Table Of Content

DRAFTVERSIONFEBRUARY2,2008 PreprinttypesetusingLATEXstyleemulateapjv.11/12/01 REST-FRAMEULTRAVIOLETSPECTRAOFZ∼3LYMANBREAKGALAXIES1 ALICEE.SHAPLEYANDCHARLESC.STEIDEL2 CaliforniaInstituteofTechnology,MS105–24,Pasadena,CA91125 MAXPETTINI InstituteofAstronomy,MadingleyRoad,CambridgeUK KURTL.ADELBERGER3 Harvard-SmithsonianCenterforAstrophysics,60GardenSt.,Cambridge,MA02139 DraftversionFebruary2,2008 3 0 ABSTRACT 0 We present the results of a systematic study of the rest-frame UV spectroscopic properties of Lyman Break 2 Galaxies(LBGs). Thedatabaseofalmost1000LBGspectraprovesusefulforconstructinghighS/Ncomposite n spectra. The compositespectrumof the entiresample revealsa wealth offeaturesattributableto hotstars, H II a regions,dust,andoutflowingneutralandionizedgas. Bygroupingthedatabaseaccordingtogalaxyparameters J suchasLyαequivalentwidth,UVspectralslope,andinterstellarkinematics,weisolatesomeofthemajortrends 3 inLBGspectrawhichareleastcompromisedbyselectioneffects.WefindthatLBGswithstrongerLyαemission 1 have bluer UV continua, weaker low-ionization interstellar absorption lines, smaller kinematic offsets between Lyα and the interstellar absorption lines, and lower star-formation rates. There is a decoupling between the 1 dependenceoflow-andhigh-ionizationoutflowfeaturesonotherspectralproperties. Additionally,galaxieswith v rest-frameW ≥20Å inemissionhaveweakerthanaveragehigh-ionizationlines, andnebularemissionlines 0 Lyα which are significantly stronger than in the sample as a whole. Most of the above trends can be explained in 3 2 termsof the propertiesof the large-scaleoutflowsseen in LBGs. Accordingto thisscenario, the appearanceof 1 LBG spectra is determinedby a combinationof the coveringfraction of outflowing neutralgas which contains 0 dust,andtherangeofvelocitiesoverwhichthisgasisabsorbing.Incontrast,thestrengthsofcollisionallyexcited 3 nebularemissionlinesshouldnotbeaffectedbythenatureoftheoutflow,andvariationsintheselinesmayindicate 0 differencesinthetemperaturesandmetallicitiesinHIIregionsofgalaxieswithverystrongLyαemission.Higher h/ sensitivityandspectralresolutionobservationsarestillrequiredforafullunderstandingofthecoveringfraction p andvelocitydispersionoftheoutflowingneutralgasinLBGs,anditsrelationshiptotheescapefractionofLyman - continuumradiationingalaxiesatz∼3. o r Subjectheadings:galaxies:active—galaxies:nuclei—galaxies:evolution—quasars:general—galaxies: t high-redshift s a : v 1. INTRODUCTION low S/N ofLBG spectra precludesanyanalysismoredetailed i thanthedeterminationofredshifts. X Untilnow,therest-frameUVspectraofLymanBreakGalax- One notable exception is the galaxy MS1512-cB58, an L ies (LBGs) havebeenused primarilyto measureredshifts. At ∗ r LBG at z=2.73 with an apparentmagnitude ofV =20.6 due a first, the spectra were used to confirmthe cosmologicallydis- tolensingbyaforegroundclusteratz=0.37(Yeeetal. 1996). tantnatureofgalaxiesphotometricallyselectedwiththeLyman Cluster lensing boosts the apparent luminosity of cB58 by a BreakTechniquetobeatz∼3(Steideletal. 1996a,b;Lowen- factorof ∼30, enablingrelativelyhigh-resolution(R≃5000) thaletal. 1997). Spectroscopicredshiftswerealso necessary studies of its rest-frame UV spectrum (Pettini et al. 2000, tostudythelarge-scalespatialdistributionandclusteringprop- 2002). The velocity profiles of low and high-ionizationinter- erties of LBGs (Adelberger et al. 1998; Steidel et al. 1998; stellarmetalabsorptionfeatureshavebeencharacterizedinde- Giavaliscoetal. 1998). Furthermore,thedeterminationofthe tail; the weakest interstellar metal transitions have been used z∼3 rest-frame UV luminosity function required knowledge togetherwith thedampedLyαabsorptionprofileto determine of the redshifts of LBGs, combined with the optical apparent theabundancepatternincB58(anα/Feenhancementindicative magnitudes and colors (Steidel et al. 1999). Most recently, ofayoungstellarpopulation,andanabundanceof∼0.4Z for the redshifts measured from rest-frame UV spectra have been theαelements); CIV andSi IVP-Cygnistellarwindpro⊙files used to study the cross-correlation of the large-scale distribu- havebeenusedasindependentprobesofthestellarpopulation tionsofgalaxiesandtheinter-galacticmediumwithinthesame andmetallicity;weakstellarabsorptionfeatureshavebeenused cosmicvolume(Adelbergeretal. 2002a). Giventhefaintna- tureofLBGs(mosthaveR =24- 25.5),thedesiretoobserve to precisely measure the systemic velocity of the stars, rela- AB tivetowhichtheredshiftsofLyαemissionandinterstellarab- alargesampleresultsinindividualspectrawithlowsignal-to- sorptionindicateoffsetsofseveralhundredkms- 1;finally,the noise (S/N) ratios and spectral resolution. In most cases, the 1 Based, inpart, ondataobtainedattheW.M.KeckObservatory, whichisoperated asascientificpartnership amongtheCalifornia InstituteofTechnology, the UniversityofCalifornia,andNASA,andwasmadepossiblebythegenerousfinancialsupportoftheW.M.KeckFoundation. 2PackardFellow 3HarvardSocietyJuniorFellow 1 2 strengthsofthestrongestinterstellarabsorptionfeatures(which et al. 1997; Lowenthal et al. 1997). The outflows represent havezerotransmissionatlinecenter)havebeenusedtoinfera not only an important feedback process which affects galaxy high coveringfraction of outflowingneutral material, through formationandevolution,butalsomayhaveaprofoundimpact which negligible Lyman continuumemission can escape. For on the metal enrichment, ionization, and physical state of the thevastmajorityofunlensedLBGs,itisunfortunatelynotpos- surroundingintergalacticmedium(Adelbergeret al. 2002a,b; sibletoobtainindividualspectrawhichcontainthesamehigh- Adelberger 2002a;Steidel, Pettini, & Adelberger 2001). So quality information about physical conditions. Since cB58 is far, most of the information about the nature of outflows at onlyoneobject,weneedtoworryabouthow“typical”itscon- high redshift comes from high-quality observations of a sin- tinuum and spectroscopic properties are, relative to the range glegalaxy–cB58–andfromstudiesofthemetalcontentand seenintheentiresampleofLBGs. HI opacity of the Lyα forest near LBGs. The current survey Even from the low S/N spectra used to measure redshifts, providesinformationwhichiscomplementarytobothtypesof we discern a large variation in the types of spectra associated observations,asitexploresoutflowpropertiesfromthegalaxy withLBGs. Mostobviously,therearelargerangesofLyαpro- perspective,butisbasedonamuchlarger(iflowerspectralres- fileshapesandUVcontinuumslopes. Therearealsovariations olution)sampleofgalaxies. among the equivalentwidths of the few strong interstellar ab- The LBG spectroscopic sample is described in §2, while sorption lines which we detect most of the time in individual §3 presents the method of generating composite spectra and spectra,andoftheredshiftoffsetmeasuredbetweenLyαemis- definingthe“rest-frame”inthepresenceoflarge-scalevelocity sionandinterstellarabsorption(forspectrainwhichbothtypes fields. §4givesanoverviewofthetypesofstellarandinterstel- of featuresare detected). While thereis nohopeof collecting larspectroscopicfeatureswhichappearinhighS/Ncomposite data of comparable quality to the cB58 spectra for individual LBGspectra,someofwhichhavenotpreviouslybeenobserved unlensed LBGs, we have assembled a sample of almost 1000 inUVspectraoflocalstar-formingregions. In§5,wepresent spectroscopically confirmed z ∼ 3 galaxies over the past six some of the important trends observed in LBG spectra, with years. Bydividingourspectroscopicdatabaseintosubsamples particularattentiontooutflow-relatedproperties. Finally,in§6 according to specific criteria, and creating high S/N compos- wepresentaphysicalpicturewhichisbroadlyconsistentwith ite spectraof eachsubsample,we hopetounderstandhowthe the observations, and highlightthe need for severalfuture ob- LBG spectroscopic properties depend in a systematic way on servationstotestthispicture. othergalaxyproperties. PastusesofcompositeLBGspectrahaveprovenveryillus- 2. THELBG SPECTROSCOPIC SAMPLE trative.Forexample,acompositeof29individualLBGspectra Theindividualgalaxyspectrausedtoconstructthecompos- athzi=3.4±0.09showssignificantresidualfluxshortwardof ite spectra in this paper were drawn from the Lyman Break theLymanlimitat912Å(Steidel,Pettini,&Adelberger 2001). Galaxy (LBG) survey of z ∼ 3 galaxies. The details of our IfthiscompositespectrumistakentoberepresentativeofLBGs surveyhavebeenpresentedelsewhere(Steideletal. 1996a,b, atz∼3,theLBGcontributiontotheionizingbackgroundcould 1999) and will be extensively summarized in a future work exceed that of QSOs at similar redshifts by as much as a fac- (Steidel et al. 2003, in preparation), so here we present only tor of 5. The appearance of this composite spectrum is very afewrelevantfeatures.ThefullLBGphotometricsamplecon- differentfromthatof cB58, with strongLyα emission, a consists of 2347galaxiesin 17 separate fields with opticalcolors tinuumslopeinthebluestquartileofthetotalLBGsample,and satisfyingthefollowingcriteria: interstellarabsorptionlinesroughlyhalfthestrengthofthosein thecB58spectrum.DeterminingtherelativenumbersofLBGs R≤25.5, G- R≤1.2, U - G≥G- R+1 (1) n thatresemblethiscompositeversusthosethatmorecloselyre- plus an additional 180 galaxies with the same color crite- semble cB58 is important for determining the overall contri- ria,butRmagnitudeswhichareasfaintasR=26(whichare bution of LBGs to the ionizing background. Composite LBG located in the LBG survey field containing the bright quasar, spectra were also constructed for galaxy subsamples grouped Q1422+239, for which we obtained significantly deeper pho- bythestellarpopulationageinferredfromoptical/near-IRpho- tometry,allowinganextensionoftheLBGselectiontechnique tometry(Shapleyetal. 2001). The“young”(t≤35Myr)and tofaintermagnitudes). “old” (t ∼1 Gyr) composite spectra exhibited systematic dif- Wehavespectroscopicallyobserved1320ofthesephotomet- ferences, including significantly stronger Lyα emission in the riccandidatesusingtheLowResolutionImagingSpectrometer “old”spectrum,andstrongerinterstellarabsorptionandstellar (LRIS)attheW.M.KeckObservatory(Okeetal. 1995).Most P-Cygni wind features in the “young” spectrum. Such differ- of the spectra were obtained using a 300 lines mm- 1 grating encesmayindicateanevolutionarysequencefortheappearance blazed at 5000 Å, and a multi-object slit mask with 1.4 slits, oftherest-frameUVspectra. ′′ providing a spectral resolution of 8- 12 Å, depending on the Basedonthepromiseofthesespecificcompositestudies,we seeing. Recent spectroscopic data, including some of the ob- are motivated to undertake a more general study of the spec- servations in the Q0933+288 and Q1422+2309 fields, and all troscopic properties of the entire LBG sample. This system- ofthedataintheQ0302- 003field,wereobtainedwiththenew atic study is in some ways a high-redshiftanalog to the work blue arm of LRIS (LRIS-B; McCarthy et al. 1998; Steidel et of Heckman et al. (1998), which characterized the region of al. 2003), and dispersed by a 300 lines mm- 1 grism blazed at UVspectroscopicparameterspaceinhabitedbylocalstarburst 5000Å.Thissetupprovidedslightlyhigherspectralresolution galaxies. Inparticular,wewouldliketogainmoredetailedin- than the 300-line grating setup, and much higher throughput. formationaboutthepropertiesofthelarge-scaleoutflowsofin- Typicalexposuretimesforbothsetupswere3×1800secwith terstellarmaterialwhichareinferredinLBGsfromblueshifted 1.0 dithers between exposures to provide for adequate sky- interstellarabsorptionandredshiftedLyαemission,relativeto ′′ subtraction.Thetwo-dimensional,sky-subtracted,andcoadded the systemic stellar redshift (Pettini et al. 2001, 2002; Franx spectrawerethenextractedtoone-dimension,andwavelength 3 calibrated using a HgNeAr arc-lamp spectrum. Spectra were has h∆vem- absi∼650 km s- 1. This velocity difference indi- flux-calibrated with observationsof standard stars taken close cates that at least one of the two sets of featuresis not at rest tothetimeofthescienceobservations.Finally,airwavelengths with respect to the stars in the galaxy. Such kinematics sug- were converted to vacuum wavelengths, in order to measure gestthatLBGsareexperiencinglarge-scaleoutflowscausedby redshiftsinthevacuumframe.TheaverageS/Nratioofspectra themechanicalenergyinputfromsupernovaexplosionswhich inthesampleis∼4perresolutionelement. are the result of vigorousmassive star-formationrates. It has From this sample of spectra, we identify 36 stars, 2 beenshownthat, inthe localuniverse,anygalaxywith astar- absorption-line galaxies at z∼0.5, 2 galaxies with z≃1.98, formationrate surface-densityΣ ≥0.1M yr- 1kpc- 2 is capa- 957objectsatz>2,and292objectsforwhichwecannotmea- bleofdrivingasuperwind(Heck∗man 200⊙2). Giventheirtyp- sure a redshift. Of the 957 galaxies at z> 2, 12 are identi- ical star-formation rates and physical sizes, LBGs easily sat- fiedasbroad-linedAGNduetothepresenceofemissionlines isfyandexceedthecriteriafordrivingasuperwind(Shapleyet withFWHM>2000 kms- 1,while16areidentifiedasnarrow- al. 2001;Giavalisco,Steidel,&Macchetto 1996b). linedAGN with strongLyαemission accompaniedbysignifi- cantCIVλ1549emission,butFWHM<2000 kms- 1(Steidel et al. 2002). We exclude from the spectroscopic sample the 28objectsidentifiedasAGNonanindividualbasis(thoughwe mayhaveincludedgalaxieswithlow-levelAGNactivitywhich is undetectedin individualspectra butmaybecomeevidentin thecomposites.Wewilltreatthispointfurtherinsection4.3.1). Furthermore,weincludeonlygalaxieswhoseredshiftswerein- dependently and securely confirmed by at least two members ofourgroup.Thislastcriterionlimitsourcompositesampleto 811spectra. 3. GENERATINGTHECOMPOSITE SPECTRA There are several steps required to generate the composite galaxy spectra presented in this work. The first step consists of carefully defining a sample of galaxy spectra to be combined. Each extracted, one-dimensional, flux-calibrated spec- truminthesampleisthenshiftedintotherestframe.Thespec- traarethenaveraged,afterbeingscaledtoacommonmodein thewavelengthrange1250-1500Å andrebinnedtoacommon dispersion of 1 Å per pixel. In order to exclude both positive and negative sky subtraction residuals and cosmic ray events, an equalnumberof positive and negativeoutliers are rejected ateachdispersionpixel,totalinglessthan10%ofthedata. FIG. 1.—Thedistribution ofvelocity offsetsbetween Lyαemissionand low-ionization interstellar absorption. The most straightforward indication 3.1. MeasuringRedshifts that LBGs are experiencing large-scale outflows of their interstellar mate- rialisthevelocity offsetmeasuredinindividual spectrabetweenLyαemis- One of the non-trivial aspects of generating the compos- sionandinterstellar absorption lines. Thishistogramshowsthedistribution ite spectra consists of defining a systemic rest frame for each of velocity offsets for the 323 galaxies with spectra in which both types of features are detected. The mean velocity offset (redshift difference) is galaxy. The low S/N ratio of typical LBG spectra precludes ∆v=650 kms- 1(∆z=0.008). much more than measuring redshifts from the very strongest rest-frameUV features. These featuresincludeH I Lyα, seen either in emission, absorption, or a combinationof both; low- ionizationresonanceinterstellarmetallinessuchasSiIIλ1260, OIλ1302+SiIIλ1304,CIIλ1334,SiIIλ1526,FeIIλ1608, Since the only features which can be detected in individ- and Al II λ1670, which are associated with the neutral inter- ual LBG spectra seem to trace outflow kinematics, the out- stellar medium; and high-ionizationmetal lines such as Si IV flows complicate our effort to assign systemic redshifts. In λλ1393,1402andCIVλλ1548,1550associatedwith ionized general, stellar systemic redshifts cannot be measured for in- interstellar gas and P-Cygni stellar wind features. In 28% of dividual galaxies because UV photospheric features from hot the rest-frame UV LBG spectra in our spectroscopic sample, starsaremuchtooweaktoseeintypicalLBGspectra.Inorder the only spectral feature visible is Lyα emission. In 32% of toestimatestellarsystemicredshiftsforindividualgalaxies,we the spectra, Lyα appearsonlyas broadabsorption, and multi- applied the formulae presented in Adelberger et al. (2002a), plelow-and/orhigh-ionizationinterstellarabsorptionlinesare whichpredictavalueofthesystemicredshiftforthreeseparate usedtomeasuretheredshift(broadLyαabsorptionisnotvery cases: whenthereisonlyaLyαemissionredshift;whenthere preciseasaredshiftindicator).Finally,intheremaining40%of is only an interstellar absorption redshift; and when there are thespectra,bothLyαemissionandinterstellarabsorptionlines bothLyαemissionandinterstellarabsorptionredshifts. arevisibleandbothcanbeusedtomeasureredshifts.In95%of Whilenoreferencewasmadetostellarphotosphericfeatures the caseswherebothLyα emissionandinterstellarabsorption intheestimateofthesystemicredshiftsofindividualgalaxies, lineshavebeenusedtomeasuretheredshiftofthegalaxy,the therest-framecompositespectrumpresentedinthenextsection Lyαemissionredshiftishigherthantheinterstellarabsorption indicates a mean systemic velocity of ∆v=- 10±35 km s- 1 redshift. Figure 1 shows the distribution of ∆vem- abs, which for the three strongest stellar features, which we have identi- 4 fied as C III λ1176, O IV λ1343, and S V λ1501. 4 It is and He II λ1640. The shape of the N V wind profile, espe- worthnotingpossiblecontributionstotheOIVλ1343absorp- cially the absorptioncomponent,is affectedby its close prox- tionfromSiIIIλ1341,andtotheSV λ1501absorptionfrom imitytotheLyαregionofthespectrum,andisthereforediffi- Si III λ1501, if there is a significant B-star componentin the cult to characterize in detail, though we do see both emission composite spectrum. However, at least the Si III λ1501 con- andabsorptionqualitativelyconsistentwithaP-Cygnitypepro- tribution will not change the inferred negligible systemic ve- file. While clear of the large-scale continuumeffects of Lyα, locity of the feature which we have identified as S V λ1501. theSiIVandCIVtransitionscontainthecombinationofstel- The insignificant velocities of the stellar features demonstrate larwindandphotosphericabsorption,plusastronginterstellar thesuccessofthesystemicredshiftestimatesfortheLBGsin- absorptioncomponent,whichare difficultto disentangle. The cluded in composite spectra, at least on average. Also, since stellarwindfeatureonlybecomesapparentinSiIVforbluegi- thestellarfeaturesappearatroughlyzerovelocity,theredshifts antandsupergiantstars,while,incontrast,theCIVwindfea- and blueshifts of other sets of spectral features measured rel- ture is strong in main sequence, giant, and supergiantO stars ative to the rest-frameof the compositespectrumshouldoffer (Walborn&Panek 1984).Consequently,theinterstellar,rather atruerepresentationoftheaveragekinematicpropertiesofthe thanthewindcomponent,seemstodominatetheSiIVdoublet large-scalegalacticoutflowsinLBGs. Theestablishmentofthe intheLBGcompositespectrum,whiletheCIVfeatureexhibits velocity zeropoint from the stellar lines in composite spectra boththeblue-shiftedbroadabsorptionandredshiftedemission representsasignificantimprovementoverthekinematicinfor- associatedwithstellarwinds,inadditiontoastrong,narrower, mation contained in individual rest-frame UV spectra, where interstellarabsorptioncomponent. Theredshiftedemissionin- only the strongest interstellar outflow-related features are de- dicatesthepresenceofstarswithM≥30M (Leitherer,Robert, tected. & Heckman 1995; Pettini et al. 2000). ⊙In Figure 3, a Star- burst99modelspectrum(Leithereretal. 1999)isplottedover 4. LBG REST-FRAME UVSPECTROSCOPIC FEATURES thezoomed-inCIVλλ1548,1550regionoftheLBGcompos- itespectrumforcomparison. Themodelspectrumisfora300 Figure 2 shows a composite spectrum which is the average Myroldepisodeofcontinuousstarformation(themedianstel- of ourentirespectroscopicsampleof 811LBGs, combinedin lar population age inferred from the optical/near-IR colors of the manner described in section 3. Rest-frame UV spectra of LBGs, Shapley et al. 2001), and is constructed from a library LBGsaredominatedbytheemissionfromOandBstarswith ofHSTFOSandSTISobservationsofmassivehotstarsinthe masseshigherthan10M andT ≥25000K. Theoverallshape Magellanic Clouds with a mean metallicity Z =0.25Z (con- of the UV spectrum is m⊙odified by dust extinction internal to sistentwiththelimitedinformationonLBGmetallicitie⊙s). The the galaxy, and, at rest wavelengths shorter than 1216 Å, by model and data agree quite well in the emission component inter-galactic HI absorption along the line of sight. Compos- of the P-Cygni profile. However, the model overpredicts the ite spectracontaintheaverageofmanydifferentlinesofsight strength of the broad wind absorption. This discrepancy may throughtheIGM.Therefore,spectralfeatureswhichareintrin- beduetoacombinationofageandmetallicityeffects. sic to the galaxy at wavelengths shorter than Lyα, and which ThecompositeLBGspectrumalsoshowsHeIIλ1640emis- canbecompletelywipedoutbyindividualLyαforestsystems sionwhichisquitestrongcomparedwithobservationsinlocal alonga specific line ofsight, becomevisiblein the composite starburstgalaxies(Heckmanetal. 1998).NarrowHeIIλ1640 spectra. While we regain spectroscopic information by aver- emissionisseeningalaxyspectrawithstrongnebularemission. agingovermanydifferentsightlines,westill,however,seethe IfmassivestarsareforminginHeIIIregions,HeIIλ1640can averagedecrementoftheLyαforest,DA. Inthefollowingsec- appear as a nebular recombination emission line (from super- tion, we describe the spectroscopic features contained in the novae remnants or superbubbles) (Leitherer, Robert, & Heck- compositespectrumofFigure2,whichtracethephotospheres man 1995). TheHeIIλ1640emissioninthecompositeLBG andwindsofmassivestars, neutralandionizedgasassociated spectrum,though,isquitebroad,withFWHM∼1500 kms- 1. withlarge-scaleoutflows,andionizedgasinHIIregionswhere Thisisvisiblybroaderthanthemostoftheotherweakemission starformationistakingplace. linesinthespectrumoriginatinginHIInebularregions,sowe favor a stellar wind originfor the He II feature. Broad stellar 4.1. StellarFeatures HeIIemissionispredominantlyproducedinfast,densewinds from Wolf-Rayet (W-R) stars, which are the evolved descen- TheCIII1176,OIV1343,andSV1501stellarphotospheric dantsofOstarsmoremassivethanM>20- 30M . Therefore, lines discussedin section 3.1aremarkedin Figure2. Also of the strength of the He II emission shouldprovide⊙information note(thoughnotmarked)isthelargenumberofweakabsorp- abouttheratioofW-RtoOstars(Schaerer&Vacca 1998). tion featuresbetween 1400and 1500Å. These includeblends In orderto interpretthe He IIline, we use Starburst99pop- of Fe V, Si II, Si III, and C III photospheric absorption lines ulation synthesis codes (Leitherer et al. 1999). Model UV from O and B stars (Bruhweiler, Kondo, & McClusky 1981; spectraofbothsolarand0.25×solarmetallicityareproduced de Mello, Leitherer, & Heckman 2000). In addition to pho- by Starburst99. The solar metallicity spectrum extends from tospheric absorption features, the spectra of the most massive 1200- 1800Å,whichincludestheHeIIline,butthesub-solar hot stars indicate the presence of stellar winds of 2000-3000 km s- 1 due to radiation pressure (Groenewegen, Lamers, & metallicitymodelonlyextendsfrom1200- 1600Å.Whilethe currentbestestimatesofLBGmetallicities(Pettinietal. 2002, Pauldrach 1989). These wind features appear as broad blue- 2001) are closer to Z = 0.25Z , we can still use the solar shiftedabsorptionforweakerwinds,orasaP-Cygnitypepro- metallicitymodelspectratomak⊙esomeinterestinginferences. file if the wind density is high enough (Leitherer, Robert, & The observedstrength of the LBG He II emission line can be Heckman 1995).Themostprominentstellarwindfeaturesare matched using a 3 Myr old instantaneous burst model with a N V λλ1238,1242, Si IV λλ1393,1402, C IV λλ1548,1550, 4TheprecisewavelengthsareCIIIλ1175.71,OIVλ1343.35,whichisablendoflinesatλ=1342.99andλ=1343.51,andSVλ1501.76 5 FIG.2.—Acompositerest-frameUVspectrumconstructedfrom811individualLBGspectra.DominatedbytheemissionfrommassiveOandBstars,theoverall shapeoftheUVcontinuumismodifiedshortwardofLyαbyadecrementduetointer-galactic HIabsorption. SeveraldifferentsetsofUVfeaturesaremarked: stellarphotosphericandwind,interstellarlow-andhigh-ionizationabsorption,nebularemissionfromHIIregions,SiII*fine-structureemissionwhoseoriginis ambiguous,andemissionandabsorptionduetointerstellarHI(LyαandLyβ).Therearenumerousweakfeatureswhicharenotmarked,aswellasseveralfeatures bluewardsofLyαwhichonlybecomevisiblebyaveragingovermanysightlinesthroughtheIGM.ThecompositeLBGspectrumisavailableinelectronicform fromhttp://www.astro.caltech.edu/∼aes/lbgspec/. 6 standardSalpeterIMF.Thisisthebriefperiod,followingaburst ofLBGs,thesefeaturesaredetectedwithhighsignificance. In of star formation, when the fractional contribution of Wolf- thissection, we describetheaveragepropertiesofthespectral Rayet stars to the integrated UV luminosity reaches a max- featuresrelatedtothelarge-scaleoutflowsinLBGs. imum. Such young burst ages have indeed been derived for Wolf-Rayetgalaxies,whosestrongopticalandUVHeIIemis- sion lines indicate the presence of numerous high-mass stars and a highW-R/O star ratio (Conti, Leitherer& Vacca 1996; Leithereretal. 1996).Itisnotclearwhichaveragestellarpopu- lationisrepresentedbythecompositespectruminFigure2,but since LBG rest-frame UV/opticalSEDs indicate a wide range of propertiesand ages, and are generallynot well-represented byinstantaneousburstmodels(Shapleyetal. 2001;Papovich, Dickinson,&Ferguson 2001),itseemsextremelyunlikelythat the 3 Myr old burst modelcan be representativeof the whole populationandaccordinglyweruleoutthisinterpretation. Ifwethenconsider300Myrcontinuousstar-formationmod- els (based on the median age derived from SED-fitting), the only way to produce a high enoughW-R/O star ratio is to in- vokeanIMFslopemuchflatterthanthestandardSalpeterform of N(m) ∝ m- α with α = 2.35. Slopes of α ≤ 1 match the He II strength, but then the C IV P-Cygni emission is over- produced by a factor of ≥ 4. The discrepancy will be even worseforthesub-solarmetallicitymodels. TheratioofW-R/O stars decreases as a strong function of decreasing metallicity, sincethelower-limitonthemassesofOstarswhichevolveinto W-R stars movestowards higher masses as the metallicity increases(Maeder 1991;Meynet 1995).Theoreticalpredictions of W-R line luminosities (Schaerer & Vacca 1998; Leitherer et al. 1999) show that, for a given IMF, 300 Myr continuous FIG. 3.—Four zoomed-in regions from the composite spectrum of Fig- ure 2. The zoomed-in vertical scale allows a more detailed look at weak star-formationmodelswithZ=0.25Z haveHeIIλ1640line nebular emission features such as Si II* λλλ1265,1309,1533, and O III] strengthsonlyhalfasstrongasinthe⊙caseofsolar-metallicity. λλ1661,1666,whichareproducedinHIIregions. Thelowerleft-handbox A more drastic adjustment to the slope of the IMF would be showsthezoomed-inregionnearCIVλλ1548,1550. Theobservedcompos- itespectrumisplottedasasolidline, whilethedashedlineisaStarburst99 requiredforthesub-solarmetallicitymodeltomatchtheHeII modelspectrumfor300Myrofcontinuousstar-formationwithZ=1Z (Lei- emissionstrength,whichwouldthenleadtoanevenlargerdis- thereretal. 1999,2001). Themodelanddataagree wellforthe4em⊙ission crepancyfortheCIVP-Cygniemission(since,empirically,we componentoftheCIVP-Cygnistellarwindline,butthemodeloverpredicts measurethattheCIVP-Cygniemissiondoesnotdependvery theamountofbroad,blue-shiftedabsorption. Thismaybeduetothelower averagemetallicityorolderageoftheLBGsincludedinthecompositespec- significantly on metallicity). Clearly, current population syn- trum,relativetothemodel. Thelowerright-handboxshowsthezoomed-in thesis models do not simultaneously reproduce the C IV and regionneartheHeIIλ1640stellarwindline,whoselargestrengthwehave He II stellar wind features in LBGs for reasonable choices of troublereproducingusingStarburst99modelswithreasonableparameters. star-formationhistory,age,metallicity,andIMFslope. Wewill be able to test this discrepancy better with high-quality spec- traforindividualLBGswhosestellarpopulationagesandstar- formation histories we know more accurately than our rough estimateofthestellarpopulationrepresentedbythecomposite 4.2.1. Low-ionizationLinesAssociatedwithNeutralGas ofalltheLBGs. The strongest low-ionization interstellar lines probing out- flowinginterstellarmaterialaremarkedinFigure2,andinclude 4.2. Outflow-relatedFeatures Si II λ1260, O I+Si II λ1303 (a blend at the spectral resolu- Thelarge-scaleoutflowofinterstellarmaterialappearstobe tionofoursample),CIIλ1334,SiIIλ1526,FeIIλ1608,and agenericfeatureofLBGs,oneimpliedbothbythetypicalLBG AlIIλ1670. Whilethereareadditionalinterstellarabsorption star-formation rate per unit area, and also by the fairly ubiq- linesatwavelengthsshorterthanLyα,aswellasahostofother, uitous observed offset in velocity between Lyα emission and weaker,low-ionizationinterstellarabsorptionlines,weconfine interstellar absorption lines when both sets of lines are seen. theanalysisinsection5tothesesixstrongestfeaturesredward Spectral features probing neutral outflowing gas are H I Lyα of Lyα, which we detect at very high significance. The fea- andLyβ,andneutralandsinglyionizedmetalabsorptionlines. tures marked have been well-studied in local starburst galax- Morehighlyionizedmetallinesprobetheionizedphaseofthe ies(GonzalezDelgadoetal. 1998;Heckmanetal. 1998),as outflow(Heckmanetal. 2001b;Pettinietal. 2002).Whilethe wellasinthelensedLBG,MS1512-cB58(Pettinietal. 2000, interstellarabsorptionlinesin∼70%ofindividualLBGspectra 2002). Relative to the stellar, systemic redshift, we measure arestrongenoughthatanabsorptionredshiftcanbeassignedto an average blueshift for the strong low-ionization interstellar atleastonelow-orhigh-ionizationfeature,itisnotpossibleto features of ∆v = - 150±60km s- 1. Using the C II λ1334 obtainrobustequivalentwidthmeasurementsformultiplefea- and Si II λ1526features, which we assume to be the least af- turesduetothetypicallowS/N(andprevalentsky-subtraction fectedbyblends,andassuminganeffectivespectralresolution residuals). However,incompositespectracontaininghundreds of2.6ÅfortheLBGcompositespectrum,wecomputetheav- 7 erage deconvolvedvelocityfull-width5 forthe low-ionization thoughwedoseeevidenceforaninterstellarabsorptioncompo- interstellar(LIS)lines,FWHM(LIS)=560±150 kms- 1. The nent.Also,itisdifficulttocharacterizethepropertiesofthehot largestuncertaintyaffectingthismeasurementistheuncertainty phasetracedbyOVI,duetothefactthattheOVIabsorption of the effectivespectralresolutionof the compositespectrum, isfairlyweak,blendedwithCIIλ1036absorption,andresides whichweconservativelyestimatetorangebetween2- 3.25Å intheredwingofthestrongLyβprofile.However,wemeasure (roughly 400- 700km s- 1). The upper bound on the spectral ∆v=- 180 km s- 1 for both members of the Si IV transition, resolution is set by the minimum FWHM value measured for which has a doublet ratio of roughly 2:1, indicating that the anyofthestronginterstellarabsorptionlines. Thelowerbound linesareonthelinearpartofthecurveofgrowth. Wemeasure of2Åissetbyourestimateofthespectralresolutionprovided ∆v=- 390 kms- 1,forwhatweisolateastheinterstellarcom- bytheoptimumobservingconditions.6 ponentoftheCIVabsorption,assumingasaturateddoubletra- We list rest-frame equivalent widths and relative systemic tioof1:1andthereforearest-framecentroidofλ=1549.479Å. redshiftsforthesixstrongestlow-ionizationinterstellarabsorp- ThemeasurementoftheCIVinterstellarvelocityisfairlyun- tionlinesinTable1.Thestrengthofthesefeaturesmakesthem certain,duetothecombinationofP-Cygniemissionandbroad idealformeasuringinterstellarabsorptionredshiftsinnoisyin- absorption, possible nebularemission, and interstellar absorp- dividual spectra. However, they are not useful for measuring tionallsuperposedononeanother. ThepropertiesoftheSiIV chemicalabundances,duetothefactthatallofthestronglines andCIVinterstellarabsorptionlinesarelistedinTable1. are saturated. The saturation of the strong lines is most eas- In light of the complexities associated with the C IV dou- ily demonstrated by comparing the equivalent widths for two blet, we emphasize the comparison between the Si IV dou- differentSi IItransitions: Si IIλ1260andλ1526. On the lin- blet and the low-ionization lines. The blueshift of the Si IV ear part of the curve of growth, W ∝ Nfλ2, where N is the doublet (∆v = - 180 km s- 1) agrees quite well with the av- column density of the ionic species and λ is the rest-frame erage blueshift of the strong low-ionization interstellar lines wavelength of the transition. According to the relative oscil- (∆v=- 150 km s- 1). Furthermore, the average deconvolved latorstrengthsandwavelengthsofthetwoSiIItransitions,the velocity full-width for the two members of the Si IV doublet ratio W (1260)/W (1526)>5 on the linear part of the curve isFWHM(SiIV)=590±140 kms- 1,verysimilartotheaver- 0 0 of growth. We measure W (1260)/W(1526)= 0.95, consis- ageFWHM forthelow-ionizationlines. Thedeconvolvedve- 0 0 tentwith a ratio ofunity,giventheuncertainties, thusdemon- locitiesassociatedwitheithertheSiIVorlow-ionizationlines strating that the Si II transitions are optically thick. There are very uncertain due to the uncertainties in spectral resolu- are weaker features detected in the composite LBG spectrum tion. Independentoftheresolutionofthecompositespectrum, which probethe linearpartof the curveof growth,andwhich however,boththeblue-shiftsandtheun-deconvolvedvelocity have been used to derive metal abundances in the outflow of widthsofthelowandhighionizationlinesareconsistentwith MS1512-cB58 (Pettini et al. 2000, 2002). These include eachother. S II λλλ1250,1253,1259, Si II λ1808, Fe II λ1144, Ni II Thepropertiesoflowandhighionizationabsorptionprofiles λ1317,λ1370,λ1703,λ1709,λ1741,andλ1751.Mostofthese were compared in the spectrum of the gravitationally lensed features are detected with only marginal significance in the LBG, MS1512-cB58, using ∼ 10 times higher spectral reso- composite LBG spectrum due to low spectral resolution. An lution (Pettini et al. 2002). In the case of cB58, which has abundance determination also requires an estimate of the H I muchstrongerthanaveragelow-ionizationinterstellarabsorp- column density, which is not easily measured from the com- tionlines,saturatedSiIVandCIVtransitions,andLyαdom- positespectrum,duetothewayinwhichitwascombined. In inated by a damped absorption profile, absorptionsfrom low- thispaper,theanalysisofthelowionsintheoutflowinggasis ionsandhigh-ionsspanthesameoverallvelocityrange. Also, confinedtothepropertiesofthestrongtransitions. thematerialwiththehighestopticaldepth(i.e. wherethelines are black) is blueshifted by roughly the same amount for the 4.2.2. High-ionizationLinesAssociatedwithIonizedGas low ions and the high ions, to within 20 km s- 1. One dis- tinction highlighted by Pettini et al. (2002) is that the high- In addition to the low-ionization features associated with ionization lines show smoother absorption profiles, while the neutral outflowing gas, we detect high-ionization interstellar low-ionizationprofilesbreakupintoanumberofdiscretecom- featuressuchasSiIVλλ1393,1402,CIV λλ1548,1550,and ponents. Composite LBG spectra do nothave sufficientspec- N V λλ1238,1242. Thesefeaturespredominantlytracegasat tralresolutiontodiscernqualitiessuchasprofilesmoothnessor T ∼> 104K, which has been ionized by a combination of radi- clumpiness. However,theoverallagreementbetweenlowand ation from massive stars and collisional processes associated highionstagesin meanblueshiftandvelocityFWHM is con- withtheoutflow.WealsodetectOVIλλ1032,1038inabsorp- sistentwiththehigh-resolutionresultsfromcB58. Incontrast, tion. Iftheradiationfieldisdominatedbythespectrumofhot Wolfe&Prochaska (2000a)finddistinctkinematicproperties stars, ratherthan an AGN, O VI is likely to arise in collision- forlowandhigh-ionizationtransitionsassociatedwithdamped allyionizedgas,indicatingthepresenceofanevenhotterphase Lyα absorbers. Specifically, the mean velocities of low and withT ∼> 105K (Heckmanetal. 2001b). intermediate(AlIII) ionizationstages differfromthose of the Section 4.1 included a discussion of the stellar winds indi- highions,althoughwithineachofthesetwosetsoflinesthereis cated by the Si IV, C IV, and N V profiles. In this section normallyverygoodinternalvelocityagreement.Also,in29out we consider the properties of the interstellar contributions to of32cases,thevelocitywidthofthehigh-ionizationabsorption the same high-ionization transitions. Again, the proximity of exceeds that of the low-ionization absorption. The kinematic NVtoLyαpreventsusfromstudyingthistransitionindetail, 5 By deconvolved velocity full-width, FWHMint, we mean the square-root of the difference in quadrature between the observed full-width, FWHMobs and the instrumentalresolution,FWHMinst:FWHMint=(FWHM2obs- FWHM2inst)1/2 6Thespectrawereobtainedthrough1′.′4slits,whichismuchlargerthantheLBGsizeintypicalseeingconditions(0′.′5- 1′.′0).Undersuchconditions,thespectral resolutionwasdictatedbytheangularsizesofobjectsfallingwithinslits(i.e.theseeing),ratherthanbytheslit-width. 8 differencesbetweenLBGsandDLAsareanothermanifestation which would have remained undetected in individual spec- ofthedifferingcharacteristicsofthesetwopopulationsofhigh tra. We detect several weak emission lines, some of which redshiftgalaxies,whichalsoexhibitdistinctclusteringproper- weattributetonebularregionsphotoionizedbyradiationfrom ties(Adelbergeretal. 2002a)andmetallicities(Pettini 2002). massive stars, and others whose origin is still ambiguous. Suchdifferenceswilleventuallyhelpusclarifythetruenature The weak emission lines in the LBG composite spectra are: ofdampedLyαabsorptionsystems. SiII*λ1265,SiII*λ1309,SiII*λ1533,OIII]λλ1661,1666, and C III] λλ1907,1909. We measure a mean velocity of 4.2.3. Lyα ∆v=100±35 kms- 1 fortheSiII*transitions. Thecentroids ofthefine-structureemissionlinesmaybebiasedtotheredby ByfarthemostprominentfeatureinindividualLBGspectra weak fine-structure absorption lines or neighboring saturated is H I Lyα. The originalsourcesof mostLyαphotonsare re- resonance absorption features (Si II λ1260, O I+Si II λ1303, combinationsinHIIregions. WhiletheLyαequivalentwidth and Si II λ1526) associated with the outflow, which attenuate is sensitive to conditionsin the H II regionssuch as tempera- theblueedgesofthefine-structureemissionprofiles. Wemea- ture,metallicity,star-formationrate,andstar-formationhistory, sureavelocityof∆v=0 kms- 1 fortheOIII]λ1663doublet, resonant scattering of Lyα in LBGs by interstellar H I makes and a velocity of ∆v=40 km s- 1 for the C III] λ1909 tran- the emergentLyα profile at least as sensitive to the geometry, sition, bothof whichagreeverywellwith the stellar systemic kinematics,anddustcontentofthelarge-scaleoutflows. When velocities(section3.1).OIII]λ1663andCIII]λ1909areboth seeninemission,Lyαcanbeusedtomeasurearedshift.Inthe collisionallyexcited,semi-forbiddentransitions,sothereisno compositeLBGspectrumshowninFigure2,wemeasureaLyα emission redshiftof ∆v=+360 km s- 1. This relativeredshift absorption from these ions in the large-scale outflow of gas. While the Al II λ1670 resonance absorption feature is fairly reflects the fact that a Lyα photon has a much better chance close toO III],the CIII]transitionshouldbeclear ofanyab- ofescapingagalaxyifitslastscatteringoccursoffofanatom sorptionline. Theremay be a nebularemission componentin whichis redshiftedwithrespecttothe bulkoftheneutralma- C IV λλ1548,1550, but it is difficult to isolate nebular C IV terial in the galaxy, impartinga Dopplershift which takes the emissionfromthestellarP-Cygniemission. Thepropertiesof Lyα photon off resonance. We measure a deconvolvedemis- sion full-width of FWHM(Lyα) = 450±150 km s- 1. When theemissionlinesaresummarizedinTable2. seen in absorption, the Lyα feature can be quite broad, with blueshifted absorption extending from zero velocity down to ∆v≤- 5000 km s- 1. Broad Lyα absorption is therefore not a precise redshift indicator. In spectra with Lyα seen only in absorption, interstellar metallines can be used to measure the redshiftmoreprecisely. AwidedistributionofLyαprofilesisseenintheLBGspec- troscopic sample, ranging from damped absorption to emission an order of magnitude stronger than the feature shown in Figure 2 (Steidel et al. 2000). In section 5.3, we will dis- cusshowseveralLBGspectroscopicpropertiesdependonLyα equivalentwidth, and whatinferencescan be drawnaboutthe physicalconditionswhichdeterminetheemergentLyαprofile. The composite spectrumshown in Figure 2 has a Lyα feature dominatedby emission. Thetotalrest-frameequivalentwidth is W =14.3 Å which includes both redshifted emission, and 0 muchweakerblueshiftedabsorption. Thisspectrumisthe average of all the LBG spectra in the spectroscopic sample, yet there are selection effects which depend on both apparent R magnitude,color,andspectroscopictype,whichbiasthespec- trumrelativetoa true“average”ofthetotalLBG photometric sample. Forexample,thespectroscopicsampleover-represents brightobjectsrelativetofaintobjects,andthenumberofobjects withLyαemissionrelativetothosewithonlyLyαabsorption. FIG. 4.—log(C/O)vs. log(O/H)forlocal HIIregions. Blue triangles Asdiscussedinsection5.1,itbecomesmuchmoredifficultat are data from dwarf irregular galaxies and the Magellanic Clouds (Garnett et al. 1995, 1997; Kobulnicky & Skillman 1998). Spiral galaxy data are fainter magnitudesto identify spectroscopicallya galaxywith shown with circles Garnett et al. (1999). Red symbols assume a shallow no Lyα emission. Therefore, as the number of spectroscop- (RV =3.1) Milky Way extinction curve, whereas green symbols indicate a ically unidentified objects increases at fainter magnitudes, so steeper(RV=5.0)one.Thesolarabundances(Holweger 2001;AllendePrieto, doestheratioofemissiontoabsorptionlinegalaxieswithmea- Lambert,&Asplund 2002)areindicatedbythelargebulls-eye.Thehorizon- tal(magenta) shadedareaindicates thelog(C/O)confidence regionderived suredredshifts. Accordingly,theLyαfeaturein Figure2may fromthetotalcompositeLBGspectrum. Thevertical(cyan)shadedareain- be biased towards stronger emission than the true average for dicatestherangeoflog(O/H)impliedbytheratioof[OIII],[OII],andHβ the total photometricsample. In order to quantifythis bias, a line-strengthsinasmallsampleofbrightLBGs(Pettinietal. 2001). moredetailedtreatmentofselectioneffectsisrequired. 4.3. EmissionLines 4.3.1. AGNContribution? One of the benefits of producing a high S/N ratio com- Eventhoughallgalaxyspectraflaggedasnarrow-orbroad- posite spectrum is that it can reveal weak spectral features linedAGNonanindividualbasiswereremovedfromtheLBG 9 compositespectrasample,thepresenceofweakemissionlines servedC/Oratiomaythusalsoreflecttheaveragetimesincethe intheLBGcompositespectrummayrevealsomeaveragelow- onsetofstar-formation. Insuchcircumstances,youngerstellar level of nuclear activity in LBGs. To address this issue, we populationsmaybedominatedbythechemicalyieldsfromthe examined the emission line ratios in a composite spectrum of mostmassivestars,whereasafterafew100Myrsincetheon- 198 LBGs with rest-frameW ≥20 Å in emission (see sec- set of star formation, an increased C/O ratio reflects the fact Lyα tion 5.3). The nebular emission lines in this composite spec- thatintermediate-massstarshavehadachancetoreleasetheir trumarealsostrongerthanthosemarkedinthetotalLBGspec- carbonintotheISM. trumofFigure2. Wecomparedtheemissionlineratiosinthis Since we detect both C III] λ1909 and O III] λ1663 in the strong-emissionspectrum with those in a composite spectrum LBG composite spectrum shown in Figure 2, we follow the of16LBGsflaggedasnarrow-linedAGNonanindividualba- analysis of Garnett et al. (1995), assuming that the electron sis(Steideletal. 2002). TheaverageLyαemissionequivalent densitiesinthese HII regionsare wellbelowthecriticallimit width in this strong-emission (yet not AGN-flagged) subsam- fortheCIII]andOIII]transitions.Inthelow-densitylimit,we pleisonlyhalfaslargetheemissionstrengthinthenarrow-line canexpresstherelativeabundancesofC+2andO+2as: AGNspectrum.Thenarrow-lineAGNspectrumhaslineinten- sityratiosofCIV/Lyα∼0.25,andCIII]/Lyα∼0.125,which C+2 1 Ω (3P,5S ) λ I(CIII]λ1909) are verysimilar to the meanratiosof C IV/Lyα=0.21±0.09 = × OIII] 2 × CIII]1909×e- 11054/T× O+2 9 Ω (1S,3P) λ I(OIII]λ1663) andCIII]/Lyα=0.10±0.05,measuredforasampleoffourlo- CIII] OIII]1663 (2) calSeyfert2galaxiesbyFerland&Osterbrock (1986).Incon- In this expression Ω (3P,5S ) and Ω (1S,3P) are the trast,thenon-AGNspectrumhasintensityratiosofCIV/Lyα≤ OIII] 2 CIII] multiplet collision strengths for the O III] λ1663 and C III] 0.02 and C III]/Lyα=0.05. The 2% represents a strict upper λ1909 doublets, which have a weak temperature dependence limitontheratioofnebularCIV/Lyα(whichisprobablymuch (see Garnett et al. 1995 for specific collision-strength val- smaller), since the total C IV emission represents the sum of ues); the prefactor of 1 represents a combination of statis- nebularemissionplusstellarP-Cygniemission. Clearly,thera- 9 tical weights; λ and λ are the effective wave- tiosofbothCIII]andCIVtoLyαaremuchsmallerinthenon- CIII]1909 OIII]1663 lengthsof the C III] and O III] doublets; T is the H II region AGN LBG spectrum than in the narrow-line AGN spectrum. electrontemperature;andI(CIII]λ1909)andI(OIII]λ1663)are Additionally,the ratioof C III]/CIV is significantly higherin the line intensities. To compute I(OIII]λ1663) we integrate thenon-AGNspectrumthanintheAGNspectrum,indicatinga the flux from both members of the resolved O III] λ1663 softer photoionizingradiationfield, more likely dominatedby doublet (the C III] doublet is unresolved). Strictly speaking, theemissionfromhotstarsratherthannon-thermalprocesses. C/O=C+2/O+2×ICF, where ICF is an ionization correction factor which takes into account the fact that while C III] and 4.3.2. C/OAbundance OIII]aresimilarionizationstates,C+2hasaslightlylowerion- Ultraviolet spectroscopic observations of H II regions in ization potential than O+2. In practice, Garnett et al. (1995) nearbyirregularandspiralgalaxieswiththeHubbleSpaceTele- find ICF only ranges between 1.06- 1.33, and so, for lack of scope (HST) have been used to study how the relative abun- anydetailedinformation,we maketheapproximationICF=1 dances of carbon and oxygen (C/O) depend on oxygen abun- and C/O≃C+2/O+2. Based on the measured ratio of C III] dance (O/H) (Garnettet al. 1995, 1997,1999;Kobulnicky& andOIII]linestrengthsinthetotalcompositeLBGspectrum, Skillman 1998). Ranging in O/H from less than 0.1 to 1.0 we measure log(C/O)=- 0.68±0.13. This confidence inter- times (O/H) , a compilation of these data is shown in fig- valismarkedbythehorizontalmagentashadedregioninFig- ure 4 and de⊙monstrates a clear trend of increasing C/O with ure 4. The quoted uncertainty corresponds to a temperature increasing O/H (Garnett et al. 1999). For example, in spiral rangeT=10000- 20000K,butdoesnotincludetheuncertainty galaxyHIIregions,log(C/O)≃- 0.7atlog(O/H)=- 4.0,and inthedeterminationofthecontinuumlevel,whichmayamount increasestolog(C/O)≃- 0.2atlog(O/H)=- 3.47. While the toanadditionalerrorofaboutafactoroftwointhedetermina- behavioratthesmallestmetallicitiesislessclear,duetheunex- tion of C/O. We measure a very similar value of log(C/O)= pectedly high C/O ratio in the extremely metal-poorgalaxy, I - 0.74±0.14intheLBGcompositespectrumconstructedfrom Zwicky18,theHIIregionsinothersub-solarmetallicityirreg- thesub-samplewhichincludesonlystrongemission-linegalax- ular galaxies show the same overall trend as the spiral galaxies (see sections 4.3.1, 5.3). Similar C/O ratios are found by ies (Garnett et al. 1995). Stellar evolution models including Garnettet al. (1999)forgalaxieswith O/H∼0.2×(O/H) . stellar winds from massive stars predict that the carbon yield TherearelargeuncertaintiesassociatedwiththeLBGC/Ome⊙a- frommassivestarsincreasesrelativetotheoxygenyield,with surements. However,wenotethatthecorrespondingO/Hval- increasingmetallicity(Maeder 1992). Thesemodelssuccess- ues are consistent with – if towards the low-end of the con- fullyreproduceboththeobservedlocaltrendinC/OwithO/H, fidence interval of – the O/H metallicity determinations for and also C/O abundance gradients in Galactic stars and H II LBGs based on the ratio of rest-frame optical nebular [O II], regions (e.g. Carigi 2000). While the rise of C/O with O/H [O III], and Hβ emission lines (∼0.1- 1×(O/H) ) (Pettini inlocalstar-formingregionsismainlyduetoyieldsfrommas- et al. 2001).8 The range in log(O/H) deduced by⊙Pettini et sivestars,thereisalsothefactthatoxygenisprimarilysynthe- al. (2001)ismarkedbytheverticalcyanshadedregioninFig- sized instarswith M>10M , whereascarbonisproducedin ure 4. In view of the large uncertainties associated with both both high and intermediate m⊙ass (2- 8M ), delaying in time C/OandO/HdeterminationsinLBGs,wedonotinterpretthese the ejection of some fraction of carbon in⊙to the ISM, relative resultsanyfurther,excepttopointoutthatthecomparisonofC tooxygen.Inrelativelyyounggalaxiesathighredshift,theob- and O emission strengths will be an interesting diagnostic of 7Thesolaroxygenabundanceislog(O/H) =- 3.26(Holweger 2001) 8Thegalaxieswithrest-frameopticalspectr⊙oscopicobservationsweredrawnfromthebrightendoftheLBGUVluminosityfunction.Itisnotclearhowthisselection criterionlimitstherangeofmetallicitiesprobed,relativetotheabundancerangeintheLBGsampleasawhole. 10 chemical evolution and stellar populations at high-redshift in system is as large as the region emitting in Si II and Si II*, future,higher-qualitydata. the sum of the resonant and fine-structure emission equiva- lentwidthsshouldbeequalin strengthto the absorptionfrom 4.3.3. SiII*Lines the same transitions. This is clearly not the case, in that the Si II *λ1265,1309,1533 fine-structure emission lines are an In additionto C III]andO III],whichhave beenstudiedin order of magnitude weaker than the Si II λ1260,1304,1526 local star-forming regions, we also detect lines at which we resonant absorption lines. We don’t even detect any resonant identify as Si II* fine-structureemission lines. Unfortunately, SiIIemission,assaturatedblue-shiftedSiIIabsorptiondomi- our spectral resolution is too coarse to determine if we detect natesoveremissionat∆v=+100kms- 1,themeanvelocityof CII*λ1335Åemissionaswell(theCIIλ1334resonanceab- the Si II* emission lines. The dominanceof resonantabsorp- sorption line from outflowing neutral gas swamps any signal tionoverfine-structureemissionmaybeduetodustattenuation atthatwavelength). Aliteraturesearchoflocalultravioletob- of Si II photons during resonant scattering. It also may indi- servations reveals very few references to Si II* fine-structure cate thatthe slit used in LBG spectroscopicobservationsonly emission lines. The International Ultraviolet Explorer (IUE) subtends a small fraction of the Si II* emitting region. One atlasofstar-forminggalaxiescompiledbyKinneyetal. (1993) problem with interpreting the Si II* emission lines as being containsa census ofultravioletemission linesassociated with produced in the outflowing gas is their kinematic properties. nebularobjectssuchasHIIregions,planetarynebulae,andsu- The outflow is optically thin to Si II* (negligible absorption pernovaremnants,butdoesnotincludetheSiII*features. The is detected in these transitions), so we expect Si II* emission spectral resolutionof IUE is ∼1000 km s- 1, coarser than the linesproducedinoutflowinggastoprobethefullrangeofap- resolutionoftheLBGcompositespectra. WhentheLBGcom- proaching and receding velocities (≥1000 km s- 1). In fact, posite spectra are smoothed to the resolution of the IUE data, theSiII*linesarebarelyresolvedinthecompositespectrum, theSiII*linesarestillvisibleat5%to10%abovethecontin- withFWHM≤500 kms- 1. WhileSiII*emissionmaybebi- uum level. Presumably, the individual IUE spectra contained ased towards positive velocities by blue-shifted fine-structure, in theKinneyetal. (1993)atlasareof insufficientS/N to see or even broad resonance, absorption, we expect that it should thesefeatures.However,theSiII*featuresarealsonotdetected beatleastasbroadastheLyαemission,whichisattenuatedon in compositestarburstspectracontaining20individualgalaxy theblueedgebymuchstrongerabsorptionoverawiderrange spectradrawnfromtheIUEatlas,eachofwhichhascontinuum of velocities. However, the Si II* lines are narrower than the S/Nofatleast10(Heckmanetal. 1998). Lyαemissionline,whichextendstomuchmoreredshiftedve- The Einstein A coefficients associated with the Si II* tran- locities. Insummary,boththeoutflowandHIIregionmodels sitionsrangefrom108- 109 s- 1, morethansixordersofmag- of the Si II* features fail to explain all of the observed prop- nitude larger than the A-values for the semi-forbidden O III] erties of the emission lines, whose true nature remains to be andCIII]transitions. Onlyinveryhigh-densityenvironments determined. (N =109- 1013cm- 3)aretheSiIIexcitedlevelpopulationsde- e terminedbycollisionalexcitationsandde-excitations(Keenan 5. LBG SPECTROSCOPIC TRENDS et al. 1992). In H II regions, where the electron densities are typically N =102- 103 cm- 3, collisions are therefore not We have characterized the basic features of the composite e spectrum formed from the z∼3 LBG spectroscopic sample. thedominantmechanismgoverningtheSiIIlevelpopulations. Wearenowreadytoexaminehowthesespectroscopicfeatures Also, when T ∼104 K and Si II and Si III have comparable varyacrossthesampleasfunctionsofdifferentgalaxyparam- abundances(as appropriatehere), it can be shown that the re- eters. Some of the parameterswhich can be measured for in- combination rate of Si III into the excited Si II levels is of dividualgalaxiesare: redshift,z;rest-frameUVapparentmag- the same order as the Si II collisional excitation rates (Shull nitude,R; rest-frameUVcolorcorrectedforIGMabsorption, & van Steenberg 1982). In order to understand the origin of (G- R) , whichcanbeparameterizedin termsofa reddening, theSiII*emissionlinesinamoresystematicway,wemodeled 0 E(B- V), given an assumed form for the intrinsic spectrum; allobservedLBGnebularemissionlinesusingtheCLOUDY96 Lyα rest-frame equivalent width, W ; and interstellar kine- softwarepackage(Ferlandetal1998). Theobservationalcon- Lyα straints are the average rest-frameoptical strengthsof [O III], matics, ∆vem- abs. Our spectroscopic sample is large enough thatwecandividethetotalsampleintoseveralsubgroupsbased [O II], and Hβ (Pettini et al. 2001), as well as the rest-frame on each of the above parameters, and still create a high S/N UV O III] and C III] strengths(section 4.3.2). We foundthat compositespectrumforeachsubgroup. Individualspectraare anymodelwhichprovidesasatisfactoryfittotheO,C,andHβ not of sufficient S/N to be able to reliably measure low- and lineratiossimultaneouslypredictsSiII*emissionlinestrengths high-ionizationinterstellar absorption equivalentwidths, W which are more than an order of magnitude weaker than ob- LIS andW . Therefore, we do not bin the sample according to served. Thisresultseemstoexcludeanorigininphotoionized HIS interstellarabsorptionlinestrength,butwecanmeasurethein- HIIregionsfortheSiII*emissionlines. terstellar absorption strengths with high significance in all of An alternative explanation for these lines is that they are thecompositespectra. produced in the large-scale outflows in LBGs. UV photons with λ=1260,1304,1526 are absorbed by ground state Si II 5.1. SelectionEffects intheoutflowingneutralgas,asindicatedbythesaturatedSiII resonance absorption transitions. Each photon absorbed in a BeforeconsideringLBGspectroscopictrends,wemustiso- resonance or fine-structure transition is re-emitted in a tran- late which parameters are sensitive to the variance in the un- sition either to the ground state or the excited ground state, derlyinggalaxypopulationandwhicharemoresensitivetoour with the relative probabilities determined by the Einstein A- photometric and spectroscopic selection criteria. To illustrate coefficientsforthedifferenttransitions. Intheabsenceofdust the importanceof selection effects, we considerhowourpho- extinction, and if the slit used to observe the galaxy/outflow tometricandspectroscopicbiaseslimittherangeofgalaxypa-