1
Constraints on the First Galaxies: z~10 Galaxy
Candidates from HST WFC3/IR
R. J. Bouwens1,2, G. D. Illingworth1, I. Labbé3, P. A. Oesch4, C. M. Carollo4, M. Trenti5, P.
G. van Dokkum6, M. Franx2, M. Stiavelli7, V. González1, D. Magee1
1
Department of Astronomy, University of California Santa Cruz,
2
Leiden Observatory, Leiden University, Netherlands
3
Carnegie Observatories, Pasadena, CA
4
Institute for Astronomy, ETH Zurich, Zurich, Switzerland
5University of Colorado, Center for Astrophysics and Space Astronomy, Boulder, CO
6
Department of Astronomy, Yale University, New Haven, CT
7
Space Telescope Science Institute, Baltimore, CA
The first galaxies likely formed a few hundred million years after the Big Bang. Until
recently, it has not been possible to detect galaxies earlier than ~750 million years
after the Big Bang. The new HST WFC3/IR camera changed this when the deepestever, near-IR image of the universe was obtained with the HUDF09 program. Here
we use this image to identify three redshift z~10 galaxy candidates in the heart of the
reionization epoch when the universe was just 500 million years old. These would be
the highest redshift galaxies yet detected, higher than the recent detection of a GRB at
z~8.2.1,2 The HUDF09 data previously revealed galaxies at z~73-7 and z~8.4-8 Galaxy
stellar population models predict substantial star formation at z>9-10.9,10 Verification
by direct observation of the existence of galaxies at z~10 is the next step. Our
2
conservative search and extensive testing for contamination and spurious images
suggests that we can set reliable constraints based upon our 3 z~10 candidates, unlike
a recent claim6 of 20 z~10 sources which appear to be spurious. The detection of
galaxies at z>8 is further enhanced by our detailed analysis of 2 other faint sources
likely at z~8.4 and z~8.7.
Our z~10 sample suggests that the luminosity function and
star formation rate density evolution found at lower redshifts11,12 continues to z~10,
and pushes back the timescale for early galaxy buildup to z>10, increasing the likely
role of galaxies in providing the UV flux needed to reionize the universe. The true
nature of these galaxies, at just 4% of the age of the universe, will remain hidden until
JWST is launched.
In August 2009 HST took the deepest near-infrared image ever obtained on the
Hubble Ultra Deep Field using the new WFC3/IR camera on HST. This near-IR image,
taken as part of the HUDF09 program (HST GO 11563: PI Illingworth), complements the
HUDF optical image taken in 2004 with the ACS. WFC3/IR is ~40X faster at finding z~7
galaxies than the NICMOS camera on HST, as recent results from the HUDF09 program
have demonstrated.3-8 These first-epoch near-IR data from the HUDF09 program reach to
28.8 AB mag (5!: 0.35"" apertures) in each of the three bands (Y105, J125, and H160) , and
provide an opportunity to explore at redshifts z>7 deep into the reionization epoch.
Detecting high redshift galaxies through their Lyman break by the photometric
“dropout” technique has resulted in a sample of over 6000 galaxies at 3<z<611 from HST
since the ACS became available in 2002. This approach has been verified through
numerous spectroscopic confirmations at redshifts from z~2 to z~6.13,14 However, the high
redshift galaxy sample drops dramatically at z~7 to just ~40, with the first robust detection
being reported in this Journal just 3 years ago. The near-IR HUDF09 data allows us to
3
increase the sample at the highest redshifts. It has already yielded the first reliable
detections of z~8 galaxies,4-8 along with larger samples of z~7 galaxies, that together with
the reports of a GRB at z~8.2,1,2 constitute the earliest objects reliably reported to date.
Yet these new near-IR data also provide an opportunity to detect z~10 galaxies in the
H-band since it is redward of the redshifted Lyman break at z~10. A distinguishing
characteristic of such galaxies would be the absence of flux in the J-band (“J-dropouts”)
and all other WFC3/IR and ACS filters. Remarkably, after a thorough search of the new
WFC3/IR HUDF09 data we report the discovery of 3 candidate z~10 J-dropout galaxies
(see Figure 1, and Figure 2 for the redshift distribution). These candidates are detected at
5! in the WFC3/IR H-band data. The sources are listed in Supplementary Table S1. Two of
the z~10 candidates also lie in the NICMOS images of the HUDF15 and are detected in a
stacked NICMOS image (see Figure 1). We consider the exciting implications of these first
constraints on the z~10 galaxy population, but first we discuss the reliability of our
detections.
Given the importance of these observations we perform extensive tests and
simulations to estimate the likelihood that these sources are real, rather than spurious or
contaminants, i.e., either stars (e.g., T-dwarfs) or lower redshift galaxies (particularly nonstar forming, low luminosity very red galaxies at z~1.5-3) or noise fluctuations (e.g., that
enhance the visibility of faint, lower redshift galaxies). The source selection, validity
testing and incompleteness estimation utilize established procedures11-12,16-18 that are
discussed in the attached Supplementary Information in sections 2, 3, 4, and 5. The
dependability of these procedures and tests has been validated by the much-improved
WFC3/IR observations3 which confirmed our previous z~7 galaxy detections using
NICMOS.17,18
4
For the HUDF09 z~10 J-dropout candidates, assessing the reliability is particularly
important since they can only be measured in a single band (the H-band). The risk of
contamination is thus higher than for the z~7 and z~8 samples where two (or more) bands
are normally used to measure the source magnitudes and colors. Fortunately, we can test
the robustness of single-band detections by selecting z~8 Y-dropouts using the J-band data
alone, mirroring the single-band approach used for our z~10 J-dropouts. It gives us an
opportunity to check the results against the robust z~8 detections using two bands8 (J and
H). We are very encouraged that we reproduce our previous result (the same 5 Y-dropouts
from Bouwens et al.8) and that we find only 1 additional source. The reason for the
robustness is the non-detection in all shorter wavelength filters (see e.g., Figure 1), which
largely eliminates contaminating objects. All the sources were also checked for any Spitzer
IRAC flux in the 3.6 µm band, including a stacking analysis. None are detected to ~27 AB
mag, strengthening the case that they are very high redshift sources rather than highlyreddened, low redshift contaminants.
Using the results of these tests and Monte-Carlo simulations we estimated a
contamination rate of ~0.1 source from spurious sources and ~0.2 from photometric scatter.
Together these give a contamination rate of only ~10%. Estimating the contamination from
lower-redshift red sources is more difficult, but the above single filter Y-dropout test
suggests that a conservative value could be ~1 source. Including this possible
contamination increases the rate to 40%. Determining the contamination rate and
minimizing it is clearly important. An unreliable sample can be obtained if inadequate
testing is performed or the search is carried out too aggressively, as for example appears to
have happened in a recent study. Here we find that most, if not all, of the 20 objects6
purported to be at z~10 in that study are likely to be spurious (see the Supplementary Info).
5
To be conservative, we base all our subsequent analyses on a volume density of z~10
sources that is decreased by 40% to account for contamination. Our volume estimates
already take into account the effect of incompleteness. Given the substantial contamination
rate and the small sample, we also consider the possibility that none of the candidates are
z~10 galaxies, i.e., if our current search at z~10 provides only upper limits. We note,
however, that the existence of galaxies at z>8.2 is strengthened by a two additional sources
that were already detected in recent searches.4-8 The updated redshift distributions from our
simulations show that these two sources8 are most likely to be at z~8.4 and z~8.7 (Figure
2).
These results have far-reaching implications. Despite the small sample, and
regardless of the contamination rate, the constraints now obtained at z~10 are striking. We
first contrast what we see at z~10 with expectations based on a “no-evolution” scenario,
i.e., the galaxy populations stay unchanged with time. We compute the “no-evolution”
estimate by using our “galaxy cloning” software19-20 to artificially redshift the observed z~6
and z~7 galaxy population to z~10, add them randomly to our data, and then repeat the
object selection process just as for the observed z~10 galaxies. We estimated that we
would find 9±3 z~10 galaxies using our z~7 detections as the baseline, and 20±5 z~10
galaxies using our z~6 detections as a baseline. This is substantially more than the
(contamination-corrected) ~2 z~10 galaxies found in our similar conservative selections.
These results suggest that there has been significant continuing evolution for galaxies that
lie at the bright end of the LF. For simple Poissonian statistics, the present samples are
inconsistent with no-evolution at 76% and 99.6% confidence, respectively (and set even
stronger limits on any “upturn” in the SFR6).
6
We estimate the field-to-field variance on the present z~10 J-dropout searches to be
39% in the HUDF09 field (see the Supplementary Material).21 While significant, cosmic
variance (“large-scale structure”) is not the dominant source of uncertainty in this case,
given the small numbers of candidates that we have identified.
The present search results also allow us to make plausible inferences about the
luminosity function at z~10 (Figure 3). The luminosity function (LF) describes the number
density of galaxies versus luminosity and is usually parameterized as "* e#L/L*(L/L*)$
where "* is the normalization, L* is the characteristic luminosity L*, and $ is the faint end
slope. While it is unclear if such a parameterization is appropriate at such high redshifts we
will utilize this form for now until larger samples at z>7 indicate otherwise. The high
redshift galaxy UV LF maintains a nearly constant form and evolves in a somewhat selfsimilar manner for more than 1.4 Gyr from z~7 to z~3, or from 0.7 Gyr to 2.1 Gyr.
Specifically, over this period, the UV luminosity function maintains a very steep faint-end
slope $ ~ #1.73.3,11,22-25 Furthermore, it is now well established that L* appears to brighten
significantly but smoothly from z~7 to z~4 (Figure 3).3,11,17,23,25 It thus seems a reasonable
assumption that the UV luminosity function at z~10 has a characteristic luminosity L* no
brighter than at z~7, with "* and $ derived from the z~4-6 fitting relations.11 With these
expectations as a baseline, the present sample at z~10 permits us to estimate the
characteristic luminosity at z~10. We find L* at z~10 to be #18.7 ± 0.4 AB mag, or L* >
#18.5 in the limit of no detected sources – indicating that evolution in the UV LF seen from
z~7 to z~4 continues to z~10.
The existence of a steep slope $ to the faint end of the UV luminosity function found
at z~6-73,11 highlights the potential importance of low luminosity galaxies in providing the
flux needed to reionize the universe. It thus becomes of great interest to estimate the UV
7
luminosity density at redshifts 7-10 where reionization most likely occurred, bounded by its
apparent completion at z~626-27 and its onset at (a very uncertain) z~11 from WMAP.28 The
recent results at z~7 and z~8 from the HUDF09 dataset provide other constraints on the UV
luminosity density.3-8 Utilizing our z~10 results we show in Figure 4 the rest-frame
continuum UV luminosity density (and the star formation rate density derived from the
extinction-corrected luminosity density by standard procedures11,12,17) at z~10 integrated
down to our approximate magnitude limit MAB ~ #18. As before, we assume that the UV
LF has the same mean slope (and normalization) as at z~6 and z~7. The star formation rate
density increases systematically and monotonically at early times from z~10 (0.5 Gyr) to
z~4 (1.6 Gyr), peaking at z~2-3 (at ~2.5 Gyr), before decreasing at z<2. The large increase
in the star formation rate density inferred recently6 at z~8-10 is not seen here (as expected,
since the sources in that analysis appear to be spurious, as noted above).
The observed luminosity density out to z~10 provides insights into the role of
galaxies in the reionization of the universe? If we compute the luminosity density implied
by our sample and extend it to #16 AB mag assuming a faint-end slope of #1.7, we find
that the UV flux that is available from galaxies with sub-solar metallicities at z~10 is ~13%
of what is needed for galaxies to be the reionizing source, with typical assumptions of an
escape fraction of ~0.5, a clumping factor of ~3 and a Salpeter IMF. While this does not
provide a definitive result regarding the source of the reionizing photons needed for
reionization, it does suggest that galaxies are playing a non-trivial role.
The detection of galaxies at z~8.4-10, just 500-600 Myr after recombination, and in
the heart of the reionization epoch provides clues about galaxy build-up at a time that must
have been as little as 200-300 Myr since the first substantial star formation (at z~15-20?).
The existence of galaxies at z~10 shows that galaxy formation was well underway at this
8
very early epoch. The new HST WFC3/IR camera has provided insights to the exciting
results that await us when the James Webb Space Telescope (JWST) characterizes galaxy
formation and evolution in the first 500 Myr.
Acknowledgements: We are grateful to all those at NASA, STScI and throughout the community who have
worked so diligently to make Hubble the remarkable observatory that it is today. The servicing missions, like
the recent SM4, have rejuvenated HST and made it an extraordinarily productive scientific facility time and
time again. We greatly appreciate the support of policymakers, and all those in the NASA flight and servicing
programs who contributed to the repeated successes of the HST servicing missions. We acknowledge our
program coordinator William Janusweski for his exceptional care in helping to set up our program and
observing configuration. We acknowledge the support of NASA grant NAG5-7697 and NASA grant HSTGO-11563.01. P.O. acknowledges support from the Swiss National Science Foundation.
Correspondence and requests for materials should be addressed to R.B. (e-mail: bouwens@ucolick.org)
1. Tanvir, N., et al. A #-ray burst at a redshift of z~8, Nature, 461, 1254
2. Salvaterra, R., et al. GRB090423 at a redshift of z~8.1, Nature, 461, 1258
3. Oesch, P.A., et al. z~7 Galaxies in the HUDF: First Epoch WFC3/IR Results,
Astrophys. J. Lett., in press, arXiv:0909.1806
4. McLure, R., et al. Galaxies at z~6-9 from the WFC3/IR imaging of the HUDF, Month.
Not. R. Astron. Soc., in press, arXiv:0909.2437
5. Bunker, A., et al. The Contribution of High Redshift Galaxies to Cosmic Reionization:
New Results from Deep WFC3 Imaging of the Hubble Ultra Deep Field, Month. Not. R.
Astron. Soc., in press, arXiv:0909.2255
9
6. Yan, H., et al. Galaxy Formation In The Reionization Epoch As Hinted By Wide Field
Camera 3 Observations Of The Hubble Ultra Deep Field, Astrophys. J., submitted,
arXiv:0910.0077
7. Finkelstein, S. et al. On the Stellar Populations and Evolution of Star-Forming Galaxies
at 6.3<z<8.6, Astrophys. J., submitted, arXiv:0912.1338.
8. Bouwens, R.J., et al. Discovery of z~8 Galaxies in the HUDF from ultra-deep WFC3/IR
Observations, Astrophys. J. Lett., in press, arXiv:0909.1803
9. Gonzalez, V. et al. Stellar Mass Density and Specific Star Formation Rates of the
Universe at z~7, Astrophys. J., submitted, arXiv:0909.3517.
10. Labbe, I., et al. Ultradeep IRAC Observations of sub-L* z~7 and z~8 Galaxies in the
HUDF: the Contribution of Low-Luminosity Galaxies to the Stellar Mass Density and
Reionization, Astrophys. J. Lett., 708, L26 (2010).
11. Bouwens, R.J., Illingworth, G.D., Franx, M., & Ford, H. UV Luminosity Functions at
z~4, 5, and 6 from the Hubble Ultra Deep Field and Other Deep Hubble Space Telescope
ACS Fields: Evolution and Star Formation History, Astrophys. J., 670, 928-958 (2007).
12. Bouwens, R., et al. UV-Continuum Slope and Dust Obscuration from z~6 to z~2: The
Star Formation Rate Density at High Redshift, Astrophys. J. 705, 936-961 (2009).
13. Steidel, C.C., Giavalisco, M., Pettini, M., Dickinson, M., & Adelberger, K.L.
Spectroscopic Confirmation of a Population of Normal Star-forming Galaxies at Redshifts
Z > 3, Astrophys. J. Lett., 462, 17-20 (1996).
14. Vanzella, E., et al. Spectroscopic Observations of Lyman Break Galaxies at Redshifts
~4, 5, and 6 in the Goods-South Field, Astrophys. J., 695, 1163 (2009).
10
15. Thompson, R., et al. The near-Infrared Camera and Multi-Object Spectrometer Ultra
Deep Field: Observations, Data Reduction, and Galaxy Photometry, Astron. J., 130, 1
(2005).
16. Bouwens, R.J., Illingworth, G.D., Blakeslee, J.P., & Franx, M. Galaxies at z~6: The
UV Luminosity Function and Luminosity Density from 506 UDF, UDF-Ps, and GOODS idropouts, Astrophys. J., 653, 53 (2006).
17. Bouwens, R.J., Illingworth, G.D., Franx, M., & Ford, H. z ~ 7-10 Galaxies in the
HUDF and GOODS Fields: UV Luminosity Functions, Astrophys. J., 686, 230 (2008).
18. Bouwens, R.J., & Illingworth, G.D. Rapid evolution of the most luminous galaxies
during the first 900 million years, Nature, 443, 189 (2006).
19. Bouwens, R.J., Broadhurst, T.J., Silk, J. Cloning Hubble Deep Fields. I. A Modelindependent Measurement of Galaxy Evolution, Astrophys. J. 506, 557-578 (1998).
20. Bouwens, R.J., Broadhurst, T.J., Illingworth, G.D. Cloning Dropouts: Implications for
Galaxy Evolution at High Redshift, Astrophys. J. 593, 640-660 (2003).
21. Trenti, M. & Stiavelli, M. Cosmic Variance and its Effect on the Luminosity Function
Determinations in Deep High-z Surveys, Astrophys. J., 676, 767-780 (2008).
22. Reddy, N., & Steidel, C.C. A Steep Faint-End Slope of the UV Luminosity Function
at z ~ 2-3: Implications for the Global Stellar Mass Density and Star Formation in LowMass Halos, Astrophys. J. 692, 778-803 (2009).
23. Yoshida, M., et al. Luminosity Functions of Lyman Break Galaxies at z~4 and z~5 in
the Subaru Deep Field, Astrophys. J. 653, 988-1003 (2006).
24. Yan, H.J. & Windhorst, R.A. Candidates of z~5.5-7 Galaxies in the Hubble Space
Telescope Ultra Deep Field, Astrophys. J. Lett., 612, 93-96 (2004).
11
25. McLure, R., Cirasuolo, M., Dunlop, J.S., Foucaud, S., Almaini, O. The luminosity
function, halo masses, and stellar masses of luminous Lyman-break galaxies at 5<z<6,
Month. Not. R. Astron. Soc., 395, 2196-2209 (2009).
26. Becker, et al. Evidence for Reionization at z~6: Detection of a Gunn-Peterson Trough
in a z=6.28 Quasar, Astronom. J., 122, 2850-2857 (2001).
27. Fan, X., et al. Evolution of the Ionizing Background and the Epoch of Reionization
from the Spectra of z~6 Quasars, Astronom. J., 123, 1247-1257 (2002).
28. Komatsu, E., et al. Five-Year Wilkinson Microwave Anisotropy Probe Observations:
Cosmological Interpretation, Astrophys. J., 180, 330-376 (2009).
29. Beckwith, S.W., et al. The Hubble Ultra Deep Field, Astrophys. J., 132, 1729-1755
(2006).
30. Madau, P., Pozzetti, L. & Dickinson, M. The Star Formation History of Field Galaxies,
Astrophys. J., 498, 106-116 (1998).
31. Schiminovich, D., et al. The GALEX-VVDS Measurement of the Evolution of the
Far-Ultraviolet Luminosity Density and the Cosmic Star Formation Rate, Astrophys. J.
Lett., 619, 47-50 (2005).
12
Figure 1. Optical and near-infrared images of the three z~10 J-dropout candidate
galaxies from the ultra-deep HUDF optical ACS (V+i+z) data29 and the
correspondingly deep HUDF09 near-IR WFC3/IR (Y, J, H) data.3,8 None of the
candidates is detected in the deep ACS BViz observations. A stacked NICMOS
image with a faint, 3! detection is shown next to the two z~10 candidates that lie in
the NICMOS data on the HUDF. Other properties of the candidates are given in
Table 1 in the Supplementary Material. Each cutout is 2.4!! " 2.4!! on a side
13
Figure 2. Predicted redshift distribution for the current z~10 J-dropout candidate
sample (red line) and the two highest-redshift z~8.5 Y-dropout candidates (blue
and green lines). The redshift distributions were derived (see Supplementary Info
§8) by adding artificial sources to the HUDF09 WFC3/IR data and reselecting them
in the same way as the actual galaxy candidates (see Supplementary Info §3-4).
The mean redshifts of these distributions are 8.4, 8.7, and 10.2. For these
simulations, the UV Luminosity Function (LF) was assumed to have an M*UV,1600Å
of !19.4 and !18.7 at z ! 8 and z ! 10, respectively, while "* and # were taken to
be 0.0012 Mpc!3 and !1.74, respectively (based on predictions from our z~4-6
fitting relation17).
14
Figure 3. Previously determined UV luminosity functions (LF) at redshift z~4, 7 and
8, with new constraints at z~10. Constraints on the stepwise UV LF at z ! 10 (black
upper limits) are derived from the J-dropout candidate galaxies over our ultra-deep
HUDF WFC3/IR field. The stepwise LF at z ! 10 is also presented as 1$ upper
limits, given the uncertainties in the nature of z ! 10 candidates. The lowest
luminosity point has been corrected for incompleteness. The UV LFs3,8,11 at z ! 4
(blue), z ! 7 (magenta), and z ! 8 (red) are shown for context.
15
Figure 4. The luminosity density and star formation rate density of the universe.
The luminosity densities are plotted down to a flux limit of –18.3 AB mag (0.07L*)
to match the current z~7 z-dropout search limit. The current result from the
contamination-corrected sample is shown at z ! 10 from the current J-dropout
search, as is the upper limit if we assume no z~10 sources are detected. The
conversion from UV luminosity to SFR is performed assuming a Salpeter IMF.30
Also included here are the recent SFR determinations at z ! 7 and z ! 8 from our
HUDF09 WFC3/IR z-dropout and Y-dropout searches,3,8 and from the literature for
z<7.11,22,31 The dust corrections at z ~ 4 are based on the estimated UV-continuum
slopes % and are already negligible by z ~ 7.12
Supplementary Information: Constraints on the First Galaxies:
z ! 10 Galaxy Candidates from HST WFC3/IR
R.J. Bouwens, G.D. Illingworth, I. Labbé, P.A. Oesch, M. Carollo, M. Trenti, P.G. van
Dokkum, M. Franx, M. Stiavelli, V. González, D. Magee
1.
Observational Data
We make use of the ultra-deep 4.7 arcmin2 WFC3/IR observations over the HUDF from
the HUDF09 program (GO 11563: PI Illingworth) and the wide-area 39.2 arcmin2 WFC3/IR
observations over the GOODS field from the Early Release Science (ERS) program (GO
11359: PI O’Connell). Our reductions of the ultra-deep first epoch WFC3/IR observations
over the HUDF from the HUDF09 program are those described in Oesch et al. (2009b) and
Bouwens et al. (2009c).
The WFC3/IR HUDF09 Y105 , J125 , and H160 data reach to 28.8, 28.8, and 28.8, respectively, in a 0.35!! -diameter aperture. The ACS/WFC B435 , V606 , i775 , and z850 -band optical
data over the HUDF (Beckwith et al. 2006) reach to 29.4, 29.8, 29.7, and 29.0 AB mag, respectively (5": 0.35!! -diameter apertures). The FWHM of the PSFs is ! 0.10!! and ! 0.16!! in
the ACS BV iz and WFC3/IR Y J H observations, respectively.
The WFC3/IR ERS Y098 , J125 , and H160 observations reach to 27.3, 27.7, and 27.4
respectively, in a 0.35!! -diameter aperture (see R.J. Bouwens 2009, in prep for more details).
The ACS/WFC B435 , V606 , i775 , and z850 -band optical data are available from the ACS
GOODS program (Giavalisco et al. 2004) and various SNe search and follow-up programs
(Riess et al. 2007). Our reductions of these data (Bouwens et al. 2006, 2007) are similar to
the GOODS v2.0 reductions, but reach !0.1-0.3 mag deeper in the z850 band (due to the
inclusion of the SNe follow-up data: Riess et al. 2007). These reductions reach to 28.0, 28.2,
27.5, and 27.4 AB mag in the B435 , V606 , i775 , and z850 bands, respectively (5": 0.35!! -diameter
apertures).
2.
Source Catalogues
Our procedure for doing object detection and photometry is essential identical to that
performed in many previous analyses by our team (e.g., Bouwens et al. 2003, 2007, 2008,
2009b). We provide a brief summary here. The SExtractor software (Bertin & Arnouts
1996) is used to do object detection and photometry. Object detection is performed using
–2–
the H160 -band image. The ACS optical images are smoothed to match the WFC3/IR images,
and colors are measured in small scalable apertures (using a Kron [1980] factor of 1.2).
Fluxes in these small scalable apertures are corrected to total magnitudes, considering the
flux in a larger scalable aperture (Kron factor of 2.5). This latter correction is made off
the H160 -band image. Finally, a correction is made for light outside this large scalable
aperture using the encircled energies tabulated for point sources in the WFC3/IR F160W
band (http://www.stsci.edu/hst/wfc3/documents/handbooks/currentIHB/c07 ir7.html)
3.
J125 -dropout Selection
We search for high redshift galaxies at and beyond z ! 9, utilizing a J125 -dropout
selection. Such a selection should permit us to identify candidate z!10 star-forming galaxies
in the ultra-deep WFC3/IR data, should they be brighter than 29.2 AB mag (>0.07 L"z=3 ).
For our selection criteria, we require candidates to have J125 ! H160 colors redder than
1.2 AB mag (see Figure 1), to be undetected (< 2") in all imaging observations blueward of
the J125 band, and to be detected at 55" detection in the H160 -band. In addition, candidates
must not be detected at >1.5" in more than one band blueward of the J125 band. Finally, to
ensure the optical imaging data are used to maximal effect to reject low-redshift interlopers,
we additional require that candidates not have a #2 value >2.5 in the B435 V606 i775 z850 Y105 band #2 image (Szalay et al. 1999). While this will reject some bona-fide z ! 10 candidates,
our primary objective with this selection is to make it as clean as possible (given the inherent
difficulty in selecting z ! 10 galaxies at modest S/N levels).
We identified 3 sources which satisfied these criteria over the ultra-deep HUDF09 WFC3/IR
data. These sources are listed in Table 1 and shown in Figure 1 of the main text. None of
these three candidates has been presented in previous z ! 10 searches – even though they
nominally satisfy the Yan et al. (2009) J125 -dropout criteria. The absence of the present
J125 -dropout candidates in the Yan et al. (2009) selection may be due to catastrophic incompleteness that sometimes afflicts catalogs made using SExtractor – where 10% of the sources
are lost due to deblending with nearby galaxies.
To ascertain the nature of the 3 J125 -dropout candidates over the HUDF, we examined
the IRAC 3.6$ and 4.5$ data (reaching to ~27.7 AB mag and ~27.1 mag, respectively,
at 1") available for these sources, after carefully modelling and subtracting the IRAC flux
from nearby neighbors (e.g., as in Labbé et al. 2006). None of the three sources show
significant (>2") 3.6$ or 4.5$ detections in the IRAC data (though UDFj-436964076 and
UDFj-38116243 are sufficiently close to nearby neighbors that photometry at 3.6$ or 4.5$ is
–3–
possible but difficult). A stack of the three J125 -dropout candidates also shows no detection
(<1!) at 3.6" or 4.5" (see Figure 3). This sets a 1! limit on the 3.6" magnitude of these
J125 -dropouts of >28.3 AB mag. This implies a H160 # 3.6" color of <0.7 mag – which is
consistent with these sources being star-forming galaxies at z!10. Even if we suppose that
one of the J125 -dropout candidates is spurious (and we discuss possible contamination in §7),
this nevertheless implies a H160 # 3.6" color <1.1 mag.
4.
Y105 -dropout Selection
We also search for galaxies at redshifts somewhat higher than 8, and in particular higher
than the redshift of the recent gamma-ray burst source found at z ! 8.2 (Tanvir et al. 2009;
Salvaterra et al. 2009). For this, we require sources to have a very strong Y105 # J125 > 1.5
breaks (see Figure 1), to be undetected (< 2!) in all bands BViz blueward of the break,
and to be detected at >~5.5! in the J125 -band to ensure they correspond to real sources.
This Y105 -dropout selection criterion is somewhat more stringent than what we applied in
our initial search over the ultra-deep HUDF WFC3/IR data and identifies galaxies in the
range z!8.2-8.8, with a mean redshift of z ! 8.5 (see Figure 2 of the main text).
We identified 2 sources which satisfied these selection criteria. These sources are listed
in Table 1 and shown in Figure 2. Both of these candidates were presented in the Bouwens
et al. (2009c) Y105 -dropout selection. These candidates were also presented in the McLure
et al. (2009b) z!6-9 search, the Bunker et al. (2009) Y105 -dropout selection, the Yan et al.
(2009) Y105 -dropout sample, and the Finkelstein et al. (2009) z!6.3-8.6 sample.
5.
J125 -dropout selection over the Early Release Science observations
We also conducted a search for z!9-10 J125 -dropout candidates over the wide-area ERS
observations. The only sources with J125 #H160 colors >1.2 were also quite bright at 5.8" and
8.0", and hence were likely z!2-3 dusty galaxies (having H160 # 5.8" colors redder than !2).
We therefore considered somewhat weaker J125 # H160 color criterion for completeness, just
to include a list of all possible bright z!9-10 galaxies in the current WFC3/IR observations.
Searching through our catalogs, we found one source with a J125 # H160 color of 0.9 – ERSj14904385 – which was consistent with being a z ! 9 galaxy. This candidate is listed in
Table 1 and Figure 2. For consistency with our HUDF09 J125 -dropout search and to be
conservative, we do not include this candidate in our LF determinations.
To better ascertain the nature of ERSj-14904385, we examined the available IRAC
–4–
Fig. 1.— Y105 ! J125 / J125 ! H160 two color selection used to identify z ~ 8.2 Y105 -dropout
galaxies. For context, the colors of star-forming galaxies with different U V -continuum slopes
" of !3, !2 versus redshift are also shown. Also presented are the colors of low-redshift
galaxy SEDs (Coleman et al. 1980) over the redshift range z ! 0 ! 2. The Y105 -dropout
selection window is presented in gray. The current Y105 -dropout selection is chosen to identify
galaxies at somewhat higher redshift than that identified in the Bouwens et al. (2009c) Y105 dropout selection (dotted black lines). The hatched green region indicates the region in color
space we would expect L,T dwarfs to lie (Knapp et al. 2004).
–5–
V
i
z
Y
J
H
UDFy!37796000
H=28.2 Y!J=2.1 J!H=!0.2
z~8.4
UDFy!38135539
H=28.0 Y!J>2.2 J!H=0.2
z~8.7
ERSj!2149044385 H=27.4 J!H= 0.9
z~9.3
Fig. 2.— V606 i775 z850 Y105 J125 H160 image cutouts of three z!8.5-9 galaxy candidates we identified in the ultra-deep HUDF WFC3/IR observations (§4-§5). The top two candidates are
Y105 -dropouts from the ultra-deep WFC3/IR observations over the HUDF (§4) while the
third candidate is a z ! 9.3 J125 -dropout candidate found over the WFC3/IR Early Release
Science observations (§5). The J125 -dropout candidates from our primary selection are given
in Figure 1 of the main text. None of the candidates is detected in the deep ACS BV iz observations. Other properties of the candidates are given in Table 1. Each cutout is 2.4!! "2.4!!
on a side.
–6–
observations. ERSj-14904385 is detected at 3.6! at 3", with an estimated magnitude of
26.5±0.4, but is undetected in the 4.5! observations. The J125 # H160 and H160 # 3.6! colors of
the candidate yield an estimated redshift of 9.3 – though a redshift of z = 2.2 also fit
the observed fluxes well. Supposing ERSj-14904385 is at z ! 9, the red H160 # 3.6! color is
presumably due to its Balmer break. If so, the candidate must not be at z > 9.5 since then
the 3.6! band would probe the rest-frame U V , and hence the H160 # 3.6! color be very blue
contrary to the observations.
6.
Possible candidates in the HUDF09 observations, with redshifts between
our J125 -dropout and Y105 -dropout selections
Of course, it seems quite likely if the HUDF09 observations contain sources with redshifts
as high as z ~ 10 and J # H colors as red as 1.2, it also contains sources at somewhat lower
redshift with J # H colors somewhat bluer than 1.2. However, our Y105 -dropout criterion
only selects sources with J # H colors as red as 0.5. Are there any sources in our HUDF09
observations with J # H colors redder than 0.5 mag, but bluer than 1.2 mag, which are
undetected in the BV izY bands? We identified one such possible candidate at 03:32:36.05
and #27:47:43.5, with an H -band magnitude of 28.9±0.2 and J # H color of 1.1 ± 0.8. While we
emphasize that the nature of this candidate is quite uncertain, given the faintness of the
source and the presence of only a modest break in the spectrum, the existence of such sources
is key. It suggests that there is a plausible population of sources at redshifts between our
z ~ 10 J125 -dropout candidates and our z~8.2-8.8 Y105 -dropout candidates.
7.
Possible Contamination
Contamination from Spurious Sources: Perhaps the most important issue in assessing
the nature of the present J125 -dropout selections is determining whether the sources are
likely real or whether some are due to random fluctuations (i.e., spurious). Since each of our
candidates is only detected in a single passband at ~5-6", one must consider the possibility
that they could be spurious.
To estimate the probable number of spurious candidates that might enter our HUDF09
J125 -dropout selection, we created H160 -band images with as realistic of noise properties as
possible. Here is a brief outline of the process. Each of the individual H160 -band exposures
in the WFC3/IR HUDF09 data set was first processed to remove signal from real sources.
This was performed by masking out all the pixels in the individual frames that correspond
–7–
Table 1. Possible z~10 J -dropout candidates and z~8.5-9 Y105 -dropout candidates
identified over the ultra-deep HUDF09 WFC3/IR observations over the HUDF
Object ID
R.A.
Dec
H160
Y105 ! J125 a
J125 ! H160
rhl b
zest c
UDFj-43696407
UDFj-35427336
UDFj-38116243
UDFy-37636015
UDFy-38135539
ERSj-14904385
03:32:43.69
03:32:35.42
03:32:38.11
03:32:37.63
03:32:38.13
03:32:14.90
-27:46:40.7
-27:47:33.6
-27:46:24.3
-27:46:01.5
-27:45:53.9
-27:44:38.5
28.9±0.2
29.1±0.2
28.9±0.2
28.2±0.2
28.0±0.1
27.5±0.2
—
>1.5
0.13""
10.2
10.2
10.2
8.4
8.7
9.3
a
—
—
2.1±0.9
>2.2
—
>1.4
>1.6
!0.2±0.3
0.2±0.2
0.9±0.4
""
0.12
0.11""
0.14""
0.15""
0.11""
Lower limits on the measured colors are the 1# limits.
b
The quoted half-light radii are as observed and are not corrected for the PSF. The half-light
radius rhl for the PSF is 0.09"".
c
Estimated redshift
Fig. 3.— Stacks of the J125 , H160 , 3.6$, and 4.5$ images (each 3.9"" %3.9"" ) for the three
J125 -dropout candidates considered here. The flux from nearby neighbors is modeled and
subtracted, before stacking the IRAC 3.6$ and 4.5$ images of each J125 -dropout candidate.
The 1# lower limit on the H160 ! 3.6$ color and H160 ! 4.5$ color for the stack is <0.7 mag
and <1.3 mag, respectively, suggesting that our J125 -dropout candidates are not likely to be
red galaxies at z ~ 2 (see §3 for details).
–8–
to sources in our HUDF09 H160 -band reductions. Empty “masked” pixels on each image
were filled in, by randomly selecting sections from other exposures. Random (1!! ) offsets
were applied to the coordinates of each noise exposure and then the entire stack of exposures
was drizzled together. Catalogues were created from the drizzled image and candidate J125 dropout sources were selected in the same way as on the actual H160 -band data. 0.1 sources
per simulation satisfy our selection criteria by chance (though this number increases to 1, 3,
and 43 spurious sources per field when the detection threshold is lowered to 4.3", 4", and 3"
significance, respectively). This is a more sophisticated version of the standard “negativeimage” test (where both the cataloguing and selection procedure is performed on the negative
images) used to assess the relevance of spurious sources to a selection (e.g., Bouwens et al.
2006).
Additional support for the reality of two of our J125 -dropout candidates – UDFj-38116243
and UDFj-35427336 – is provided by the deep NICMOS H160 -band observations over the
HUDF (Thompson et al. 2005; Bouwens et al. 2008; Oesch et al. 2009a). Both candidates
are formally detected at 1.5" significance in the NICMOS data, with H160 -band magnitudes
consistent with those measured in the WFC3/IR data. A stack of these two candidates
shows an apparent 3" detection and is shown in Figure 1 of the main paper. The H160 -band
magnitude from the stack is 28.5+0.4
!0.3 AB mag. This magnitude is 0.5 mag brighter than
measured in our WFC3/IR observations, but nonetheless consistent.
Contamination from Red Low-Redshift Galaxies: An equally important concern for the
present J125 -dropout selection is low-redshift, very red sources (typically because of age or
dust). It is plausible that such sources could satisfy our J125 -dropout selection criteria, with
J # H colors redder than 1.2 and showing very little flux in the ultra-deep optical data.
Obviously, we might expect such sources to be somewhat brighter at redder wavelengths,
and so the IRAC data provide us with a way of addressing this possibility. As we noted in
the previous section, each of our candidates is undetected in the 3.6$ band, allowing us to a
set a H160 # 3.6$ < 1.3 limit on its color blueward of the J # H break – providing some
evidence that J125 -dropouts in our selection are not just red galaxies at z ~ 2.
An independent way of examining whether red galaxies are a concern for our J125 dropout selection is performing an analogous selection for z ~ 8 Y105 -dropouts – assuming
that no H160 -band observations are available. We therefore select Y105 -dropouts using two
criteria: being undetected in the optical data and satisfying a single color Y105 # J125 > 0.8
cut (and no J # H criterion). In performing this experiment, we only remove sources that
are bright enough in the optical that they would be detected if the source had the same H160 band magnitude as our J125 -dropout candidates (which are ~1 mag fainter). Hence, we only
remove sources that show 4" detections in the optical. We only identify one source in this
–9–
Y105 -dropout selection not present in the Bouwens et al. (2009c) z ~ 8 Y105 -dropout selection,
03:32:43.41, !27:46:36.1. This source did not make the Bouwens et al. (2009c) selection,
because it shows a 2" detection in the z850 -band in our catalogs. While obviously there is
some uncertainty as to its precise nature (e.g., it is found in the Yan et al. 2009 z ~ 8 selection
as z8-B094), we will conservatively assume that it represents a small level of contamination
for a single-band Y105 -dropout selection. By analogy, this suggests a contamination rate of
~1 source per field for our J125 -dropout selection.
Of course, if such a source were in the present J125 -dropout selection, it would be detected at 1" in both the Y105 and J125 bands, and none of the present J125 -dropout candidates
are so detected, suggesting that the estimated contamination level of ~1 source per field is
likely higher than in reality. So, we are being conservative. In any case, this test demonstrates how unique the colors of z ~ 7 ! 10 LBGs really are, relative to other faint sources in
deep field data. It also suggests that other techniques that do not take advantage of information about the actual prevalence of sources with specific colors at faint magnitudes may
actually overestimate the contamination rate for high redshift selections (e.g., techniques
that compute P (z) based upon #2 fits to various SED templates versus redshift).
Contamination from Photometric Scatter: Another possibly significant source of contamination for the current J125 -dropout selection are sources that satisfy our J125 -dropout
criterion due to noise in the photometry. To assess this possibility, we looped over all of the
sources in our HUDF09 H160 -band catalog, randomly selecting a source in the magnitude
range 26.5 to 27.5 AB mag (taken to be representative of the color distribution at fainter
magnitudes), rescaling its fluxes to match the fainter source in our H160 -band catalog, adding
noise, and finally reapplying our J125 -dropout selection criteria. Repeating this simulation
many times, we found that ~0.2 contaminants per field satisfy our J125 -dropout criterion,
simply as a result of noise.
Other Sources of Contamination: Other possible sources of contamination for our J125 dropout selection include transient sources like SNe or stars. Contamination by SNe seem
extremely unlikely since the J125 and H160 observations were taken within 5 days of each
other. Contamination by stars also seem rather implausible, since the only stars which are
red enough to satisfy the J ! H > 1.2 criterion are extreme carbon stars or Mira variables
(e.g., Whitelock et al. 1995) – which given their luminosity would need to be located well
outside our galaxy (see discussion in Dickinson et al. 2000).
Lastly, all of our candidates are sufficiently separated from extended foreground sources
that they are unlikely subcomponents of these sources. A somewhat parallel concern in
doing candidate selection near extended sources is that the detection significance of possible
sources is enhanced by light from the extended wings. As we show in §11 (Figure 7), it
– 10 –
seems quite probable that an independent selection of z~8-9 candidates (Yan et al. 2009) is
subject to significant contamination as a result of this issue.
Summary: The above simulations suggest we would find ~1.3 contaminants in the
z~10 J125 -dropout selection we perform over the HUDF09 observations: ~0.1 from spurious
sources, ~0.2 based upon photometric scatter, and ~1 from red sources. This implies that
1.7 J125 -dropout candidates from our selection correspond to z ~ 10 galaxies. Of course, it
is important to remember that such simulations can often be uncertain at the factor of ~2
level, and so we will also consider the case that the contamination rate is at the 100% level.
In the sections which follow, we will explore the implications (1) if 1.7 J125 -dropout
candidates here are at z~10 and (2) if none are.
Contamination in our J125 -dropout selection over the ERS observations: We also repeated the above simulations for the J125 -dropout selection over the ERS observations. ~0.6
contaminants per arcmin2 field per simulation are predicted (most from photometric scattering experiments). This suggests that our only J125 -dropout candidate over the ERS observations – ERSj-14904385 – is at z ~ 2 . The case for the candidate being at z~9 seems
even less likely in light of the results from the next section – where only 0.1 J125 -dropout
candidates are expected to be found over the ERS observations for plausible z~9-10 LFs.
Contamination in Y105 -dropout selections: The only important contaminant for Y105 dropout selections (see Bouwens et al. 2009c) seems to be sources that enter the selection as
a result of noise, and even for this source of contamination, we derive a contamination rate
of < 4% from Monte-Carlo simulations.
8.
Expected Numbers
To provide a baseline for interpreting the number of high-redshift dropout candidates
we have found over our search fields, we compare the present findings with what we might
expect extrapolating lower-redshift results to z ~ 10 assuming no evolution.
We use a similar procedure as described in previous work (e.g., Bouwens & Illingworth
2006; Bouwens et al. 2009b) to account for the many different selection effects (and incompleteness). We start by generating catalogues based upon the model LFs, creating realistic
pixel-by-pixel simulations of the model sources, adding these simulated images to the real
data, and then processing the images (and selecting the sources) in the same way as on
the real data. The spatial profiles of the sources in the simulations are modeled using the
pixel-by-pixel morphologies of similar-luminosity z ~ 4 B-dropout galaxies from the HUDF
– 11 –
(e.g., from Bouwens et al. 2007) scaled in size to match the (1 + z)!1 size-redshift scaling
(Oesch et al. 2009c; Bouwens et al. 2004; Ferguson et al. 2004). The sources are also taken
to have U V -continuum slope distribution with a mean " equal to !3 with a scatter of 0.4
to match the observed trends with redshift (Bouwens et al. 2009d).
We predict 20 and 10 J125 -dropouts over the HUDF and ERS fields using the z ~ 6 LF
(Bouwens et al. 2007: but see also McLure et al. 2009a), 9 and 3 dropouts over the HUDF
and ERS fields using the z ~ 7 LF (Oesch et al. 2009b: see also Bouwens et al. 2008), and
6 and 1 dropouts over the HUDF and ERS fields using the z ~ 8 LF (Bouwens et al. 2009c:
R.J. Bouwens et al. 2009, in prep). We can also provide a prediction at z ~10 based upon
the LF-fitting formula in Bouwens et al. (2008) where M*UV,AB = !18.7, #* = 0.0012 Mpc!3 ,
and $ = !1.74. Based upon this LF, we predict 2 J125 -dropouts over the HUDF at z ~ 10
and 0.1 J125 -dropouts at z ~ 9 over the ERS.
The predicted J125 -dropout surface densities are presented in Figure 4 for the various U
V LFs. The observed numbers (surface densities) are zero at magnitudes where significant
numbers would be expected based upon the U V LF at z ~ 6 and z ~ 7 if there is no
evolution. We only detect galaxies at the faint end of our two searches (i.e., H160,AB ~ 29 for
the HUDF09 observations or H160,AB ~ 27.5 for the ERS observations). In the HUDF, the
surface density of J125 -dropout candidates observed is comparable to that predicted based
upon lower redshift LFs, though the uncertainties are very large. If our estimate of 1.7
J125 -dropout candidates is confirmed with subsequent observations, it would imply that the
volume density of z ~ 10 galaxies is high (and the UV LF at z ~ 10 very steep). However,
given that the contamination rate also increases rapidly near the search limits, there is a
reasonable probability that more than 40% of our J125 -dropout candidates are contaminants.
We expect the field-to-field variance on the present J125 -dropout searches to be 39% and
40% in our HUDF09 and ERS searches – assuming a redshift selection window with width
%z ~ 1.3; a search area of 4.7 arcmin2 and 39.2 arcmin2 , respectively; and bias factors of ~8
and ~12 (corresponding to source volume densities of ~10!3 Mpc!3 and ~3&10!5 Mpc!3 ,
respectively).
The predicted redshift distributions for our J125 -dropout selection is given in Figure 2 of
the main text. For this calculation, we assume that the U V LF at z ~ 10 has a M * U V = !18.7,
$ = !1.74, and #* = 0.0012 Mpc!3 , i.e., using the Bouwens et al. (2008) LF fitting formula
to extrapolate the z ~ 4 ! 7 LF results to z ~ 10. The mean redshift for our z ~ 10 J125 dropout selection is 10.2. Also presented in Figure 2 of the main text is the predicted redshift
distribution of the present Y105 -dropout selection (assuming M *U V = !19.4, $ = !1.7, and
#* = 0.0012 Mpc!3 ) for each candidate found. The J ! H color for UDFy-38135539 is 0.3
mag redder than UDFy-37636015 – suggesting probable redshifts of 8.7 and 8.4, respectively
– 12 –
Fig. 4.— Observed surface densities of z~9-10 J125 -dropout candidates (histograms shown
with thick black lines ) in the wide-area ERS observations (top panel ) and ultra-deep HUDF09
observations (bottom panel ). The observed surface densities have been corrected assuming
~40% and 60% contamination in the HUDF09 and ERS observations, respectively (see §7).
For comparison, we have also plotted the surface densities projecting the UV LF at z ~ 6
(Bouwens et al. 2007: red histogram ) and z ~ 7 (Bouwens et al. 2008: magenta histogram )
to z ~ 8 assuming no evolution. Also shown are the surface densities expected (dotted
histograms) based upon the predicted LFs at z ~ 8 and z ~ 10 (these predicted LFs are
based upon the extrapolation of the Bouwens et al. 2008 results to z> ~ 8). Note that the
H160,AB ~ 29 magnitude bin suffers from substantial incompleteness (~50%).
– 13 –
for the sources. While the latter redshift estimate is somewhat higher (!z~0.2-0.3) than the
McLure et al. (2009b) estimate for these two Y105 -dropouts, the present estimate adopts a
somewhat bluer U V -continuum slope " distribution (as suggested by Bouwens et al. 2009d)
than may be considered by the SED templates McLure et al. (2009b) use for their redshift
estimates.
9.
Implications for the U V LF at z ~ 10
The depth and area of the present WFC3/IR observations have provided us with our
best opportunity yet to find star-forming galaxies at z ~ 10, and therefore it is not surprising
that we were able to find 3 promising z~10 J125 -dropout candidates.
The present search results permit us to significantly improve the constraints we have on
the LF at z~10. We begin by exploring stepwise constraints on the UV LF. The simulations
described in §8 are used to estimate the effective search volume as a function of absolute
magnitude. Both the wide-area ERS and ultra-deep HUDF searches are included in these
effective volume estimates. These volumes implicitly include the effect of incompleteness.
The LF is then equal to the number of z ~ 10 candidates found divided by these effective
volumes. Based upon our contamination estimates (§7), we assume that 1.7 J125 -dropout
candidates in our HUDF09 selection are probable z ~ 10 galaxies, but also consider the
possibility that none correspond to z ~ 10 galaxies. Our stepwise LFs are presented in
Figure 3 of the main text.
It is also interesting to examine the present LF results in terms of a Schechter parametrization, given the utility such parametrizations have had in modelling the LF over a wide range
in redshifts. Again we consider the case that 1.7 J125 -dropout candidates from our HUDF09
selection correspond to z ~ 10 galaxies and no candidates do. To quantify the constraints
we can set on the Schechter parameters at z ~ 10, we calculate the number of J125 -dropouts
we would expect to find to these magnitude limits, using the selection efficiencies we can
infer from the simulation results in §8. For simplicity, we consider the constraints on the
characteristic luminosity M * at z ~ 10, if the normalization #* and faint-end slope $ is
equal to the same values ~0.0012 Mpc%3 and ~ %1.74 as has been found at lower redshift
z ~ 4 % 6 (Bouwens et al. 2007, 2008). In detail, we find that M * U V = %18.7 ± 0.4 AB
mag at z ~ 10, assuming that 1.7 J125 -dropout candidates from the present sample are at
z ~ 10. In the case that none are, M * U V > %18.5 AB mag at z ~ 10, which is ~2 mag fainter
than the MU* V,AB ~ %21 AB mag derived at z ~ 4 (e.g., Yoshida et al. 2006; Bouwens et
al. 2007) and the MU* V,AB ~ %19.4 ± 0.4 estimated at z ~ 8 (Bouwens et al. 2009c; R.J.
Bouwens et al. 2009, in prep) – illustrating how dramatically the U V LF is evolving at high
– 14 –
Fig. 5.— Constraints on the characteristic luminosity M *U V,AB at z ~ 10 from the present
J125 -dropout search – assuming the ! = 0.0012 Mpc"3 and # = "1.74 parameters found
consistent with the observations at z~3-8 (Bouwens et al. 2008; R.J. Bouwens et al. 2009, in
prep). The z ~ 10 constraint is shown as a solid circle based upon an estimate of 1.7 probable
J125 -dropout candidates and as an upper limit (assuming none of our candidates correspond
to z ~ 10 galaxies). Also shown are the characteristic luminosities estimated for the UV LF
at z ~ 8 (Bouwens et al. 2007, 2008, 2009c; Oesch et al. 2009b; Arnouts et al. 2005). The
solid black line shows the evolution predicted from the halo mass function (Sheth & Tormen
1999) assuming a mass-to-light ratio which is constant while the dashed black line shows this
evolution assuming a mass-to-light ratio which varies as (1 + z)"1 . These lines are identical
to those in Figure 10 of Bouwens et al. (2008). While it is unclear whether the U V LF at
z ~ 10 is well represented by a Schechter function and whether !* and # are consistent with
the assumed values, the characteristic luminosity M is a convenient one-parameter way of
characterizing the evolution of the UV LF over the range z ~ 8 to z ~ 3 at least. From
this figure, it is clear that observed evolution in the U V LF reported from z ~ 8 to z ~ 4
continues out to z ~ 10.
– 15 –
redshift. While we have no evidence that the UV LF maintains a Schechter parametrization
out to z ~ 10 with a relatively constant !* and ", parametrizing the evolution in terms of
the characteristic luminosity M * does provide us with a convenient way of quantifying the
overall rate of evolution in the U V LF.
10.
Constraints on the U V Luminosity Density/SFR density at z ~ 10
The present J125 -dropout searches permit us to set constraints on the luminosity density
at z ~ 10 brightward of 0.07 L z=3 . To derive these constraints, we integrate up the LF results
from §9 to the survey limits. Again, we consider the case that 1.7 J125 -dropout candidates
from our selection correspond to z ~ 10 galaxies and the case that none do. The derived
U V luminosity densities are then converted into the equivalent SFR densities, adopting the
Madau et al. (1998) conversion factor. The Madau et al. (1998) conversion factor assumes
a constant rate of star formation for >~100 Myr. The results are presented in Table 2 and
Figure 4 of the main text. Since essentally all galaxies in the luminosity range of our samples
(i.e., ~ #19.5 AB mag) show very blue U V -continuum slopes $ ($ <~#2.5) at z > ~ 5 (e.g.,
Bouwens et al. 2009c, 2009d), no dust obscuration is assumed in the SFR density estimates.
11.
Previous J110 /J125 -dropout selections
Already there have been a variety of different searches for z ~ 10 galaxies in the literature. Searches for z ~ 10 J110 -dropout candidates in deep NICMOS data were considered
by Bouwens et al. (2005), Bouwens et al. (2008), Richard et al. (2008), Bouwens et al.
(2009a), and Henry et al. (2009). None of the candidates in those selections has proven to
be particularly compelling, however, after the acquisition of follow-up observations.
More recently though, it has been possible to search for z~9-10 J125 -dropout candidates
in ultra-deep WFC3/IR observations.
Both Yan et al. (2009) and Bunker et al. (2009)
consider such a search. Bunker et al. (2009) report no robust J125 -dropout candidates to
H160,AB ~ 28.5 while Yan et al. (2009) report 20 candidates to H160,AB ~29, 3 in the
magnitude range H160,AB ~28-28.5 and 17 in the range H160,AB ~28.5-29.
It seems worthwhile to focus on the Yan et al. (2009) z~9-10 J125 -dropout results briefly,
given that their results are quite different from our own. The Yan et al. (2009) z~9-10 J125 dropout selection adopt a fairly aggressive 3% threshold in the H160 -band for identifying
sources. For such an aggressive selection, contamination is clearly a significant concern, not
only from spurious sources, but also from sources scattering into the selection as a result
– 16 –
Fig. 6.— H160 -band images of 9 z ~ 9 J125 -dropout candidates of Yan et al. (2009) located
very close to bright extended foreground galaxies. The Yan et al. (2009) J125 -dropout candidates are almost exclusively found very close to bright extended foreground galaxies (see
Figure 7). The probability that this would occur at random is 5 ! 10"7 . While it is unclear
why Yan et al. (2009) identify such a large number of sources nearby bright galaxies, almost
all explanations (e.g., substructure in the outer regions of foreground galaxies, substructure
in the PSF, or an enhancement in spurious source surface density due to flux from the wings
of the bright foreground galaxies) suggest that the contamination rate in the Yan et al.
(2009) selection must be large. Lensing does not provide a plausible explanation for this
association with foreground galaxies – since one would then predict ~5 even more strongly
lensed galaxies at ~28 mag (see §11).
– 17 –
Fig. 7.— Distances of z~9-10 J125 -dropout candidates in the Yan et al. (2009) sample to the
closest bright foreground galaxy (solid blue histogram) in the HUDF09 observations. Also
shown is the distribution for our z ~ 10 candidates (open red histogram). The solid black
line shows the distribution for blank regions on the HUDF09 image. Clearly, we would not
expect z ~ 9 galaxies to be correlated with bright foreground sources, and so their proximity to
bright galaxies should be the same as blank regions on the image. This expectation is not
borne out for the z~9-10 J125 -dropout selection. Almost all of their candidates (blue
histogram) are found around bright foreground sources. The probability this would occur at
random is almost zero (i.e., 5 ! 10"7 : determined from a KS test) and suggests that most
of the Yan et al. (2009) z ~ 9 J125 -dropout candidates are associated with the extended
foreground sources (Figure 6 shows 9 examples) and hence the contamination rate must be
substantial (see §11).
– 18 –
Fig. 8.— Distances of the Oesch et al. (2009b) z ~ 7 z-dropout sample and the Bouwens et al.
(2009c) z ~ 8 Y -dropout sample from bright foreground galaxies (open red histogram) in the
HUDF09 observations. Also shown is the distribution for the Yan & Windhorst candidates
(solid red histogram). Otherwise as in Figure 7. The z ~ 7 Oesch et al. (2009b) and z ~ 8
Bouwens et al. (2009b) show a similar distribution in distances from extended foreground
sources, as do blank regions on the image (black line ). The situation is different for the
9 additional Y105 -dropout galaxies that Yan et al. (2009) identify – not in the selections
of Bouwens et al. (2009c), McLure et al. (2009b), or Bunker et al. (2009). Most of those
candidates are close to bright foreground galaxies (as was the situation for the Yan et al.
2009 z ~ 9 sample: see Figure 7). The probability that this would occur at random is 0.8%
(see §11).
– 19 –
of noise. Using the simulations from §7, we expect ~17 spurious sources per HUDF09 field
using a 3! threshold and H160,AB < 29.0 magnitude limit. We also predict that 3 sources will
scatter into the Yan et al. (2009) J125 -dropout selection due to noise (photometric scatter).
Both calculations suggest that the Yan et al. (2009) z~9-10 selection suffers from substantial
contamination.
However, there is a much more direct and model-independent test we can perform to
assess the reliability of the z ~ 9 J125 -dropout selection given in Yan et al. (2009). It is with
regard to the apparent association of many of their candidates with bright foreground sources
and hence proximity to these sources (see Figure 6). If their z ~ 9 selection identified real
z~9-10 galaxies, we would expect the distribution in distances relative to bright foreground
sources to be the same as for blank regions on the image. To ensure that our identification of
extended foreground sources was objective, these sources were taken to be those objects which
after smoothing the HUDF WFC3/IR H160 -band image with a gaussian kernel (!=0.09"" )
had a surface brightness greater than 27.7 AB mag arcsec#2 over an area with radius $0.54"" .
In Figure 7, the distances of Yan et al. (2009) J125 -dropouts relative to the closest foreground
source are presented. Also presented is that found for blank regions on the HUDF09 image.
Clearly, the Yan et al. (2009) z ~ 9 J125 -dropout distribution is skewed towards very
small distances, in significant contrast to that found for blank regions in the HUDF09 image.
Use of the Kolmogorov-Smirnov test shows that the probability for drawing such a distribution at random is ~ 5 % 10#7 . By contrast, the other dropout samples selected by our team
and other teams (e.g., Oesch et al. 2009c) seem to be quite consistent with that found for
blank regions on the image (Figure 8). One possible exception is the 9 additional z ~ 8
Y105 -dropout candidates identified by Yan et al. (2009) – not present in the Bouwens et al.
(2009c), McLure et al. (2009b), or Bunker et al. (2009) selections. A good fraction of these
9 candidates are also quite close to foreground sources (blue histogram on Figure 8). The
probability of obtaining this distance distribution for these sources at random is just 0.8%.
Perhaps the only explanation for this effect would be due to lensing from foreground
galaxies, but it is easy to see that this explanation is not very satisfactory. The typical magnification factors (from the foreground galaxies) at the radii (~1.5-4"" ) where these sources
are found are just ~2 – meaning that the intrinsic magnitudes of most Yan et al. (2009) z ~ 9
candidates are ~0.6 mag fainter than observed, or ~29.4 AB mag. Of course, one would
expect the z ~ 10 J -dropout population – if they possessed the huge volume density implied
at ~29.4 – to not simply be found in surveys with magnification factors of ~2, but also to
be found in modest numbers at brighter magnitudes with larger magnification factors, i.e.,
~5-8 sources at ~27.5-28.0 if one assumes that the foreground galaxies are well described
by an isothermal sphere mass profile (e.g., Gavazzi et al. 2007). However, such sources are
– 20 –
Table 2. U V Luminosity Densities and Star Formation Rate Densities.a
Dropout Sample
<z>
log10 L
(ergs s!1
Hz!1 Mpc!3 )
z (Oesch et al. 2009b)
Y (Bouwens et al. 2009c)
J (This work: 1.7 candidates)
J (This work: no candidates)
6.8
8.2
10.2
10.2
25.73±0.16
25.18±0.24
24.80±0.64
<24.57c
a
log10 SFR density
(M Mpc!3 yr!1 )
Uncorrected
Correctedb
!2.17 ± 0.16
!2.72 ± 0.24
!3.10±0.64
< !3.33c
!2.17 ± 0.16
!2.72 ± 0.24
!3.10±0.64
< !3.33c
Integrated down to ~!18.3 AB mag.
b
Based upon the blue U V -continuum slopes " ~ !2.5 (Bouwens et al. 2009c,d; Oesch et
al. 2009), no dust correction is assumed.
c
Upper limits here are 1# (68% confidence).
– 21 –
not found in even modest numbers at such bright magnitudes. This suggests that lensing
by foreground galaxies does not provide a plausible explanation for the distribution of z ~ 9
candidates around foreground sources.
It seems quite clear from this discussion that the Yan et al. (2009) selection suffers
from significant contamination concerns – particularly in the vicinity of bright sources. The
situation appears severe enough as to cast substantial doubt on the entire Yan et al. (2009)
z~9-10 selection – raising questions about whether any sources in their sample are credible.
Such concerns seem particularly relevant given their use of a very aggressive >~3! threshold
for source detection.
REFERENCES
Arnouts, S., et al. 2005, ApJ, 619, L43
Beckwith, S. V. W., et al. 2006, AJ, 132, 1729
Bertin, E. and Arnouts, S. 1996, A&AS, 117, 39
Bouwens, R. J., Illingworth, G. D., Blakeslee, J. P., Broadhurst, T. J., & Franx, M. 2004,
ApJ, 611, L1
Bouwens, R. J., Illingworth, G.D., Thompson, R.I., & Franx, M. 2005, ApJ, 624, L5
Bouwens, R.J., Illingworth, G.D., Blakeslee, J.P., & Franx, M. 2006, ApJ, 653, 53
Bouwens, R. J., Illingworth, G. D., Franx, M., & Ford, H. 2007, ApJ, 670, 928
Bouwens, R. J., Illingworth, G. D., Franx, M., & Ford, H. 2008, ApJ, 686, 230
Bouwens, R. J., et al. 2009a, ApJ, 690, 1764
Bouwens, R. J., et al. 2009b, ApJ, 705, 936
Bouwens, R.J., et al. 2009c, ApJ, in press, arXiv:0909.1803
Bouwens, R.J., et al. 2009d, ApJ, 708, L69
Bunker, A., et al. 2009, MNRAS, in press, arXiv:0909.2255
Dickinson, M., et al. 2000, ApJ, 531, 624
Ferguson, H. C. et al. 2004, ApJ, 600, L107
– 22 –
Finkelstein, S. L., Papovich, C., Giavalisco, M., Reddy, N. A., Ferguson, H. C., Koekemoer,
A. M., & Dickinson, M. 2009, ApJ, submitted, arXiv:0912.1338
Gavazzi, R., Treu, T., Rhodes, J. D., Koopmans, L. V. E., Bolton, A. S., Burles, S., Massey,
R. J., & Moustakas, L. A. 2007, ApJ, 667, 176
Giavalisco, M., et al. 2004b, ApJ, 600, L103
Henry, A. L., et al. 2009, ApJ, 697, 1128
Knapp, G. R., et al. 2004, AJ, 127, 3553
Kron, R. G. 1980, ApJS, 43, 305
Labbé, I., Bouwens, R., Illingworth, G. D., & Franx, M. 2006, ApJ, 649, 67
Madau, P., Pozzetti, L. & Dickinson, M. 1998, ApJ, 498, 106
McLure, R. J., Cirasuolo, M., Dunlop, J. S., Foucaud, S., & Almaini, O. 2009a, MNRAS,
395, 2196
McLure, R., et al. 2009b, MNRAS, in press, arXiv:0909.2437
Oesch, P. A., et al. 2009a, ApJ, 690, 1350
Oesch, P.A., et al. 2009b, ApJ, in press, arXiv:0909.1806
Oesch, P.A., et al. 2009c, ApJ, in press, arXiv:0909.5183
Reddy, N. A., & Steidel, C. C. 2009, ApJ, 692, 778
Richard, J., Stark, D. P., Ellis, R. S., George, M. R., Egami, E., Kneib, J.-P., & Smith, G. P.
2008, ApJ, 685, 705
Riess, A. G., et al. 2007, ApJ, 659, 98
Salvaterra, R., et al. 2009, Nature, 461, 1258
Schiminovich, D., et al. 2005, ApJ, 619, L47
Sheth, R. K. & Tormen, G. 1999, MNRAS, 308, 119
Szalay, A. S., Connolly, A. J., & Szokoly, G. P. 1999, AJ, 117, 68
Tanvir, N. R., et al. 2009, Nature, 461, 1254
– 23 –
Thompson, R. I., et al. 2005, AJ, 130, 1
Trenti, M., & Stiavelli, M. 2008, ApJ, 676, 767
Whitelock, P., Menzies, J., Feast, M., Catchpole, R., Marang, F., & Carter, B. 1995, MNRAS, 276, 219
Yan, H., Windhorst, R., Hathi, N., Cohen, S., Ryan, R., O’Connell, R., & McCarthy, P.
2009, ApJ, submitted, arXiv:0910.0077
Yoshida, M., et al. 2006, ApJ, 653, 988
This preprint was prepared with the AAS LATEX macros v5.2.