Build-up and Evolution in the First 2 Billion of the Universe

Rychard Bouwens Leiden Observatory

Special thanks to my collaborators Pascal Oesch, Ivo Labbe, Garth Illingworth, Renske Smit, Benne Holwerda...

Santa Cruz Galaxy Formation Workshop 2014

Santa Cruz, California August 11, 2014 State-of-the-Art Infrared Instrumentation is allowing for Great Progress

WFC3 camera on the IRAC Camera Very Wide-Area Hubble Space Telescope Spitzer Space Ground-based Telescope Cameras WFC3/IR HST Spitzer HAWK-I

VISTA -- 4 arcmin field of view (6x larger than NICMOS) -- excellent sensitivity (3-4x better than NICMOS) -- excellent spatial resolution (2x higher than NICMOS) State-of-the-Art Infrared Instrumentation is allowing for Great Progress

WFC3 camera on the IRAC Camera Very Wide-Area Hubble Space Telescope Spitzer Space Ground-based Telescope Cameras WFC3/IR HST Spitzer HAWK-I

WFC3/IR increased search efficiency by 40x VISTA -- 4 arcmin field of view (6x larger than NICMOS) -- excellent sensitivity (3-4x better than NICMOS) -- excellent spatial resolution (2x higher than NICMOS) HUDF NICMOS J110+H160

144 4 HUDF WFC3/IR Y105+J125+JH140+H160

4 z > 6.5 (before WFC3/IR) (first 850 Myr of universe)

120 z > 6.5 galaxies (after WFC3/IR) (first 850 Myr of universe)

ALL FIELDS

15 z > 6.5 galaxies (before WFC3/IR)

255 orbits > 800 z > 6.5 galaxies (after 5WFC3/IR) Why are studies of galaxies at very high redshifts interesting?

-- It is when galaxies first form... (halos of L* and sub-L* galaxies built up from z~30+ to z~3)

-- It is when the universe was reionized... (galaxies are most likely driver, so by studying the formation of first galaxies perhaps we can gain insight) Focus of this Presentation:

Galaxy Growth from z~10 to z~4 What are the different regimes to study galaxy growth?

“Normal” Population of Faint Galaxies (Most in universe form here) How rapidly do faint (low mass) galaxies grow up? ⇒ Study using the Hubble Ultra Deep Field (and similar fields)

Rare Population of Bright Galaxies How rapidly do bright (massive) galaxies grow up? How Bright / Massive can Galaxies Become?

⇒ Study using very wide-area fields

What can we learn based on current wide-area fields? Full CANDELS Program (+ BORG + ERS) provides an ideal dataCANDELS set to study Observations the properties completed of the August most 2013 luminous galaxies GOODS-N 14’x10’ GOODS-N

EGS 30’x6’

COSMOS

WFC3 - 5 Orbits WFC3 - 2 Orbits 22’x8’

HUDF

UDS 22’x8’ GOODS-S Would like to use all 5 fields! 10’x13’ Grogin+ 11 SameCredit: fields Ferguson covered & CANDELSwith WFC3 team Grism in AGHAST & 3D-HST Koekemoer+ 11 CANDELS- Maximizing the NumberUDS+EGS of z>6 Sources at High Luminosity +COSMOS

z~7 Galaxies

CANDELS- CANDELS- North+South UDS+EGS + ERS +COSMOS

Ultra-Deep UDS + HUDF + Ultravista HUDF09-Ps Bowler+2014

BRIGHT FAINT UV Luminosity Bouwens+2014 Large Areas Required to Overcome Large Field-to-Field Variance Observed at High Redshift

Estimated field-to- field variance for z~4-8 samples.

Field-to-field variance is substantial, especially at high redshifts and at the bright end of the LF.

Bouwens+2014 With lots of great data sets.... what can we learn about galaxy growth?

Let’s look at the z>=4 LFs z~4-10 LFs from all CANDELS + HUDF + other legacy fields (Bouwens et al. 2014, arXiv:1403.4295, 48 pages) First Two CANDELS CANDELS CANDELS CANDELS CANDELS Frontier HUDF BORG + parallels South North UDS COSMOS EGS Fields z = 4

z = 5 Bouwens+2007

z = 6 Bouwens+2014 B14 z = 7 McLure McLure+2013 +2013 McLure z = 8 +2013 z = 10 Oesch+2014 Possible to Make Use of Whole CANDELS area?

CANDELS-North/South: Deep HST data over a contiguous wavelength range Possible to Make Use of Whole CANDELS area?

CANDELS-North/South: Deep HST data over a contiguous wavelength range

CANDELS-UDS DEEP Y DATA 27 mag (5σ) Credit: Fontana / HUGS

CANDELS-COSMOS DEEP Y DATA 26 mag (5σ) Credit: McCracken / UltraVISTA

COSMOS-EGS: BUT deep ground-based data fills in wavelength gaps (most >27 mag)! z~4-10 LFs from all CANDELS + HUDF + other legacy fields (Bouwens et al. 2014, arXiv:1403.4295, 48 pages) First Two CANDELS CANDELS CANDELS CANDELS CANDELS Frontier HUDF BORG + parallels South North UDS COSMOS EGS Fields z = 4

z = 5 Bouwens+2007 Slightly More Challenging z = 6 (deep ground-based data are useful) Bouwens+2014 B14 z = 7 McLure McLure+2013 +2013 McLure z = 8 +2013 z = 10 Oesch+2014 New determinations of UV LF at z~4, 5, 6, 7, 8, 10 from all HST Legacy Fields (Bouwens et al. 2014, arXiv:1403.4295) Highlights of Bouwens+2014:

> 800 likely z~7-8 galaxies > 11000 z~4-10 galaxies

Provide First Determination of z~10 Luminosity Function (together with Oesch+2014) How Bright Can Galaxies Become at z~9-10?

Amazingly, ~10-20x Brighter than the z~9-10 Galaxies in the HUDF

Some of our best z~10 candidates are as bright as L* galaxies found at z ~ 3 by Steidel et al. 6 Oesch et al.

TABLE 2 Flux Densities of z>9 LBG Candidates in the GOODS-N Field

Filter GNDJ-625464314 GNDJ-722744224 GNWJ-604094296 GNDJ-652258424

B435 7 ± 93± 6 −2 ± 6 −11 ± 11 V606 2 ± 7 −3 ± 5 −5 ± 52± 8 i775 5 ± 10 6 ± 76± 7 −9 ± 11 I814 3 ± 71± 5 −2 ± 80± 9 z850 17 ± 11 −7 ± 6 −7 ± 8 −14 ± 13 Y105 −7 ± 9 −7 ± 7 −2 ± 10 −18 ± 8 J125 11 ± 812± 723± 836± 9 JH140 102 ± 47 85 ± 34 ··· 86 ± 54 H160 152 ± 10 68 ± 973± 882± 11 K137± 67 −45 ± 51 85 ± 261 76 ± 55 IRAC 3.6 µm139± 20 (< 81)* 39 ± 21 65 ± 18 IRAC 4.5 µm122± 21 (< 119)* 93 ± 21 125 ± 20 Note.—MeasurementsaregiveninnJywith1σ uncertainties. * 3σ upper limit due to uncertainties in the neighbor flux subtraction. 6 Oesch et al.

0 24 10 ric redshifts with templates that include emission lines 10 GNDJ−625464314 TABLEas 2 included in the v1.1 distribution of the code .The −2 26 Flux Densities of z>9 LBG10 Candidatesbest-fit in EAZY the GOODS-N redshifts Field are all within 0.1 of the ZEBRA

AB values listed in Table 1. p(z) m −4 Filter28 GNDJ-625464314 GNDJ-72274422410 GNWJ-604094296 GNDJ-652258424 z = 10.2 (χ2 = 4.9) best 3.3. Possible Sample Contamination B435 z = 2.1 (7χ2 ±= 47.5)93± 6 −2 ± 6 −11 ± 11 low −6 V60630 2 ± 7 10−3 ± 5 As we will show−5 in± Section52 4.1, the detection± 8 of such i775 5 ± 10z 6 ± 76∼ − ± 7 −9 ± 11 GNDJ−722744224 bright z 9 10 galaxy candidates in the GOODS-N I814 3 ± 71± 5 −2 ± 80± 9 −2 dataset is surprising given previous constraints on UV z85026 17 ± 11 10−7 ± 6 −7 ± 8 −14 ± 13 LFs at z>8. A detailed analysis of possible contam-

Y105AB −7 ± 9 −7 ± 7 −2 ± 10 −18 ± 8 p(z) J125m 11 ± 812−4 ± 723ination is therefore particularly± 836 important. We± 9 discuss 28 10 JH140 z = 9.9102 (χ2 =± 9.4)47 85 ± 34 ··· 86 ± 54 best several possible sources of contamination. H160 z = 2.6152 (χ2 =± 24.2)10 68 ± 973± 882± 11 low −6 K13730 ± 67 −1045 ± 513.3.1. 85 ±Emission261 Line Galaxies 76 ± 55 IRAC 3.6 µm139± 20z (< 81)* 39 ± 21 65 ± 18 GNWJ−604094296 * Strong emission line galaxies have long been known IRACOne of 4.5 Sixµ m122Bright z~9-10 Galaxies in CANDELS± 21 (<−2119) 93 ± 21 125 ± 20 26 Bright z ∼ 9 − 10 Galaxy Candidates in GOODS-North 5 10 to potentially contaminate very high-redshift sample se-

NoteAB .—MeasurementsaregiveninnJywith1σ uncertainties. lection. These are a particular concern in datasets p(z) m −4 * 3σ28upper limit due to uncertainties in the neighbor flux subtrac10 tion. which do not have very deep optical data to establish a z = 9.5 (χ2 = 3.9) best z = 0.6 (χ2 = 16.5) strong spectral break through non-detections (see, e.g., low −06 3024 10 ricAtek redshifts et al. 2011; with templates van der Wel that et al. include 2011; emission Hayes et lines al. z 10 GNDJ−652258424625464314 as2012). included Sources in the with v1.1 extreme distribution rest-frame of the code optical.The line −2 26 z = 10.2 10 best-fitemission EAZY may redshifts also contaminate are all withinz ! 0.19 samples of the ZEBRA if the AB AB valuesz ∼ 10 listed candidate in Table UDFj-39546284 1. (Bouwens et al. 2011a; p(z) p(z) m m −4 Oeschet al. 2012a) is any guide. In that case, the 28 2 10 z = 9.210.2 ( χ(2χ = = 8.3) 4.9) best 3.3. Possible Sample Contamination 2 extremely deep supporting data did not result in any z = 2.22.1 (χ2 = 30.0)47.5) low −6 detectionAs we will shortward show in of Section the H 4.1,160 band, the detection but other of such evi- Fig. 2.— 6′′×6′′images30 of the four z ≥ 9galaxycandidatesidentifiedintheCANDELSGOODS-Ndata.From left to right, the images 10 show a stack of all optical bands, Y105, JH140, J125, H160,MOIRCSK,andneighbor-subtractedIRAC3.6µmand4.5µmimages.The stamps are sorted from high to lower0.5 photometric redshift1.0 from SED fits (indicated2.0 in the lower left, see4.0 also Table6.0 1). The0IRAC2.5 neighbor- 5 z7.5 10 12.5 brightdence (tentativez ∼ 9 − 10 detection galaxy candidates of an emission in the line GOODS-N at 1.6 µm subtraction works well for all sourcesGNDJ except for− GNDJ-722772274422444224, where the nearby foreground source is too bright, and clear residuals are visible atOesch+2014 the location of the candidate. Only IRACobserved upper limits arewavelength therefore included for [µ thism] source in the following analysis. Clearly, z all other sources show significant (> 4.5σ)detectionsinthe4.5µmchannel.Thebrightestsource(GNDJ-625464314)isalsodetected at 6.9σ in the 3.6 µmchannel.Withtheexceptionofthebrightestcandidate,which is weakly detected in the K-band (at 2σ), the MOIRCS −2 datasetand the is high surprising luminosity given of previous UDFj-39546284) constraints indicates on UV K-band data provide only26 upper limits. 10 Fig. 3.— Spectral energy distribution fits to the HST and LFsthat at anz> extreme8. A emission detailed analysis line galaxy of possible at z ∼ contam-2.2 is its flat continuum longwardAB of 1.6 µm. Taken together, the photometric redshift code ZEBRA (Feldmann et al. ∼ − these resultsSpitzer/IRAC suggest that the most likely interpretation photometry2006; Oesch of et the al. 2010b) four using GOODS-N a large library of stel- z 9 10 galaxy p(z) m inationa more is likely therefore interpretation particularly of important. the current We data discuss (see is that GNDJ-625464314 is at high redshift.left lar population synthesis template models based on the −4 Stamps of thecandidates four28 viable high-redshift ( candidates)togetherwiththeredshiftlikelihoodfunctions are library of Bruzual & Charlot2 (2003). Additionally, we 10 presented in Figure 2, and their positions and photome- added nebularz = line 9.9 and (χ continuum = 9.4) emission to these tem- severalBouwens possible et al. 2013a; sources Ellis of contamination. et al. 2013; Brammer et al. try are listed( inright Tables 1 and). 2. The measurementsplate SEDsbest and in a self-consistent their manner, upper i.e., by limits convert- (2σ)areshown As is evident from Fig. 2, the four sources are all de- ing ionizingz photons = 2.6 to(χ H2 = and 24.2) He recombination lines (see low 2013; Capak et al. 2013). tected at ≥ 7inσ in dark the H160 band. red. The Best-fit brightest source SEDsalso, e.g., are Schaerer shown & de Barros as 2009). blue Emission solid lines of lines, in addition−6 is 15σ. Furthermore,30 all sources are seen in observations other elements were added based on line ratios relative to 10 3.3.1. Emission Line Galaxies at other wavelengths,to the albeit best at lower significance. low redshift With Hβ solutionstabulated by Anders in & Fritze-v. gray. Alvensleben The (2003). correspondingSED In our SED analysis in Section 3.2, we specifically in- the exception of the brightest one, all show weak detec- The template library adopted for the SED analysis is z tions in J ,magnitudes and two are evenGNWJ seen weakly are−604094296 in theshown very based as on filled both constant circles. and exponentially For declining all - sources, the z ≥ 9 125 cludedStrong line emission emission line in galaxies order to have test for long contamination been known shallow JH140 data. Furthermore, the brightest source formation histories of varying star-formation timescales is detected atsolution 2σ in the ground-based fits K-band the data. observed(τ =108 to fluxes 1010 yr), as also significantly used to model lower redshift better than anyof−2 Neighbor-subtraction26 was applied to the IRAC data of sources. All models assume a Chabrier initial mass func- 10 fromto potentially strong emission contaminate line sources. very high-redshift Indeed, for sample two of the se- all four z ! 9the galaxy possible candidates. The resulting low-redshift cleaned tion and SEDs. a metallicity The of 0.5Z⊙ best-fit,andtheagesrangefrom low redshift solutions IRAC images are shownAB in the two right-hand columns t = 10 Myr to 13 Gyr. However, only SEDs−4 with ages lection. These are a particular concern in datasets

p(z) candidates, the best-fit low-redshift photometric redshift of Figure 2.are As canm be clearly seen, three of ruled these sources out are withless than likelihoods the age of the universe< at a10 given redshiftfor are three of the four clearly detected in at least one IRAC band. For source allowed in the fit. Dust extinction is modeled following −4 GNDJ-722744224,sources. the28 residuals of For the bright GNWJ-604094296, foreground Calzetti et al. (2000). the low-redshift solutions have10 a which do not have very deep optical data to establish a z = 9.5 (χ2 = 3.9) solutions are obtained from a combination of extreme neighbor are still visible and its IRAC flux measure- As is evidentbest in Figure 3, all candidates have a best- likelihood of just 0.2% (see text). ≥ ments are therefore highly uncertain. In order to pro- fit photometric redshift2 at z 9 with uncertainties of strong spectral break through non-detections (see, e.g., vide some photometric constraints for this source from σz =0.3-0z .4(1 =σ 0.6). These (χ = uncertainties 16.5) could be reduced emission lines and high dust extinction. However, all low IRAC, we use conservative upper limits based on the with deeper JH140 imaging data in the future. With the −6 RMS fluctuations in the30 residual image at the position of exception of one source, the best-fit low redshift solu- 10 candidatesAtek et al. are 2011; detected van der (although Wel et al. sometimes 2011; Hayes faintly) et al. in the bright foreground source. All flux measurements for tions are securely ruled out with likelihoods < 10−4 (see z these sources,such together with an the SED uncertainties really are listed is.right panels However, of Figure 3). For GNWJ-604094296, deeper theY low- data or spec- in Table 2. GNDJ−652258424 redshift solutions have a likelihood of 0.2%. As can105 be several2012). non-overlapping Sources with extreme filters. For rest-frame example, with optical the line ex- seen from Figure 3, the low-redshift SED is a combina- −2 3.2.troscopicPhotometric26 Redshift observations Analysis tion of high could dust extinction rule and extreme out emission such lines, an SED. Some10 ceptionemission of may GNDJ-625464314, also contaminate all sourcesz ! 9 show samples some if flux the which line up to boost the fluxes in the H160 and IRAC Figure 3 shows the SED fits to the fluxes of the four 4.5 µm bands. It is unclear how likely the occurrence of high-redshiftfirst galaxyAB candidates. spectroscopic These are derived with constraints are already available from inz ∼ the10J candidate125 filter, UDFj-39546284 as well as a clear (Bouwens detection et in al.H 2011a;160.It p(z) shallowm WFC3 grism observations (see Section 3.3.2).−4 28 10 Oeschetis therefore al. unlikely 2012a) is that any the guide. detected InHST thatflux case, origi- the z = 9.2 (χ2 = 8.3) As a cross-check, webest also tested and confirmed the extremelynates from deep emission supporting lines alone. data Furthermore,did not result three in any of z = 2.2 (χ2 = 30.0) low high-redshift solutions with the photo-z code EAZY−6 detection shortward of the H band, but other evi- 30 10 10 160 (Brammer0.5 et al.1.0 2008).2.0 In particular,4.0 6.0 0 2.5 we5 fit7.5 photomet-10 12.5 denceavailable (tentative at http://code.google.com/p/eazy-photoz/ detection of an emission line at 1.6 µm observed wavelength [µm] z and the high luminosity of UDFj-39546284) indicates Fig. 3.— Spectral energy distribution fits to the HST and that an extreme emission line galaxy at z ∼ 2.2 is Spitzer/IRAC photometry of the four GOODS-N z ∼ 9−10 galaxy a more likely interpretation of the current data (see candidates (left)togetherwiththeredshiftlikelihoodfunctions (right). The measurements and their upper limits (2σ)areshown Bouwens et al. 2013a; Ellis et al. 2013; Brammer et al. in dark red. Best-fit SEDs are shown as blue solid lines, in addition 2013; Capak et al. 2013). to the best low redshift solutions in gray. The correspondingSED In our SED analysis in Section 3.2, we specifically in- magnitudes are shown as filled circles. For all sources, the z ≥ 9 cluded line emission in order to test for contamination solution fits the observed fluxes significantly better than anyof the possible low-redshift SEDs. The best-fit low redshift solutions from strong emission line sources. Indeed, for two of the are clearly ruled out with likelihoods < 10−4 for three of the four candidates, the best-fit low-redshift photometric redshift sources. For GNWJ-604094296, the low-redshift solutions have a solutions are obtained from a combination of extreme likelihood of just 0.2% (see text). emission lines and high dust extinction. However, all candidates are detected (although sometimes faintly) in such an SED really is. However, deeper Y105 data or spec- several non-overlapping filters. For example, with the ex- troscopic observations could rule out such an SED. Some ception of GNDJ-625464314, all sources show some flux first spectroscopic constraints are already available from in the J125 filter, as well as a clear detection in H160.It shallow WFC3 grism observations (see Section 3.3.2). is therefore unlikely that the detected HST flux origi- As a cross-check, we also tested and confirmed the nates from emission lines alone. Furthermore, three of high-redshift solutions with the photo-z code EAZY 10 (Brammer et al. 2008). In particular, we fit photomet- available at http://code.google.com/p/eazy-photoz/ 6 Oesch et al.

TABLE 2 Flux Densities of z>9 LBG Candidates in the GOODS-N Field

Filter GNDJ-625464314 GNDJ-722744224 GNWJ-604094296 GNDJ-652258424

B435 7 ± 93± 6 −2 ± 6 −11 ± 11 V606 2 ± 7 −3 ± 5 −5 ± 52± 8 i775 5 ± 10 6 ± 76± 7 −9 ± 11 I814 3 ± 71± 5 −2 ± 80± 9 z850 17 ± 11 −7 ± 6 −7 ± 8 −14 ± 13 Y105 −7 ± 9 −7 ± 7 −2 ± 10 −18 ± 8 J125 11 ± 812± 723± 836± 9 JH140 102 ± 47 85 ± 34 ··· 86 ± 54 H160 152 ± 10 68 ± 973± 882± 11 K137± 67 −45 ± 51 85 ± 261 76 ± 55 IRAC 3.6 µm139± 20 (< 81)* 39 ± 21 65 ± 18 IRAC 4.5 µm122± 21 (< 119)* 93 ± 21 125 ± 20 Note.—MeasurementsaregiveninnJywith1σ uncertainties. * 3σ upper limit due to uncertainties in the neighbor flux subtraction.

0 24 10 ric redshifts with templates that include emission lines 10 GNDJ−625464314 as included in the v1.1 distribution of the code .The −2 26 10 best-fit EAZY redshifts are all within 0.1 of the ZEBRA

AB values listed in Table 1. p(z) m −4 28 10 z = 10.2 (χ2 = 4.9) best 3.3. Possible Sample Contamination z = 2.1 (χ2 = 47.5) low −6 30 10 As we will show in Section 4.1, the detection of such z ∼ − GNDJ−722744224 bright z 9 10 galaxy candidates in the GOODS-N −2 dataset is surprising given previous constraints on UV 26 10 Bright z ∼ 9 − 10 Galaxy Candidates in GOODS-North 5 LFs at z>8. A detailed analysis of possible contam- AB p(z) m −4 ination is therefore particularly important. We discuss 28 10 z = 9.9 (χ2 = 9.4) best several possible sources of contamination. z = 2.6 (χ2 = 24.2) low −6 30 10 3.3.1. Emission Line Galaxies z GNWJ−604094296 Strong emission line galaxies have long been known Another Bright z~9-10 Galaxy in CANDELS −2 26 Bright z ∼ 9 − 10 Galaxy Candidates in GOODS-North 5 10 to potentially contaminate very high-redshift sample se-

AB lection. These are a particular concern in datasets p(z) m −4 28 10 which do not have very deep optical data to establish a z = 9.5 (χ2 = 3.9) best z = 0.6 (χ2 = 16.5) strong spectral break through non-detections (see, e.g., low −6 Fig. 2.— 6′′×6′′images30 of the four z ≥ 9galaxycandidatesidentifiedintheCANDELSGOODS-Ndata.From left to right, the images 10 Atek et al. 2011; van der Wel et al. 2011; Hayes et al. show a stack of all optical bands, Y105, JH140, J125, H160,MOIRCSK,andneighbor-subtractedIRAC3.6µmand4.5µmimages.The stamps are sorted from high to lower photometric redshift from SED fits (indicated in the lower left, see also Table 1). The IRAC neighbor- z subtraction works well for all sourcesGNDJ except for− GNDJ-722765225842444224, where the nearby foreground source is too bright, and clear residuals are 2012). Sources with extreme rest-frame optical line visible at the location of the candidate. Only IRAC upper limits are therefore included for this source in the following analysis. Clearly, all other sources show significant (> 4.5σ)detectionsinthe4.5µmchannel.Thebrightestsource(GNDJ-625464314)isalsodetected at −2 6.9σ in the 3.6 µmchannel.Withtheexceptionofthebrightestcandidate,whz = 9.2 ich is weakly detected in the K-band (at 2σ), the MOIRCS emission may also contaminate z ! 9 samples if the K-band data provide only26 upper limits. z = 9.2 10

its flat continuum longwardAB of 1.6 µm. Taken together, the photometric redshift code ZEBRA (Feldmann et al. z ∼ 10 candidate UDFj-39546284 (Bouwens et al. 2011a;

these results suggest that the most likely interpretation 2006; Oesch et al. 2010b) using a large library of stel- p(z) m is that GNDJ-625464314 is at high redshift. lar population synthesis template models based on the −4 Oeschet al. 2012a) is any guide. In that case, the Stamps of the four28 viable high-redshift candidates are library of Bruzual & Charlot2 (2003). Additionally, we 10 presented in Figure 2, and their positions and photome- added nebularz = line 9.2 and (χ continuum = 8.3) emission to these tem- best try are listed in Tables 1 and 2. plate SEDs in a self-consistent2 manner, i.e., by convert- extremely deep supporting data did not result in any As is evident from Fig. 2, the four sources are all de- ing ionizingz photons = 2.2 ( toχ H = and 30.0) He recombination lines (see low tected at ≥ 7σ in the H160 band. The brightest source also, e.g., Schaerer & de Barros 2009). Emission lines of −6 detection shortward of the H160 band, but other evi- isFig. 15σ 2.—. Furthermore,6′′×6′′images30 all of the sources four z are≥ 9galaxycandidatesidentifiedintheCANDELSGOODS-Ndata.F seen in observations other elements were added basedrom on line left to ratios right, relative the images to 10 atshow other a stack wavelengths, of all optical albeit bands, atY0.5105 lower, JH140 significance., J125, H1601.0,MOIRCSK,andneighbor-subtractedIRAC3.6 With Hβ tabulated2.0 by Anders4.0 & Fritze-v.µmand4.56.0 Alvensleben0µmimages.The2.5 (2003).5 7.5 10 12.5 thestamps exception are sorted of from the high brightest to lower one, photometric all show redshift weak fr detec-om SED fits (indicatedThe template in the lower library left, see adopted also Table for 1). the The SEDIRAC analysis neighbor- is dence (tentative detection of an emission line at 1.6 µm subtraction works well for all sources except for GNDJ-722744224, where the nearby foreground source is too bright, and clear residuals are tions in J125, and two are even seen weaklyobserved in the very wavelengthbased on both constant [ m] and exponentially declining star- z visible atOesch+2014 the location of the candidate. Only IRAC upper limits are therefore included forµ this source in the following analysis. Clearly, shallowall other sourcesJH140 showdata. significant Furthermore, (> 4.5σ)detectionsinthe4.5 the brightest sourceµmchannel.Thebrightestsource(GNDJ-625464314)isalsodeformation histories of varying star-formation timescalestected at and the high luminosity of UDFj-39546284) indicates is6.9 detectedσ in the 3.6 atµmchannel.Withtheexceptionofthebrightestcandidate,wh 2σ in the ground-based K-band data. (τ =10ich8 isto weakly 1010 yr), detected as also in the used K-band to model (at 2σ lower), the MOIRCS redshift K-bandNeighbor-subtraction data provide onlyFig. upper was appliedlimits. 3.— to the IRAC data of sources. All models assume a Chabrier initial mass func- all four z ! 9 galaxy candidates. The resultingSpectral cleaned tion energy and a metallicity distribution of 0.5Z⊙,andtheagesrangefrom fits to the HST and that an extreme emission line galaxy at z ∼ 2.2 is IRACits flat images continuumSpitzer/IRAC are shown longward in the of 1.6 twoµm. right-hand Taken photometry together, columns tthe= photometric 10 Myr of to the 13 redshift Gyr. four However, code ZEBRA GOODS-N only (Feldmann SEDs with et ages al.z ∼ 9−10 galaxy ofthese Figure results 2. Assuggest can be that seen, the three most oflikely these interpretation sources are less2006; than Oesch the et age al. of 2010b) the universe using a at large a given library redshift of stel- are a more likely interpretation of the current data (see clearlyis that GNDJ-625464314detected in at least is one at high IRAC redshift. band. For source allowedlar population in the fit. synthesis Dust extinction template models is modeled based following on the GNDJ-722744224,Stamps of thecandidates four the viable residuals high-redshift of the ( brightleft candidates foreground)togetherwiththeredshiftlikelihoodfunctions are Calzettilibrary of et Bruzual al. (2000). & Charlot (2003). Additionally, we neighborpresented are in Figure stillright visible 2, and and their its positions IRAC flux and photome- measure- addedAs is nebular evident line in andFigure continuum 3, all candidates emission to have these a best- tem- Bouwens et al. 2013a; Ellis et al. 2013; Brammer et al. mentstry are are listed therefore( in Tables highly 1 and). uncertain. 2. The In measurements order to pro- fitplate photometric SEDs and in a redshift self-consistent their at z ≥ manner, upper9 with uncertaintiesi.e., by limits convert- of (2σ)areshown videAs some is evident photometricin from dark Fig. constraints 2, the red. four for sources thisBest-fit source are all from de- SEDsσingz =0 ionizing.3-0 are.4(1 photonsσ shown). These to H and uncertainties He as recombination blue could be solid lines reduced (see lines, in addition 2013; Capak et al. 2013). IRAC,tected at we≥ use7σ conservativein the H160 band. upper The limits brightest based on source the withalso, deeper e.g., SchaererJH140 imaging & de Barros data 2009). in the Emissionfuture. With lines the of RMSis 15σ fluctuations. Furthermore, in the all residual sources are image seen at in the observations position of exceptionother elements of one were source, added the based best-fit on line low ratios redshift relative solu- to theat other bright wavelengths, foregroundto the source. albeit best at All lower flux significance.low measurements redshift With for tionsHβ solutionstabulated are securely by Anders ruled in out & Fritze-v. with gray. likelihoods Alvensleben The< 10− (2003).4 correspondin(see gSED In our SED analysis in Section 3.2, we specifically in- thesethe exception sources,magnitudes of together the brightest with the one, uncertainties all show are weak are shown detec- listed right asThe panels template filled of Figure library circles. 3). adopted For GNWJ-604094296, for the For SED allanalysis the low- sources, is the z ≥ 9 intions Table in J 2.125, and two are even seen weakly in the very redshiftbased on solutions both constant have a and likelihood exponentially of 0.2%. declining As can star- be cluded line emission in order to test for contamination shallow JH140 data. Furthermore, the brightest source seenformation from Figure histories 3, of the varying low-redshift star-formation SED is a timescales combina- solution fits the observed8 fluxes10 significantly better than anyof is detected3.2. at 2σPhotometricin the ground-based Redshift Analysis K-band data. tion(τ =10 of highto 10 dustyr), extinction as also used and extreme to model emission lower redshift lines, Neighbor-subtraction was applied to the IRAC data of whichsources. line All up models to boost assume the fluxes a Chabrier in the initialH massand IRAC func- from strong emission line sources. Indeed, for two of the Figure 3 showsthe the possible SED fits to the fluxes low-redshift of the four SEDs. The best-fit160 low redshift solutions all four z ! 9 galaxy candidates. The resulting cleaned 4.5tionµm and bands. a metallicity It is unclear of 0.5Z how⊙,andtheagesrangefrom likely the occurrence of high-redshift galaxy candidates. These are derived with −4 IRAC images are shown in the two right-hand columns t = 10 Myr to 13 Gyr. However, only SEDs with ages candidates, the best-fit low-redshift photometric redshift of Figure 2.are As can be clearly seen, three of ruled these sources out are withless than likelihoods the age of the universe< at a10 given redshiftfor are three of the four clearly detected in at least one IRAC band. For source allowed in the fit. Dust extinction is modeled following GNDJ-722744224,sources. the residuals of For the bright GNWJ-604094296, foreground Calzetti et al. (2000). the low-redshift solutions have a solutions are obtained from a combination of extreme neighbor arelikelihood still visible and its IRAC of just flux measure- 0.2% (seeAs is evident text). in Figure 3, all candidates have a best- ments are therefore highly uncertain. In order to pro- fit photometric redshift at z ≥ 9 with uncertainties of emission lines and high dust extinction. However, all vide some photometric constraints for this source from σz =0.3-0.4(1σ). These uncertainties could be reduced IRAC, we use conservative upper limits based on the with deeper JH140 imaging data in the future. With the RMS fluctuations in the residual image at the position of exception of one source, the best-fit low redshift solu- candidates are detected (although sometimes faintly) in the bright foreground source. All flux measurements for tions are securely ruled out with likelihoods < 10−4 (see these sources,such together with an the SED uncertainties really are listed is.right panels However, of Figure 3). For GNWJ-604094296, deeper theY low-105 data or spec- several non-overlapping filters. For example, with the ex- in Table 2. redshift solutions have a likelihood of 0.2%. As can be troscopic observationsseen from could Figure 3, the rule low-redshift out SED is such a combina- an SED. Some 3.2. Photometric Redshift Analysis tion of high dust extinction and extreme emission lines, ception of GNDJ-625464314, all sources show some flux which line up to boost the fluxes in the H160 and IRAC Figure 3 showsfirst the SED spectroscopic fits to the fluxes of the four constraints4.5 µm bands. It is unclear are how likely already the occurrence of available from high-redshift galaxy candidates. These are derived with in the J125 filter, as well as a clear detection in H160.It shallow WFC3 grism observations (see Section 3.3.2). is therefore unlikely that the detected HST flux origi- As a cross-check, we also tested and confirmed the nates from emission lines alone. Furthermore, three of high-redshift solutions with the photo-z code EAZY 10 (Brammer et al. 2008). In particular, we fit photomet- available at http://code.google.com/p/eazy-photoz/ The Astrophysical Journal,785:1(19pp),2014??? Oesch et al.

0 The Astrophysical Journal,785:1(19pp),2014???24 Oesch et al. 10 GS−z10−1 p(z<5) = 0.12% Another Bright z~9-10 Galaxy in CANDELS −2 26 Bright z ∼ 9 − 10 Galaxy Candidates in GOODS-North 5 10 AB p(z) m −4 28 10 z = 9.9 (χ2 = 4.0) best z = 2.5 (χ2 = 17.7) Total of 6 bright z~9-10low Figure 12. 6 6 negative images of the two new z 9galaxycandidatesidentifiedinourreanalysisoftheCANDELSGOODS-Sdata.Fromlefttoright,the −6 ′′ × ′′ 30 ! 10 images show a stack of all optical bands, Y105, J125, Hcandidates160, HAWKI K, and neighbor-subtracted in CANDELS- IRAC 3.6 µmand4.5µmimages.TheK-band image is a very deep stack (26.5 mag, 5σ )ofESO/VLT HAWK-I data from the HUGS survey (PI: Fontana). Both sources are only weakly detected in these data. z (A color version of this figure is available in the onlineGS− journal.)z9−1 North + South p(z<5) = 0.08% z = 9.3 Table 7 −2 Coordinates26 and Basic Photometry of Two New z>9LBGCandidatesintheGOODS-SField 10

J H H . z NameAB ID R.A. Decl. H160 125 160 160 [4 5] phot − − p(z) GS-z10-1 GSDJ-2269746283m 03:32:26.97These sources27:46:28.3 26 .88are0.15 quite 1.7 0.6 0.4 0.69.9 0.5 − ± ± − ± ± −4 GS-z9-1 GSDJ-232055041728 a 03:32:32.05 rare27:50:41.7 though! 26.61 0.18 1.1 0.50.5 0.39.3 0.5 10 − ±z = 9.3 ±(χ2 = 5.8) ± ± a best Note. The source GS-z9-1 does not satisfy the criterion J125 H160 > 1.2 and is not included in the UV LF analysis. − z = 2.2 (χ2 = 20.9) low −6 color cut ofFig.J125 2.— H6′′160×6′′>images0.5,30 of as the we four didz ≥ in9galaxycandidatesidentifiedintheCANDELSGOODS-Ndata.F GOODS-N, rather Tablerom8 left to right, the images 10 show a stack− of all optical bands, Y105, JH140, J125, H160,MOIRCSK,andneighbor-subtractedIRAC3.6Flux Densities of Two New z>9LBGCandidatesintheGOODS-SFieldµmand4.5µmimages.The than the more conservative cut of J125 0.5H160 > 1.2asadopted1.0 2.0 4.0 6.0 0 2.5 5 7.5 10 12.5 stamps are sorted from high to lower− photometric redshift from SED fits (indicated in the lower left, see also Table 1). The IRAC neighbor- in our previoussubtraction work works (e.g., well Oesch for all sourceset al. 2013a except). forobserved GNDJ-722744224, wherewavelengthFilter the nearby foreground [ m] source is too GS-z10-1 bright, and clear residuals are GS-z9-1z These newvisible catalogs atOesch+2014 the location revealed of the two candidate. possible, Only IRAC bright upperz> limits9 are therefore included forµ this source in the following analysis. Clearly, all other sources show significant (> 4.5σ)detectionsinthe4.5µmchannel.Thebrightestsource(GNDJ-625464314)isalsodeB435 1 97tected at 10 galaxy candidates6.9σ in the 3.6 inµmchannel.Withtheexceptionofthebrightestcandidate,wh the CANDELS GOODS-S data set, ich is weakly detected in the K-band− ± (at 2σ), the MOIRCS ± V606 1 608 Figure 13. Spectral energyK-band data distribution provide only upper fits limits. (left) and redshift likelihood functions (right)± for the two bright± z 9–10 sources in GOODS-S. In the left panel, GS-z9-1 and GS-z10-1. They have magnitudes of H160 i775 6 9 5 12 = − ± − ± ∼ photometry and26 2σ.6 upper0.2and limitsH160 are26 shown.9 0.2, in respectively. dark red. The Best-fit latter SEDsI814 are shown as blue solid5 6 lines, in addition3 9 to the best low redshift solutions in gray. The legend ± its flat continuum= longward± of 1.6 µm. Taken together, the photometric redshift code ZEBRA± (Feldmann et al. − ± candidatethese also shows results asuggest color2 that of J the125 mostH160 likely> interpretation1.2(namely 2006;z850 Oesch et al. 2010b) using4 a large9 library of stel- 5 16 lists their respective redshift and χ values. The− redshift likelihood distributions from SED− ± fitting in the right− panels± are shown on logarithmic axes. For both sources, 1.7 0.6),is while that GNDJ-625464314 the first is only slightly is at high too redshift. blue to satisfy this larY105 population synthesis template0 models6 based on the 14 9 ± ± − ± the likelihood of a low redshiftJ StampsH of solution the four. viable. is only high-redshift1% candidates as indicated are bylibraryJ125 the of labels. Bruzual & Charlot (2003).13 729 Additionally, we 11 criterion ( 125 160 1 1 0 5). ± ± presented− in= Figure± 2, and their positions∼ and photome- addedJH140 nebular line and continuum12 emission23 to these tem- 55 33 Given its red color, GS-z10-1 could already have been in the ± ± (A color version of thistry figure are listed is available in Tables 1 and in 2. the online journal.) plateH160 SEDs in a self-consistent manner,66 985 i.e., by convert- 14 previous catalog of Oesch et al. (2013a)whoanalyzedthesame ± ± As is evident from Fig. 2, the four sources are all de- ingK ionizingHAWKI photons to H and He 33 recombination19 lines (see 54 18 CANDELS GOODS-South data set. The reason this source − ± ± tected at ≥ 7σ in the H160 band. The brightest source also,IRAC e.g., 3.6 µ Schaererm32 & de Barros 2009).17 Emission lines of 58 24 was not previously selected is due to a very faint neighbor ± ± is 15σ. Furthermore, all sources are seen in observations otherIRAC elements4.5 µm44 were added based on line22 ratios relative to 131 23 that was includedat other in wavelengths, the Kron aperture albeit at in lower the earlier significance. SExtractor With Hβ tabulated by Anders & Fritze-v.± Alvensleben (2003). ± the exception of the brightest one, all show weak detec- The template library adopted for the SED analysis is catalog. This caused the candidate to be rejected due to apparent Note. Measurements are given in nJy with 1σ uncertainties. optical fluxtions in the in J aperture.125, and two With are careful even seen visual weakly inspection in the very we based on both constant and exponentially declining star- shallow JH140 data. Furthermore, the brightest source formation histories of varying star-formation timescales assessed thatis detected the optical at 2 fluxσ in in the the ground-based previous aperture K-band was data. due to (τ =108 to 1010 yr), as also used to model lower redshift a faint neighboringNeighbor-subtraction galaxy and is was not applied likely associated to the IRAC with data the of sources.IRAC detections All models in assume both a 3.6 Chabrier and 4.5 initialµmbandswithfluxes mass func- high-z candidate.all four z With! 9 the galaxy new candidates. deblending The parameters resulting for cleaned our tionconsistent and a metallicity with a significant of 0.5Z⊙ Balmer,andtheagesrangefrom break at z 9, giving added SExtractorIRAC run, this images source are isshown now in confirmed the two right-hand to be a legitimate columns tweight= 10 Myr to our to identification 13 Gyr. However, of this only source SEDs as with∼ a probable ages z 9 z>9galaxycandidate.Itsphotometricredshiftisfoundtoof Figure 2. As can be seen, three of these sources are lesscandidate. than the From age of SED the fitting universe we at find a given a photometric redshift are redshift∼ of clearly detected in at least one IRAC band. For source allowed in the fit. Dust extinction is modeled following be zphot 9.9 0.5. We thus include this candidate in the full zphot 9.3 0.5forthissource. =GNDJ-722744224,± the residuals of the bright foreground Calzetti= et al.± (2000). analysis ofneighbor the main are body still of this visible paper. and We its have IRAC verified flux measure- that the AsImages is evident of both in Figure new 3, GOODS-S all candidates candidates have a best- are shown in GOODS-Nments data are returns therefore the same highly candidates uncertain. when In order using to these pro- fitFigure photometric12,andtheirSEDfitsandphotometricredshiftlikelihood redshift at z ≥ 9 with uncertainties of updated deblendingvide some parameters. photometric constraints for this source from σfunctionsz =0.3-0.4(1 areσ shown). These in uncertainties Figure 13.Table could be7 reducedlists the basic The inclusionIRAC, of we this usez conservative10 candidate upper does limits not significantly based on the withinformation deeper JH of140 theseimaging sources, data in and the Table future.8 Withlist all the their flux RMS fluctuations∼ in the residual image at the position of exception of one source, the best-fit low redshift solu- change thethe results. bright For foreground instance, source. including All flux this measurements candidate only for tionsmeasurements. are securely ruled out with likelihoods < 10−4 (see causes athese change sources, of 0.1dexin together withφ thewhen uncertainties assuming are density listed right panels of Figure 3). For GNWJ-604094296, the low- evolutionorachangeofonly0.1inin Table 2. M∗ for luminosity evolution. redshift solutions have aAPPENDIX likelihood of 0.2%.B As can be The total cosmic SFRD changes by only∗ 0.02 dex, because this seen from Figure 3, the low-redshift SED is a combina- is dominated by large3.2. fluxPhotometric from lower Redshift luminosity Analysis sources as tion of high dust extinctionIRAC Neighbor and extreme Subtraction emission lines, which line up to boost the fluxes in the H160 and IRAC indicated byFigure the faintest 3 shows candidate the SED in fits the to XDF the (and fluxes by of the the steep four 4.5Theµm bands. point-spread It is unclear function how of likelySpitzer the/IRAC occurrence is 10 of broader slopes foundhigh-redshift at slightly galaxy later times candidates. at z These7–8). are derived with ∼ × ∼ than for WFC3/IR. A crucial aspect of using the Spitzer/IRAC The other source, GS-z9-1, was already in the previous data to constrain the rest-frame optical fluxes of faint galaxies SExtractor catalogs. However, it was not included in the analysis at high redshift is therefore to reliably subtract neighboring due to its bluer color of J125 H160 < 1.2. For completeness, we − sources to deal with source confusion. Several teams have present this source here as well, particularly since it is so close to developed techniques to perform efficient neighbor subtraction our z 10 color cutoff. Interestingly, it also shows significant ∼ based on modeling the IRAC fluxes from the high-resolution 17

Figure 14. Demonstration of the IRAC neighbor subtraction routine. The left column shows the 13 13 stamps of the original H160 image of the four GOODS-N ′′ × ′′ high-redshift candidates. The second column is the original IRAC image at 4.5 µm. We use a kernel to convolve the H160 image to the PSF of this IRAC image and fit the combined light profile of each source in the H160-band image. The third column shows the result of this fit for all the neighbors of the target source. This model is subsequently subtracted from the original IRAC image which results in a residual image (right column), on which we perform aperture photometry of the source of interest. (A color version of this figure is available in the online journal.)

18 Very Low Formal Probability of Contamination (CANDELS z~9-10 sample much more robust than HUDF z~9-10 sample)

HUDF/XDF Sample Oesch+2014 Ellis+2013 / CANDELS sample Oesch+2013

Oesch+2014 ~0.05% ~5% How does the observed z~10 LF compare with extrapolations from lower redshift?

Cai+2014 model

Cai+2014; Oesch+2014; Bouwens+2014 How does the observed z~10 LF compare with extrapolations from lower redshift?

Cai+2014 prediction z=10

Cai+2014; Oesch+2014; Bouwens+2014 How does the observed z~10 LF compare with extrapolations from lower redshift?

Similarly good fit from conditional LF model Bouwens +2008

Cai+2014; Oesch+2014; Bouwens+2014 The Sizes of z=9-10 Candidate Galaxies 5

1. While most redshift z 9 10 candidate galaxies ⇠ are unambiguously resolved (re > 000. 1) with HST CANDELS or XDF F160W data (Figure 1), the brighter sources in our z 9 10 CANDELS sam- ⇠ ple are larger (=000. 16) and more resolved than the fainter z 10 candidates in the XDF ⇠ (=000. 09), allowing for a more optimal con- straints on the size evolution of galaxies to z 10. ⇠ 2. The mean e↵ective radius (=000. 16 observed, 0.6 kpc physical) for bright sources in the Oesch et al. (2013b) z =9 10 sample conforms to 1 expected size-redshift relation (e.g., (1 + z) or The Sizes of z=9-10 Candidate Galaxies1.5 5 (1+z) ) for bright star-forming galaxies. This is substantially smaller than the 0. 59 mean sizes de- 1. While most redshift z 9 1000 candidate galaxies The sizes of these z~9-10 candidates are exactly rived for intrinsically-red⇠ contaminants that other- are unambiguously resolved (re > 000. 1) with HST what we would expect... wise satisfy z 10 J125-dropout selection criteria. TheCANDELS present or size⇠ XDF measurementsF160W data therefore (Figure provide 1), the brighter sources in our z 9 10 CANDELS sam- additional evidence that⇠ these bright z=9-10 can- didatesple are largerare indeed ( the=0 indicated00. 16) and redshift. more resolved than the fainter z 10 candidates in the XDF ⇠ 3. The(=0900. 09),10 allowing candidate for galaxiesa more optimal are smaller con- andstraints brighter⇠ on the than sizez evolution7 galaxies, of galaxies a factor to ofz 1.610. growth in e↵ective radius⇠ for fixed luminosity⇠ (Fig-⇠ 2. Theure 2). mean e↵ective radius (=000. 16 observed, 0.6 kpc physical) for bright sources in the Oesch 4. Theet al. implied (2013b) SFRz surface=9 density10 sample for the conformsz 9 10 to ⇠1 1 2 Derive from expectedpopulation size-redshift is on average relation⌃SFR =4M (e.g., (1yr +z)kpc or 1.5 HST data (1+but withz) a) wide for bright range star-forming in individual galaxies. values (⌃ ThisSFR is 1 2 ⇠ substantially1 10 M yr smaller kpc than , Figurethe 000. 59 2). mean These sizes sur- de- rivedface densities for intrinsically-red are similar, albeit contaminants slightly higher that other- than wisethose satisfy foundz by Oesch10 J125 et-dropout al. (2010) selection and Ono criteria. et al. The(2013) present at lower size⇠ redshifts. measurements therefore provide additional evidence that these bright z=9-10 can- 5. Thedidates size-mass are indeed relation at the we indicated find for our redshift.z 10 sam- ple is very similar to the one found by Mosleh⇠ et al. 3. The(2012)z at z9 610 with candidate these galaxies galaxies approximately are smaller ⇠ ⇠ Holwerda+2014Figure 4. The (resubmitted e↵ective to radius ApJL after as responding a function to referee of redshift report) for our andhalf brighterthe size of than star-formingz 7 galaxies, galaxies a factor at z of2 and1.6 growthz = 0 of in comparable e↵ective radius⇠ mass for (Figure fixed luminosity 3). ⇠ (Fig-⇠ sample for both bright (L>0.3Lz⇤=3,toppanel)andlower- luminosity galaxies (L<0.3Lz⇤=3,bottompanel).Forcomparison, ure 2). we show the mean sizes from earlier epochs from Bouwens et al. 6. Including the Oesch et al. (2013b) z 10 candi- ⇠ (2004); Oesch et al. (2010); Ono et al. (2013). The mean size of 4. Thedates, implied we find SFR that surface the sizes density of > for0.3 theLz⇤=3z galaxies9 10 the six potential interlopers to a z 9 10 selection (see 4.1) is 1.0 0.1 ⇠1 2 ⇠ § exhibitpopulation a (1 is + onz) average± scaling⌃SFR =4M with redshift,yr kpc con- well above any expected relation at z 9. We do not include the Bouwens et al. (2011a) z 2/z 12 candidate⇠ as there is consider- sistentbut with with a wide what range Bouwens in individual et al. (2004, values 2006) (⌃SFR and ⇠ ⇠ 1 2 ⇠ able doubt as to whether it is at z 12 (Ellis et al. 2013; Brammer 1Oesch10 et M al. (2010)yr kpc previously , Figure found 2). (Figure These 4). sur- et al. 2013; Bouwens et al. 2013; Capak⇠ et al. 2013; Pirzkal et al. face densities are similar, albeit slightly higher than 2013). The dotted line shows the best fits from Oesch et al. (2010). ACKNOWLEDGEMENTS Dashed lines are our fits to the Bouwens et al. and Oesch et al. those found by Oesch et al. (2010) and Ono et al. values combined with our mean size constraints at z 9 10. The This(2013) work at is lower based redshifts. on observations taken by the 1⇠ mean size of L>0.3L⇤ galaxies scale as (1 + z) . CANDELS Multi-Cycle Treasury Program with the z=3 5. The size-mass relation we find for our z 10 sam- NASA/ESA HST, which is operated by the⇠ Associa- tion ofple Universities is very similar for Research to the one in found Astronomy, by Mosleh Inc., et un- al. (2012) at z 6 with these galaxies approximately der NASA contract⇠ NAS5-26555 and HST GO-11563. In this letter we take advantage of the recent discovery half the size of star-forming galaxies at z 2 and Figure 4. The e↵ective radius as a function of redshift for our This work is based on observations taken by⇠ the 3D- ofsample six bright for bothz bright9 10 (L> candidate0.3L⇤ galaxies,toppanel)andlower- within CAN- HST Treasuryz = 0 of comparable Program (GO mass 12177 (Figure and 3).12328) with the ⇠ z=3 DELSluminosity (Oesch galaxies et ( al.L< 2013b)0.3Lz⇤=3 to,bottompanel).Forcomparison, explore the size-luminosity NASA/ESA HST, which is operated by the Association andwe show size-mass the mean relations sizes from out earlier to epochs the earliest from Bouwens observable et al. 6. Including the Oesch et al. (2013b) z 10 candi- (2004); Oesch et al. (2010); Ono et al. (2013). The mean size of of Universitiesdates, we for find Research that the sizesin Astronomy, of > 0.3L⇠ Inc.,galaxies under . By comparing sizes measured for sources in our NASA contract NAS5-26555. This researchz⇤=3 has made the six potential interlopers to a z 9 10 selection (see 4.1) is exhibit a (1 + z) 1.0 0.1 scaling with redshift, con- wellsample above with any similar expected size relation measures at⇠z 9. at We lower do not redshifts, include§ the we use of NASA’s Astrophysics ± Data System. We acknowl- Bouwens et al. (2011a) z 2/z 12 candidate⇠ as there is consider- sistent with what Bouwens et al. (2004, 2006) and quantify the size evolution⇠ of⇠ galaxies. We establish these edge the support of ERC grant HIGHZ #227749, and a sizesable doubt by fitting as to whether the light it is profiles at z 12 of (Ellis the et galaxies al. 2013; from Brammer the Oesch et al. (2010) previously found (Figure 4). et al. 2013; Bouwens et al. 2013; Capak⇠ et al. 2013; Pirzkal et al. NWO “Vrije Competitie” grant 600.065.140.11N211. 2013).XDF The and dotted CANDELS line shows WFC3/IR the best fitsF160W from Oeschobservations. et al. (2010). ACKNOWLEDGEMENTS ToDashed ensure lines that are our our fits size to the measurements Bouwens et al. are and accurate, Oesch et we al. REFERENCES values combined with our mean size constraints at z 9 10. The This work is based on observations taken by the conducted an extensive set of comparisons against1⇠ previ- mean size of L>0.3L⇤ galaxies scale as (1 + z) . CANDELS Multi-Cycle Treasury Program with the ous size measurements.z=3 Baldry, I. K., Driver, S. P., Loveday, J., et al. 2012, MNRAS, Our conclusions are as follows: NASA/ESA421, 621 HST, which is operated by the Associa- tion of Universities for Research in Astronomy, Inc., un- der NASA contract NAS5-26555 and HST GO-11563. In this letter we take advantage of the recent discovery This work is based on observations taken by the 3D- of six bright z 9 10 candidate galaxies within CAN- HST Treasury Program (GO 12177 and 12328) with the DELS (Oesch et⇠ al. 2013b) to explore the size-luminosity NASA/ESA HST, which is operated by the Association and size-mass relations out to the earliest observable of Universities for Research in Astronomy, Inc., under epoch. By comparing sizes measured for sources in our NASA contract NAS5-26555. This research has made sample with similar size measures at lower redshifts, we use of NASA’s Astrophysics Data System. We acknowl- quantify the size evolution of galaxies. We establish these edge the support of ERC grant HIGHZ #227749, and a sizes by fitting the light profiles of the galaxies from the NWO “Vrije Competitie” grant 600.065.140.11N211. XDF and CANDELS WFC3/IR F160W observations. To ensure that our size measurements are accurate, we REFERENCES conducted an extensive set of comparisons against previ- ous size measurements. Baldry, I. K., Driver, S. P., Loveday, J., et al. 2012, MNRAS, Our conclusions are as follows: 421, 621 The Sizes of z=9-10 Candidate Galaxies 5

1. While most redshift z 9 10 candidate galaxies ⇠ are unambiguously resolved (re > 000. 1) with HST CANDELS or XDF F160W data (Figure 1), the brighter sources in our z 9 10 CANDELS sam- ⇠ ple are larger (=000. 16) and more resolved than the fainter z 10 candidates in the XDF ⇠ (=000. 09), allowing for a more optimal con- straints on the size evolution of galaxies to z 10. ⇠ 2. The mean e↵ective radius (=000. 16 observed, 0.6 kpc physical) for bright sources in the Oesch et al. (2013b) z =9 10 sample conforms to 1 expected size-redshift relation (e.g., (1 + z) or The Sizes of z=9-10 Candidate Galaxies1.5 5 (1+z) ) for bright star-forming galaxies. This is substantially smaller than the 0. 59 mean sizes de- 1. While most redshift z 9 1000 candidate galaxies The sizes of these z~9-10 candidates are exactly rived for intrinsically-red⇠ contaminants that other- are unambiguously resolved (re > 000. 1) with HST what we would expect... wise satisfy z 10 J125-dropout selection criteria. TheCANDELS present or size⇠ XDF measurementsF160W data therefore (Figure provide 1), the brighter sources in our z 9 10 CANDELS sam- additional evidence that⇠ these bright z=9-10 can- didatesple are largerare indeed ( the=0 indicated00. 16) and redshift. more resolved than the fainter z 10 candidates in the XDF ⇠ 3. The(=0900. 09),10 allowing candidate for galaxiesa more optimal are smaller con- IRAC-red andstraints brighter⇠ on the than sizez evolution7 galaxies, of galaxies a factor to ofz 1.610. interlopers growth in e↵ective radius⇠ for fixed luminosity⇠ (Fig-⇠ 2. Theure 2). mean e↵ective radius (=000. 16 observed, 0.6 kpc physical) for bright sources in the Oesch 4. Theet al. implied (2013b) SFRz surface=9 density10 sample for the conformsz 9 10 to ⇠1 1 2 expectedpopulation size-redshift is on average relation⌃SFR =4M (e.g., (1yr +z)kpc or 1.5 (1+but withz) a) wide for bright range star-forming in individual galaxies. values (⌃ ThisSFR is 1 2 ⇠ substantially1 10 M yr smaller kpc than , Figurethe 000. 59 2). mean These sizes sur- de- rivedface densities for intrinsically-red are similar, albeit contaminants slightly higher that other- than wisethose satisfy foundz by Oesch10 J125 et-dropout al. (2010) selection and Ono criteria. et al. The(2013) present at lower size⇠ redshifts. measurements therefore provide additional evidence that these bright z=9-10 can- 5. Thedidates size-mass are indeed relation at the we indicated find for our redshift.z 10 sam- ple is very similar to the one found by Mosleh⇠ et al. 3. The(2012)z at z9 610 with candidate these galaxies galaxies approximately are smaller ⇠ ⇠ Holwerda+2014Figure 4. The (resubmitted e↵ective to radius ApJL after as responding a function to referee of redshift report) for our andhalf brighterthe size of than star-formingz 7 galaxies, galaxies a factor at z of2 and1.6 growthz = 0 of in comparable e↵ective radius⇠ mass for (Figure fixed luminosity 3). ⇠ (Fig-⇠ sample for both bright (L>0.3Lz⇤=3,toppanel)andlower- luminosity galaxies (L<0.3Lz⇤=3,bottompanel).Forcomparison, ure 2). we show the mean sizes from earlier epochs from Bouwens et al. 6. Including the Oesch et al. (2013b) z 10 candi- ⇠ (2004); Oesch et al. (2010); Ono et al. (2013). The mean size of 4. Thedates, implied we find SFR that surface the sizes density of > for0.3 theLz⇤=3z galaxies9 10 the six potential interlopers to a z 9 10 selection (see 4.1) is 1.0 0.1 ⇠1 2 ⇠ § exhibitpopulation a (1 is + onz) average± scaling⌃SFR =4M with redshift,yr kpc con- well above any expected relation at z 9. We do not include the Bouwens et al. (2011a) z 2/z 12 candidate⇠ as there is consider- sistentbut with with a wide what range Bouwens in individual et al. (2004, values 2006) (⌃SFR and ⇠ ⇠ 1 2 ⇠ able doubt as to whether it is at z 12 (Ellis et al. 2013; Brammer 1Oesch10 et M al. (2010)yr kpc previously , Figure found 2). (Figure These 4). sur- et al. 2013; Bouwens et al. 2013; Capak⇠ et al. 2013; Pirzkal et al. face densities are similar, albeit slightly higher than 2013). The dotted line shows the best fits from Oesch et al. (2010). ACKNOWLEDGEMENTS Dashed lines are our fits to the Bouwens et al. and Oesch et al. those found by Oesch et al. (2010) and Ono et al. values combined with our mean size constraints at z 9 10. The This(2013) work at is lower based redshifts. on observations taken by the 1⇠ mean size of L>0.3L⇤ galaxies scale as (1 + z) . CANDELS Multi-Cycle Treasury Program with the z=3 5. The size-mass relation we find for our z 10 sam- NASA/ESA HST, which is operated by the⇠ Associa- tion ofple Universities is very similar for Research to the one in found Astronomy, by Mosleh Inc., et un- al. (2012) at z 6 with these galaxies approximately der NASA contract⇠ NAS5-26555 and HST GO-11563. In this letter we take advantage of the recent discovery half the size of star-forming galaxies at z 2 and Figure 4. The e↵ective radius as a function of redshift for our This work is based on observations taken by⇠ the 3D- ofsample six bright for bothz bright9 10 (L> candidate0.3L⇤ galaxies,toppanel)andlower- within CAN- HST Treasuryz = 0 of comparable Program (GO mass 12177 (Figure and 3).12328) with the ⇠ z=3 DELSluminosity (Oesch galaxies et ( al.L< 2013b)0.3Lz⇤=3 to,bottompanel).Forcomparison, explore the size-luminosity NASA/ESA HST, which is operated by the Association andwe show size-mass the mean relations sizes from out earlier to epochs the earliest from Bouwens observable et al. 6. Including the Oesch et al. (2013b) z 10 candi- (2004); Oesch et al. (2010); Ono et al. (2013). The mean size of of Universitiesdates, we for find Research that the sizesin Astronomy, of > 0.3L⇠ Inc.,galaxies under epoch. By comparing sizes measured for sources in our NASA contract NAS5-26555. This researchz⇤=3 has made the six potential interlopers to a z 9 10 selection (see 4.1) is exhibit a (1 + z) 1.0 0.1 scaling with redshift, con- wellsample above with any similar expected size relation measures at⇠z 9. at We lower do not redshifts, include§ the we use of NASA’s Astrophysics ± Data System. We acknowl- Bouwens et al. (2011a) z 2/z 12 candidate⇠ as there is consider- sistent with what Bouwens et al. (2004, 2006) and quantify the size evolution⇠ of⇠ galaxies. We establish these edge the support of ERC grant HIGHZ #227749, and a sizesable doubt by fitting as to whether the light it is profiles at z 12 of (Ellis the et galaxies al. 2013; from Brammer the Oesch et al. (2010) previously found (Figure 4). et al. 2013; Bouwens et al. 2013; Capak⇠ et al. 2013; Pirzkal et al. NWO “Vrije Competitie” grant 600.065.140.11N211. 2013).XDF The and dotted CANDELS line shows WFC3/IR the best fitsF160W from Oeschobservations. et al. (2010). ACKNOWLEDGEMENTS ToDashed ensure lines that are our our fits size to the measurements Bouwens et al. are and accurate, Oesch et we al. REFERENCES values combined with our mean size constraints at z 9 10. The This work is based on observations taken by the conducted an extensive set of comparisons against1⇠ previ- mean size of L>0.3L⇤ galaxies scale as (1 + z) . CANDELS Multi-Cycle Treasury Program with the ous size measurements.z=3 Baldry, I. K., Driver, S. P., Loveday, J., et al. 2012, MNRAS, Our conclusions are as follows: NASA/ESA421, 621 HST, which is operated by the Associa- tion of Universities for Research in Astronomy, Inc., un- der NASA contract NAS5-26555 and HST GO-11563. In this letter we take advantage of the recent discovery This work is based on observations taken by the 3D- of six bright z 9 10 candidate galaxies within CAN- HST Treasury Program (GO 12177 and 12328) with the DELS (Oesch et⇠ al. 2013b) to explore the size-luminosity NASA/ESA HST, which is operated by the Association and size-mass relations out to the earliest observable of Universities for Research in Astronomy, Inc., under epoch. By comparing sizes measured for sources in our NASA contract NAS5-26555. This research has made sample with similar size measures at lower redshifts, we use of NASA’s Astrophysics Data System. We acknowl- quantify the size evolution of galaxies. We establish these edge the support of ERC grant HIGHZ #227749, and a sizes by fitting the light profiles of the galaxies from the NWO “Vrije Competitie” grant 600.065.140.11N211. XDF and CANDELS WFC3/IR F160W observations. To ensure that our size measurements are accurate, we REFERENCES conducted an extensive set of comparisons against previ- ous size measurements. Baldry, I. K., Driver, S. P., Loveday, J., et al. 2012, MNRAS, Our conclusions are as follows: 421, 621 The colors of these z~9-10 candidates are exactly what we would expect... Derive from HST+Spitzer data Bright z~9-10 UV colors of Candidates similarly bright z~5 galaxies z~10 candidates are slightly bluer as expected

Red Blue Oesch+2014; Wilkins et al. 2014, in prep BEFORE OESCH+2014

How do the Oesch+2014 z~10 galaxies rank among the most distant galaxies ever discovered?

Name Redshift Discoverer

MACS0647-JD 10.8 Coe et al. (2013)

XDFj-38113333 9.8 Oesch et al. (2013) + Bouwens et al. (2011) MACS1149-JDLensed CLASH9.6 ...Zheng et al. (2012) HUDF12-42657049 9.5 Ellis et al. (2013) HUDF09-2_247HUDF129.4 sourcesMcLure et al. (2013) HUDF09-2_50104 9.0 McLure et al. (2013) Revived Bouwens+2011 HUDF12-42657049 8.8 Ellis et al. (2013) BEFORE OESCH+2014 How do the Oesch+2014 z~10 galaxies rank among the most distant galaxies ever discovered?

Name Redshift Discoverer

MACS0647-JD 10.8 Coe et al. (2013)

GN-z10-1 10.2 Oesch et al. (2014)

GN-z10-2 9.9 Oesch et al. (2014)

GS-z10-1 9.9 Oesch et al. (2014)

XDFj-38113333 9.8 Oesch et al. (2013) + Three of the Four Most DistantBouwens et al. (2011) MACS1149-JD Galaxies9.6 Known!Zheng et al. (2012)

GN-z10-3 9.5 Oesch et al. (2014) What new z~9-10 science can we expect in Cycle 22?

Follow-up bright z~10 galaxy with Expected WFC3/IR the HST Grism Spectrum PI: P. Oesch 0.03 1.1 1.2 1.3 1.4 1.5 1.6 1.7 0.03 1.1 1.2 1.3 1.4 1.5 1.6 1.7 z~10.2 galaxy LyWavelengthα Break µm F160W emission line contaminantWavelength µm 0.02 0.02

0.01 0.01 counts [e/s] 0 counts [e/s] 0

−0.01 −0.01 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.1 1.2 1.3 1.4 1.5 1.6 1.7 Wavelength [µm] Wavelength [µm]

Figure 4: Simulated 2D (top) and 1D (bottom) grism spectra for the proposed 10 grism observation Spectroscopic Confirmation of the z 10 candidate GN-z10-1 (2 orbits are F140W pre-imaging). The left panels show the spectrum in the case⇠ of a z = 10.2 galaxy (the best-fit photometric redshift), while the right panels show the spectrum Together with HST spectroscopy on the Coe+2013 of the best-fit lower redshift SED atαzlowz =2.1 shown in Fig 1. The gray lines represent the 1D spectra candidate,extracted these across could 000. 36 kill in the the spatialgame to direction. secure Thethe dark blue line is the same smoothed over a 230 A˚ most windowdistant inspectroscopically-confirmedin the the dispersion universe direction. with Ly The dark galaxies red line represents the input signal. The blue region indicates the coverage of the H160 band in which the source was detected and selected. The grism data will be able to definitely unveil the true nature of this candidate. Even with this relatively short exposure, we will for the first time be able to measure a redshift for a normal z>8 galaxy based on a continuum break in a spectrum. Since this region around 1.3-1.4 µm is blocked at the ground by the atmosphere (Figure 1), HST provides the only opportunity to measure the redshift of this extremely luminous z 10 galaxy candidate. ⇠

Data Processing and Release: The grism data will be reduced and analyzed in the same way as we did for all the existing HST WFC3/IR grism observations over the CANDELS fields from the 3D-HST program. Over the past two years we have developed data reduction tools that do not introduce noise correlations between pixels and provide nearly perfect background subtraction and extraction (see Brammer et al. 2013). We are thus perfectly set up to eciently work with these data. The reduced and stacked spectra of all sources within this pointing will be made publicly available to the community in a similar fashion as the recent 3D-HST team release of the processed grism data over the HUDF1. Special Requirements

In our analysis of the existing grism spectra over GOODS-N (see Fig 3), we found that the background varied rapidly within individual exposures by a factor > 3 , resulting in the loss of almost half of the data over our particular candidate. This⇠ happens⇥ due to strong airglow in the HeI 1.08 µm line when HST observes GOODS-N too close to the bright Earth limb (see Brammer et al. 2014). We will work closely with the WFC3 instrument team and schedulers to ensure that our observations will not be a↵ected by this significant source of background. Only under these conditions will it be possible to obtain the required S/N in the rather small number of 12 orbits for such a faint, distant, and unique galaxy that cannot be observed by any other telescope.

1http://3dhst.research.yale.edu/Data.html

7 What new z~9-10 science can we expect in Cycle 22?

Follow-up bright ~7 more bright z~9-10 ~9 more bright z~9-10 z~10 galaxy with candidates using candidates using ambitious the HST Grism remaining CANDELS fields pure-parallel program PI: P. Oesch PI: R. Bouwens ExamplePI: M. BrightTrenti z~9 Galaxy over CANDELS COSMOS 15 more bright z~9-10 galaxies from 500 orbits!

Use All of Spectroscopic Confirmation CANDELS MoreSightlines Random How Bright Can Galaxies Become at z~4-8?

UV luminosities can reach ~3-4 L*(z=3)

Can approximately quantify using characteristic luminosity from Schechter fit

Best Derived Using All Wide-Area CANDELS fields

Bouwens+2014 How Bright Can Galaxies Become at z~4-8?

Relatively Limited

Characteristic Luminosity Characteristic Evolution “Maximum” UV Luminosity “Maximum”

Bouwens+2014 But individual galaxies become more UV luminous with cosmic time (Bouwens+2007), no?

upsizing of galaxies in UV luminosity first quantified by Bouwens+2007

and later quantified in rest-frame optical (Stark+2009)

this upsizing of galaxies was “rediscovered” and more properly quantified by Papovich+2011 and Lundgren+2014 using a cumulative number-density matching formalism Bouwens+2007 luminosity evolution is still true... but only for normal galaxies...

Evolution of Bouwens+2014 (using simple-mindedBouwens+2007/2008 luminosity UV evolution model [using M*]) LuminosityAlmost exactly the same evolution as (for a “normal” before for “normal” galaxies! galaxy)

Bouwens+2014 One might guess that the brightest and rarest objects brighten in the same way in the UV?

But this does not seem to occur! However, More Limited Evolution in UV luminosity for the brightest galaxies

Relatively Limited Evolution “Maximum” UV Luminosity “Maximum”

Bouwens+2014 Why Little Evolution in Maximum Brightness of Galaxies in UV? Probably Dust Extinction Idea that L*(UV) was set by dust extinction first proposed by Bouwens+2009 and Reddy+2010

Observed UV Saturates Luminosity

Correlated

Star Formation Rate Bouwens+2009 Credit: Elbaz, Obergurgl 2007 If the UV luminosity of galaxies saturates at a maximum value, how would the UV LF evolve?

COMMON Develop simple LF model here

FINALLY Assume galaxy light traces DENSITY the the halo mass EVOLUTION evolution

but UV light cannot exceed INITIALLY some maximum value LUMINOSITY EVOLUTION?

RARE mass UV BRIGHT FAINT Saturation Luminosity What about the bolometric LF?

COMMON Develop simple LF model here

Assume galaxy light traces the the halo mass Quenching likely evolution becomes important at some mass (of course) LUMINOSITY EVOLUTION?

RARE

Bolometric BRIGHT FAINT Could dust set an upper limit on UV luminosity of galaxies at z>=4 ? Let’s first look at the UVSome colors phenomenon of z~4 is galaxies causing the depend on UVbrightest luminosity objects to become particularly red...

red Dust?

Steeper Dependence at Bright Luminosities Dust? Color trend is flatter... Dust not important? UV continuum slope slope continuum UV (“color”) blue

“bright” “faint”

Bouwens et al. 2013; see also Bouwens et al. 2009, 2010, 2012; Wilkins et al. 2011; Dunlop et al. 2012; Castellano et al. 2012; Finkelstein et al. 2012 UV Luminosity Saturates in Bright Galaxies

Combination of UV slope and UV-to-optical colors indicate that MUV saturates due to increasing dust extinction.

Trend between UV colors and luminosity ") becomes clear vs. optical luminosities # "!'( BC!

$'( >?+@1A(B*>?+C'(1A" SMC "! $ z=4 4-=7>?@A

;< CA $'( ""'( -:-7 # +, #'( ""

23*45678699.*:;5<=* ! Steeper Dependence at "#'( !'( Bright Luminosities Dust? Color trend is flatter... 563-.7865-9-7 " Dust not important? !" !! !# !$ #% #& "# ) *+,-*./01 Oesch et al. 2013 “bright” %$$$ “faint” 7D.BCE4 $'( !! !" !# "$ "% "& * -./0-1234 +, Characteristic luminosity of UV LF may not be determined by characteristic SFR, but rather by upturn in dust extinction. (see Reddy+10 for analog results at z~2-3)

Ringberg, June 2013 P. Oesch, UCSC UCO/Lick Observatory 27 z~5-6 galaxies show a similar dependence on UV luminosity as at z~4...

red

Steeper Dependence at Bright Luminosities Steeper Dependence at blue Bright Luminosities Dust? Dust?

bright faint bright faint bright faint

Bouwens et al. 2013 More generally that z~4-8 galaxies have similar colors as a function of luminosity

red

Dusty, high SFR systems are generally here at z~4...

Could dust set L*(UV) as high as z~7? blue

Bouwens et al. 2013 Bouwens+2014 How does the faint-end slope of the LF change with redshift?

Shallow

Faint-end Slope

Steep

see also Bouwens+2011, Oesch+2012, Bradley+2012, McLure+2013, Bouwens+2014 Schenker+2013; Schmidt+2014 How does the faint-end slope of the LF change with redshift?

Shallow Evolution significant at 4.5σ

Almost the same Faint-end steepening as the Slope halo mass function

Steep

see also Bouwens+2011, Oesch+2012, Bradley+2012, McLure+2013, Bouwens+2014 Schenker+2013; Schmidt+2014 How does the shape of the LF change with redshift?

Also evident in parameter free way:

Renormalize LFs vertically to agree here

Bouwens+2014 Evolution of the faint-end slope in the simple LF model shown earlier

Faint-end slope of the LF is COMMON much steeper at earlier times

RARE mass UV BRIGHT FAINT Saturation Luminosity New determinations of UV LF at z~4, 5, 6, 7, 8, 10 from all HST Legacy Fields (Bouwens et al. 2014, arXiv:1403.4295) Summary / Conclusions

Current HST data sets allow us to identify as many as >800 galaxies at z~7-8 and 15 galaxies at z~9-10...

Six luminous (~L*(z=3)) galaxy candidates at z~9-10 have been identified over CANDELS (Oesch+2014). These candidates have exactly the volume density, colors, and sizes we would expect if they were z~9-10 galaxies.

Amazingly, the newly discovered population of bright z~9-10 galaxies appear to be more robust than the fainter z~9-10 candidates in the HUDF.

Our large samples of bright z~4-7 galaxies from the 5 CANDELS + BORG fields allow us to set robust constraints on the volume density of bright galaxies. The characteristic luminosity M* at z~7 appears to be similar to z~3. We speculate this is due to the UV light saturating at a certain SFR.

The UV LF shows strong evidence (4.5σ) for being progressively steeper at high redshift to faint-end slope of −2.06 ± 0.12 at z=7. The observed evolution similar to expected evolution in halo mass function.