the metallicity evolution of star-forming galaxies …€¦ · and =-((+)/ –

The Astrophysical Journal 7639 (26pp) 2013 January 20 doi1010880004-637X76319Ccopy 2013 The American Astronomical Society All rights reserved Printed in the USA

THE METALLICITY EVOLUTION OF STAR-FORMING GALAXIES FROM REDSHIFT 0 TO 3COMBINING MAGNITUDE-LIMITED SURVEY WITH GRAVITATIONAL LENSING

T-T Yuan12 L J Kewley124 and J Richard31 Institute for Astronomy University of Hawaii 2680 Woodlawn Drive Honolulu HI 96822 USA

2 Research School of Astronomy and Astrophysics The Australian National University Cotter Road Weston Creek ACT 2611 Australia3 CRAL Observatoire de Lyon Universite Lyon 1 9 Avenue Charles Andre F-69561 Saint Genis Laval Cedex France

Received 2012 September 23 accepted 2012 November 26 published 2012 December 27

ABSTRACT

We present a comprehensive observational study of the gas-phase metallicity of star-forming galaxies fromz sim 0 rarr 3 We combine our new sample of gravitationally lensed galaxies with existing lensed andnon-lensed samples to conduct a large investigation into the massndashmetallicity (MZ) relation at z gt 1 We applya self-consistent metallicity calibration scheme to investigate the metallicity evolution of star-forming galaxies asa function of redshift The lensing magnification ensures that our sample spans an unprecedented range of stellarmass (3times107 to 6times1010 M) We find that at the median redshift of z = 207 the median metallicity of the lensedsample is 035 dex lower than the local SDSS star-forming galaxies and 018 dex lower than the z sim 08 DEEP2galaxies We also present the z sim 2 MZ relation using 19 lensed galaxies A more rapid evolution is seen betweenz sim 1 rarr 3 than z sim 0 rarr 1 for the high-mass galaxies (1095 M lt M lt 1011 M) with almost twice asmuch enrichment between z sim 1 rarr 3 than between z sim 1 rarr 0 We compare this evolution with the mostrecent cosmological hydrodynamic simulations with momentum-driven winds We find that the model metallicityis consistent with the observed metallicity within the observational error for the low-mass bins However for highermasses the model overpredicts the metallicity at all redshifts The overprediction is most significant in the highestmass bin of 1010ndash1011 M

Key words galaxies abundances ndash galaxies evolution ndash galaxies high-redshift ndash gravitational lensing strong

Online-only material color figures

1 INTRODUCTION

Soon after the pristine clouds of primordial gas collapsed toassemble a protogalaxy star formation ensued leading to theproduction of heavy elements (metals) Metals were synthesizedexclusively in stars and were ejected into the interstellarmedium through stellar winds or supernova explosions Tracingthe heavy element abundance (metallicity) in star-forminggalaxies provides a ldquofossil recordrdquo of galaxy formation andevolution

When considered as a closed system the metal content of agalaxy is directly related to the yield and gas fraction (Searleamp Sargent 1972 Pagel amp Patchett 1975 Pagel amp Edmunds1981 Edmunds 1990) In reality a galaxy interacts with itssurrounding intergalactic medium hence both the overall andlocal metallicity distribution of a galaxy is modified by feedbackprocesses such as galactic winds inflows and gas accretions(eg Lacey amp Fall 1985 Edmunds amp Greenhow 1995 Koppenamp Edmunds 1999 Dalcanton 2007) Therefore observations ofthe chemical abundances in galaxies offer crucial constraints onthe star formation history and various mechanisms responsiblefor galactic inflows and outflows

The well-known correlation between galaxy mass (lu-minosity) and metallicity was first proposed by Lequeuxet al (1979) Subsequent studies confirmed the existence ofthe luminosityndashmetallicity relation (eg Rubin et al 1984Skillman et al 1989 Zaritsky et al 1994 Garnett 2002) Lumi-nosity was used as a proxy for stellar mass in these studies asluminosity is a direct observable Aided by new sophisticatedstellar population models stellar mass can be robustly calcu-lated and a tighter correlation is found in the massndashmetallicity

4 ARC Future Fellow

(MZ) relation Tremonti et al (2004) have established the MZrelation for local star-forming galaxies based on sim5 times 105 SloanDigital Sky Survey (SDSS) galaxies At intermediate redshifts(04 lt z lt 1) the MZ relation has also been observed for a largenumber of galaxies (gt100 eg Savaglio et al 2005 Cowie ampBarger 2008 Lamareille et al 2009) Zahid et al (2011) derivedthe MZ relation for sim103 galaxies from the Deep Extragalac-tic Evolutionary Probe 2 (DEEP2) survey validating the MZrelation on a statistically significant level at z sim 08

Current cosmological hydrodynamic simulations and semi-analytical models can predict the metallicity history of galax-ies on a cosmic timescale (Nagamine et al 2001 De Luciaet al 2004 Bertone et al 2007 Brooks et al 2007 Dave ampOppenheimer 2007 Dave et al 2011a 2011b) These modelsshow that the shape of the MZ relation is particularly sensitive tothe adopted feedback mechanisms The cosmological hydrody-namic simulations with momentum-driven wind models providea better match with observations than energy-driven wind mod-els (Oppenheimer amp Dave 2008 Finlator amp Dave 2008 Daveet al 2011a) However these models have not been tested thor-oughly in observations especially at high redshifts (z gt 1)where the MZ relation is still largely uncertain

As we move to higher redshifts selection effects and small-number statistics haunt observational metallicity history studiesThe difficulty becomes more severe in the so-called redshiftdesert (1 z 3) where the metallicity sensitive opticalemission lines have shifted to the sky-background dominatednear-infrared (NIR) Ironically this redshift range harbors therichest information about galaxy evolution It is during thisredshift period (sim2ndash6 Gyr after the big bang) that the firstmassive structures condensed the star formation rate (SFR)major merger activity and black hole accretion rate peakedmuch of todayrsquos stellar mass was assembled and heavy elements

1

The Astrophysical Journal 7639 (26pp) 2013 January 20 Yuan Kewley amp Richard

were produced (Fan et al 2001 Dickinson et al 2003 Chapmanet al 2005 Hopkins amp Beacom 2006 Grazian et al 2007Conselice et al 2007 Reddy et al 2008) It is therefore ofcrucial importance to explore NIR spectra for galaxies in thisredshift range

Many spectroscopic redshift surveys have been carried outin recent years to study star-forming galaxies at z gt1 (egSteidel et al 2004 Law et al 2009) However due to thelow efficiency in the NIR those spectroscopic surveys almostinevitably have to rely on color-selection criteria and the biasesin UV-selected galaxies tend to select the most massive andless dusty systems (eg Capak et al 2004 Steidel et al 2004Reddy et al 2006) Space telescopes can observe much deeperin the NIR and are able to probe a wider mass range Forexample the narrowband Hα surveys based on the new WFC3camera onboard the Hubble Space Telescope (HST) have locatedhundreds of Hα emitters up to z = 223 finding much faintersystems than observed from the ground (Sobral et al 2009)However the low-resolution spectra from the narrow band filtersforbid derivations of physical properties such as metallicitiesthat can only currently be acquired from ground-based spectralanalysis

Due to the advent of long-slitmulti-slit NIR spectrographson 8ndash10 m class telescopes enormous progress has been madein the last decade to capture galaxies in the redshift desert Forchemical abundance studies a full coverage of rest-frame opticalspectra (4000ndash9000 Aring) is usually mandatory for the most robustdiagnostic analysis For 15 z 3 the rest-frame opticalspectra have shifted into the J H and K bands It remainschallenging and observationally expensive to obtain high signal-to-noise (SN) NIR spectra from the ground especially forldquotypicalrdquo targets at high-z that are less massive than conventionalcolor-selected galaxies Therefore previous investigations intothe metallicity properties between 1 z 3 focused on stackedspectra samples of massive luminous individual galaxies orvery small numbers of lower-mass galaxies (eg Erb et al2006 2010 Forster Schreiber et al 2006 Law et al 2009 Yabeet al 2012)

The first MZ relation for galaxies at z sim 2 was found by Erbet al (2006 hereafter Erb06) using the stacked spectra of 87 UVselected galaxies divided into six mass bins Subsequently massand metallicity measurements have been reported for numerousindividual galaxies at 15 lt z lt 3 (Forster Schreiber et al 2006Genzel et al 2008 Hayashi et al 2009 Law et al 2009 Erbet al 2010) These galaxies are selected using broadband colorsin the UV (Lyman break technique Steidel et al 1996 2003)or using B- z- and K-band colors (BzK selection Daddi et al2004) The Lyman break and BzK selection techniques favorgalaxies that are luminous in the UV or blue and may thereforebe biased against low-luminosity (low-metallicity) galaxies anddusty (potentially metal-rich) galaxies Because of these biasesgalaxies selected in this way may not sample the full range inmetallicity at redshift z gt 1

A powerful alternative method to avoid these selection effectsis to use strong gravitationally lensed galaxies In the case ofgalaxy cluster lensing the total luminosity and area of thebackground sources can easily be boosted by sim10ndash50 timesproviding invaluable opportunities to obtain high SN spectraand probe intrinsically fainter systems within a reasonableamount of telescope time In some cases sufficient SN can evenbe obtained for spatially resolved pixels to study the resolvedmetallicity of high-z galaxies (Swinbank et al 2009 Jones et al2010 2012 Yuan et al 2011) Before 2011 metallicities have

been reported for a handful of individually lensed galaxies usingoptical emission lines at 15 lt z lt 3 (Pettini et al 2001Lemoine-Busserolle et al 2003 Stark et al 2008 Quider et al2009 Yuan amp Kewley 2009 Jones et al 2010) Fortunatelylensed galaxy samples with metallicity measurements haveincreased significantly due to reliable lensing mass modelingand larger dedicated spectroscopic surveys of lensed galaxieson 8ndash10 m telescopes (Richard et al 2011 Wuyts et al 2012Christensen et al 2012)

In 2008 we began a spectroscopic observational surveydesigned specifically to capture metallicity sensitive lines forlensed galaxies Taking advantage of the multi-object cryogenicNIR spectrograph (MOIRCS) on Subaru we targeted well-known strong lensing galaxy clusters to obtain metallicities forgalaxies between 08 lt z lt 3 In this paper we present the firstmetallicity measurement results from our survey

Combining our new data with existing data from the literaturewe present a coherent observational picture of the metallicityhistory and massndashmetallicity evolution of star-forming galaxiesfrom z sim 0 to z sim 3 Kewley amp Ellison (2008) have shownthat the metallicity offsets in the diagnostic methods can easilyexceed the intrinsic trends It is of paramount importanceto make sure that relative metallicities are compared on thesame metallicity calibration scale In MZ relation studies themethods used to derive the stellar mass can also cause systematicoffsets (Zahid et al 2011) Different spectral energy distribution(SED) fitting codes can yield a non-negligible mass offsethence mimicking or hiding evolution in the MZ relation Inthis paper we derive the mass and metallicity of all samplesusing the same methods ensuring that the observational data arecompared in a self-consistent way We compare our observedmetallicity history with the latest prediction from cosmologicalhydrodynamical simulations

Throughout this paper we use a standard ΛCDM cosmologywith H0 = 70 km sminus1 Mpcminus1 ΩM = 030 and ΩΛ = 070 Weuse solar oxygen abundance 12 + log(OH) = 869 (Asplundet al 2009)

The paper is organized as follows Section 2 describes ourlensed sample survey and observations Data reduction andanalysis are summarized in Section 3 Section 4 presents anoverview of all the samples we use in this study Section 5describes the methodology of derived quantities The metallicityevolution of star-forming galaxies with redshift is presentedin Section 6 Section 7 presents the massndashmetallicity relationfor our lensed galaxies Section 8 compares our results withprevious work in literature Section 9 summarizes our resultsIn the Appendix we show the morphology slit layout andreduced one-dimensional (1D) spectra for the lensed galaxiesreported in our survey

2 THE LEGMS SURVEY AND OBSERVATIONS

21 The Lensed Emission-line GalaxyMetallicity Survey (LEGMS)

Our survey (LEGMS) aims to obtain oxygen abundance oflensed galaxies at 08 ltz lt 3 LEGMS has taken enormousadvantage of the state-of-the-art instruments on Mauna KeaFour instruments have been utilized so far (1) the Multi-ObjectInfraRed Camera and Spectrograph (MOIRCS Ichikawa et al2006) on Subaru (2) the OH-Suppressing Infra-Red ImagingSpectrograph (OSIRIS Larkin et al 2006) on Keck II (3) theNear Infrared Spectrograph (NIRSPEC McLean et al 1998)on Keck II (4) the Low Dispersion Imaging Spectrograph

2


Table 1MOIRCS Observation Summary

Target Dates Exposure Time PA Seeing (Ks) Slit Width FilterGrism(ks) (deg) (primeprime) (primeprime)

A1689 2011 Apr 28 500 60 05ndash08 middot middot middot Ks ImagingA1689 2010 Apr 28 156 minus60 05ndash08 08 HK500A1689 2010 Apr 29 192 45 05ndash06 08 HK500A1689 2010 Mar 24 168 20 05ndash06 08 HK500A1689 2010 Mar 25 120 minus20 06ndash07 08 HK500A1689 2008 Apr 23 Mar 24 156 60 05ndash08 08 zJ500A68 2009 Sep 29ndash30 120 60 06ndash10 10 HK500 zJ500

Notes Log of the observations We use a dithering length of 2primeprime5 for all the spectroscopic observations

(LRIS Oke et al 1995) on Keck I The scientific objectiveof each instrument is as follows MOIRCS is used to obtainthe NIR images and spectra for multiple targets behind lensingclusters NIRSPEC is used to capture occasional single fieldlensed targets (especially galaxy-scale lenses) LRIS is used toobtain the [O ii] λ3727 to [O iii] λ5007 spectral range for targetswith z lt 15 From the slit spectra we select targets that havesufficient fluxes and angular sizes to be spatially resolved withOSIRIS In this paper we focus on the MOIRCS observationsof the lensing cluster A1689 for targets between redshifts 15 z 3 Observations for other clusters are ongoing and will bepresented in future papers

The first step to construct a lensed sample for slit spectroscopyis to find the lensed candidates (arcs) that have spectroscopicredshifts from optical surveys The number of known spectro-scopically identified lensed galaxies at z gt 1 is still on the orderof a few tens The limited number of lensed candidates makes itimpractical to build a sample that is complete and well defined inmass A mass complete sample is the future goal of this projectOur strategy for now is to observe as many arcs with knownredshifts as possible If we assume the active galactic nucleus(AGN) fraction is similar to local star-forming galaxies thenwe expect sim10 of our targets to be AGN dominated (Kewleyet al 2004) Naturally lensed sample is biased toward highlymagnified sources However because the largest magnificationsare not biased toward intrinsically bright targets lensed samplesare less biased toward the intrinsically most luminous galaxies

A1689 is chosen as the primary target for MOIRCS ob-servations because it has the largest number (sim100 arcs orsim30 source galaxies) of spectroscopically identified lensed arcs(Broadhurst et al 2005 Frye et al 2007 Limousin et al 2007)

Multi-slit spectroscopy of NIR lensing surveys greatly en-hances the efficiency of spectroscopy of lensed galaxies in clus-ters Theoretically sim40 slits can be observed simultaneously onthe two chips of MOIRCS with a total field of view (FOV) of4prime times 7prime In practice the number of lensed targets on the slits isrestricted by the strong lensing area slit orientations and spec-tral coverage For A1689 the lensed candidates cover an area ofsim2prime times 2prime well within the FOV of one chip We design slit masksfor chip 2 which has better sensitivity and fewer bad pixels thanchip 1 There are sim40 lensed images (sim25 individual galaxies)that fall in the range of 15 z 3 in our slit masks We use theMOIRCS low-resolution (R sim 500) grisms that have a spectralcoverage of 09ndash178 μm in ZJ and 13ndash25 μm in HK To max-imize the detection efficiency we give priority to targets withthe specific redshift range such that all the strong emission linesfrom [O ii] λ3727 to [N ii] λ6584 can be captured in one grismconfiguration For instance the redshift range of 25 z 3 is

optimized for the HK500 grism and 15 z 17 is optimizedfor the ZJ500 grism

From UT 2008 March to UT 2010 April we used eightMOIRCS nights (six usable nights) with four position angles(PAs) and six masks to observe 25 galaxies Metallicity qualityspectra were obtained for 12 of the 25 targets We also includeone z gt 15 galaxy from our observations of A685 The PA ischosen to optimize the slit orientation along the targeted arcsrsquoelongated directions For arcs that are not oriented to match thePA the slits are configured to center on the brightest knotsof the arcs We use slit widths of 0primeprime8 and 1primeprime0 with a varietyof slit lengths for each lensed arc For each mask a brightgalaxystar is placed on one of the slits to trace the slit curvatureand determine the offsets among individual exposures Typicalintegrations for individual frames are 400 s 600 s and 900 sdepending on levels of skyline saturation We use an ABBAdithering sequence along the slit direction with a ditheringlength of 2primeprime5 The observational logs are summarized in Table 1

3 DATA REDUCTION AND ANALYSIS

31 Reduce 1D Spectrum

The data reduction procedures from the raw mask data to thefinal wavelength and flux calibrated 1D spectra were realized bya set of IDL codes called MOIRCSMOSRED The codes werescripted originally by Youichi Ohyama T-T Yuan extended thecode to incorporate new skyline subtraction (see eg Henryet al 2010 for a description of utilizing MOIRCSMOSRED)

We use the newest version (2011 April) of MOIRC-SMOSRED to reduce the data in this work The sky subtractionis optimized as follows For each Ai frame we subtract a skyframe denoted as α((Biminus1+Bi+1)2) where Biminus1 and Bi+1 arethe science frames before and after the Ai exposure The scaleparameter α is obtained by searching through a parameter rangeof 05ndash20 with an increment of 00001 The best α is ob-tained where the root mean square (rms) of the residual R =Ai- α((Biminus1+Bi+1)2) is minimal for a user defined wavelengthregion λ1 and λ2 We find that this sky subtraction method yieldssmaller sky OH line residuals (sim20) than conventional AndashBmethods We also compare with other skyline subtraction meth-ods in literature (Kelson 2003 Davies 2007) We find the skyresiduals from our method are comparable to those from theKelson (2003) and Davies (2007) methods within 5 in general

5 Most of the candidates in A68 are at z lt 1 Due to the low spectralresolution in this observation Hα and [N ii] are not resolved at z lt 1 We donot have sufficient data to obtain reliable metallicities for the z lt 1 targets inA68 and therefore exclude them from this study

3


cases However in cases where the emission line falls on top ofa strong skyline our method is more stable and improves theskyline residual by sim10 than the other two methods

Wavelength calibration is carried out by identifying skylinesfor the ZJ grism For the HK grism we use argon lines tocalibrate the wavelength since only a few skylines are availablein the HK band The argon-line calibrated wavelength is thenre-calibrated with the available skylines in HK to determinethe instrumentation shifts between lamp and science exposuresNote that the rms of the wavelength calibration using a third-order polynomial fitting is sim10ndash20 Aring corresponding to asystematic redshift uncertainty of 0006

A sample of A0 stars selected from the UKIRT photometricstandards were observed at a similar airmass as the targetsThese stars were used for both telluric absorption correctionsand flux calibrations We use the prescriptions of Erb et al(2003) for flux calibration As noted in Erb et al (2003) theabsolute flux calibration in the NIR is difficult with typicaluncertainties of sim20 We note that this uncertainty is evenlarger for lensed samples observed in multi-slits because ofthe complicated aperture effects The uncertainties in the fluxcalibration are not a concern for our metallicity analysis whereonly line ratios are involved However these errors are a majorconcern for calculating SFRs The uncertainties from the multi-slit aperture effects can cause the SFRs to change by a factor of2ndash3 For this reason we refrain from any quantitative analysisof SFRs in this work

32 Line Fitting

The emission lines are fitted with Gaussian profiles Forthe spatially unresolved spectra the aperture used to extractthe spectrum is determined by measuring the Gaussian profileof the wavelength collapsed spectrum Some of the lensedtargets (sim10) are elongated and spatially resolved in theslit spectra however because of the low surface brightnessand thus very low SN per pixel we are unable to obtainusable spatially resolved spectra For those targets we makean initial guess for the width of the spatial profile and forcea Gaussian fit then we extract the integrated spectrum usingthe aperture determined from the FWHM of the Gaussianprofile

For widely separated lines such as [O ii] λ3727 Hβ λ4861single Gaussian functions are fitted with four free parametersthe centroid (or the redshift) the line width the line flux andthe continuum The doublet [O iii] λ λ49595007 are initiallyfitted as a double Gaussian function with six free parametersthe centroids 1 and 2 line widths 1 and 2 fluxes 1 and 2and the continuum In cases where the [O iii] λ4959 line istoo weak its centroid and line velocity width are fixed tobe the same as [O iii] λ5007 and the flux is fixed to be one-third of the [O iii] λ5007 line (Osterbrock 1989) A triple-Gaussian function is fitted simultaneously to the three adjacentemission lines [N ii] λ6548 6583 and Hα The centroid andvelocity width of [N ii] λ6548 6583 lines are constrained by thevelocity width of Hα λ6563 and the ratio of [N ii] λ6548 and[N ii] λ6583 is constrained to be the theoretical value of 13given in Osterbrock (1989) The line profile fitting is conductedusing a χ2 minimization procedure which uses the inverse ofthe sky OH emission as the weighting function The SN perpixel is calculated from the χ2 of the fitting The final reduced1D spectra are shown in the Appendix

33 Lensing Magnification

Because the lensing magnification (μ) is not a direct functionof wavelength line ratio measurements do not require pre-knowledge of the lensing magnification However μ is neededfor inferring other physical properties such as the intrinsic fluxesmasses and source morphologies Parametric models of the massdistribution in the clusters A68 and A1689 were constructedusing the Lenstool software Lenstool6 (Kneib et al 1993 Julloet al 2007) The best-fit models have been previously publishedin Richard et al (2007) and Limousin et al (2007) As detailed inLimousin et al (2007) Lenstool uses Bayesian optimizationwith a Monte Carlo Markov Chain sampler which providesa family of best models sampling the posterior probabilitydistribution of each parameter In particular we use this familyof best models to derive the magnification and relative error onmagnification μ associated to each lensed source Typical errorson μ are sim10 for A1689 and A68

34 Photometry

We determine the photometry for the lensed galaxies inA1689 using four-band HST imaging data one-band MOIRCSimaging data and two-channel Spitzer IRAC data at 36 and45 μm

We obtained a 5000 s image exposure for A1689 on theMOIRCS Ks filter at a depth of 24 mag using a scale of0primeprime117 pixelminus1 The image was reduced using MCSRED in IRAFwritten by the MOIRCS supporting astronomer Ichi Tanaka7

The photometry is calibrated using the Two Micron All SkySurvey stars located in the field

The ACS F475W F625W F775W F850LP data are obtainedfrom the HST archive The HST photometry are determinedusing SExtractor (Bertin amp Arnouts 1996) with parametersadjusted to detect the faint background sources The F775Wfilter is used as the detection image using a 1primeprime0 aperture

The IRAC data are obtained from the Spitzer archive and arereduced and drizzled to a pixel scale of 0primeprime6 pixelminus1 In orderto include the IRAC photometry we convolved the HST andMOIRCS images with the IRAC point-spread functions derivedfrom unsaturated stars All photometric data are measured usinga 3primeprime0 radius aperture Note that we only consider sources thatare not contaminated by nearby bright galaxies sim70 of oursources have IRAC photometry Typical errors for the IRACband photometry are 03 mag with uncertainties mainly from theaperture correction and contamination of neighboring galaxiesTypical errors for the ACS and MOIRCS bands are 015 magwith uncertainties mainly from the Poisson noise and absolutezero-point uncertainties (Wuyts et al 2012) We refer to JRichard et al (2013 in preparation) for the full catalog of thelensing magnification and photometry of the lensed sources inA1689

4 SUPPLEMENTARY SAMPLES

In addition to our lensed targets observed in LEGMS we alsoinclude literature data for complementary lensed and non-lensedsamples at both local and high-z (Table 2) The observationaldata for individually measured metallicities at z gt 15 arestill scarce and caution needs to be taken when using themfor comparison The different metallicity and mass derivationmethods used in different samples can give large systematic

6 httpwwwoampfrcosmologylenstool7 httpwwwnaojorgstaffichiMCSREDmcsredhtml

4


Table 2MedianMean Redshift and Metallicity of the Samples

Sample Redshift Metallicity (12 + log(OH))

gt107M (all) gt109M 109ndash1095 M 1095ndash1011 M 109ndash1010 M 1010ndash1011 MMean

SDSS 0071 plusmn 0016 8589 plusmn 0001 8616 plusmn 0001 8475 plusmn 0002 8666 plusmn 0001 8589 plusmn 0001 8731 plusmn 0001DEEP2 0782 plusmn 0018 8459 plusmn 0004 8464 plusmn 0004 8373 plusmn 0006 8512 plusmn 0005 8425 plusmn 0004 8585 plusmn 0006Erb06 226 plusmn 017 8418 plusmn 0051 8418 plusmn 0050 8265 plusmn 0046 8495 plusmn 0030 8316 plusmn 0052 8520 plusmn 0028Lensed 191 plusmn 063 8274 plusmn 0045 8309 plusmn 0049 8296 plusmn 0090 8336 plusmn 0066 8313 plusmn 0083 8309 plusmn 0086

Median

SDSS 0072 8631 plusmn 0001 8646 plusmn 0001 8475 plusmn 0003 8677 plusmn 0001 8617 plusmn 0001 8730 plusmn 0001DEEP2 0783 8465 plusmn 0005 8472 plusmn 0006 8362 plusmn 0009 8537 plusmn 0008 8421 plusmn 0008 8614 plusmn 0006Erb06 middot middot middot 8459 plusmn 0065 8459 plusmn 0065 8297 plusmn 0056 8515 plusmn 0048 8319 plusmn 0008 8521 plusmn 0043Lensed 207 8286 plusmn 0059 8335 plusmn 0063 8303 plusmn 0106 8346 plusmn 0085 8313 plusmn 0083 8379 plusmn 0094

Notes The errors for the redshift are the 1σ standard deviation of the sample redshift distribution (not the σ of the meanmedian) The errors for the metallicity arethe 1σ standard deviation of the meanmedian from bootstrapping

Table 3Fit to the SFRndashStellar-mass Relation

Sample Redshift (Mean) δ γ

SDSS 0072 0317 plusmn 0003 071 plusmn 001DEEP2 078 0795 plusmn 0009 069 plusmn 002Erb06 226 1657 plusmn 0027 048 plusmn 006Lensed (Wuyts12) 169 293 plusmn 128 147 plusmn 014Lensed (all) 207 202 plusmn 083 069 plusmn 009

Notes The SFR versus stellar mass relations at different redshifts can becharacterized by two parameters δ(z) and γ (z) where δ(z) is the logarithmof the SFR at 1010 M and γ (z) is the power-law index The best fits for thenon-lensed samples are adopted from Zahid et al (2012) The best fits for thelensed sample are calculated for the Wuyts et al (2012) sample and the wholelensed sample separately

discrepancies and provide misleading results For this reason weonly include the literature data that have robust measurementsand sufficient data for consistently recalculating the stellar massand metallicities using our own methods Thus in generalstacked data objects with lowerupper limits in either line ratiosor masses are not chosen The one exception is the stacked dataof Erb06 as it is the most widely used comparison sample atz sim 2 (Table 3)

The samples used in this work are as follows

1 The Sloan Digital Sky Survey sample (z sim 007)We use the SDSS sample (Abazajian et al 2009httpwwwmpa-garchingmpgdeSDSSDR7) defined byZahid et al (2011) The mass derivation method used inZahid et al (2011) is the same as we use in this workAll SDSS metallicities are recalculated using the PP04N2method which uses an empirical fit to the [N ii] and Hαline ratios of H ii regions (Pettini amp Pagel 2004)

2 The Deep Extragalactic Evolutionary Probe 2 sample(z sim 08) The DEEP2 sample (Davis et al 2003httpwwwdeeppsucieduDR3) is defined in Zahid et al(2011) At z sim 08 the [N ii] and Hα lines are not availablein the optical We convert the KK04 R23 metallicity to thePP04N2 metallicity using the prescriptions of Kewley ampEllison (2008)

3 The UV-selected sample (z sim 2) We use the stacked dataof Erb06 The metallicity diagnostic used by Erb06 is thePP04N2 method and no recalculation is needed We offsetthe stellar mass scale of Erb06 by minus03 dex to match the

mass derivation method used in this work (Zahid et al2012) This offset accounts for the different initial massfunction (IMF) and stellar evolution model parametersapplied by Erb06

4 The lensed sample (1 lt z lt 3) Besides the 11 lensedgalaxies from our LEGMS survey in A1689 we include 1lensed source (z = 1762) from our MOIRCS data on A68and 1 lensed spiral (z = 149) from Yuan et al (2011) Wealso include 10 lensed galaxies from Wuyts et al (2012)and 3 lensed galaxies from Richard et al (2011) sincethese 13 galaxies have [N ii] and Hα measurements as wellas photometric data for recalculating stellar masses Werequire all emission lines from the literature to have SN gt3 for quantifying the metallicity of 1 lt z lt 3 galaxiesUpper-limit metallicities are found for 6 of the lensedtargets from our LEGMS survey Altogether the lensedsample is composed of 25 sources 12 (612 upper limits)of which are new observations from this work Upper-limitmetallicities are not used in our quantitative analysis

The methods used to derive stellar mass and metallicityare discussed in detail in Section 5

5 DERIVED QUANTITIES

51 Optical Classification

We use the standard optical diagnostic diagram (BPT) toexclude targets that are dominated by AGNs (Baldwin et al1981 Veilleux amp Osterbrock 1987 Kewley et al 2006) Forall 26 lensed targets in our LEGMS sample we find 1 targetthat could be contaminated by AGNs (B82) The fraction ofAGNs in our sample is therefore sim8 which is similar to thefraction (sim7) of the local SDSS sample (Kewley et al 2006)We also find that the line ratios of the high-z lensed sample has asystematic offset on the BPT diagram as found in Shapley et al(2005) Erb06 Kriek et al (2007) Brinchmann et al (2008) Liuet al (2008) and Richard et al (2011) The redshift evolutionof the BPT diagram will be reported in L J Kewley et al (2013in preparation)

52 Stellar Masses

We use the software LE PHARE8 (Ilbert et al 2009) todetermine the stellar mass LE PHARE is a photometric redshift

8 wwwcfhthawaiiedusimarnoutsLEPHARElepharehtml

5


and simulation package based on the population synthesismodels of Bruzual amp Charlot (2003) If the redshift is knownand held fixed LE PHARE finds the best-fitted SED on a χ2

minimization process and returns physical parameters suchas stellar mass SFR and extinction We choose the IMF byChabrier (2003) and the Calzetti et al (2000) attenuation lawwith E(B minus V ) ranging from 0 to 2 and an exponentiallydecreasing SFR (SFR prop eminustτ ) with τ varying between 0 and13 Gyr The errors caused by emission-line contamination aretaken into account by manually increasing the uncertainties inthe photometric bands where emission lines are located Theuncertainties are scaled according to the emission-line fluxesmeasured by MOIRCS The stellar masses derived from theemission-line-corrected photometry are consistent with thosewithout emission-line correction albeit with larger errors in afew cases (sim01 dex in log space) We use the emission-line-corrected photometric stellar masses in the following analysis

53 Metallicity Diagnostics

The abundance of oxygen (12 + log(OH)) is used as a proxyfor the overall metallicity of H ii regions in galaxies The oxygenabundance can be inferred from the strong recombinationlines of hydrogen atoms and collisionally excited metal lines(eg Kewley amp Dopita 2002) Before doing any metallicitycomparisons across different samples and redshifts it is essentialto convert all metallicities to the same base calibration Thediscrepancy among different diagnostics can be as large as07 dex for a given mass large enough to mimic or hide anyintrinsic observational trends Kewley amp Ellison (2008 hereafterKE08) have shown that both the shape and the amplitude of theMZ relation change substantially with different diagnostics Forthis work we convert all metallicities to the PP04N2 methodusing the prescriptions from KE08

For our lensed targets with only [N ii] and Hα we usethe N2 = log([N ii] λ6583Hα) index as calibrated by Pettiniamp Pagel (2004 the PP04N2 method) All lines are requiredto have SN gt 3 for reliable metallicity estimations Linesthat have SN lt 3 are presented as 3σ upper limits Fortargets with only [O ii] to [O iii] lines we use the indicatorR23 = ([O ii] λ3727 + [O iii] λλ4959 5007)Hβ to calculatemetallicity The formalization is given in Kobulnicky amp Kewley(2004 KK04 method) The upper and lower branch degeneracyof R23 can be broken by the valueupper limit of [N ii]Hα Ifthe upper limit of [N ii]Hα is not sufficient or available to breakthe degeneracy we calculate both the upper and lower branchmetallicities and assign the statistical errors of the metallicitiesas the range of the upper and lower branches The KK04 R23metallicity is then converted to the PP04N2 method using theKE08 prescriptions The line fluxes and metallicity are listed inTable 4 and stellar masses are in Table 5 For the literature datawe have recalculated the metallicities in the PP04N2 scheme

The statistical metallicity uncertainties are calculated bypropagating the flux errors of the [N ii] and Hα lines Themetallicity calibration of the PP04N2 method itself has a1σ dispersion of 018 dex (Pettini amp Pagel 2004 Erb06Therefore for individual galaxies that have statistical metallicityuncertainties of less than 018 dex we assign errors of 018 dex

Note that we are not comparing absolute metallicities be-tween galaxies as they depend on the accuracy of the calibrationmethods However by re-calculating all metallicities to the samecalibration diagnostic relative metallicities can be compared re-liably The systematic error of relative metallicities is lt007 dexfor strong-line methods (KE08)

6 THE COSMIC EVOLUTION OF METALLICITYFOR STAR-FORMING GALAXIES

61 The Zz Relation

In this section we present the observational investigation intothe cosmic evolution of metallicity for star-forming galaxiesfrom redshift 0 to 3 The metallicity in the local universeis represented by the SDSS sample (20577 objects 〈z〉 =0072 plusmn 0016) The metallicity in the intermediate-redshiftuniverse is represented by the DEEP2 sample (1635 objects〈z〉 = 078 plusmn 002) For redshift 1 z 3 we use 19 lensedgalaxies (plus six upper limit measurements 〈z〉 = 191plusmn061)to infer the metallicity range

The redshift distributions for the SDSS and DEEP2 samplesare very narrow (Δz sim 002) and the mean and median redshiftsare identical within 0001 dex However for the lensed samplethe median redshift is 207 and is 016 dex higher than the meanredshift There are two z sim 09 objects in the lensed sample andif these two objects are excluded the mean and median redshiftsfor the lensed sample are 〈z〉 = 203 plusmn 054 zmedian = 209 (seeTable 2)

The overall metallicity distributions of the SDSS DEEP2 andlensed samples are shown in Figure 1 Since the z gt 1 samplesize is 2ndash3 orders of magnitude smaller than the z lt 1 sampleswe use a bootstrapping process to derive the mean and medianmetallicities of each sample Assuming the measured metallicitydistribution of each sample is representative of their parentpopulation we draw from the initial sample a random subset andrepeat the process for 50000 times We use the 50000 replicatedsamples to measure the mean median and standard deviationsof the initial sample This method prevents artifacts from small-number statistics and provides robust estimation of the medianmean and errors especially for the high-z lensed sample

The fraction of low-mass (M lt109 M) galaxies is largest(31) in the lensed sample compared to 9 and 5 in theSDSS and DEEP2 samples respectively Excluding the low-mass galaxies does not notably change the median metallicityof the SDSS and DEEP2 samples (sim001 dex) while it increasesthe median metallicity of the lensed sample by sim005 dexTo investigate whether the metallicity evolution is different forvarious stellar mass ranges we separate the samples in differentmass ranges and derive the mean and median metallicities(Table 2) The mass bins of 109 M lt M lt1095 M and1095 M lt M lt1011 M are chosen such that there aresimilar number of lensed galaxies in each bin Alternativelythe mass bins of 109 M lt M lt1010 M and 1010 M ltM lt1011 M are chosen to span equal mass scales

We plot the metallicity (Z) of all samples as a function ofredshift z in Figure 2 (called the Zz plot hereafter) The firstpanel shows the complete observational data used in this studyThe following three panels show the data and model predictionsin different mass ranges The samples at local and intermediateredshifts are large enough such that the 1σ errors of the meanand median metallicity are smaller than the symbol sizes onthe Zz plot (0001ndash0006 dex) Although the z gt 1 samplesare still composed of a relatively small number of objects wesuggest that the lensed galaxies and their bootstrapped mean andmedian values more closely represent the average metallicitiesof star-forming galaxies at z gt 1 than Lyman break orB-band magnitude-limited samples because the lensed galaxiesare selected based on magnification rather than colors Howeverwe do note that there is still a magnitude limit and flux limit foreach lensed galaxy

6


Figure 1 Left panel the metallicity distribution of the local SDSS (blue) intermediate-z DEEP2 (black) and high-z lensed galaxy samples (red) Right panel thestellar mass distribution of the same samples To present all three samples on the same figure the SDSS (20577 points) and DEEP2 (1635 points) samples arenormalized to 500 and the lensed sample (25 points) is normalized to 100

(A color version of this figure is available in the online journal)

We derive the metallicity evolution in units of ldquodex perredshiftrdquo and ldquodex per Gyrrdquo using both the mean and medianvalues The metallicity evolution can be characterized by theslope (dZdz) of the Zz plot We compute dZdz for two redshiftranges z sim 0 rarr 08 (SDSS to DEEP2) and z sim 08 rarrsim25 (DEEP2 to lensed galaxies) As a comparison we alsoderive dZdz from z sim 08 to 25 using the DEEP2 and theErb06 samples (yellow circleslines in Figure 3) We deriveseparate evolutions for different mass bins We show our resultin Figure 3

A positive metallicity evolution ie metals enrich galaxiesfrom high-z to the local universe is robustly found in all massbins from z sim 08 rarr 0 This positive evolution is indicated bythe negative values of dZdz (or dZdz(Gyr)) in Figure 3 Thenegative signs (both mean and median) of dZdz are significantat gt5σ of the measurement errors from z sim 08 rarr 0 Fromz sim 25 to 08 however dZdz is marginally smaller thanzero at the sim1σ level from the lensed rarr DEEP2 samples Ifusing the Erb06 rarr DEEP2 samples the metallicity evolution(dZdz) from z sim 25 to 08 is consistent with zero withinsim1σ of the measurement errors The reason that there is nometallicity evolution from the z sim 2 Erb06 rarr z sim 08 DEEP2samples may be due to the UV-selected sample of Erb06 beingbiased toward more metal-rich galaxies

The right column of Figure 3 is used to interpret thedecelerationacceleration in metal enrichment Decelerationmeans the metal enrichment rate (dZdz(Gyr) = Δ dex Gyrminus1)is dropping from high-z to low-z Using our lensed galaxiesthe mean rise in metallicity is 0055 plusmn 0014 dex Gyrminus1 forz sim 25 rarr 08 and 0022 plusmn 0001 dex Gyrminus1 for z sim 08 rarr 0The MannndashWhitney test shows that the mean rises in metallicityare larger for z sim 25 rarr 08 than for z sim 08 rarr 0 at asignificance level of 95 for the high-mass bins (1095 M ltM lt1011 M) For lower mass bins the hypothesis that themetal enrichment rates are the same for z sim 25 rarr 08 andz sim 08 rarr 0 cannot be rejected at the 95 confidence level

ie there is no difference in the metal enrichment rates for thelower mass bin Interestingly if the Erb06 sample is used insteadof the lensed sample the hypothesis that the metal enrichmentrates are the same for z sim 25 rarr 08 and z sim 08 rarr 0 cannotbe rejected at the 95 confidence level for all mass bins Thismeans that statistically the metal enrichment rates are the samefor z sim 25 rarr 08 and z sim 08 rarr 0 for all mass bins from theErb06 rarr DEEP2 rarr SDSS samples

The clear trend of the averagemedian metallicity in galaxiesrising from high-redshift to the local universe is not surprisingObservations based on absorption lines have shown a continuingfall in metallicity using the damped Lyα absorption galaxiesat higher redshifts (z sim 2ndash5 eg Songaila amp Cowie 2002Rafelski et al 2012) There are several physical reasons toexpect that high-z galaxies are less metal-enriched (1) high-zgalaxies are younger have higher gas fractions and have gonethrough fewer generations of star formation than local galaxies(2) high-z galaxies may be still accreting a large amount ofmetal-poor pristine gas from the environment and hence havelower average metallicities (3) high-z galaxies may have morepowerful outflows that drive the metals out of the galaxy It islikely that all of these mechanisms have played a role in dilutingthe metal content at high redshifts

62 Comparison between the Zz Relation and Theory

We compare our observations with model predictions fromthe cosmological hydrodynamic simulations of Dave et al(2011a 2011b) These models are built within a canonical hi-erarchical structure formation context The models take intoaccount the important feedback of outflows by implement-ing an observation-motivated momentum-driven wind model(Oppenheimer amp Dave 2008) The effect of inflows and merg-ers are included in the hierarchical structure formation ofthe simulations Galactic outflows are dealt specifically in themomentum-driven wind models Dave amp Oppenheimer (2007)

7


Figure 2 Zz plot metallicity history of star-forming galaxies from redshift 0 to 3 The SDSS and DEEP2 samples (black dots) are taken from Zahid et al (2011) TheSDSS data are plotted in bins to reduce visual crowdedness The lensed galaxies are plotted in blue (upper-limit objects in green arrows) with different lensed samplesshowing in different symbols (see Figure 6 for the legends of the different lensed samples) The purple ldquobowtiesrdquo show the bootstrapping mean (filled symbol) andmedian (empty symbol) metallicities and the 1σ standard deviation of the mean and median whereas the orange dashed error bars show the 1σ scatter of the dataFor the SDSS and DEEP2 samples the 1σ errors of the median metallicities are 0001 and 0006 (indiscernible from the figure) whereas for the lensed sample the 1σ

scatter of the median metallicity is 0067 Upper limits are excluded from the median and error calculations For comparison we also show the mean metallicity of theUV-selected galaxies from Erb06 (symbol the black bowtie) The six panels show samples in different mass ranges The red dotted and dashed lines are the modelpredicted median and 1σ scatter (defined as including 68 of the data) of the SFR-weighted gas metallicity in simulated galaxies (Dave et al 2011b)


8


xFigure 3 Cosmic metal enrichment rate (dZdz) in two redshift (cosmic time) epochs dZdz is defined as the slope of the Zz relation The left column shows dZdzin units of Δdex per redshift whereas the right column is in units of Δdex per Gyr We derive dZdz for the SDSS to the DEEP2 (z sim 0ndash08) and the DEEP2 to thelensed (z sim 08ndash20) samples respectively (black squareslines) As a comparison we also derive dZdz from z sim 08 to 20 using the DEEP2 and the Erb06 samples(yellow circleslines) The filled and empty squares are the results from the mean and median quantities The model prediction (using median) from the cosmologicalhydrodynamical simulation of Dave et al (2011a) is shown in red stars The second to fifth rows show dZdz in different mass ranges The first row illustrates theinterpretation of the dZdz in redshift and cosmic time frames A negative value of dZdz means a positive metal enrichment from high-redshift to local universeThe negative slope of dZdz vs cosmic time (right column) indicates a deceleration in metal enrichment from high-z to low-z


9


found that the outflows are key to regulating metallicity whileinflows play a second-order regulation role

The model of Dave et al (2011a) focuses on the metal contentof star-forming galaxies Compared with the previous workof Dave amp Oppenheimer (2007) the new simulations employthe most up-to-date treatment for supernova and AGB starenrichment and include an improved version of the momentum-driven wind models (the vzw model) where the wind propertiesare derived based on host galaxy masses (Oppenheimer amp Dave2008) The model metallicity in Dave et al (2011a) is defined asthe SFR-weighted metallicity of all gas particles in the identifiedsimulated galaxies This model metallicity can be directlycompared with the metallicity we observe in star-forminggalaxies after a constant offset normalization to account for theuncertainty in the absolute metallicity scale (KE08) The offsetis obtained by matching the model metallicity with the SDSSmetallicity Note that the model has a galaxy mass resolutionlimit of M sim109 M For the Zz plot we normalize the modelmetallicity with the median SDSS metallicity computed fromall SDSS galaxies gt109 M For the MZ relation in Section 7we normalize the model metallicity with the SDSS metallicityat the stellar mass of 1010 M

We compute the median metallicities of the Dave et al(2011a) model outputs in redshift bins from z = 0 to z = 3with an increment of 01 The median metallicities with 1σspread (defined as including 68 of the data) of the model ateach redshift are overlaid on the observational data in the Zzplot

We compare our observations with the model prediction inthree ways

1 We compare the observed median metallicity with themodel median metallicity We see that for the lower massbins (109ndash1095 109ndash1010 M) the median of the modelmetallicity is consistent with the median of the observedmetallicity within the observational errors However forhigher mass bins the model overpredicts the metallicityat all redshifts The overprediction is most significant inthe highest mass bin of 1010ndash1011 M where the Studentrsquost-statistic shows that the model distributions have signif-icantly different means than the observational data at allredshifts with a probability of being a chance difference oflt10minus8 lt10minus8 17 57 for SDSS DEEP2 the lensedand the Erb06 samples respectively For the alternativehigh-mass bin of 1095ndash1011 M the model also overpre-dicts the observed metallicity except for the Erb06 samplewith a chance difference between the model and obser-vations of lt10minus8 lt10minus8 17 89 93 for SDSSDEEP2 the lensed and the Erb06 samples respectively

2 We compare the scatter of the observed metallicity (orangeerror bars on Zz plot) with the scatter of the models (reddashed lines) For all the samples the 1σ scatter of the datafrom the SDSS (z sim 0) DEEP2 (z sim 08) and the lensedsample (z sim 2) are 013 015 and 015 dex whereas the1σ model scatter is 023 019 and 014 dex We find thatthe observed metallicity scatter is increasing systematicallyas a function of redshift for the high-mass bins whereas themodel does not predict such a trend 010 014 017 dex cfmodel 017 015 012 dex 1011ndash1011 M and 007 012018 dex cf model 012 011 010 dex 1010ndash1011 Mfrom SDSS rarr DEEP2 rarr the lensed sample Our observedscatter is in tune with the work of Nagamine et al (2001) inwhich the predicted stellar metallicity scatter increases withredshift Note that our lensed samples are still small and

have large measurement errors in metallicity (sim02 dex)The discrepancy between the observed scatter and modelsneeds to be further confirmed with a larger sample

3 We compare the observed slope (dZdz) of the Zz plot withthe model predictions (Figure 3) We find the observeddZdz is consistent with the model prediction within theobservational errors for the undivided sample of all massesgt1090 M However when divided into mass bins themodel predicts a slower enrichment than observations fromz sim 0 rarr 08 for the lower mass bin of 1090ndash1095 Mand from z sim 08 rarr 25 for the higher mass bin of1095ndash1011 M at a 95 significance level

Dave et al (2011a) showed that their models overpredictthe metallicities for the highest mass galaxies in the SDSSThey suggested that either (1) an additional feedbackmechanism might be needed to suppress star formation inthe most massive galaxies or (2) wind recycling may bebringing in highly enriched material that elevates the galaxymetallicities It is unclear from our data which (if any)of these interpretations is correct Additional theoreticalinvestigations specifically focusing on metallicities in themost massive active galaxies are needed to determine thetrue nature of this discrepancy

7 EVOLUTION OF THEMASSndashMETALLICITY RELATION

71 The Observational Limit of the MassndashMetallicity Relation

For the N2-based metallicity there is a limiting metallicitybelow which the [N ii] line is too weak to be detected Since [N ii]is the weakest of the Hα +[N ii] lines it is therefore the flux of[N ii] that drives the metallicity detection limit Thus for a giveninstrument sensitivity there is a region on the massndashmetallicityrelation that is observationally unobtainable Based on a fewsimple assumptions we can derive the boundary of this regionas follows

Observations have shown that there is a positive correlationbetween the stellar mass M and SFR (Noeske et al 2007bElbaz et al 2011 Wuyts et al 2011) One explanation for theM versus SFR relation is that more massive galaxies haveearlier onset of initial star formation with shorter timescales ofexponential decay (Noeske et al 2007a Zahid et al 2012) Theshape and amplitude of the SFR versus M relation at differentredshift z can be characterized by two parameters δ(z) and γ (z)where δ(z) is the logarithm of the SFR at 1010 M and γ (z) isthe power-law index (Zahid et al 2012)

The relationship between the SFR and M then becomes

log10(SFR(z)) = δ(z) + γ (z)[log10(MM) minus 10] (1)

As an example we show in Figure 4 the SFR versus M

relation at three redshifts (z sim 0 08 2) The best-fit values ofδ(z) and γ (z) are listed in Table 3

Using the Kennicutt (1998) relation between SFR and Hα

SFR = 79 times 10minus42L(Hα)(erg sminus1) (2)

and the N2 metallicity calibration (Pettini amp Pagel 2004)

12 + log(OH) = 890 + 057 times log10[N ii]Hα (3)

we can then derive a metallicity detection limit We combineEquations (1) (2) and (3) and assume the [N ii] flux is greater

10


Figure 4 SFR vs stellar mass relation The light blue blue and red lines showthe best-fit SFR vs stellar mass relation from the SDSS DEEP2 and Erb06samples respectively (Zahid et al 2011 see also Table 3) The back dots are thelensed sample used in this work The SFR for the lensed sample is derived fromthe Hα flux with dust extinction corrected from the SED fitting The errors onthe SFR of the lensed sample are statistical errors of the Hα fluxes Systematicerrors of the SFR can be large (a factor of 2ndash3) for our lensed galaxies due tocomplicated aperture effects (Section 31)


than the instrument flux detection limit We provide the detectionlimit for the PP04N2 diagnosed MZ relation

Zmet [log10(finstμ) + 2 log10 DL(z) minus γ (z)

M minus β(z) + log10(4π )]057 + 89(4)

whereβ(z) equiv δ(z) minus γ (z)10 + 42 minus log10 79 (5)

δ(z) γ (z) are defined in Equation (1) finst is the instrument fluxdetection limit in erg sminus1 cmminus2 μ is the lensing magnificationin flux DL(z) is the luminosity distance in cm

The slope of the massndashmetallicity detection limit is related tothe slope of the SFRndashmass relation whereas the y-interceptof the slope depends on the instrument flux limit (andflux magnification for gravitational lensing) redshift and they-intercept of the SFRndashmass relation

Note that the exact location of the boundary depends on theinput parameters of Equation (4) As an example we use the δ(z)and γ (z) values of the Erb06 and lensed samples respectively(Figure 5 Table 3) We show the detection boundary for threecurrent and future NIR instruments SubaruMOIRCS KECKNIRSPEC and JWSTNIRSpec The instrument flux detectionlimit is based on background-limited estimation in 105 s (fluxin units of 10minus18 erg sminus1 cmminus2 below) For SubaruMOIRCS(low-resolution mode HK500) we adopt finst = 230 based onthe 1σ uncertainty of our MOIRCS spectrum (flux = 46 in10 hr) scaled to 3σ in 105 s For KECKNIRSPEC we usefinst = 120 based on the 1σ uncertainty of Erb06 (flux = 30in 15 hr) scaled to 3σ in 105 s For JWSTNIRSpec we usefinst = 017 scaled to 3σ in 105 s9

Since lensing flux magnification is equivalent to loweringthe instrument flux detection limit we see that with a lensingmagnification of sim55 we reach the sensitivity of JWST usingKECKNIRSPEC Stacking can also push the observationsbelow the instrument flux limit For instance the z sim 2 Erb06sample was obtained from stacking the NIRSPEC spectra of

9 httpwwwstsciedujwstinstrumentsnirspecsensitivity

Figure 5 Instrument detection limit on the MZ relation We give the dependence of this detection limit in Equation (4) Shown here are examples of the detection limitbased on given parameters specified as follows The solid lines use the parameters based on the massndashSFR relation of the Erb06 sample δ = 1657 and γ = 048 atz = 226 The dashed lines use the parameters based on the massndashSFR relation of the lensed sample δ = 202 and γ = 069 at z = 207 (see Figure 4 Table 3) Theparameters adopted for the instrument flux limit are given in Section 71 The lensing magnification (μ) are fixed at 10 (ie non-lensing cases) for SubaruMOIRCS(blue lines) and JWSTNIRSpec (light blue) The red lines show the detection limits for KECKNIRSPEC with different magnifications The black filled trianglesshow the Erb et al (2006) sample We show that stacking andor lensing magnification can help to push the observational boundary of the MZ relation to lower massand metallicity regions For example Erb06 used stacked NIRSPEC spectra with N sim 15 spectra in each mass bin The effect of stacking (N sim 15 per bin) is similarto observing with a lensing magnification of μ sim 4


11


Table 4Measured Emission-line Fluxes

ID [O ii] λ3727 Hβ [O iii] λ5007 Hα [N ii] λ6584 KK04(rarrPP04N2) Branch PP04N2 E(B minus V )a Final Adoptedb

B111pa20 2143 plusmn 260 542 plusmn 163 2112 plusmn 243 3354 plusmn 238 lt459 838(816) plusmn 014 Up lt841 073 plusmn 029 848 plusmn 018c

B111pa-20 2202 plusmn 488 940 plusmn 275 1478 plusmn 221 2813 plusmn 256 890 plusmn 089 874(854) plusmn 014 Up 861 plusmn 005 005 plusmn 029B112pa-60 lt605 lt646 lt603 536 plusmn 409 lt1712 middot middot middot middot middot middot lt862 middot middot middotB112pa45 7365 plusmn 1201 lt3423 6106 plusmn 74 8046 plusmn 99 lt1308 lt874(854) Up lt873 0

B21pa20 lt282 lt63 947 plusmn 056 766 plusmn 069 lt067 middot middot middot middot middot middot lt830 middot middot middot lt830B22pa20 lt773 lt206 232 plusmn 30 lt545 lt64 middot middot middot middot middot middot middot middot middot middot middot middot middot middot middotMS1pa20 lt308 lt45 67 plusmn 09 middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middotB293pa20 middot middot middot middot middot middot middot middot middot 1705 plusmn 16 lt31 middot middot middot middot middot middot lt848 middot middot middot lt848

G3pan20 middot middot middot lt40 60 plusmn 07 middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middot middotMS-Jm7pan20 1916 plusmn 274 125 plusmn 36 5812 plusmn 164 2303 plusmn 16 lt922 869(833) plusmn 012 Up lt867 0 825 plusmn 018

823(819) plusmn 012 Low

B53pan20 middot middot middot middot middot middot middot middot middot 907 plusmn 07 lt294 middot middot middot middot middot middot lt862 middot middot middot lt862B51pan20 middot middot middot middot middot middot middot middot middot 3038 plusmn 46 lt1334 middot middot middot middot middot middot lt870 middot middot middotB51pa45 middot middot middot middot middot middot middot middot middot 6439 plusmn 39 lt595 middot middot middot middot middot middot lt888 middot middot middotG2pan20 lt47 684 plusmn 074 3649 plusmn 125 middot middot middot middot middot middot lt862(841) Up middot middot middot middot middot middot lt841G2pan60 lt258 lt88 lt987 1009 plusmn 10 lt31 middot middot middot middot middot middot lt860 middot middot middotLowz136pan60 middot middot middot middot middot middot middot middot middot 5919 plusmn 71 lt882 middot middot middot middot middot middot lt843 middot middot middot lt843

MSnewz3pa45 3694 plusmn 115 4402 plusmn 1106 3003 plusmn 178 middot middot middot middot middot middot 85(829) plusmn 011 Up middot middot middot middot middot middot 823 plusmn 018812(816) plusmn 011 Low middot middot middot middot middot middot

B122pa45 lt7158 lt6779 14101 plusmn 1007 9045 plusmn 695 lt106 middot middot middot middot middot middot lt8369d middot middot middot lt8369

B82pa45 402 plusmn 118 lt177 757 plusmn 66 11526 plusmn 125 lt7213 lt851(829) Up lt878e gt12 lt829lt811(816) Low middot middot middot middot middot middot

B223pa60 1621 plusmn 203 1460 plusmn 292 9423 plusmn 628 7344 plusmn 565 lt365 813(817) plusmn 012 Low lt822 054 plusmn 022 810 plusmn 018

A68-C27pa60 31701 plusmn 179 1492 plusmn 175 8844 plusmn 589 8146 plusmn 218 404 plusmn 1092 826(825) plusmn 006 Low 816 plusmn 007 062 plusmn 011 816 plusmn 018

Notes Observed emission-line fluxes for the lensed background galaxies in A1689 Fluxes are in units of 10minus17 erg sminus1 cmminus2 without lensing magnification correction Some lines are notdetected because of the severe telluric absorptiona E(B minus V ) calculated from Balmer decrement if possibleb Final adopted metallicity converted to PP04N2 base and extinction corrected using E(B minus V ) values from Balmer decrement if available otherwise E(B minus V ) returned from SED fittingare assumedc This galaxy shows significant [N ii] Hα ratios in slit position pa-20 The final metallicity is based on the average spectrum over all slit positionsd Based on NIRSPEC spectrum at KECK II (L J Kewley et al 2013 in preparation)e Possible AGN contamination

87 galaxies with sim15 spectra in each mass bin thus the Erb06sample has been able to probe sim4 times deeper than the nominaldetection boundary of NIRSPEC

The observational detection limit on the MZ relation isimportant for understanding the incompleteness and biases ofsamples due to observational constraints However we cautionthat the relation between Zmet and M in Equation (4) will havesignificant intrinsic dispersion due to variations in the observedproperties of individual galaxies This includes scatter in theMndashSFR relation the N2 metallicity calibration the amountof dust extinction and variable slit losses in spectroscopicobservations For example a scatter of 08 dex in δ for thelensed sample (Table 3) implies a scatter of approximately05 dex in Zmet In addition Equations (2) and (4) includeimplicit assumptions of zero dust extinction and no slit losssuch that the derived line flux is overestimated (and Zmet isunderestimated) Because of the above uncertainties and biasesin the assumptions we made Equation (4) should be used withdue caution

72 The Evolution of the MZ Relation

Figure 6 shows the mass and metallicity measured from theSDSS DEEP2 and our lensed samples The Erb06 stacked dataare also included for comparison We highlight a few interestingfeatures in Figure 7

1 To first order the MZ relation still exists at z sim 2 iemore massive systems are more metal rich The Pearsoncorrelation coefficient is r = 033349 with a probability ofbeing a chance correlation of P = 17 A simple linear fit

to the lensed sample yields a slope of 0164 plusmn 0033 witha y-intercept of 68 plusmn 03

2 All z gt 1 samples show evidence of evolution to lowermetallicities at fixed stellar masses At high stellar mass(M gt 1010 M) the lensed sample has a mean metallicityand a standard deviation of the mean of 841 plusmn 005whereas the mean and standard deviation of the mean forthe Erb06 sample is 852 plusmn 003 The lensed sample is offsetto lower metallicity by 011 plusmn 006 dex compared to theErb06 sample This slight offset may indicate the selectiondifference between the UV-selected (potentially more dustyand metal rich) sample and the lensed sample (less biasedtoward UV bright systems)

3 At lower mass (M lt1094 M) our lensed sample provides12 individual metallicity measurements at z gt 1 The meanmetallicity of the galaxies with M lt1094 M is 825 plusmn005 roughly consistent with the lt820 upper limit of thestacked metallicity of the lowest mass bin (M sim1091 M)of the Erb06 galaxies

4 Compared with the Erb06 galaxies there is a lack of thehighest mass galaxies in our lensed sample We note thatthere is only one object with M gt 10104 M amongall three lensed samples combined The lensed sampleis less affected by the color selection and may be morerepresentative of the mass distribution of high-z galaxies Inthe hierarchical galaxy formation paradigm galaxies growtheir masses with time The number density of massivegalaxies at high redshift is smaller than at z sim 0 thusthe number of massive lensed galaxies is small Selectioncriteria such as the UV-color selection of the Erb06 and

12


Figure 6 Left the observed MZ relation The black symbols are the lensed galaxy sample at z gt 1 Specifically the squares are from this work the stars are fromWuyts et al (2012) and the diamonds are from Richard et al (2011) The orange triangles show the Erb06 sample The local SDSS relation and its 1σ range are drawnin purple lines The z sim 08 DEEP2 relations from Zahid et al (2011) are drawn in purple dots Right the best fit to the MZ relation A second-degree polynomialfunction is fit to the SDSS DEEP2 and Erb06 samples A simple linear function is fit to the lensed sample The z gt 1 lensed data are binned in five mass bins (symbolred star) and the median and 1σ standard deviation of each bin are plotted on top of the linear fit


Table 5Physical Properties of the Lensed Sample

ID1 ID2a RA Decl (J2000) Redshift lg(SFR)b Lensing Magnification log(MlowastM)(M yrminus1) (flux)

B111pa20 888_351 131133336 minus01210694 2540 plusmn 0006 108 plusmn 01 118 plusmn 27 91+02minus03

B112pa45 middot middot middot 131129053 minus01200126 2540 plusmn 0006 142 plusmn 011 131 plusmn 18


B22pa20 middot middot middot 131132961 minus01202531 2537 plusmn 0006 middot middot middot 150 plusmn 20

MS1pa20 869_328 131128684 minus01194262 2534 plusmn 001 middot middot middot 583 plusmn 28 85+01minus01


c

G3pan20 middot middot middot 131126219 minus01210964 2540 plusmn 001 middot middot middot 77 plusmn 01 middot middot middotMS-Jm7pan20 865_359 131127600 minus01213500 2588 plusmn 0006 middot middot middot 185 plusmn 32 80+05

minus04c

B53pan20 892_339 131134109 minus01202090 2636 plusmn 0004 047 plusmn 005 142 plusmn 13 91+04minus02

B51pan60 870_346 131129064 minus01204833 2641 plusmn 0004 10 plusmn 005 143 plusmn 03

G2 894_332 131134730 minus01195553 1643 plusmn 001 045 plusmn 009 167 plusmn 31 80+03minus04

Lowz136 891_321 131133957 minus01191590 1363 plusmn 001 067 plusmn 011 116 plusmn 27 89+03minus03

MSnewz3pa45 middot middot middot 131124276 minus01195208 3007 plusmn 0003 065 plusmn 055 29 plusmn 17 86+03minus04

c

B122pa45 863_348 131127212 minus01205189 1834 plusmn 0006 100 plusmn 005d 560 plusmn 44 74+02minus00

B82pa45 middot middot middot 131127212 minus01205189 2662 plusmn 0006 136 plusmn 007e 237 plusmn 30 82+05minus06

c

B223pa60f middot middot middot 1311324150 minus012115917 1703 plusmn 0006 188 plusmn 004 155 plusmn 03 85+02minus02

A68-C27pa60 middot middot middot 003704866+09102926 1762 plusmn 0006 246 plusmn 01 49 plusmn 11 96+01minus01

Notes The redshift errors in Table 5 is determined from rms of different emission-line centroids If the rms is smaller than 0006 (for most targets) or if there is onlyone line fitted we adopt the systematic error of 0006 as a conservative estimation for absolute redshift measurementsa ID used in J Richard et al (2013 in preparation) The name tags of the objects are chosen to be consistent with the Broadhurst et al (2005) conventions if overlappingb Corrected for lensing magnification but without dust extinction correction We note that the systematic errors of SFR in this work are extremely uncertain due tocomplicated aperture correction and flux calibration in the multi-slit of MOIRCSc The IRAC photometry for these sources are not included in the stellar mass calculation due to the difficulty in resolving the lensed image from the adjacent foregroundgalaxiesd Based on NIRSPEC observatione Possible AGN contaminationf See also Yuan amp Kewley (2009)

13


Figure 7 Model predictions of the MZ relation The data symbols are the same as those used in Figure 6 The small green and light blue dots are the cosmologicalhydrodynamic simulations with momentum-conserving wind models from Dave et al (2011a) The difference between the left and right panels are the differentnormalization methods used The left panel normalizes the model metallicity to the observed SDSS values by applying a constant offset at Mstar sim 1010 M whereasthe right panel normalizes the model metallicity to the observed SDSS metallicity by allowing a constant shift in the slope amplitude and stellar mass Note that themodel has a mass cutoff at 11 times 109 M


Figure 8 ldquoMean evolved metallicityrdquo as a function of redshift for two mass bins (indicated by four colors) The dashed lines show the median and 1σ scatter of themodel prediction from Dave et al (2011a) The observed data from DEEP2 and our lensed sample are plotted as filled circles


SINs (Genzel et al 2011) galaxies can be applied to targetthe high-mass galaxies on the MZ relation at high redshift

73 Comparison with Theoretical MZ Relations

Understanding the origins of the MZ relation has beenthe driver of copious theoretical work Based on the ideathat metallicities are mainly driven by an equilibrium amongstellar enrichment infall and outflow Finlator amp Dave (2008)developed smoothed particle hydrodynamic simulations Theyfound that the inclusion of a momentum-driven wind model(vzw) fits best to the z sim 2 MZ relations compared to otheroutflowwind models The updated version of their vzw modelis described in detail in Dave et al (2011a) We overlay the

Dave et al (2011a) vzw model outputs on the MZ relation inFigure 7 We find that the model does not reproduce the MZredshift evolution seen in our observations We provide possibleexplanations as follows

KE08 found that both the shape and scatter of the MZ relationvary significantly among different metallicity diagnostics Thisposes a tricky normalization problem when comparing modelsto observations For example a model output may fit the MZrelation slope from one strong-line diagnostic but fail to fit theMZ relation from another diagnostic which may have a verydifferent slope This is exactly what we are seeing on the leftpanel of Figure 7 Dave et al (2011a) applied a constant offsetof the model metallicities by matching the amplitude of themodel MZ relation at z sim 0 with the observed local MZ relation

14


Figure 9 z = 2540 B111 MOIRCS J H band spectra Detailed descriptions are given in the Appendix Note that the dashed box indicates the sim01 arcsec alignmenterror of MOIRCS


of Tremonti et al (2004 hereafter T04) at the stellar mass of1010 M Dave et al (2011a) found that the characteristic shapeand scatter of the MZ relation from the vzw model matches theT04 MZ relation between 1090 M lt M lt10110 within the 1σmodel and observational scatter However since both the slopeand amplitude of the T04 SDSS MZ relation are significantlylarger than the SDSS MZ relation derived using the PP04N2

method (KE08) the PP04N2-normalized MZ relation from themodel does not recover the local MZ relation within 1σ

In addition the stellar mass measurements from differentmethods may cause a systematic offsets in the x-direction of theMZ relation (Zahid et al 2011) As a result even though theshape scatter and evolution with redshifts are independent pre-dictions from the model systematic uncertainties in metallicity

15


Figure 10 z = 2540 B112 MOIRCS J H band spectra Detailed descriptions are given in the Appendix


diagnostics and stellar mass estimates do not allow the shape tobe constrained separately

In the right panel of Figure 7 we allow the model slope(α) metallicity amplitude (Z) and stellar mass (Mlowast) to changeslightly so that it fits the local SDSS MZ relation Assumingthat this change in slope (Δα) and x y amplitudes (ΔZ ΔMlowast)are caused by the systematic offsets in observations then thesame Δα ΔZ and ΔMlowast can be applied to model MZ relationsat other redshifts Although normalizing the model MZ relation

in this way will make the model lose prediction power for theshape of the MZ relation it at least leaves the redshift evolutionof the MZ relation as a testable model output

Despite the normalization correction we see from Figure 7that the models predict less evolution from z sim 2 to z sim 0 thanthe observed MZ relation To quantify we divide the model datainto two mass bins and derive the mean and 1σ scatter in eachmass bin as a function of redshift We define the ldquomean evolvedmetallicityrdquo on the MZ relation as the difference between the

16




mean metallicity at redshift z and the mean metallicity at z sim 0at a fixed stellar mass (log (OH) [z sim 0] minus log (OH) [z sim 2])The ldquomean evolved metallicityrdquo errors are calculated based onthe standard errors of the mean

In Figure 8 we plot the ldquomean evolved metallicityrdquo as a func-tion of redshift for two mass bins 1090 M lt M lt1095 M1095 M lt M lt1011 M We calculate the observed ldquomeanevolved metallicityrdquo for DEEP2 and our lensed sample in thesame mass bins We see that the observed mean evolution of thelensed sample are largely uncertain and no conclusion betweenthe model and observational data can be drawn However theDEEP2 data are well constrained and can be compared with themodel

We find that at z sim 08 the mean evolved metallicity of thehigh-mass galaxies are consistent with the mean evolved metal-licity of the models The observed mean evolved metallicity ofthe low-mass bin galaxies is sim012 dex larger than the meanevolved metallicity of the models in the same mass bins

8 COMPARE WITH PREVIOUS WORK IN LITERATURE

In this section we compare our findings with previous workon the evolution of the MZ relation

For low masses (109 M) we find a larger enrichment (iesmaller decrease in metallicity) between z sim 2 rarr 0 than

either the non-lensed sample of Maiolino et al (2008 015 dexcf 06 dex) or the lensed sample of Wuyts et al (2012) andRichard et al (2011 04 dex) These discrepancies may reflectdifferences in metallicity calibrations applied It is clear thata larger sample is required to characterize the true mean andspread in metallicities at intermediate redshift Note that thelensed samples are still small and have large measurement errorsin both stellar masses (01ndash05 dex) and metallicity (sim02 dex)

For high masses (1010 M) we find similar enrichment(04 dex) between z sim 2 rarr 0 compare to the non-lensed sampleof Maiolino et al (2008) and the lensed sample of Wuyts et al(2012) and Richard et al (2011)

We find in Section 61 that the deceleration in metal en-richment is significant in the highest mass bin (1095 M ltM lt1011 M) of our samples The deceleration in metal en-richment from z sim 2 rarr 08 to z sim 08 rarr 0 is consistentwith the picture that the star formation and mass assembly peakbetween redshift 1 and 3 (Hopkins amp Beacom 2006) The de-celeration is larger by 0019 plusmn 0013 dex Gyrminus2 in the high-mass bin suggesting a possible mass-dependence in chemi-cal enrichment similar to the ldquodownsizingrdquo mass-dependentgrowth of stellar mass (Cowie et al 1996 Bundy et al 2006)In the downsizing picture more massive galaxies formed theirstars earlier and on shorter timescales compared with less mas-sive galaxies (Noeske et al 2007a) Our observation of the

17




Figure 13 z = 254 MS1 MOIRCS J H band spectra Detailed descriptions are given in the Appendix


18




Figure 15 z = 2540 G3 MOIRCS J H band spectra Detailed descriptions are given in the Appendix


chemical downsizing is consistent with previous metallicity evo-lution work (Panter et al 2008 Maiolino et al 2008 Richardet al 2011 Wuyts et al 2012)

We find that for higher mass bins the model of Daveet al (2011a) overpredicts the metallicity at all redshifts Theoverprediction is most significant in the highest mass bin of1010ndash1011 M This conclusion similar to the findings in Daveet al (2011a 2011b) In addition we point out that whencomparing the model metallicity with the observed metallicitythere is a normalization problem stemming from the discrepancyamong different metallicity calibrations (Section 73)

We note the evolution of the MZ relation is based on anensemble of the averaged SFR-weighted metallicity of the star-forming galaxies at each epoch The MZ relation does not reflectan evolutionary track of individual galaxies We are probablyseeing a different population of galaxies at each redshift (Brookset al 2007 Conroy et al 2008) For example a sim10105 M

massive galaxy at z sim 2 will most likely evolve into anelliptical galaxy in the local universe and will not appear onthe local MZ relation On the other hand to trace the progenitorof a sim1011 M massive galaxy today we need to observe asim1095 M galaxy at z sim2 (Zahid et al 2012)

It is clear that gravitational lensing has the power to probelower stellar masses than current color selection techniquesLarger lensed samples with high-quality observations are re-quired to reduce the measurement errors

9 SUMMARY

To study the evolution of the overall metallicity and MZrelation as a function of redshift it is critical to remove thesystematics among different redshift samples The major caveatsin current MZ relation studies at z gt 1 are (1) metallicityis not based on the same diagnostic method (2) stellar mass

19


Figure 16 z = 2588 Jm7 MOIRCS J H band spectra Detailed descriptions are given in the Appendix


is not derived using the same method (3) the samples areselected differently and selection effects on mass and metallicityare poorly understood In this paper we attempt to minimizethese issues by re-calculating the stellar mass and metallicityconsistently and by expanding the lens-selected sample atz gt 1 We aim to present a reliable observational picture ofthe metallicity evolution of star-forming galaxies as a functionof stellar mass between 0 lt z lt 3 We find that

1 There is a clear evolution in the mean and median metallic-ities of star-forming galaxies as a function of redshift Themean metallicity falls by sim018 dex from redshift 0 to 1and falls further by sim016 dex from redshift 1 to 2

2 A more rapid evolution is seen between z sim 1 rarr 3than z sim 0 rarr 1 for the high-mass galaxies (1095 M ltM lt1011 M) with almost twice as much enrichmentbetween z sim 1 rarr 3 than between z sim 1 rarr 0

3 The deceleration in metal enrichment from z sim 2 rarr 08to z sim 08 rarr 0 is significant in the high-mass galaxies(1095 M lt M lt1011 M) consistent with a mass-dependent chemical enrichment

4 We compare the metallicity evolution of star-forming galax-ies from z = 0 rarr 3 with the most recent cosmological hy-drodynamic simulations We see that the model metallicityis consistent with the observed metallicity within the obser-

vational error for the low-mass bins However for highermass bins the model overpredicts the metallicity at all red-shifts The overprediction is most significant in the highestmass bin of 1010ndash1011 M Further theoretical investigationinto the metallicity of the highest mass galaxies is requiredto determine the cause of this discrepancy

5 The median metallicity of the lensed sample is 035 plusmn006 dex lower than local SDSS galaxies and 028 plusmn006 dex lower than the z sim 08 DEEP2 galaxies

6 Cosmological hydrodynamic simulation (Dave et al 2011a)does not agree with the evolutions of the observed MZrelation based on the PP04N2 diagnostic Whether themodel fits the slope of the MZ relation depends on thenormalization methods used

This study is based on six clear nights of observations on an8 m telescope highlighting the efficiency in using lens-selectedtargets However the lensed sample at z gt 1 is still small Weaim to significantly increase the sample size over the years

We thank the referee for an excellent report that significantlyimproved this paper T-TY thanks the MOIRCS supporting as-tronomer Ichi Tanaka and Kentaro Aoki for their enormous sup-port on the MOIRCS observations We thank Youichi Ohyamafor scripting the original MOIRCS data reduction pipeline

20


Figure 17 z = 2641 B51 MOIRCS J H band spectra Note that the reason that the flux of B51 in slit position PAn60 is less than PA45 (B51+B52) is that thedithering length of PAn60 was smaller than the separation of 51 and 52 thus part of the flux of PA45 (B51+B52) was canceled out during the dithering processDetailed descriptions are given in the Appendix




21




We are grateful to Dave Rommel for providing and explain-ing to us his most recent models T-Y thanks Jabran Zahidfor the SDSS and DEEP2 data and many insightful discus-sions T-Y acknowledges a Soroptimist Founder Region Fel-lowship for Women LK acknowledges an NSF Early CA-REER Award AST 0748559 and an ARC Future Fellowshipaward FT110101052 JR is supported by the Marie Curie Ca-reer Integration Grant 294074 We wish to recognize and ac-knowledge the very significant cultural role and reverence thatthe summit of Mauna Kea has always had within the indigenousHawaiian community

Facility Subaru (MOIRCS)

APPENDIX

SLIT LAYOUT AND SPECTRAFOR THE LENSED SAMPLE

This section presents the slit layouts reduced and fittedspectra for the newly observed lensed objects in this work Theline-fitting procedure is described in Section 32 For each targetthe top panel shows the HST ACS 475W broadband image of thelensed target The slit layouts with different positional anglesare drawn in white boxes The bottom panel(s) show(s) the final

reduced 1D spectrum(a) zoomed in for emission-line vicinitiesThe black line is the observed spectrum for the target The cyanline is the noise spectrum extracted from object-free pixels ofthe final two-dimensional spectrum The tilted gray mesh linesindicate spectral ranges where the sky absorption is severeEmission lines falling in these spectral windows suffer fromlarge uncertainties in telluric absorption correction The bluehorizontal line is the continuum fit using first-order polynomialfunction after blanking out the severe sky absorption regionThe red lines overplotted on the emission lines are the overallGaussian fit with the blue lines showing individual componentsof the multiple Gaussian functions The vertical dashed linesshow the center of the Gaussian profile for each emission lineThe SN of each line are marked under the emission-line labelsNote that for lines with SN lt 3 the fit is rejected and a 3σupper limit is derived

Brief remarks on individual objects (see also Tables 4 and 5for more information) are as follows

1 Figures 9 and 10 B11 (888_351) This is a resolvedgalaxy with spiral-like structure at z = 2540 plusmn 0006As reported in Broadhurst et al (2005) it is likely to bethe most distant known spiral galaxy so far B11 has threemultiple images We have observed B111 and B112 with

22


Figure 20 z = 1834 plusmn 0002 B122 MOIRCS J H band spectra Detailed descriptions are given in the Appendix


Figure 21 z = 1363 low-z MOIRCS J H band spectra Detailed descriptions are given in the Appendix


23


Figure 22 z = 3003 new target MOIRCS J H band spectra Detailed descriptions are given in the Appendix




24


Figure 24 z = 1763 A68C27 MOIRCS J H band spectra Detailed descriptions are given in the Appendix


two slit orientations on each image respectively Differentslit orientation yields very different line ratios implyingpossible gradients Our IFU follow-up observations are inprogress to reveal the details of this 26 Gyr old spiral

2 Figures 11 and 12 B2 (860_331) This is one of theinteresting systems reported in Frye et al (2007) It has fivemultiple images and is only 2primeprime away from another five-image lensed system ldquoThe Sextet Arcsrdquo at z = 3038 Wehave observed B21 and B22 and detected strong Hα and[O iii] lines in both of them yielding a redshift of 2537 plusmn0006 consistent with the redshift z = 2534 measuredfrom the absorption lines ([C ii] λ1334 [Si ii] λ1527) inFrye et al (2007)

3 Figure 13 MS1 (869_328) We have detected a 7σ [O iii]line and determined its redshift to be z = 2534 plusmn 0010

4 Figure 14 B29 (884_331) This is a lensed system with fivemultiple images We observed B293 the brightest of thefive images The overall surface brightness of the B293 arcis very low We have observed a 10σ Hα and an upper limitfor [N ii] placing it at z = 2633 plusmn 0010

5 Figure 15 G3 This lensed arc with a bright knot hasno recorded redshift before this study It was put onone of the extra slits during mask designing We havedetected an 8σ [O iii] line and determined its redshift tobe z = 2540 plusmn 0010

6 Figure 16 Ms-Jm7 (865_359) We detected [O ii] Hβ [O iii]Hα and an upper limit for [N ii] placing it at redshiftz = 2588 plusmn 0006

7 Figures 17 and 18 B5 (892_339 870_346) It has threemultiple images of which we observed B51 and B53Two slit orientations were observed for B51 the finalspectrum for B51 has combined the two slit orientationsweighted by the SN of Hα Strong Hα and upper limit of[N ii] were obtained in both images yielding a redshift ofz = 2636 plusmn 0004

8 Figure 19 G2 (894_332) Two slit orientations were avail-able for G2 with detections of Hβ [O iii] Hα and up-per limits for [O ii] and [N ii] The redshift measured isz = 1643 plusmn 0010

9 Figure 20 B12 This blue giant arc has five multiple imagesand we observed B122 It shows a series of strong emissionlines with an average redshift of z = 1834 plusmn 0006

10 Figure 21 Lensz136 (891_321) It has a very strongHα and [N ii] is at noise level from Hα we derivez = 1363 plusmn 0010

11 Figure 22 MSnewz3 This is a new target observed inA1689 we detect [O ii] Hβ and [O iii] at a significantlevel yielding z = 3007 plusmn 0003

12 Figure 23 B8 This arc has five multiple images in totaland we observed B82 detection of [O ii] [O iii] Hα with

25


Hβ and [N ii] as upper limit yields an average redshift ofz = 2662 plusmn 0006

13 B223 A three-image lensed system at z = 1703 plusmn 0004this is the first object reported from our LEGMS programsee Yuan amp Kewley (2009)

14 Figure 24 A68-C27 This is the only object chosen fromour unfinished observations on A68 This target has manystrong emission lines z = 1762 plusmn 0006 The morphologyof C27 shows signs of merger IFU observation on thistarget is in process

REFERENCES

Abazajian K N Adelman-McCarthy J K Agueros M A et al 2009 ApJS182 543

Asplund M Grevesse N Sauval A J amp Scott P 2009 ARAampA 47 481Baldwin J A Phillips M M amp Terlevich R 1981 PASP 93 5Bertin E amp Arnouts S 1996 AampAS 117 393Bertone S De Lucia G amp Thomas P A 2007 MNRAS 379 1143Brinchmann J Pettini M amp Charlot S 2008 MNRAS 385 769Broadhurst T Bentez N Coe D et al 2005 ApJ 621 53Brooks A M Governato F Booth C M et al 2007 ApJL 655 17Bruzual G amp Charlot S 2003 MNRAS 344 1000Bundy K Ellis R S Conselice C J et al 2006 ApJ 651 120Calzetti D Armus L Bohlin R C et al 2000 ApJ 533 682Capak P Cowie L L Hu E M et al 2004 AJ 127 180Chabrier G 2003 PASP 115 763Chapman S C Blain A W Smail I amp Ivison R J 2005 ApJ 622 772Christensen L Laursen P Richard J et al 2012 arXiv12090775Conroy C Shapley A E Tinker J L Santos M R amp Lemson G 2008 ApJ

679 1192Conselice C J Bundy K Trujillo I et al 2007 MNRAS 381 962Cowie L L amp Barger A J 2008 ApJ 686 72Cowie L L Songaila A Hu E M amp Cohen J G 1996 AJ 112 839Daddi E Cimatti A Renzini A et al 2004 ApJ 617 746Dalcanton J J 2007 ApJ 658 941Dave R Finlator K amp Oppenheimer B D 2011a MNRAS 416 1354Dave R amp Oppenheimer B D 2007 MNRAS 374 427Dave R Oppenheimer B D amp Finlator K 2011b MNRAS 415 11Davies R I 2007 MNRAS 375 1099Davis M Faber S M Newman J et al 2003 Proc SPIE 4834 161De Lucia G Kauffmann G amp White S D M 2004 MNRAS 349 1101Dickinson M Papovich C Ferguson H C amp Budavari T 2003 ApJ

587 25Edmunds M G 1990 MNRAS 246 678Edmunds M G amp Greenhow R M 1995 MNRAS 272 241Elbaz D Dickinson M Hwang H S et al 2011 AampA 533 A119Erb D K Pettini M Shapley A E et al 2010 ApJ 719 1168Erb D K Shapley A E Pettini M et al 2006 ApJ 644 813Erb D K Shapley A E Steidel C C et al 2003 ApJ 591 101Fan X Strauss M A Schneider D P et al 2001 AJ 121 54Finlator K amp Dave R 2008 MNRAS 385 2181Forster Schreiber N M Genzel R Lehnert M D et al 2006 ApJ 645 1062Frye B L Coe D Bowen D V et al 2007 ApJ 665 921Garnett D R 2002 ApJ 581 1019Genzel R Burkert A Bouche N et al 2008 ApJ 687 59Genzel R Newman S Jones T et al 2011 ApJ 733 101Grazian A Salimbeni S Pentericci L et al 2007 AampA 465 393Hayashi M Motohara K Shimasaku K et al 2009 ApJ 691 140Henry J P Salvato M Finoguenov A et al 2010 ApJ 725 615Hopkins A M amp Beacom J F 2006 ApJ 651 142Ichikawa T Suzuki R Tokoku C et al 2006 Proc SPIE 6269 626916Ilbert O Salvato M Le Flocrsquoh E et al 2010 ApJ 709 644Jones T Ellis R Jullo E amp Richard J 2010 ApJL 725 176Jones T Ellis R S Richard J amp Jullo E 2012 arXiv12074489Jullo E Kneib J-P Limousin M et al 2007 NJPh 9 447Kelson D D 2003 PASP 115 688

Kennicutt R C Jr 1998 ARAampA 36 189Kewley L J amp Dopita M A 2002 ApJS 142 35Kewley L J amp Ellison S L 2008 ApJ 681 1183Kewley L J Geller M J amp Jansen R A 2004 AJ 127 2002Kewley L J Groves B Kauffmann G amp Heckman T 2006 MNRAS

372 961Kneib J P Mellier Y Fort B amp Mathez G 1993 AampA 273 367Kobulnicky H A amp Kewley L J 2004 ApJ 617 240Koppen J amp Edmunds M G 1999 MNRAS 306 317Kriek M van Dokkum P G Franx M et al 2007 ApJ 669 776Lacey C G amp Fall S M 1985 ApJ 290 154Lamareille F Brinchmann J Contini T et al 2009 AampA 495 53Larkin J Barczys M Krabbe A et al 2006 NewAR 50 362Law D R Steidel C C Erb D K et al 2009 ApJ 697 2057Lemoine-Busserolle M Contini T Pello R et al 2003 AampA 397 839Lequeux J Peimbert M Rayo J F Serrano A amp Torres-Peimbert S 1979

AampA 80 155Limousin M Richard J Jullo E et al 2007 ApJ 668 643Liu X Shapley A E Coil A L Brinchmann J amp Ma C-P 2008 ApJ

678 758Maiolino R Nagao T Grazian A et al 2008 AampA 488 463McLean I S Becklin E E Bendiksen O et al 1998 Proc SPIE 3354 566Nagamine K Fukugita M Cen R amp Ostriker J P 2001 ApJ 558 497Noeske K G Faber S M Weiner B J et al 2007a ApJL 660 47Noeske K G Weiner B J Faber S M et al 2007b ApJL 660 43Oke J B Cohen J G Carr M et al 1995 PASP 107 375Oppenheimer B D amp Dave R 2008 MNRAS 387 577Osterbrock D E 1989 Astrophysics of Gaseous Nebulae and Active Galactic

Nuclei (Research Supported by the University of California John SimonGuggenheim Memorial Foundation University of Minnesota et al MillValley CA Univ Science Books) 422

Pagel B E J amp Edmunds M G 1981 ARAampA 19 77Pagel B E J amp Patchett B E 1975 MNRAS 172 13Panter B Jimenez R Heavens A F amp Charlot S 2008 MNRAS 391 1117Pettini M amp Pagel B E J 2004 MNRAS 348 L59Pettini M Shapley A E Steidel C C et al 2001 ApJ 554 981Quider A M Pettini M Shapley A E amp Steidel C C 2009 MNRAS 398

1263Rafelski M Wolfe A M Prochaska J X Neeleman M amp Mendez A J

2012 ApJ 755 89Reddy N A Steidel C C Erb D K Shapley A E amp Pettini M 2006 ApJ

653 1004Reddy N A Steidel C C Pettini M et al 2008 ApJS 175 48Richard J Jones T Ellis R et al 2011 MNRAS 126Richard J Kneib J-P Jullo E et al 2007 ApJ 662 781Rubin V C Ford W K Jr amp Whitmore B C 1984 ApJL 281 21Savaglio S Glazebrook K Le Borgne D et al 2005 ApJ 635 260Searle L amp Sargent W L W 1972 ApJ 173 25Shapley A E Coil A L Ma C-P amp Bundy K 2005 ApJ 635 1006Skillman E D Kennicutt R C amp Hodge P W 1989 ApJ 347 875Sobral D Best P N Geach J E et al 2009 MNRAS 398 75Songaila A amp Cowie L L 2002 AJ 123 2183Stark D P Swinbank A M Ellis R S et al 2008 Natur 455 775Steidel C C Adelberger K L Shapley A E et al 2003 ApJ 592 728Steidel C C Giavalisco M Pettini M Dickinson M amp Adelberger K L

1996 ApJL 462 17Steidel C C Shapley A E Pettini M et al 2004 ApJ 604 534Swinbank A M Webb T M Richard J et al 2009 MNRAS 400 1121Tremonti C A Heckman T M Kauffmann G et al 2004 ApJ 613 898Veilleux S amp Osterbrock D E 1987 ApJS 63 295Wuyts E Rigby J R Sharon K amp Gladders M D 2012 ApJ 755 73Wuyts S Forster Schreiber N M van der Wel A et al 2011 ApJ 742 96Yabe K Ohta K Iwamuro F et al 2012 PASJ 64 60Yates R M Kauffmann G amp Guo Q 2012 MNRAS 422 215Yuan T-T amp Kewley L J 2009 ApJL 699 161Yuan T-T Kewley L J Swinbank A M Richard J amp Livermore R C

2011 ApJL 732 14Zahid H J Dima G I Kewley L J Erb D K amp Dave R 2012 ApJ 757

54Zahid H J Kewley L J amp Bresolin F 2011 ApJ 730 137Zaritsky D Kennicutt R C Jr amp Huchra J P 1994 ApJ 420 87

26

1 INTRODUCTION


21 The Lensed Emission-line Galaxy Metallicity Survey (LEGMS)



32 Line Fitting


34 Photometry




52 Stellar Masses


6 THE COSMIC EVOLUTION OF METALLICITY FOR STAR-FORMING GALAXIES

61 The Zz Relation


7 EVOLUTION OF THE MASSndashMETALLICITY RELATION





9 SUMMARY

APPENDIX SLIT LAYOUT AND SPECTRA FOR THE LENSED SAMPLE

REFERENCES


were produced (Fan et al 2001 Dickinson et al 2003 Chapmanet al 2005 Hopkins amp Beacom 2006 Grazian et al 2007Conselice et al 2007 Reddy et al 2008) It is therefore ofcrucial importance to explore NIR spectra for galaxies in thisredshift range

Many spectroscopic redshift surveys have been carried outin recent years to study star-forming galaxies at z gt1 (egSteidel et al 2004 Law et al 2009) However due to thelow efficiency in the NIR those spectroscopic surveys almostinevitably have to rely on color-selection criteria and the biasesin UV-selected galaxies tend to select the most massive andless dusty systems (eg Capak et al 2004 Steidel et al 2004Reddy et al 2006) Space telescopes can observe much deeperin the NIR and are able to probe a wider mass range Forexample the narrowband Hα surveys based on the new WFC3camera onboard the Hubble Space Telescope (HST) have locatedhundreds of Hα emitters up to z = 223 finding much faintersystems than observed from the ground (Sobral et al 2009)However the low-resolution spectra from the narrow band filtersforbid derivations of physical properties such as metallicitiesthat can only currently be acquired from ground-based spectralanalysis

Due to the advent of long-slitmulti-slit NIR spectrographson 8ndash10 m class telescopes enormous progress has been madein the last decade to capture galaxies in the redshift desert Forchemical abundance studies a full coverage of rest-frame opticalspectra (4000ndash9000 Aring) is usually mandatory for the most robustdiagnostic analysis For 15 z 3 the rest-frame opticalspectra have shifted into the J H and K bands It remainschallenging and observationally expensive to obtain high signal-to-noise (SN) NIR spectra from the ground especially forldquotypicalrdquo targets at high-z that are less massive than conventionalcolor-selected galaxies Therefore previous investigations intothe metallicity properties between 1 z 3 focused on stackedspectra samples of massive luminous individual galaxies orvery small numbers of lower-mass galaxies (eg Erb et al2006 2010 Forster Schreiber et al 2006 Law et al 2009 Yabeet al 2012)

The first MZ relation for galaxies at z sim 2 was found by Erbet al (2006 hereafter Erb06) using the stacked spectra of 87 UVselected galaxies divided into six mass bins Subsequently massand metallicity measurements have been reported for numerousindividual galaxies at 15 lt z lt 3 (Forster Schreiber et al 2006Genzel et al 2008 Hayashi et al 2009 Law et al 2009 Erbet al 2010) These galaxies are selected using broadband colorsin the UV (Lyman break technique Steidel et al 1996 2003)or using B- z- and K-band colors (BzK selection Daddi et al2004) The Lyman break and BzK selection techniques favorgalaxies that are luminous in the UV or blue and may thereforebe biased against low-luminosity (low-metallicity) galaxies anddusty (potentially metal-rich) galaxies Because of these biasesgalaxies selected in this way may not sample the full range inmetallicity at redshift z gt 1

A powerful alternative method to avoid these selection effectsis to use strong gravitationally lensed galaxies In the case ofgalaxy cluster lensing the total luminosity and area of thebackground sources can easily be boosted by sim10ndash50 timesproviding invaluable opportunities to obtain high SN spectraand probe intrinsically fainter systems within a reasonableamount of telescope time In some cases sufficient SN can evenbe obtained for spatially resolved pixels to study the resolvedmetallicity of high-z galaxies (Swinbank et al 2009 Jones et al2010 2012 Yuan et al 2011) Before 2011 metallicities have

been reported for a handful of individually lensed galaxies usingoptical emission lines at 15 lt z lt 3 (Pettini et al 2001Lemoine-Busserolle et al 2003 Stark et al 2008 Quider et al2009 Yuan amp Kewley 2009 Jones et al 2010) Fortunatelylensed galaxy samples with metallicity measurements haveincreased significantly due to reliable lensing mass modelingand larger dedicated spectroscopic surveys of lensed galaxieson 8ndash10 m telescopes (Richard et al 2011 Wuyts et al 2012Christensen et al 2012)

In 2008 we began a spectroscopic observational surveydesigned specifically to capture metallicity sensitive lines forlensed galaxies Taking advantage of the multi-object cryogenicNIR spectrograph (MOIRCS) on Subaru we targeted well-known strong lensing galaxy clusters to obtain metallicities forgalaxies between 08 lt z lt 3 In this paper we present the firstmetallicity measurement results from our survey

Combining our new data with existing data from the literaturewe present a coherent observational picture of the metallicityhistory and massndashmetallicity evolution of star-forming galaxiesfrom z sim 0 to z sim 3 Kewley amp Ellison (2008) have shownthat the metallicity offsets in the diagnostic methods can easilyexceed the intrinsic trends It is of paramount importanceto make sure that relative metallicities are compared on thesame metallicity calibration scale In MZ relation studies themethods used to derive the stellar mass can also cause systematicoffsets (Zahid et al 2011) Different spectral energy distribution(SED) fitting codes can yield a non-negligible mass offsethence mimicking or hiding evolution in the MZ relation Inthis paper we derive the mass and metallicity of all samplesusing the same methods ensuring that the observational data arecompared in a self-consistent way We compare our observedmetallicity history with the latest prediction from cosmologicalhydrodynamical simulations

Throughout this paper we use a standard ΛCDM cosmologywith H0 = 70 km sminus1 Mpcminus1 ΩM = 030 and ΩΛ = 070 Weuse solar oxygen abundance 12 + log(OH) = 869 (Asplundet al 2009)

The paper is organized as follows Section 2 describes ourlensed sample survey and observations Data reduction andanalysis are summarized in Section 3 Section 4 presents anoverview of all the samples we use in this study Section 5describes the methodology of derived quantities The metallicityevolution of star-forming galaxies with redshift is presentedin Section 6 Section 7 presents the massndashmetallicity relationfor our lensed galaxies Section 8 compares our results withprevious work in literature Section 9 summarizes our resultsIn the Appendix we show the morphology slit layout andreduced one-dimensional (1D) spectra for the lensed galaxiesreported in our survey


21 The Lensed Emission-line GalaxyMetallicity Survey (LEGMS)

Our survey (LEGMS) aims to obtain oxygen abundance oflensed galaxies at 08 ltz lt 3 LEGMS has taken enormousadvantage of the state-of-the-art instruments on Mauna KeaFour instruments have been utilized so far (1) the Multi-ObjectInfraRed Camera and Spectrograph (MOIRCS Ichikawa et al2006) on Subaru (2) the OH-Suppressing Infra-Red ImagingSpectrograph (OSIRIS Larkin et al 2006) on Keck II (3) theNear Infrared Spectrograph (NIRSPEC McLean et al 1998)on Keck II (4) the Low Dispersion Imaging Spectrograph

2

















3





32 Line Fitting





34 Photometry









4






Median


















52 Stellar Masses



5










61 The Zz Relation






6











7





8




9
























10






M minus β(z) + log10(4π )]057 + 89(4)









11








823(819) plusmn 012 Low

















12












c


minus04c






c



c




13











14







15








16













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES

















3





32 Line Fitting





34 Photometry









4






Median


















52 Stellar Masses



5










61 The Zz Relation






6











7





8




9
























10






M minus β(z) + log10(4π )]057 + 89(4)









11








823(819) plusmn 012 Low

















12












c


minus04c






c



c




13











14







15








16













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES





32 Line Fitting





34 Photometry









4






Median


















52 Stellar Masses



5










61 The Zz Relation






6











7





8




9
























10






M minus β(z) + log10(4π )]057 + 89(4)









11








823(819) plusmn 012 Low

















12












c


minus04c






c



c




13











14







15








16













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES






Median


















52 Stellar Masses



5










61 The Zz Relation






6











7





8




9
























10






M minus β(z) + log10(4π )]057 + 89(4)









11








823(819) plusmn 012 Low

















12












c


minus04c






c



c




13











14







15








16













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES










61 The Zz Relation






6











7





8




9
























10






M minus β(z) + log10(4π )]057 + 89(4)









11








823(819) plusmn 012 Low

















12












c


minus04c






c



c




13











14







15








16













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES











7





8




9
























10






M minus β(z) + log10(4π )]057 + 89(4)









11








823(819) plusmn 012 Low

















12












c


minus04c






c



c




13











14







15








16













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES





8




9
























10






M minus β(z) + log10(4π )]057 + 89(4)









11








823(819) plusmn 012 Low

















12












c


minus04c






c



c




13











14







15








16













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES




9
























10






M minus β(z) + log10(4π )]057 + 89(4)









11








823(819) plusmn 012 Low

















12












c


minus04c






c



c




13











14







15








16













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES
























10






M minus β(z) + log10(4π )]057 + 89(4)









11








823(819) plusmn 012 Low

















12












c


minus04c






c



c




13











14







15








16













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES






M minus β(z) + log10(4π )]057 + 89(4)









11








823(819) plusmn 012 Low

















12












c


minus04c






c



c




13











14







15








16













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES








823(819) plusmn 012 Low

















12












c


minus04c






c



c




13











14







15








16













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES












c


minus04c






c



c




13











14







15








16













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES











14







15








16













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES







15








16













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES








16













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES













17






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES






18











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES











9 SUMMARY


19














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES














20






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES






21






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES






APPENDIX






22






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES






23






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES






24
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES
















25





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES





REFERENCES

















26

1 INTRODUCTION





32 Line Fitting


34 Photometry




52 Stellar Masses



61 The Zz Relation







9 SUMMARY


REFERENCES

the metallicity evolution of star-forming galaxies …€¦ · and =-((+)/ –

Documents