric, in dense environments it can be interstellar aswell. Therefore we suggest to use it with caution, and onlyin the spectra where this emission is absent or weak enoughto not compromise the measurements of the EW.We nevertheless present the calibrations obtained forthis feature, and discuss the results found for the observedgalaxies using it as metallicity indicator.The obtained calibrations are:
Log ( Z/Z ⊙ ) = − .
845 + 1 . EW − . EW (2)in the case of real continuum, and Log ( Z/Z ⊙ ) = − .
893 + 1 . EW − . EW (3)in case of “pseudo-continuum”.The results found using this feature are in Sec. 3.4. In previous sections we defined two new photospheric linessensitive to stellar metallicity and independent to the otherstellar parameter (such ad age and IMF), ∼ ∼ ∼ ∼ ∼ Starburts99 consider two different cases of star forma-tion: an instantaneous burst and a continuos star formationat constant rate: in the former the spectrum changes rapidlywith time, in the latter the equilibrium is reached after few Myr, and then the spectrum changes little with the time.In case of one single burst we expect to observe galaxieswhere the O-B stars are already died, however galaxies withcontinuos star formation rate are dominated by bright andhot massive stars which determine the UV continuum (seeAdelberger et al. 2004). Therefore, we decided to considerthe continuos star formation rate to create models with
Starburts99 because it seems to be the better descriptionfor most star forming galaxies.Another critical issue in this kind of work is the deter-mination of the continuum. It is worth noticing that, unlikewhat was done in other similar works (see Rix et al. 2004)the values of EW are relative to the real, theoretical contin-uum. However, often low- and medium resolution spectrado not have many spectral regions free from absorptionlines, and the estimate of such a continuum is not straight-forward. For this reason it is common to define a “pseudo”continuum: this is derived by fitting a spline curve throughthe mean flux in some spectral windows relatively free fromemission and absorption lines, and a normalized spectrumis obtained dividing by this fit . This definition tends tounderestimate the real continuum because no spectral win-dow is totally free from absorptions, but using the samedefinition of “pseudo” continuum for both calibrations andobservations, this uncertainty tends to cancel out.When observing faint, high-redshift galaxies, this methodis difficult to use, either because not enough signal-to-noiseratio is present in the narrow wavelength ranges (1-2˚A, Rixet al. 2004) used to define the “pseudo” continuum, or be-cause bright sky-lines present at λ > z ∼ continuum and in other cases is better to estimate the realcontinuum, we provide both the calibrations.The coefficients for the calibrations using the definitionof the “pseudo-continuum” by Rix et al. (2004) (see Table 3in their paper for the regions used to define the continuum)and using the last version of Starburts99 are in Table 3.Fig. 3 (right panel) show the relations obtained betweenmetallicity and EW.A side effect of these two different procedures is thatthese calibrations have no or very weak dependency onspectral resolution, as far it is high enough to well samplethe region of interest, i.e., 10˚A rest-frame. Only the F1978index depends strongly on resolution, as already noticed byHalliday et al. (2008). In addition this index depends alsoon the IMF, as shown Fig. 3.It is worth noticing that these latter calibrations aremore easy to use because the definition of the “pseudo con-tinuum” is unambiguous: in fact only some tiny and definedregions are used to defined the “pseudo continuum”, andnot the entire continuum is considered, as in the other case.Therefore, we recommend to use the “pseudo continuum”in case of spectra with high signal-to-noise ratio, where thecontinuum is less affected by bright sky lines and emissionor absorption interstellar lines, and in case of low signal-to-noise spectra we suggest to use the first relations anddefine the real continuum for each spectrum in order toselect regions not affected by strong, bright sky lines.
3. Stellar metallicity in high redshift galaxies in theAMAZE sample.
In this section we apply the method and the relations pre-sented above to a sample of five galaxies at z ∼ . . , in orderto find, for the first, the stellar mass-metallicity relation atthis redshift. For the present investigation we selected a sub-sample ofAMAZE galaxies at z ∼ . − The data reduction was performed using the ESO FORS-pipeline. The default reduction of spectroscopic sciencedata is as follows: the data are first corrected for bias,then an extracted mask, containing the positions of thelong slit and the spatial curvatures of the spectra, is ap-plied to the science data; they are then flat-fielded andre-mapped eliminating the cosmic-ray and the optical dis-tortions. Afterwards they are rebinned to constant wave-length steps. The wavelength calibration is adjusted usingsky emission lines: this allows to correct for shifts betweennight-time science and day-time calibration data.The sky background is obtained using a median of the pixelfree from emission lines, and the sky is subtracted beforeremapping, i.e. when the spectra are still in the originalCCD coordinate system. The sky is determined with a ro-bust linear fitting, which allows for a linear spatial gradientin the background.Flux calibration was performed using spectra of spec-trophotometric standard stars obtained each night. Thesespectra were reduced with the pipeline, as for the sciencedata, and used to convert the ADUs into flux units. Theobserved spectra of the standard stars are divided by thecorresponding stellar spectra taken from the literature inorder to obtain response curves and calibration factors.Representative correction curves are created for each runcombining the individual response curves and smoothingthe result by a spline interpolation. Finally we applied thecorrection curves and calibration factor obtained at all theobserved spectra.At the end we obtained 120 flux calibrated spectra for eachgalaxy.We combined all the reduced data and finally we used theIRAF task apall, with the aperture size of 5 pixel, to ex-tract one-dimensional spectra for the objects. The list ofthe sample of galaxies selected for this project are listedin Table 4. The table lists the properties of the objects:ID, RA, DEC, R magnitude, spectroscopic redshift, signalto noise ratio of the summed spectra, stellar mass and gasphase metallicity.Because of the low signal to noise ratio of some reduceddata, we can not use all the spectra for our analysis. After acareful visual inspection, we removed from our sample thegalaxy CDFS-5161 because its spectrum presents too lowsignal-to-noise ratio to detect the single lines. Moreover,we combined the three galaxies CDFS-12361, CDFS-9313,and CDFS-6664 in order to improve the signal of the singlespectra (hereafter we call it CDFS-comb spectrum). Thespectrum of the CDFS-4417 galaxy was good enough to beused individually. IRAF (Image Reduction and Analysis Facility) is distributedby the National Optical Astronomy Observatories, which areoperated by the Association of Universities for Research inAstronomy, Inc., under co- operative agreement with theNational Science Foundation. 7eronica Sommariva et al.: Stellar metallicity of star-forming galaxies at z ∼ Table 4.
Properties of objects observed with FORS: Col. 1 object name in the MUSIC catalog (Grazian et al. 2006),Cols. 2,3, coordinates (J2000), Col. 4 R-band magnitude, Col. 5 spectroscopic redshift, Col. 6 the signal to noise ratio,Col. 7 Stellar Mass and Col. 8 gas metallicity.
ID RA(J2000) DEC(J2000) Rmag z SNR a LogM
12 +
Log ( O/H ) gas CDFS-12631 03 32 18.1 -27 45 19.0 24.72 3.709 7 9 . +0 . − . . +0 . − . b CDFS-9313 03 32 17.2 -27 47 54.4 24.82 3.654 8 9 . +0 . − . . +0 . − . b CDFS-6664 03 32 33.3 -27 50 07.4 24.80 3.797 8 8 . +0 . − . . +0 . − . c CDFS-5161 03 32 22.6 -27 51 18.0 24.96 3.660 4 9 . +0 . − . . +0 . − . b CDFS-4417 03 32 23.3 -27 51 56.8 23.42 3.473 14 10 . +0 . − . . +0 . − . c Notes. ( a ) Signal-to-noise ratio in the wavelength used ( b ) Troncoso et al. in preparation ( c ) Maiolino et al. 2008.
Defining the continuum is a critical step in measuring theEWs. First, the signal-to-noise ratio of our spectra is gen-erally limited (see Tab. 4) and changes significantly withwavelength because of the presence of sky-lines. Second, atour resolution most of the spectrum is affected by absorp-tion lines. Taking into account these two problems, we fitthe continuum by using a third- or fifth-order polynomial,depending on the continuum shape, by using only regionsthat are expected to be free from deep absorption lines.To define the continuum, we use the
Starburst99 spectra,excluding all the regions where absorption larger than 5%are expected for solar metallicities. This method defines acontinuum which is slightly underestimated. The amount ofcorrection needed can be measured by applying the sametechnique to theoretical spectra, where the continuum levelis well know. We obtain that our fitting method definesa continuum level that is 3% ±
2% below the real one, de-pending on the spectrum used. This means that the con-tinuum bias can be removed by multiplying the normalizedspectra by 1.03. A series of simulations were also used toestimate the uncertainties on the continuum level in eachtarget galaxy. Random gaussian noise was added to the
Starburst99 spectra to obtain the same signal-to-noise ob-served in each spectrum, and the continuum was fitted,repeating the procedure many times. We obtain that thecontinuum level is uncertain of about 5% for CDFS-4417,and 10% for the combined spectrum, and that this uncer-tainty is fairly constant with wavelength.The EW are measured with respect to this fitted con-tinuum, after excluding regions affected by bright sky lines.The uncertainties on the EWs derive from both the poissonnoise on the used pixels, and on the uncertainties on thecontinuum level described above. The latter contribution isdominant in all cases, and can become very large for F1978which is 85˚A wide. On the contrary, F1501 index is defineon the narrowest wavelength range and is less affected bythis contribution. As a result, usually F1501 is the mostreliable index for our galaxies.
The EWs measured as described above are compared withthe calibrations in Tab. 2 to obtain the stellar metallicities.We will not consider neither F1370, which is not measur-able in our spectra of limited signal-to-noise, nor F1978,which, for our redshifts, falls in wavelength regions covered by many sky lines. Therefore for the observed galaxies wecan use only F1425, F1460, and F1501.Fig. 4 and Fig. 5 show the comparison of the observedspectrum of CDFS-4417 (solid line) and the composite spec-trum CDFS-comb (solid line) with the theoretical model(dotted line), computed for five different metallicities forthe indices considered to compute the stellar metallicity.Among the five models plotted in the figures, the metallici-ties that most closely match the observations are 0.2 Z ⊙ , inall the spectra regions considered.This agreement can be made more quantitative usingthe equations reported in Table 2 and computing the metal-licity for the observed objects. In Table 5 we present thevalue found for each available features with the associatederror calculated applying the theory of propagation of theerror, i.e. we propagated the uncertainties both in the ob-servations and the models to get the final value. The lastcolumn of the Table 5 represents the final value of metallic-ity from weighted average of the value of the single featuresand the associated weighted error.Because in our spectra we did not observe any emis-sion at 1533 λ , we calculated the metallicity of the spectrumCDFS-4417 applying the Equation 2, and we found a valueof Z = 0 . Z ⊙ ± .
12. This value is in good agreement withwhat found using the other indicators (0.23 ± . λ as metallicity indicatorwhen the spectra is free from the Si II* fine-structure emis-sion line. Tthe final value of the stellar metallicity does notchange even computing the weighted average using also thisfeature, where possible.The gas phase metallicity of the CDFS-4417 was takenfrom Maiolino et al. (2008), while for the combined FORSspectrum we combined rest frame optical spectrum in thesame way we did for the rest frame UV one, in order toderive consistent properties. We measured the gas phasemetallicity with the R using the line fluxes measured onthe combined spectrum.
4. Other data from literature
To enlarge the sample we collected from the literature otherrest frame UV spectra suitable for this kind of work.To compute the gas-phase metallicity we consider one ofthe most frequently used metallicity diagnostics, the R parameter, defined as: R = F ([OII] λ F ([OIII] λ F ([OIII] λ F (H β λ z ∼ Fig. 4.
Comparison of the observed spectrum of the object CDFS-4417 at z = 3 .
47 (black line) with the theoreticalspectra (dotted grey line) produced by
Starburts99 for five different metallicities, for the each index used to computethe stellar metallicity. In red are highlighted the metallicity indicators. All the spectra are normalized by the fittedcontinuum.
Table 5.
Metallicity measured in the observed galaxies with the errors computed used all the available indicator, andweighted average.
ID F1425 F1460 F1501 WeightedAverageCDFS-4417 0.14 ± .
44 0.17 ± .
47 0.30 ± .
32 0.23 ± . ± .
74 0.37 ± .
62 – 0.24 ± . where F([O II] λ λ λ λ R was proposed by Pagel et al.(1979), and its calibration to the oxygen abundance hasbeen improved by both photoionization model calculations(e.g., McGaugh 1991; Kewley & Dopita 2002), and em-pirical calibrations (Nagao et al. 2006). For consistencywith the FORS galaxies discussed above, and to comparewith the data of Mannucci et al. (2010), we used the mostrecent R calibration provided by Maiolino et al. (2008). The spectrum of Cosmic Horseshoe, a gravitationally lensedgalaxy at z = 2 .
38, was analyzed by Quider et al. (2009).They measured the EW of the F1425 and following the defi-nition and obtained the metallicity Z = 0 . Z ⊙ . Their value of the EW are obtained considering the pseudo continuumas defined in Rix et al. (2004). To make a consistent com-parison of the Cosmic Horseshoes metallicity with the othergalaxies, we measured the EWs using the definition of thereal continuum described in 3.3, and computing the metal-licity with the equation 1. With our procedure the metal-licity found is Z = 0 . Z ⊙ , i.e. 12 + log( O/H ) = 8 . ± R index with the emission linefluxes taken from Hainline et al. (2009) and the metallicitycalibration of Maiolino et al. (2008). In this way we ob-tained 12 + log( O/H ) = 8 . ± . µ m and 4.5 µ m by using Spitzer archive images andusing a photometric aperture of 8 arcsec and subtractingthe flux from the central lensing galaxy. The U, G andI photometry were taken from Belokurov et al. (2007). z ∼ Fig. 5.
Comparison of the observed combined spectrum of the object CDFS-comb at z = 3 .
71 (black line) with thetheoretical spectra (dotted grey line) produced by
Starburts99 for five different metallicities, for the each index used tocompute the stellar metallicity. In red are highlighted the metallicity indicators. All the spectra are normalized by thefitted continuum.We used the Hyperzmass code (Pozzetti et al. 2007 andBolzanella et al. 2000), with Bruzual & Charlot (2003) li-braries, assuming a Chabrier IMF (Chabrier et al. 2003)with an upper mass limit of 100 M ⊙ , smooth exponentiallydecreasing Star Formation Histories (SFHs) with time scale τ = [0 . , ∞ ] and age t = [0 . , Log ( M/M ⊙ ) = 10 . ± .
19, corrected for the magnifica-tion.
The lensed galaxy Cosmic Eye at z = 3 . Z ∼ . Z ⊙ , comparing the observed spec-trum with the models. Computing the EW of the otherindices and applying our calibration, we found metallicity Z = 0 . Z ⊙ , i.e. 12 + log( O/H ) = 8 . ± . O/H ) = 8 . ± . R from Stark et al. (2008)and applying it in the metallicity calibrations of Maiolinoet al. (2008).The stellar mass of the Cosmic Eye derived from SED fit- ting is Log ( M/M ⊙ ) = 9 . ± .
14 (Troncoso et al. in prepa-ration).
The last spectrum that we obtained belong to the MS 1512-cB58 (Pettini et al. 2000), a gravitational lensed galaxiesat z = 2 .
72 that, thanks to the magnification, presents aspectrum with very high signal to noise ratio. As for theprevious galaxies, we measured the EWs for all the definedindicators on the spectra using the real continuum and andwe obtained a stellar metallicity from equation 1 of Z =0 . Z ⊙ , i.e. 12 + log( O/H ) = 8 . ± . R calibration of Maiolino et al. (2008). The result is 12 +log( O/H ) = 8 . ± . Log ( M/M ⊙ ) = 8 . ± .
15 (Siana et al. 2008).
For five additional high redshift galaxies it was not possi-ble to obtain the spectra, but only the value of the EWsfrom the literature. This is the case of the lensed galaxy 8o’clock arc at z = 2 .
73 studied by Dessauges-Zavadsky et z ∼ al. (2010), as well as for the four galaxies of the Fors DeepField with redshift 2 . < z < . Z = 0 . Z ⊙ , i.e. 12 + log( O/H ) = 8 . R index with the emis-sion line fluxes taken from Finkelstein et al. (2009) and us-ing the calibrations of Maiolino et al. (2008). We obtained12 + log( O/H ) = 8 . ± . Log ( M/M ⊙ ) = 10 . +2 . − . (Richard et al. 2011). We can not give the errors for the stel-lar metallicity of this galaxy because Dessauges-Zavadskyet al. (2010) did not give any indication of the uncertaintiesrelative to the EWs measurements.In the same way, we calculated the stellar metallicity forfour FORS Deep Field galaxies studied by Mehlert et al.(2006). It was not possible to compute the gas phase metal-licity because the emission line fluxes for these galaxies arenot available. We reported the results in Fig. 7, where wehighlighted these objects with black crosses. The massesof these galaxies were provided by Drory et al. (2005 andprivate communication).
5. Comparison between stellar and gas-phasemetallicities
In this section we compare the stellar metallicity obtainedusing the empirical calibrations described above with thegas phase ones.Small differences are expected between the stellar metal-licities as measured using UV absorption features and thegas phase metallicities obtained by strong optical emissionlines. Nevertheless, if the galaxy is experiencing a rapidmetallicity evolution, differences could be related to thelonger lifetimes of the star dominating UV emission withrespect to the more massive stars dominating line emission.In fact, the stars responsible of the UV emission are young,hot O-B stars (with a life time of ∼ − yr), whichwere formed from interstellar gas with very similar proper-ties of the one seen in emission given their short lifetime.On the other hand, larger differences are expected usingoptical absorption features, dominated by longer lived stars(e.g. Lick indices, as found in Galazzi et al. (2005) andPanter et al. (2008) for local SDSS galaxies).In Table 6 we report the comparison between stellarand gas phase metallicity for the objects that we discussedin the previous sections. The differences between the stellarand the gas-phase metallicity are plotted in Fig. 6. For eachgalaxies the error are computed by combining the errorsfrom the gas-phase and stellar metallicities in quadrature.As expected, we do not find large discrepancy betweenthe two quantities: the average difference is -0.16 and theuncertainty is 0.14, therefore the difference has a signifi-cance of about 1.1sigma. This does not allow us to claimany significant difference, given the large systematic uncer-tainties associated with both methods. Also Halliday et al.(2008), analyzing star forming galaxies at z ∼ ∼ .
6. The stellar mass-metallicity relation
Our data allow for the first time the study of the stellarmass-metallicity relation at high redshift.Fig. 7 shows the position and the relative error of theobserved galaxies (black squares) at z > z ∼ z ∼ .
07, themagenta dotted line represents the relation obtained byErb et al. (2006) at z ∼ .
2, and the blue dotted line showsthe behavior of such relation inferred from the initialsample of LSD and AMAZE sources at z ∼ z ∼
3, while no evolution isfound at lower redshifts (see also Cresci et al. 2011).As shown in Fig. 6, the stellar metallicities derived for z ∼ z ∼ Table 6.
Comparison between the stellar metallicity obtained with this work and gas phase metallicity. In all the caseswe assume a solar value of log ( Z/Z ⊙ = 12 + log ( O/H ) − .
69 (Allande Prieto et al. 2001). The Mass is log(
M/M ⊙ ). ID Mass redshift stellar metallicity gas metallicityCDFS-4417 10.38 3.47 8.05 ± .
22 8.46 ± . ( a ) CDFS-comb 9.38 3.71 7.92 ± .
24 7.98 ± . ( b ) horseshoe 10.56 2.37 8.26 ± .
29 8.48 ± . ( c ) − − − ± . ( d ) Cosmic eye 9.60 3.07 8.36 ± .
20 8.60 − − − ( b ) MS 1525 cB58 8.94 2.72 8.33 ± .
25 8.35 ± . ( e ) FDF-3173 9.96 3.27 8.54 − − − − − − − − − −
FDF-3810 10.95 2.37 8.90 − − − − − − − − − −
FDF-5903 11.01 2.77 8.24 − − − − − − − − − −
FDF-6934 11.31 2.44 8.57 − − − − − − − − − −
Notes.
The gas metallicity was calculated using the calibration of Maiolino et al. (2008), and the emission line fluxes taken from ( a ) Maiolino et al. (2008) ( b ) Troncoso et al. in prep. ( c ) Hainline et al. (2009) ( d ) Finkelstein et al. (2009) ( e ) Teplitzt al. (2000).respectively.
Fig. 6.
Difference between the stellar and the gas-phasemetallicity of the galaxies. The redshift of each source isreported in the labels. The dotted lines are the mean valueof the differences and the 1- σ deviation.Panter et al. (2008), their stellar metallicities for the galax-ies with a younger stellar populations ( ≤ z > . z=2.73 z=2.38z=3.07 z=3.47z=3.71z=2.72 z=3.27 z=2.37z=3.39z=2.44 Fig. 7.
Stellar mass-metallicity relation for the FORSgalaxies at z ∼ . z ∼ z = 0 .
07, the magenta line was found by Erb et al. (2006)at z = 2 ,
2, and the blue dotted line shows the relation fromthe LSD and AMAZE inferred by Mannucci et al. (2009)at z ∼ .
5. The red line is stellar mass-metallicity relationfound by Panter et al. (2008) in local universe.
7. Conclusions
In this paper, we have investigated for the first time thestellar mass-metallicity relation at high redshift, z ∼ Starburts99 , we looked for photospheric ab- z ∼ sorption lines to be used as indicators of stellar metallicity.First we tested the line indices proposed by Leitherer et al.(2001) and Rix et al. (2004), the F1370, F1425 and F1978using the last version of Starburts99 , although the F1978shows a strong dependence from the resolution and theIMF.Then we defined two new photospheric lines, F1460 andF1501, and we found that these lines are sensitive to themetallicity and almost independent of the age and theIMF, and therefore useful stellar metallicity indicators.The F1501 index seems to be the most promising becauseit is defined on the narrowest wavelength range and lessaffected by the uncertainties on the continuum definition.We provided the metallicity calibrations, see Fig. 3,with two different definitions of the continuum: the firstrelations are referred to the real continuum, that wesuggest to use in case of spectra with low signal-to-noiseratio, the others were obtained using the definition of the“pseudo-continuum” provided by Rix et al. (2004), thatwe recommend in case of high signal-to-noise spectra, seeSec. 2.3.We applied the relations on one galaxy and a compositespectra comprised of three additional galaxies of theAMAZE sample at z ∼ .
3, for which the gas phasemetallicity and the galaxy masses were already know.We took from the literature the spectra of eight additionalgalaxies, and we recompute their stellar metallicity usingthe new calibrations.At the end we compared the results found with the gasphase metallicity for each object, see Fig. 6. The mainconclusion of this work is that within the errors, the stellarand the gas phase metallicity are consistent, althoughthere seems to be a tendency to find stellar metallicitylower than the gas phase one by ∼ . dex , as already foundby Halliday et al. (2008). This result supports the lowmetal content derived for the gas phase of high-z galaxiesfrom optical strong line ratios, as well as an evolution ofthe Fundamental Metallicity Relation at z > z > .
5, see Fig. 7. We notice that the stellarmetallicities found at high redshift is comparable withthose found by Panter et al. (2008) for local galaxies,although the two are not straightforwardly comparableas in high redshift galaxies the stellar metallicity arecomputed for hot, young stars, while in the local galaxiesfor cold, older stellar population.In summary, the rest-frame UV is rich in metallicitydependent features, which are able to provide a measureof stellar metallicity in high redshift galaxies. This repre-sent an independent measure of the chemical abundancesin galaxies with respect to the more widespread gas phasemetallicities, which can provide important constraints tothe star formation histories of galaxies in the early Universe.Although this technique is currently limited to very brightor lensed galaxies by the high S/N required, the adventof next generation of telescopes will give us much higherquality spectra for high redshift galaxies, and the stellarmetallicity indicators will play a more important role inchemical abundances studies at high redshift.
Acknowledgements.
GC acknowledges financial support from ASI-INAF grant I/009/10/0. Thanking Chuck Steidel for his spectrum ofMS1512-cB58, and Alice Shapley for her spectra of some lensed galax- ies. We thank Max Pettini and Claus Litherer for useful comment andsuggestions. We are grateful to Drory for providing the masses of theFDF galaxies.
References