ebook img

The GLARE Survey II. Faint z=6 Ly-alpha Line Emitters in the HUDF PDF

0.4 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview The GLARE Survey II. Faint z=6 Ly-alpha Line Emitters in the HUDF

Mon.Not.R.Astron.Soc.000,1–14() Printed5February2008 (MNLATEXstylefilev2.2) z ≈ 6 α The GLARE Survey II. Faint Ly- Line Emitters in the HUDF Elizabeth R. Stanway1⋆, Andrew J. Bunker2,3, Karl Glazebrook4, Roberto G. Abraham5, James Rhoads6, Sangeeta Malhotra6, 7 David Crampton7, Matthew Colless8, Kuenley Chiu,3,4 0 0 1 Astronomy Department, Universityof Wisconsin-Madison, Sterling Hall, Madison, WI, 53726, USA 2 2 Institute of Astronomy, Madingley Road, Cambridge, CB3OHA, UK 3 School of Physics, University of Exeter, Stocker Road, Exeter, EX44QL, UK n 4 Department of Physics and Astronomy, John Hopkins University,3400 N Charles St, Baltimore, MD 21218, USA a 5 Dept of Astronomy & Astrophysics, Universityof Toronto, 60 St. George St, Toronto, ON, M5S 3H8, Canada J 6 Arizona State University, Department of Physics & Astronomy, Box 871504, Temp, AZ85287, USA 8 7 Dominion Astrophysical Observatory, 5071 W Saanich Rd, Victoria, V9E 2E7, Canada 8 Anglo-Australian Observatory, P.O.Box 296, Epping, NSW 1710, Australia 1 v 1 1 Accepted .Received;inoriginalform 2 1 0 ABSTRACT 7 The galaxy population at z ≈ 6 has been the subject of intense study in recent 0 years, culminating in the Hubble Ultra Deep Field (HUDF) – the deepest imaging / survey yet. A large number of high redshift galaxy candidates have been identified h p within the HUDF, but until now analysis of their properties has been hampered by - thedifficultyofobtainingspectroscopicredshiftsforthesefaintgalaxies.Our“Gemini o Lyman-AlphaatReionisationEra”(GLARE)projecthasbeendesignedtoundertake r spectroscopic follow up of faint (z′ < 28.5) i′-drop galaxies at z ≈ 6 in the HUDF. t s In a previous paper we presented preliminary results from the first 7.5 hours of data a from GLARE. In this paper we detail the complete survey.We have now obtained 36 : v hoursofspectroscopyonasingleGMOSslitmaskfromGemini-South,with aspectral Xi resolutionofλ/∆λFWHM ≈1000.WeidentifyfivestrongLyman-αemittersatz >5.5, anda furthernine possible lineemitters withdetections atlowersignificance.We also r ′ a placetightconstraintsontheequivalentwidthofLyman-αemissionforafurtherteni- dropgalaxiesandexaminetheequivalentwidthdistributionofthisfaintspectroscopic sampleofz ≈6galaxies.Wefindthatthefractionofgalaxieswithlittleornoemission issimilartothatatz ≈3,butthatthez ≈6populationhasatailofsourceswithhigh rest frame equivalent widths. Possible explanations for this effect include a tendency towardsstrongerlineemissioninfaintsources,whichmayarisefromextremeyouthor lowmetallicityintheLyman-breakpopulationathighredshift,orpossiblyatop-heavy initial mass function. Key words: 1 INTRODUCTION around6withthez′-bandfilterandthei′-dropLymanbreak selection technique(e.g.Bunkeret al.2004),theHUDFex- TheHubbleUltraDeepField(HUDF,Beckwith et al.2006) plored the end of the reionization epoch signaled by the openedanewwindowontheearlyuniverse.Theexception- Gunn-PetersonHIabsorptiontroughinQSOs(Becker et al. ally deep, multiwavelength data provided opportunities to 2001). studycolour-selected samplesof highredshift galaxy candi- Nonetheless, the use of the Lyman break colour se- dates with modest luminosities more typical of the general lection criterion to isolate star-forming galaxies at a spe- galaxy population. Previously, only the most luminous ‘tip cific redshift, first developed to study z ≈ 3 galaxies of the iceberg’ had been accessible. By pushing to redshifts (Steidelet al. 1995) and extended to z ≈ 6 candidates in analysis of the Great Observatories Origins Deep Sur- ⋆ Current Address: H H Wills Physics Laboratory, Bristol, BS8 vey (GOODS) and HUDF fields (e.g. Stanway et al. 2003; 1TL,UK Bouwens et al. 2006), presents challenges. Before the prop- 2 E R Stanway et al. erties of the population can be meaningfully discussed, the selectionfunctionitselfmustbeunderstood.Estimatesmust 40 beobtainedofthecontaminantfractionandtheredshiftdis- tribution of i′-drop galaxies. Many colour-selected galaxies are significantly fainter GLARE than the conventional spectroscopic limit of today’s large 45 telescopes.TheHUDFreacheslimitsofz′ =28.5(10σ)– asodneapbtlheethxpato,suunretitliJmWesStToboebctoamineasahvigahilasbiAgleBn,arl-etqou-inroeisseunraretaio- Dec -27 +50 HUDF (S/N) spectrum for the continuum. However, line emission (e.g.Lyman-α)maybewithinreachofultra-deepmoderate- dispersion spectroscopy on 8-10m telescopes even for the faintestgalaxies,providedtheequivalentwidthofthelineis 55 GOODS-S large enough. If the properties of the star-forming popula- tionatz≈6aresimilartothoseoftheLymanBreakGalaxy (LBG) population at z ≈ 3 (Steidelet al. 2003), then half 60 40 20 0 RA 03 32 + the sources are expected to show the resonant Lyman-α transitioninemission.DetectionofaLyman-αemissionline Figure1.ThegeometryoftheUDFandGOODS-Sfields(taken fixes the redshift of a source, while detection or constraints at two position angles), and the layout of the GLARE mask. onnearbyhighionisationemissionlinescanquantifycontri- GLAREencompassestheentireUDFfield,andsomepartsofthe bution by an AGN. Furthermore, spectroscopy can enable GOODS-Sfield theidentificationoflower-redshiftcontaminantsinthesam- ple which will not emit Lyman-α photons at these wave- lengths, but may show other emission lines. When a pop- ing low equivalent width limits on Lyman-α emission for ulation of Lyman-α emitters is studied, the distribution of a uniformly-selected sample of i′-drop Lyman-break galax- equivalentwidthsissensitivetothestellarinitialmassfunc- ies(Stanway et al.2003;Bouwens et al. 2004;Bunkeret al. tion (since the Lyman-α transition is excited by emission 2004). The i′-drop colour selection is sensitive to galaxies from hot, massive, short-lived stars), and also to the frac- at 5.6 < z < 7.0, although the selection sensitivity falls tion of neutral gas in the intergalactic medium (IGM) and off rapidly above z ≈ 6.5 (see Stanway et al. 2003). With theemergent photon fraction (Malhotra & Rhoads2004b). this data we aim to probe galaxies at the end of the epoch In this paper we present results from the Gemini ofreionisation, andcomparetheirpropertiestocomparable Lyman-Alpha at Reionisation Era (GLARE) project. This galaxysamplesatlowerredshift,inordertoexplorepossible program used the 8-m Gemini South telescope to obtain evolution in the nature of the star-forming galaxy popula- extremely deep spectroscopy on a single slitmask, centered tion. ontheHUDF(Figure1).Byobtainingextremelylongexpo- Initial results obtained from this survey were reported suresusingatelescopewithalargecollectingarea,weaimed in Stanway et al. (2004b), which presented 7.5hour spec- tostudycontinuum-selectedgalaxiesfainterthanthosetar- tra of three Lyman-α line emitters observed on the 2003 getedbyanyothersurvey,andtoquantifythelineemission GLAREslitmask. Selection of targetsfor our2003 observa- propertiesofouri′-dropsampleintheHUDF.Wepresented tions was made possible by the early release of a list of red initialresultsfromthisprogrammeinStanway et al.(2004b, sources in the HUDF, based on observations to one third hereafterPaperI),whichwasbasedon7.5hoursofon-source of its final depth, and supplemented by i′-drop galaxies se- exposure taken at low spectral resolution (λ/∆λ≈500). In lected from the shallower, but wider field, GOODS survey thispaperwepresent an analysis of thetwenty-fouri′-drop of thesame region (Stanway 2004). selected 5.6 < z < 7.0 candidates targeted for 36hours of Between observations of our 2003 GLARE mask, and spectroscopy at higher resolving power (λ/∆λ ≈ 1200). In thestartofsemester2004B,thefullHUDFdataweremade section 2 we describe the GLARE program, and in section public and the GLARE slitmask was redesigned accord- 3 we present the results of our observing campaign. In sec- ingly. Sources with (i′ −z′)AB > 1.3 and zA′B 6 28.5 in tion 4 we analyse the equivalent width distribution of the the HUDF imaging (i.e. the catalog of Bunkeret al. 2004) GLARElineemitters,andinsection5wediscusstheimpli- were assigned the highest priority for spectroscopic follow- cations of this distribution in the context of other work in up and fifteen such sources were placed on the slitmask. this field.Finally in section 6 we present our conclusions. The agreement is excellent between the HUDF i′-drop dis- Weadoptthefollowing cosmology:aflatUniversewith covery catalog of Bunkeret al. (2004), which we use as our ΩΛ = 0.7, ΩM = 0.3 and H0 = 70h70kms−1Mpc−1. All GLAREsourcelist,andtherecentdatapaperbytheHUDF magnitudes are quoted in the AB system (Oke & Gunn team (Beckwith et al. 2006). All the robust i′-drops tar- 1983). getted on the GLARE slitmask were are also identified by Beckwith et al. In order to maximise use of the mask, additional ob- 2 THE GLARE PROJECT jects were targetted. Seven slits were placed on candidates (including the two emitters identified in Paper I, GLARE 2.1 Candidate Selection 3001&3011)fromabrighterselectionwith(i′−z′)AB >1.3 TheobjectivesoftheGLAREprojectweretodeterminethe and z′ 6 27.5, selected from the GOODS imaging (Stan- AB redshift and line emission properties of our targets, reach- way 2004), primarily at the edges of the slitmask and ly- Faint z ≈ 6 Ly-α Line Emitters in the HUDF 3 ID NoofSlits Description sampledthetypicalseeingandspectralresolution,theimage wasbinnedat 2×2pixelssoastoreducethereadoutnoise 1XXX 15 HUDF,i′−z′>1.3,z′628.5 and improve theS/N. After this binning, the spectral scale 2 HUDF,otherredsources 2XXX 5 AlignmentStars was0.94˚A/pixel,and0.146arcsec/pixelspatiallyonthede- 3XXX 7 GOODSi′−z′>1.3,z′627.5 tector.Theslitwidthwas1.0arcsec,whichproducedaspec- 4XXX 2 GOODSv−i′ selected, z=5cands tralresolutionof6.5˚AFWHMforobjectswhichfilltheslit. 5XXX 2 Known,lowz,[OII]emitters Both mask and slits were oriented due north. The higher 6XXX 6 Blanksskyslits spectralresolutionofthe2004GLAREmask(λ/∆λ≈1200 7XXX 3 GOODSmarginali′-dropcandidates comparedwith500for2003GLARE)decreasesthefraction ofthewavelength rangeadverselyaffected byOHskylines, 42 TotalNumberofSlitsonMask and also enables us to better study the profiles of emis- Table 1.Thecomposition of the 2004 GLARE slitmask.1XXX sion lines from the targets. The OG515 filter was used to indicatesaGLAREidentifiernumberintherange1000-1999,etc. suppresssecondorderlightfromshorterwavelengths.Inor- der to fill CCD chip gaps and ensure full wavelength cov- erage in the range λ ≈ 7000−10000˚A three different cen- ing outside the HUDF region. A further three slits were tral wavelength settings were observed (8580˚A, 8700˚A and placed on candidates with colours lying marginally out- 8820˚A). The shortest wavelength reached was 6420˚A, while side of our GOODS selection criteria, either in colour (i.e. thelongest wavelenth surveyedwas 10950˚A. 1.0<i′−z′<1.3)ormagnitude(i.e.z′>27.5),orinnoisy Wavelengthswerecalibratedfromthenightskylinesin regionsoftheGOODSimages,andtwoslitsonsourcesfrom each slit, leading toasolution with an rmserrorof approx- the original UDF early release list of red sources that did imately 0.3˚A. Fluxes were calibrated from the broadband not meet our final criteria. Two slits were placed on z ≈ 5 magnitudesofthealignmentstarsonthemask,andchecked v-band dropout candidates, and two on known low redshift against both line emission of known [OII] emitters also ob- [OII]3727˚A emitters, previously observed by VLT/FORS2 served in VLT/FORS2 spectroscopy that had been placed (Vanzella et al. 2006) and used as a check on flux calibra- on our mask for verification purposes, and existing spec- tion.Finally,fiveslitswereplacedonalignmentstarstoen- troscopy for the known z =5.83 Lyman-α emitter GLARE sureaccuratepositioningofthemask,andsixslitswereused 1042 (Stanway et al. 2004b,a;Dickinson et al 2004). Wees- toconductablankskysurveyforserendipitoussources.The timate a 20% error on the flux calibration, associated with composition of the2004 GLAREslitmask is summarised in slit losses and centroiding uncertainty in thenarrow slits. table 1. To optimise the subtraction of night sky emis- Both the z′ = 27.5 cut applied in the GOODS data AB sion lines (which occupy a large fraction of the spec- and the z′ = 28.5 cut applied in the HUDF correspond AB trum at > 8000˚A) we used the instrument in ‘nod to a signal to noise of approximately 8. We chose to work & shuffle’ mode (Glazebrook & Bland-Hawthorn 2001; well above the detection limit in order to have confidence Abraham et al. 2004). Each exposure was 30 minutes long, in the reality and natureof our candidatesources (many of noddingevery60seconds.Henceweareabletosuppresssky which are detected only in this one band). In the event of emission that varies on timescales longer than one minute. non-detectioninthei′-bandweusethelimitingmagnitudes The total exposure time on this mask was 36 hours. i′ =29.15(GOODS)andi′ =30.4(HUDF)correspond- AB AB The reduction of nod & shuffle data involves the sub- ing to 3σ variation in the sky background, as measured on traction of positive and negative spectra, observed in al- the images. All sources were required to be undetected in ternate 60 second exposures and offset spatially by the theavailableb(F435W) imaging butfaint detectionsinthe telescope nodding. We employed slits 2.47 arcseconds in deep v (F606W) filter (which lies above the Lyman limit length, nodding by 1.25 arcseconds between sub-exposures. at z ≈ 6, and which is present in several spectroscopically Our queue scheduled observations were constrained by the confirmedz>5.6galaxies)werepermitted.Completelyun- requirement that the seeing was less than 0.5 arcseconds resolvedsourceswereomittedfromtheselectionalthoughat FWHM,andthenoddistancesetsuchthatthesignalswere faint magnitudes, the dividing line between unresolved and separatedbyatleasttwicetheseeingdisk.Hencethecharac- slightly-resolved sources becomes blurred. teristicsignal ofan emission line comprises a‘dipole’ signal Of the line emitter candidates presented in section 3, of positive and negative emission at the same wavelength, only GLARE 1042 and GLARE 1040 lie within the NIC- spatially offset by 1.25 arcseconds. In visually identifying MOS HUDFfield (Thompson et al. 2005). Both have near- candidate line emitters, we looked for this dipole signature infrared colours consistent with a high redshift interpreta- withpositiveandnegativechannelsofcomparablestrengths; tion, as indeed did all the i′-drop sources in this field dis- this requirement effectively increases the sensitivity of the cussed byStanway et al. (2005). spectroscopy beyond the formal limit, since random back- ground fluctuations are unlikely to produce simultaneous signals in both positive and negative channels. 2.2 Observations and Data Analysis Dipolesignalslyingunderstrongemission skylinesare The2004GLAREslitmaskwasobservedinsemester2004B treated with caution; as well as having more Poisson noise, using the GMOS spectrograph on Gemini-South (Hook et skylinevariabilityandchargediffusionattheredendofthe al.2003).Thismaskwasobservedathigherspectralresolu- spectrum may lead to a spurious signal. tion thanthe2003 GLAREmask,usingtheGMOS-SR600 A number of charge traps and CCD artifacts were grating rather than the R150, blazed at an angle of 48.5◦ also masked when theindividual exposures were combined. giving a central wavelength of ∼8700˚A. As the CCD over- These charge traps affect some regions of the CCD more 4 E R Stanway et al. severely than others, and so the noise properties vary from ofthisspectroscopy,wecanruleoutalowredshiftinterpre- slitlet to slitlet, and also with wavelength. However, the tationforthesesourcesandconfirmthemasz ≈6Lyman-α 1 sigma rms pixel-to-pixel variation in the background at emitters.ThespectralresolutionoftheGMOSconfiguration 8500˚Awas1.4×10−19ergcm−2s−1˚A−1 forthe2×2binned used in our 2004 GLARE mask is sufficient to resolve the pixels, measured between skylines on the nod & shuffle [OII]λrest =3727,3729˚Adoublet(anddoessointhecaseof background-subtracted spectra. Hence, our sensitivity to the two known [OII] emitters on our 2004 GLARE mask), an emission line extending over 500kms−1 × 1arcsec is and we would expect to identify weaker emission lines such 2.5×10−18ergcm−2s−1 (3σ combiningbothnodpositions, as [NII], [SII] within our observed redshift range if the de- or 2σ per nod channel). tected line was Hα (λrest=6563˚A). The other strong opti- cal emission lines (Hβ4861, [OIII]4959,5007) would always yield other strong lines within our spectral range. Sources at these low redshifts are also unlikely to satisfy the colour 3 LINE EMITTERS IN THE 2004 GLARE cut criterion used for target selection. MASK All five detected strong Lyman-α emission lines also show significant asymmetry as expected for high redshift Inthe36hourexposureofthe2004GLAREmask,weiden- emitters (in which the blue wing of the line is significantly tify five strong Lyman-α emission line sources (including self absorbed, and the red wing broadened by re-emission). the three candidates tentatively proposed in Stanway et al. This phenomenon is well known at z ≈ 3 and is believed 2004b).Weidentify afurtherfoursourceswhich havelower to arise in outflows from the actively star-forming galax- significancedetections,butwhichareconsideredpossiblyto ies (Adelberger et al. 2003). Similar asymmetry has now be real emission lines due to their dipole natures and sep- been observed in all z > 5.5 galaxies confirmed to date by aration from possible sky line residuals. Finally we identify spectroscopy that resolves the Lyman-α emission line (e.g. fivesources with tentativeemission line detections that are Bunkeret al.2003),suggestingthatsimilaroutflowsarepro- considered unlikely to bereal. duced at higher redshift galaxies. 3.1 Strong Emission Lines 3.2 Possible Emission Lines Table 2 lists the properties of the five z ≈ 6 sources Weidentifyafurtherfoursources(listed intable3) aspos- with strong Lyman-α line emission. Figure 2 presents the siblelineemitters.Inmostcases,thedetectionineachchan- two dimensional spectra obtained for these sources, and nelisoflowsignificance,butthecoincidenceofpositiveand the summed flux from positive and negative spectral chan- negative signals suggests that the emission lines are real. nels. We also placed two known lower-redshift galaxies on Alternatelysomecandidatelinesmaylieon topofskylines, the 2004 GLARE mask, placing [OII] within our spec- leading to the suspicion that these represent skyline resid- tral range, as a check on our sensitivity and calibration. uals. These sources are illustrated in figure 3. All but one These are galaxies GDS J033235.79-274734.7 (z = 1.223) of the candidate emission lines lie shortward of the lower and GDS J033229.63-274511.3 (z = 1.033) from the ESO redshift limit selected by the i′-drop technique. Sources at VLT/FORS2 survey of Vanzella et al. (2005). We detect these redshifts would be expected in a v-drop rather than the [OII] emission at λ = 8285˚A&7577˚A, with line fluxes i′-drop selection. It is possible for sources with large errors 1.53&1.52 ×10−17ergcm−2s−1, respectively. on their i′−z′ colour to scatter into the i′-drop selection, FortheLyman-αdetections,GLAREsources1042and although it is unlikely that photometric scatter alone could 3001 were previously identified as line emitters in Paper explain this discrepancy. I from the 2003 GLARE mask, Stanway et al. (2004b). Three of these sources were identified in the HUDF i′- GLARE 3011 was tentatively identified as a line emitter, drop sample, one (GLARE 3000) from the GOODS sam- and this identification is now strongly confirmed. GLARE ple.GLARE3000wasfirstidentifiedasani′-dropsourcein 1054 and 1008 are presented for the first time here. Stanway et al.(2003,,SBM03#05) butconsideredlikelyto Since itsdiscovery (SBM03#1 in Stanway et al. 2003) be an M or L class Galactic dwarf star, as it is unresolved GLARE 1042 at z = 5.83 has been spectroscopically in HST imaging. The candidate emission line in this source confirmed (Stanway et al. 2004a,b; Dickinson et al 2004, – lies beside a strong sky line residual. FORS2 spectroscopy SiD002) and extensively studied, including in the infrared of this source by Vanzella et al. (2004) also interpreted the with Spitzer (Eyles et al. 2005; Yan et al. 2005, – #1ab). spectrum obtained as that of a Galactic star. The stellar GLARE1054wasidentifiedfromthei′-dropselectioninthe identification is almost certainly correct, given the broad- HUDF,GLARE3001and3011fromthesomewhatbrighter bandcoloursandunresolvedhalflightradiusofthissource, GOODS selection. GLARE 1008 is a source selected from suggesting that possible emission lines at this significance theinitialearlydatareleaselistofredsourcesintheHUDF should be considered highly suspect. survey. It lies outside the final overlap region of the HUDF Therearetwopossibleemissionlinesofsimilarstrength whichhasbeenusedforcatalogueconstructionandanalysis, in GLARE 1067 (at 7037˚A & 7099˚A), rendering it unlikely butwithinthenoisyouterregionsoftheHUDFmosaic.Itis thatthissourcelies at high redshift.Theobservedline sep- technicallybelowthedetectionlimit oftheGOODSsurvey, aration may be consistent with [OIII] emission (λrest = although it is faintly detected in the GOODS z′-band. It is 4959,5007˚A) at z = 0.418, although a galaxy at this red- faintly detectedinthez′-bandin theshallower edgesof the shiftispredictedtohavecoloursthataresignificantlybluer HUDF. (i.e. i′ −z′ < 0.5 and detected in the b band, as opposed Giventhehigherresolutionandmoresensitivefluxlimit to the observed i′−z′ = 1.4±0.2 and no b detection). In Faint z ≈ 6 Ly-α Line Emitters in the HUDF 5 (a) (a) (a) 8280 8300 8320 8340 8360 8400 8420 8440 8460 8480 8640 8660 8680 8700 8720 (b) (b) (b) 8280 8300 8320 8340 8360 8400 8420 8440 8460 8480 8640 8660 8680 8700 8720 Flux / x 10 erg cm s A-19-2-1-1 -111-055050(c) 1042 Flux / x 10 erg cm s A-19-2-1-1 -111-050550(c) 1054 Flux / x 10 erg cm s A-19-2-1-1 -111-050550(c) 1008 8280 8300 8320 8340 8360 8400 8420 8440 8460 8480 8640 8660 8680 8700 8720 Wavelength / Angstroms Wavelength / Angstroms Wavelength / Angstroms (a) (a) 8220 8240 8260 8280 8300 8400 8420 8440 8460 8480 (b) (b) 8220 8240 8260 8280 8300 8400 8420 8440 8460 8480 Flux / x 10 erg cm s A-19-2-1-1 -111-055050(c) 3001 Flux / x 10 erg cm s A-19-2-1-1 -111-050550(c) 3011 8220 8240 8260 8280 8300 8400 8420 8440 8460 8480 Wavelength / Angstroms Wavelength / Angstroms Figure 2. One and two dimensional spectra of the strong emission line candidates. The upper panel (a) shows the two dimensional spectrum,thecentralpanel(b)aspectrumsmoothedoverthreepixels,andthelowerpanel(c)theonedimensionalcombinedspectrum extractedfrombothpositiveandnegativechannels.ThesourceIDisshownontherightofpanel(c)ineachcase.The1sigmastandard deviation in the background, smoothed over 15˚A and including the Poisson noise on the emission line flux is indicated with a dotted line.Thearrowindicatesthewavelength oftheputative emissionline. ID RA&Declination Alternate z zA′B i′−z′ LineFlux WrLeystα LLyα J(2000) ID ergcm−2s−1 ˚A 1042ergs−1 1042† 033240.01-274814.9 20104 5.83 25.35±0.02 1.64±0.04 1.58×10−17 22.9 5.90 1054 033233.20-274643.3 42414 5.93 27.65±0.07 1.45±0.17 0.68×10−17 120 2.63 1008 033247.97-274705.1 6.13 28.51±0.18 >1.35 0.43×10−17 159 1.80 3001 033246.04-274929.7 5.79 26.69±0.06 1.67±0.20 0.77×10−17 44.1 2.83 3011 033243.18-274517.6 5.93 27.47±0.12 1.86±0.50 1.13×10−17 242 4.39 † GLARE1042hasapanoplyofalternatenames.ItisSBM03#3inStanwayetal.(2003),20104in(Bunkeretal.2004)andis SiD0002,thez=5.83lineemitterofDickinsonetal(2004). Table2.Stronglineemittersonthe2004GLAREmask.Errorsonlinefluxesandequivalentwidthsareapproximately20%.Equivalent widths are calculated from broadband magnitudes, accounting for line contamination and Lyman-α forest blanketing. 2σ limits on magnitudes aregiven where appropriate. Alternate ID is taken from Bunkeretal. (2004). Note that GLARE 1042 lies inthe chipgap ofoneofthethreewavelength settingsobserved,andhenceinaregionwithslightlylowersignaltonoise. addition,theredward lineof this[OIII]doublet isexpected firmlyintheredshiftrangeselectedbythei′-droptechnique, tobethreetimesstrongerthanthebluewardline,whilethe and may well be a high redshift emission line, although the observedemission peaksare ofcomparable strength.Anal- detection is too weak torule out a low redshift [OII] expla- ternateexplanationmightbethattwogalaxies,separatedby nation, or to detect line asymmetry. 3000kms−1 liewithintheslit.Whilethissourcehasaclose neighbour (at αJ2000 =03:32:35.8, δJ2000 =-27:48:49, with a separation of < 1 arcsecond), the neighbouring galaxy is blue in colour (i′−z′ =0.2), and likely lies at significantly Finally, we identify signals in a further five sources as lower redshift. Therefore the most probable interpretation ‘unlikely’ emission lines. These are shown in table 4 and of these lines (if they are real) is that they represent [OIII] emissionarisingnotinthetargetedgalaxybutratherinthe figure4.Althoughthedipolesignalfromthefirsttwosources appears strong, they lie on bright skylines and so may be neighbouring source. partlyorwhollyduetoskysubtractionresiduals.Theymay GLARE 1040 and GLARE 1086 are the most convinc- also arrise as a result of charge diffusion from the adjacent ingcandidatesinthiscategory.GLARE1040, isanisolated slit. Both candidates lie at λ > 10000˚A (not far from the source that clearly drops between the i′ and z′-bands, and 10500˚ASibandgap)wherethediffusionisstrong.Thesignal isundetectedintheHUDFv-band.Thecandidateemission intheremainingcandidatesisweak.SlitGLARE6050wasa lineliesatz=5.2,justshortwardofthei′-dropselection,al- blankskyslit,and thereisnothingvisiblein anywaveband thoughitispossibleforafaintgalaxysuchasthistoscatter at the depth of the GOODS imaging of this region. The upwardsinto theselection window. possible serendipitous line emitter is offset from the centre Thecandidatelineinthefinalsource,GLARE1086,lies of theslit by approximately 0′.′2 to thenorth. 6 E R Stanway et al. (a) (a) (a) 7520 7540 7560 7580 7600 7060 7080 7100 7120 7140 7000 7020 7040 7060 (b) (b) (b) 7520 7540 7560 7580 7600 7060 7080 7100 7120 7140 7000 7020 7040 7060 Flux / x 10 erg cm s A-19-2-1-1 -111-055050(c) 1040 Flux / x 10 erg cm s A-19-2-1-1 -111-050550(c) 1067 Flux / x 10 erg cm s A-19-2-1-1 -111-050550(c) 1067 7520 7540 7560 7580 7600 7060 7080 7100 7120 7140 7000 7020 7040 7060 Wavelength / Angstroms Wavelength / Angstroms Wavelength / Angstroms (a) (a) 8620 8640 8660 8680 7360 7380 7400 7420 7440 (b) (b) 8620 8640 8660 8680 7360 7380 7400 7420 7440 Flux / x 10 erg cm s A-19-2-1-1 -111-055050(c) 1086 Flux / x 10 erg cm s A-19-2-1-1 -111-050550(c) 3000 8620 8640 8660 8680 7360 7380 7400 7420 7440 Wavelength / Angstroms Wavelength / Angstroms Figure3.Onedimensionalcombinedspectrumextractedfrombothchannelsoftheimage,andtwodimensionalspectra,ofthepossible emissionlinecandidates. Asinfigure2. Slit RA&Declination Alternate z zA′B i′−z′ LineFlux WrLeystα LLyα J(2000) ID ergcm−2s−1 ˚A 1042ergs−1 1040 033238.28-274751.3 24458 5.21? 27.51±0.07 1.60±0.17 0.17×10−17 20.2 0.49 1067 033235.83-274848.9 14210 4.84?† 28.08±0.10 1.43±0.24 0.30×10−17 81.5 0.72 4.78?† 0.31×10−17 88.0 0.73 1086 033230.14-274728.4 30359 6.10? 28.13±0.11 1.46±0.25 0.37×10−17 68.1 1.53 3000∗ 033238.80-274953.7 5.08? 25.65±0.03 >3.50 0.11×10−17 2.24 0.30 † GLARE1067hastwoemissionlines,whichareconsistentwith[OIII]atz=0.418;iftheseareinsteadLyman-α,bothareatz<5. ∗ GLARE3000ismostlikelyalow-massGalacticstar. Table3.Possiblelineemittersonthe2004GLAREmask.Asintable2.TheredshiftassumesthattheemissiondetectedistheLyman-α lineatλrest =1215.7˚A. (a) (a) (a) 9980 10000 10020 10040 10060 10260 10280 10300 10320 10340 8080 8100 8120 8140 8160 (b) (b) (b) 9980 10000 10020 10040 10060 10260 10280 10300 10320 10340 8080 8100 8120 8140 8160 Flux / x 10 erg cm s A-19-2-1-1 -111-505005(c) 1034 Flux / x 10 erg cm s A-19-2-1-1 -111-050550(c) 1030 Flux / x 10 erg cm s A-19-2-1-1 -111-050550(c) 3015 9980 10000 10020 10040 10060 10260 10280 10300 10320 10340 8080 8100 8120 8140 8160 Wavelength / Angstroms Wavelength / Angstroms Wavelength / Angstroms (a) (a) 9760 9780 9800 9820 9840 6900 6920 6940 6960 (b) (b) 9760 9780 9800 9820 9840 6900 6920 6940 6960 Flux / x 10 erg cm s A-19-2-1-1 -111-055050(c) 6050 Flux / x 10 erg cm s A-19-2-1-1 -111-050550(c) 8005 9760 9780 9800 9820 9840 6900 6920 6940 6960 Wavelength / Angstroms Wavelength / Angstroms Figure 4.Oneandtwodimensionalspectraofthe‘unlikely’emissionlinecandidates presentedintable4.Asinfigure2. Faint z ≈ 6 Ly-α Line Emitters in the HUDF 7 Slit RA&Declination Alternate z z′ i′−z′ Wrest AB Lyα J(2000) ID ˚A 1034 033239.454-274543.42 52086 7.24? 27.97±0.09 >2.43 10.6 1030 033241.184-274914.81 10188 7.46? 27.10±0.05 2.04±0.16 4.4 3015 033227.910-274941.98 5.67? 27.30±0.10 >1.85 16.7 6050∗ 033237.450-274947.60 7.05? - - >104 8005 033234.392-274533.01 4.69? 27.11±0.16 1.15±0.36 66.6 ∗ 6050isa‘blanksky’slitlet Table4.Thefiveemissionlinecandidatesclassifiedas‘unlikely’tobereal.Alimithasbeenplacedonthepossibleserendipitousemitter GLARE6050basedonnon-detection inthez′ bandtoourimagingdepthofz′ =28.5.Columnheadings asintable2 AB 3.3 Sources with No Line Emission We will discuss the equivalent width limits on our i′- drop galaxies in Section 4. Theremaining21sciencetargetsontheslitmaskshowedno evidenceforlineemissioninthewavelengthrangeobserved, λobs ≈7800−10000˚A (correspondingto5.4<z<7.2,with 3.4 Agreement with Other Spectroscopic Surveys some slit to slit variation dependingon slit location). Of these, 11 formed part of our high priority i′-drop A subset of sources in our sample have also been observed selection, with marginal targets (5 sources) and sky slits (5 spectroscopicallyaspartoftheconcertedcampaignoffollow slits) constituting the remainder. We discuss the implica- up observations to theGOODS survey. tions of thesenon-detections of line emission below. Sources in the GOODS-S field have been targeted for 8m spectroscopy by FORS2 on the VLT (Vanzella et al, It is possible that, despite optimising our experimen- 2006),byDEIMOSonKeck(Stanway et al.2004a,,Bunker tal setup for sky line subtraction, we are missing flux from etal2006)andbyGMOSonGemini(PaperIandthiswork). emission lines that fall in regions dense with skylines. The In addition, this field was surveyed with the HST/ACS noise in such areas is greater than the slit average, making grismaspartoftheGRAPESsurvey(Malhotra et al2005). identificationof linecandidatesmoredifficult.Some35% of This slitless spectroscopy has been obtained to varying the wavelength coverage of this spectroscopy lies undersky depths, and at different spectral resolutions. In particular, lines(definedasregionswhereskyemissionexceeds150%of the GRAPES grism spectroscopy is of too low a resolu- theinter-lineskycontinuumlevel)whichleavePoissonnoise tiontodetectallbutthehighestequivalentwidthLyman-α andskybrightnessfluctuationresidualsofvaryingstrength, emission lines adjacent to the Lyman-α forest continuum althoughonly10%liesunderstrongskylines(skylineemis- break (compare, for example, the DEIMOS spectrum of sion > 5 times sky continuum). The noise under the resid- SBM03#1/GLARE1042 inStanwayetal.2004showingthe uals of bright sky lines is a factor of 2.5 greater than that strong Lyman-α line, with the HST/ACS spectrum of the between lines, and might also be prone to systematics in sameobject,SiD002,infig.2aofDickinsonetal.2004,which theskysubtractionleadingtopotentiallyspuriousemission showsa continuumbreakbutthelineiswashed out).How- lines. ever,GRAPESdoesprovidearedshiftestimatebylocalising Only half of Lyman Break Galaxies at z ≈ 3 show thewavelength of the Lyman-αbreak in thespectrum. Lyman-α in emission (Shapleyet al. 2003), and one quar- Where redshifts have been determined by multiple ter have WrLeystα >20˚A. Given that we reach a limiting rest- groups, our results are generally in reasonable agreement frame equivalent width of 66˚A for the majority of our tar- with previous observations, given the difficulty in obtain- gets,wemightna¨ıvelyexpecttoobserveemissioninhalfour ing precise redshifts from the self-absorbed and resonantly- slits, less ten percentlost tostrong skyline residuals. How- broadenedLyman-αlinesatmoderatespectralresolution.In ever, the fraction of Lyman Break Galaxies with emission thecaseofGLARE1042, ourmeasuredredshift ofz =5.83 at z ≈ 6 is still unknown, and it is likely that a large frac- is in close agreement with that measured by FORS2 (z = tion of z ≈ 6 i′-drop galaxies will not be spectroscopically 5.82, Vanzella et al. 2004), by the GRAPES team (z =5.8, confirmed until the continuum level is reached with either Malhotra et al2005)and,usingKeck/DEIMOSspectra,by longerexposuresormoresensitivetelescopes, andinterstel- Stanway et al.(2004a, z =5.83).GLARE3001wasspectro- lar absorption lines can be used for redshift determination scopicallyidentifiedbyFORS2asalineemitteratz=5.78, (see, for example, Dow-Hygelundet al. 2005, for rest-frame agreeing with ourredshift of z=5.79. UV continuumspectroscopy of a bright z =5.5 galaxy). GLARE 1054 and 1030 were both placed at z = 5.7 Finally in the case of the marginal candidates used to by GRAPES spectroscopy (Malhotra et al 2005). This con- fill the slitmask, it is possible that the true redshift of the trasts with our measured redshifts of z = 5.9 for GLARE sources lies above z = 4.5 (set by the dual requirements of 1054 and (tentatively) z = 7.4 for GLARE 1030. The dis- non-detectionin theb band,andthei′−z′ colours ofthese crepancy for GLARE 1054 is within the expected level of sources), but below the limit of our spectroscopy. Two of agreement for GRAPES grism spectroscopy and so these the slits without detections were placed on v-drop galaxies results are consistent. Given thelow significance of our line expected to lie in this range, while a further four slits were candidateinGLARE1030,andthei′-dropredshiftselection placed on sources slightly too blue to meet our strict se- function, theGRAPES redshift remains themore likely. lection criteria. It seems unlikely that photometric scatter The remaining source, GLARE 3000 was identified by could place these sources in our detectable redshift range. the VLT/FORS2 observations as a Galactic star on the 8 E R Stanway et al. basis of weakly detected [OI] and [NII] lines. This source, that the emission line has not been lost behind a the sub- which corresponds to the unresolved candidate SBM03#5 traction residual of a bright night sky line. inStanway et al.(2003),falls inour‘possible’ category.We Combining our line limits with those sources for which notethattheFORSIIspectrumisflaggedwiththeirquality emission lines have been identified or tentatively proposed class ‘C’ and that there is no secure continuum detection. formsasampleof24sourcesuniformlyselectedfromtheiri′- Despite this, a Galactic star is still the most likely identifi- drop colours. The resultant distribution of Lyman-αequiv- cation of thistarget, illustrating thecaution with which we alent widthsis plotted in figure5. present our fainter line candidates. If all the sources for which candidate emission lines Noothertargetonour2004GLAREmaskhasbeenre- are identified in this paper prove to be z ≈ 6 Lyman-α portedasalineemitterbyotherteams,orhasanestimated emitters, then the escape fraction of Lyman-α photons at redshift from GRAPES spectroscopy. z ≈ 6 appears qualitatively similar to that at z ≈ 3. From our z ≈ 6 sample, 66% of the sources (16 out of 24) have Lyman-α equivalent widths < 25˚A, compared with 75% at 3.5 Limits on NV and other Emission Lines z ≈ 3 (Shapley et al. 2003), although the lower redshift sample loses a smaller fraction due to skyline contamina- Lyman break galaxies at z ≈ 3 show few other emission tion.Thesamplesprobetosimilar pointsontheluminosity lines in the rest frame ultraviolet. The composite spectrum function in both cases (approximately 0.1L∗). These frac- of ≈1000 such galaxies produced by Shapley et al. (2003) shows weak emission features due to SiII* (λrest =1265˚A, tions are consistent within the errors on our small number 1309˚A, 1533˚A), OIII] (λ = 1661˚A, 1666˚A) and CIII] statistics. Harder to explain in comparison with lower red- (λ =1908˚A), and absorption features due to stellar winds, shift galaxies is the tail stretching to very high equivalent widths (> 200˚A) observed in this survey, a trait also ob- primarily SiII and CIV. served in some narrowband selected sources at this redshift The presence of an AGN in our target galaxies could (e.g. Malhotra & Rhoads 2002) and in other i-dropout Ly- also lead to thepresence of emission lines due to high exci- tation states, primarily NV (λrest =1238.8˚A, 1242.8˚A) and man break galaxies (e.g. Dow-Hygelund et al. 2006, who SiIV (λ=1394˚A, 1403˚A), and rarely OV](λ=1218˚A). find one source with W0 =150˚A). Although the number statistics are small, weobserve four line emitters (17±8%) Foragalaxy at z =6,ourspectraextendtorest frame wavelengthsof≈1428˚A.Giventhatwearenotabletomea- withequivalentwidthsintherange50−100˚A,andafurther sureahighsignaltonoisecontinuumonanyonetarget,we fourwithW0 >100˚A(threeofwhichcomefromour‘robust’ list of line emitters). This contrasts with the Lyman Break are unable to measure absorption features in the spectra, Galaxy population at z≈3in which <5% of galaxies have and therefore confine ourselves to searching for evidence of otheremissionfeaturesinthespectraofourLyman-αemit- line emission with W0>100˚A (Shapley et al. 2003). While high redshift galaxies at both redshifts are se- ters. lected on their rest-frame ultraviolet continuum and the A careful inspection of our five good Lyman-α emis- spectralbreakcausedbyLyman-αabsorption,thetwopop- sionlinecandidatesdoesnotprovideevidenceforanyother ulations are not identical. emission lines at theredshift of Lyman-α.While this is not The sample discussed here reaches some two magni- surprisinggiventheweakness ofsecondary lines,thefailure tudes fainter than the tentatively proposed and still uncer- todetectNVinthisspectrscopysuggestsalargeLy-α/NV tain typical luminosity L∗ at z ≈ 6 (Bouwens et al. 2006; ratio. Using our 3σ limit on undetected lines as an upper Bunkeret al. 2004). This contrasts with a sample reaching constraint on NV flux we determine that f(Lyα)/f(NV)> justonemagnitudebelowL∗ atz =3.Shapley et al.(2003) [10.5,4.5,2.8,5.1,7.5](3σ) respectively for GLARE targets consideredsubsamplesatz =3,dividingtheirspectroscopic [1042, 1054, 1008, 3001, 3011] Typical line ratios for AGN data into quartiles based on rest-frame equivalent width. aref(Lyα)/f(NV)=4.0(Osterbrock1989),whilethosefor They found a modest trend in thestrength of line emission galaxies are greater than this. The limits we determine dis- with magnitude. Galaxies in their highest equivalent width favour an AGN origin to the Lyman-α flux in all but the quartile are some 0.2 magnitudes fainter on average than faintest target (in whichtheconstraint istooweak tomake theirquartileofweaktomoderatelineemission. Itispossi- a firm statement). These limits constrain AGN activity, if blethatthestrong lineemission observed hereismore typ- present,toonlyaweakcontributiontotherestframeultra- ical of sources with faint rest-frame ultraviolet continuum violet flux. suggesting that faint sources differ physically from brighter members of thepopulation. A second difference between the samples is intrin- 4 THE EQUIVALENT WIDTH DISTRIBUTION sic rather than arising from a selection effect. Several OF GLARE LINE EMITTERS authors (e.g. Stanway et al. 2005; Bouwens et al. 2006; Lehnert & Bremer 2003) have now observed that Lyman 4.1 The Observed Equivalent Width Distribution break galaxies at z > 5 have steeper rest-frame ultraviolet Using the mean variance in the background of the exposed slopes than those at z ≈3. Bouwens et al. (2006) interpret slits, and the broadband magnitudes of the targeted galax- thisasindicatingthatthedustpropertiesofthispopulation ies,weareabletocalculate limitson therestframe equiva- evolveoverredshift.Whileasteeprest-ultravioletslopecan lentwidth W0 forthosesources whichsatisfy ourcolourse- also arise dueto low metallicity or a top-heavyinitial mass lection criteria and yet are undetected in our spectroscopy. function(asdiscussedbelow)dustevolutionisanaturalin- Thesearepresentedintable5.Ineachcaseweassumethat terpretation. At z ≈ 6 the universe is less than 1 gigayear the galaxy lies at the mean i′-drop redshift (z = 6.0) and old and galaxies may not have had time to develop a high Faint z ≈ 6 Ly-α Line Emitters in the HUDF 9 dustcontent.Shapley et al.(2003)foundthatz≈3galaxies withhighequivalentwidthsinLyman-αalsohadlowermean dust extinction. Lyman-α photons are resonantly scattered 15 by dust and hence the line is preferentially absorbed with respecttotherest-frameultravioletcontinuum.Ifthedistri- bution of Lyman-α equivalent widths in the z ≈ 3 popula- tion is truncated by dust absorption, this could producean 10 apparent ‘excess’ of strong lines at high redshift. However, mber even zero dust absorption cannot explain equivalentwidths Nu exceeding 200˚A unless the sources are also very young and very low in metallicity. The full explanation for the equiv- 5 alent width distribution observed in the GLARE data may well be a combination of these effects and those discussed below. Several other possible explanations exist for both the 0 0 50 100 150 200 steepening of the rest frame UV slope and thedifference in WLya(rest) equivalentwidthdistributions.Aninterpretationofcontam- inant galaxies at lower redshifts seems unlikely due to our i′-drop selection criteria; low redshift sources with strong Figure5.Thedistributionofrestframeequivalentwidthsforthe GLAREmask,dividedintobinsof25˚A.Lineemissioncandidates spectral breaks are likely to have more than one emission areshownwithsolidlines,sourceswithequivalentwidths>150˚A line in ourobserved redshift range. are shown at 150˚A, and those for which only upper and lower A high equivalent width line can arise if the observed limitsareavailable are added to the distribution to produce the Lyman-α photons excited by a population of AGN rather dottedlines. than by young, hot stars. The luminosity function of AGN ispoorlyconstrainedatthesemagnitudesandredshifts,but However at least two of our good line emission candidates thespacedensityofsuchsourcesisexpectedtobeverylow have rest frame equivalent widths exceeding 200˚A. This is (e.g., thez >6SDSSQSOs,Fanet al. 2003). Atz≈6, the possible if the galaxy is in the first few Myrs of an ongoing deep 2Ms X-ray exposure of the UDF and surrounding re- starburst,butmayalsoprovideevidenceforvariationinthe gion(Alexanderet al.2003)woulddetectonlythebrightest AGN(L>1042ergss−1cm−2).NoneoftheGLAREtargets initial mass function of star formation. High equivalent widths of Lyman-α emission can arise are detected in this X-ray observation. AGN would also be from a “top-heavy” initial mass function (i.e. star forma- expectedtoshowemissionlinesthatarenotjuststrongbut tion weighted towards a population of high mass stars). also broad, while none of the GLARE line candidates are Malhotra & Rhoads(2004b)calculatedtheLyman-αequiv- broad.Thereisalsoanabsenceofhigh-ionizationlinessuch as NV1240˚A which are common in AGN. alent width expected for a metal-enriched population with a very extreme IMF, weighted towards massive stars (i.e. The tail of line emitters extendingto higherequivalent IMFslopeα=0.5).Asisthecaseforlowmetallicity popu- widths may also arise from a difference in the fundamental lations,thefluxisweightedtowardsaharderspectrum,and properties of the stellar population between z ≈6 and that Lyman-αemissionisstrengthened.Theyfoundthatsuchan at lower redshifts. IMF could explain line equivalent widths of up to 240˚A at Modeling of emission from metal-free Population III agesofafewMyr,withhigherIMFspossibleforveryyoung galaxies predicts rest frame Lyman-α equivalent widths >1000˚Aforyoungstarbursts(<2Myr),decreasingtoW(Ly- bursts. α)∼500˚A for older bursts (Schaerer 2002). These very high The hard rest-frame ultraviolet spectrum associated withsuchanIMFmayalsobeconsistentwiththesteeprest equivalent widths arise from the relatively hard spectrum frame ultraviolet spectral slopes observed in z ≈ 6 galax- of metal-free reactions in the most massive stars. How- ies (e.g. Stanway et al. 2005). While the evidence from the ever it is unlikely that zero metallicity (population III) GLAREstudyislimited,withthenumberofhighequivalent starsarestillcontributingsignificantlytotheemissionfrom width sources small, the existence of such sources suggests massive stars almost a billion years after the Big Bang, thattheenvironmentofstarformationatz≈6islessmetal particularly given the identification of stellar populations enhanced,orweightedtowardsmoremassivestarsthanthat >100Myroldinsomez>5sources(e.g.Egami et al.2005; at z≈3. Eyles et al. 2005). Further evidence for moderate metallic- ityatz>5hasbeenobservedinthespectroscopyofbright AGN from the Sloan Digital Sky Survey. Metals including 4.2 Implications of High Equivalent Widths iron (Barth et al. 2003) and carbon (Maiolino et al. 2005) have been detected from even the highest redshift quasar The presence of a tail of large equivalent width emission (at z=6.4). linesourceshasimplicationsfortheredshiftdistributionex- By contrast, even 1/20th solar metallicity leads to a pected from an i′-drop selection. sharpreductioninthepeak(zeroage)equivalentwidthpre- The redshift distribution and number densities of i′- dictedto∼300˚A,withamoretypicalW(Ly-α)∼100˚Abyan drop galaxies haveusually been calculated assuming a neg- age of 10-100Myr (Malhotra & Rhoads 2002). Most of the ligiblecontributionfromlineemissiontothemeasuredmag- candidate emission lines presented in this study can hence nitudes and colours. In the case of line emitters with rest be explained with normal, if low metallicity, populations. frame equivalent widths W0 < 30˚A this is a reasonable as- 10 E R Stanway et al. ID Alt RA&Declination z′ i′−z′ Wrest /˚A AB Lyα 1001 48989 033241.43-274601.2 28.26±0.12 >2.1(2σ) <6.5 1004 46223 033239.86-274619.1 28.03±0.10 >2.37(2σ) <5.2 1009 12988 033238.50-274857.8 28.11±0.11 >2.29(2σ) <5.6 1045 21111 033242.60-274808.8 28.02±0.11 1.67±0.26 <5.2 1047 35271 033238.79-274710.8 28.44±0.14 1.33±0.30 <7.6 1060 11370 033240.06-274907.5 28.13±0.11 >2.28(2σ) <5.7 1077 16258 033236.44-274834.2 27.64±0.07 1.42±0.16 <3.7 3002 033243.35-274920.4 26.89±0.07 1.42±0.20 <1.8 3030 033248.94-274651.4 27.04±0.08 1.41±0.23 <2.1 3033 033249.08-274627.7 27.18±0.09 1.40±0.27 <2.4 Table5.The2004GLAREtargetsforwhichnolineemissionwasobserved.Limitsontheequivalentwidtharecalculatedusingthez′- bandmagnitudetodeterminethecontinuumlevel,the3σstandarddeviationinthebackgroundastheminimumlinefluxandaccounting forIGMabsorption,assumingthegalaxyliesatz=6.‘Alt’indicatesanalternateIDinBunkeretal.(2004) sumption, with the contribution to z′ flux amounting to a 7.0 few percent. Forlineemitterswithlargerequivalentwidthstherecan dfibgtrauleatlceataerexddi(seibis.bgey.ny.Ttizfihfihc<egiauslni6rntee.0effe)6eeff,.cmettIhcifsfteastnilhoolstnenh,ieetnamhtine′oid−ssstsiezhool′renueccreotcliileonmosnueawriffniaoutlfnhlrstcehhtgiiieniogmnhgtaehesloeqafxauiisyi′′v--baiidsllalreurnonesdpt-- n-alpha 666...468 i’-z’A>AL1N.L3YO, z WE’AWEB<D28.5 EW > 0 ALLOWED EW > 25 ALLOWEDEW > 50 ALLOWEDEW > 100 ALLOWEDEW > 150 ALLOWED ma widthsfalloutofthecolour-magnitudeselectionwindow.A y L source at our z′ detection limit with a rest frame Lyman-α of 6.2 equivalentwidthof100˚Acanbeasblueasi′−z′ =0.67(AB) hift ds andanintrinsiclinewidthof150˚Awould lead toacolourof Re 6.0 justi′−z′=0.46(AB).Thesecoloursaresimilartothoseof EW > 100 NOT ALLOWED EW > 50 NOT ALLOWED muchlower redshift galaxies. Ensuring acomplete selection 5.8 of line emitting galaxies at 5.6 < z < 6.0 is therefore im- EW > 25 NOT ALLOWED NOT possible using a simple colour-selected sample without also 5.6 NO LINE EMISSION ALLOWED ALLOWED including a great many lower redshift contaminant sources. 25 26 27 28 29 30 31 Ifthelineemissionfallsinz′filter,thei′−z′colourisen- Continuum z’ Magnitude hanced andgalaxies with continuumfluxbelow thez′ limit are promoted into the selection window. Hence at z > 6.5 the i′-drop criterion can select a population of low contin- Figure 6. The effects of Lyman-α equivalent width on the i- dropselectionfunction.Contoursshowtheequivalent widthline uum, strong line emission sources rather than the contin- requiredtoplaceagalaxywiththegivenredshiftandcontinuum uum break sources it targets. There is also a redshift range magnitude into the selection window. Solid lines indicate mini- in which the line emission would fall in the overlap region mum equivalent width required, dotted lines indicate maximum between filtersand both effects compete. equivalent width permitted. In the absence of line emission, all This effect may lead to a bimodal redshift distribution galaxies above the limiting observed magnitude and at z > 5.6 for i′-drop galaxies, with weak emission line sources prefer- aredetected. In the presence of strong line emission,galaxies at entiallyselectedtowardslowerredshift,andhighequivalent z<6.0falloutoftheselectionastheemissionlineliesinthei′- width sources selected at higher redshifts. Itis necessary to band,whilefaintergalaxiesathighredshiftcanbepromotedinto quantifytheequivalentwidthdistribution,combininggalax- the selection due to brightening of the measured z′ magnitude. Thepropertiesoftheseven i′-dropsources withredshiftsinthis ies at the faint limits explored here with brighter galaxies, rangeareshownwithdiamonds. in order to tightly constrain thegalaxy luminosity function at any given redshift. Asfigure6illustrates,ourgoodlineemissioncandidates all lie in regions of parameter space affected by equivalent missing sources with high emission line equivalent widths widthselectionbiases.Ourcandidateslieinparameterspace at the low redshift end of our survey (where our redshift theoreticallyallowedfortheirequivalentwidthwithtwoex- selection function is at its peak), then our conclusion that ceptions-GLARE1054and3011bothhavestrongemission the distribution of line emission at high redshift extends to lines in the tail of the i′ filter that should haveforced their larger equivalent widthsis strengthened. i′−z′coloursoutofourselectionfunction.Wenote,however, Thiseffecthasinterestingimplicationsfortheluminos- that that both of these sources have i′−z′ colours within ity functions presented in the literature for z ≈ 6 galaxies. 1σofourminimumselectioncutoff,andmayhavescattered Ifthei′-dropcriteriaareoverestimatingthenumberdensity up into out selection. Only one source is boosted into the of faint continuum sources (due to contamination by a tail selection function by virtue of its line emission rather than in the distribution of strong line emitters) then it is pos- continuum flux,suggesting that oursample is, as expected, sible that the faint end slope of the luminosity function is dominated by continuum-selected sources. If we are indeed shallower than hithertoreported. If thereis a large number

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.