EXPANDED SEARCH FOR z ∼ 10 GALAXIES FROM HUDF09, ERS, AND CANDELS DATA: EVIDENCE FOR ACCELERATED EVOLUTION AT z > 8?^*

The HUDF09 program (PI: Illingworth; Bouwens et al. 2011b) consists of 192 HST orbits to provide ultra-deep WFC3/IR imaging over three pointings centered on the HUDF (Beckwith et al. 2006) and its two parallel fields from the UDF05 program (PI: Stiavelli; Oesch et al. 2007). The program has been completed, providing the deepest IR images ever taken. It comprises 3 × 4.7 arcmin² imaging in the three filters Y₁₀₅, J₁₂₅, and H₁₆₀, reaching down to H_{160, AB} = 29.4, 28.6, and 28.9 (5σ in 05 diameter apertures) for the HUDF, HUDF09-1, and HUDF09-2, respectively (see Figure 1). For a more detailed description of this data set and the data reduction see Bouwens et al. (2011b).

In addition to the ultra-deep HUDF09 data, we also analyzed shallower, wider area WFC3/IR imaging from the ERS and CANDELS programs, in order to constrain the volume density of more luminous star-forming galaxies at z ∼ 10.

The ERS data provide WFC3/IR imaging of ∼41 arcmin² of the northern part of the GOODS South field. Two orbits of WFC3/IR imaging were obtained in each of the filters Y₀₉₈, J₁₂₅, and H₁₆₀, over a 2 × 5 grid of pointings (60 orbits in total). These data are reduced in an analogous way to our HUDF09 data and are aligned and drizzled to the GOODS ACS mosaics after rebinning to a 006 pixel scale. These data reach to H₁₆₀ = 27.6 (see also Bouwens et al. 2011b). For a more detailed description of this data set see Windhorst et al. (2011).

The last two fields included in our analysis are obtained as part of the multi-cycle treasury program CANDELS (PI: Faber/Ferguson; Grogin et al. 2011; Koekemoer et al. 2011). In particular, we include the first six visits of the CANDELS-Deep program (obtained until 2011 August 6), which covers the central part of GOODS South in 3 × 5 tiles with ∼6000 s exposures in both J₁₂₅ and H₁₆₀ in a total of 92 orbits. These data cover ∼63 arcmin² and reach to H_{160, AB} ∼ 27.5 mag. Additionally, we also included the imaging data of the supernova follow-up program of CANDELS (PI: Riess), which adds imaging over two pointings over CANDELS-Deep (one of which is essentially centered on the HUDF). Finally, we made use of the 29 orbits of WFC3/IR data of the CANDELS-Wide survey (Y₁₀₅, J₁₂₅, and H₁₆₀ obtained until 2011 March 29). These comprise nine WFC3/IR pointings (∼42 arcmin²), completing the coverage of the GOODS South field, and reach to H_{160, AB} = 26.9 (∼2000 s exposures). As for the ERS, the WFC3/IR data of the CANDELS program have been aligned to the GOODS ACS mosaics with a pixel scale of 006. The part of the CANDELS field overlapping with the WFC3/IR HUDF has been omitted when analyzing this data set in order not to duplicate the analysis of that area.

We will subsequently refer to the combination of the ERS and the two CANDELS fields as "WIDE Fields."

Source catalogs are derived with the SExtractor program (Bertin & Arnouts 1996), which is used to detect galaxies in the H₁₆₀ images and perform matched aperture photometry on point-spread function (PSF) matched images. The colors used here are based on isophotal apertures derived from the H₁₆₀ images, and total magnitudes are measured in standard 2.5 Kron apertures, corrected by 0.2 mag in order to account for flux loss in the PSF wings.

The detection significance of sources was established in 025 radius apertures. The rms maps were scaled based on the detected variance in 1000 random apertures for each WFC3/IR frame on empty sky regions after 3σ clipping. This procedure ensures that the SExtractor weight maps correctly reproduce the actual noise in the images. Subsequently, only sources with signal-to-noise ratios (S/Ns) larger than 5 (in 025 radius apertures) in H₁₆₀ are considered.

Galaxies at z > 9.5 are expected to exhibit very red J₁₂₅ − H₁₆₀ colors since the redshifted, strong Lyα absorption (by the predominantly neutral intergalactic hydrogen) cuts into the flux in the J₁₂₅ filter (see Figure 2). This makes such high-redshift galaxies completely invisible blueward of J₁₂₅, and we use this fact for their identification.

As can be seen from Figure 2, however, also passively evolving or dusty galaxies at intermediate redshifts (z ∼ 2.5–4) can exhibit similarly red colors in J₁₂₅ − H₁₆₀. While the requirement of optical non-detections removes the bulk of lower redshift contamination, certain intermediate-redshift galaxies with evolved or dusty stellar populations can still be included due to the fact that the optical data do not reach deep enough, if at similar depth as the IR (see Figure 3). Deep Spitzer IRAC data provide a way to identify contaminating galaxies. These are expected to exhibit very red H₁₆₀ − [3.6 μm] colors, which discriminate them from genuine z ∼ 10 candidates.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To exclude possible low-redshift contamination, we thus use two steps to select z ≳ 9.5 galaxy candidates. For the first step, the primary criteria are based on HST data only:

$$(J_{125}-H_{160})>1.2$$

$$\rm{S/N}(H_{160}) > 5 \quad \wedge \quad \rm{S/N}({<}J_{125})<2 \quad \wedge \quad \chi ^2_{{\rm opt}}<\chi ^2_{{\rm cut}}.$$

Additional to excluding objects that are detected in any band blueward of J₁₂₅ at more than 2σ, we include a cut in the optical χ²_opt value of a galaxy (see, e.g., Bouwens et al. 2011a, 2011b). This is computed from the 025 radius aperture fluxes as χ²_opt = ∑_isign(f_i)(f_i/σ_i)², where the sum runs over all the bands available in the given data set blueward of J₁₂₅, i.e., it includes all the available optical data as well as the NIR band Y₁₀₅ for the HUDF09 data and Y₀₉₈ in the ERS. The relatively large apertures were chosen in order to sample >70% of the light of point-like sources.

The limiting χ²_cut are derived from photometric scatter simulations. They are set to exclude the majority of interlopers which remain undetected at 2σ purely due to photometric noise, but not to cut a substantial fraction of galaxies with real zero flux in the optical bands. The scatter simulations utilize all galaxies in our catalogs that are 1–3 mag above the completeness limit, applying photometric Gaussian noise from 1 mag fainter sources. From these simulations it is clear that contamination is mainly an issue at 0.75 mag above the completeness limits, but that ∼60%–80% of contaminants can be eliminated by using a χ²_opt limit of χ²_cut = 2.8 or 2.4, for five filters or four filters, respectively. In the HUDF09 data, the resulting number of expected contaminants due to photometric scatter is thus reduced from ∼0.5 source per WFC3/IR field to ∼0.1 source.

On the other hand, the adopted χ²_cut limits do remove an additional ∼20% of sources with real zero flux, simply due to Gaussian statistics. This reduction of the real galaxy sample is reflected in our subsequent analysis in the reduction of the selection volume.

All galaxies passing the above selection criteria, using both the ACS and WFC3/IR data, are retained and analyzed individually. These total 17 sources with H_{160, AB} in the range 23.6–28.8 mag; one source in the HUDF, none in the parallel HUDF09 fields, three in the ERS, and eight and five in the CANDELS-Deep and -Wide, respectively (see Tables 2 and 4).

Interestingly, only one source (the previously reported galaxy with H₁₆₀ ∼ 29 mag from Bouwens et al. 2011a) did pass our selection in the three deep HUDF09 fields, while the shallower CANDELS and ERS fields contribute a total of 16 sources (all with H_{160, AB} ≲ 26 mag). Upon inspection of their images, it turns out that all these brighter sources are very well detected in the IRAC data, even in the shallow 8.0 μm band. Their measured H₁₆₀ − [3.6] colors are in the range 1.6–4.3, which, for a z ∼ 10 source, would correspond to a UV continuum slope β ≳ −0.2 or a dust reddening with A_V > 1.6 mag. Given that galaxies at the bright end of the z ∼ 7 population are measured to have very low extinction values and continuum slopes of β ≃ −2.0 ± 0.2 (see, e.g., Bouwens et al. 2010b; Finkelstein et al. 2010; Dunlop et al. 2011; Wilkins et al. 2011b), the extremely red colors of these galaxies rule out z ≳ 9 solutions with any sensible spectral energy distribution (SED).

All sources with IRAC detections are thus removed from our sample of potential z ∼ 10 galaxies, reducing the sample to one single candidate in the HUDF, previously reported in Bouwens et al. (2011a, see Figure 4 and Table 2). The 16 removed sources are shown the Appendix in Figure 10 and listed in Table 4.

We do note the existence of one potential candidate in the HUDF, which is only marginally below our detection limit (seen at 4.8σ). The source is located at R.A. = 03:32:43.01, decl. =−27:46:53.3 (see also Yan et al. 2010), and could be checked in analyses by others, or as further WFC3/IR data become available over this field. Note, however, that the i₇₇₅-band data show very faint positive flux (∼1.5σ) exactly at the location of this source, which increases the chance that it is at lower redshift. We will not discuss this candidate further. However, this demonstrates the value of even deeper optical imaging over these fields for more robust high-redshift galaxy selections.

We show images of the z ∼ 10 galaxy candidate with S/N > 5 in Figure 4. The source is detected at 6.3σ in H₁₆₀ (measured in circular apertures of 025 radius). This is higher but completely consistent with the Bouwens et al. (2011a) significance estimates, which are based on smaller apertures. As can be seen, the source is not significantly detected in any other band. Its value of χ²_opt = 2.77 is very close, but just below the limit of χ²_cut = 2.8. This is mainly due to a 1.5σ flux excess in i₇₇₅, which appears to be due to an extended structure in the background of that image. When adopting smaller apertures, the χ²_opt value is found to be reduced, indicating also that this excess of flux is not associated with the source itself.

Zoom In Zoom Out Reset image size

The Spitzer IRAC 3.6 μm data show some flux from a nearby source. However, after subtraction of all the neighboring sources in IRAC, the candidate is undetected (0.09σ), with a 2σ upper limit on its IRAC [3.6] magnitude of >27.2 mag AB (V. Gonzalez et al. 2012, in preparation). This thus corresponds to H₁₆₀ − [3.6] < 1.6 at 2σ, which is much bluer (by >0.4 mag) than the typical low-redshift contaminants that we culled from our sample (see also the next section). After adding the newly acquired IRAC data from the IUDF10 program (Spitzer proposal 70145, PI: Labbe) to the GOODS IRAC data and removing neighboring sources, the z ∼ 10 candidate is also undetected at [4.5] (S/N = 0.3) providing added weight to the likelihood of it being at high rather than low redshift.

We derive the photometric redshift of the candidate using the code ZEBRA (Feldmann et al. 2006; Oesch et al. 2010b) with synthetic stellar population models from Bruzual & Charlot (2003) to which we added nebular continuum and line emission following, e.g., Schaerer & de Barros (2009). Using the full 11 band fluxes and flux errors, we derive a photometric redshift for this source of z_phot = 10.4^+0.5_−0.4, with a likelihood of a low-redshift solution at z < 8 of <6%. The full SED of the source and its redshift likelihood function are shown in Figure 5. The best-fit SED corresponds to a very young, dust-free starburst (see also V. Gonzalez et al. 2012, in preparation).

Zoom In Zoom Out Reset image size

The best low-redshift solution is found at z_lowz = 2.7, for an evolved, very low mass galaxy SED (M = 3 × 10⁸ M_☉) with moderate extinction. Interestingly, this SED is expected to be detected only at ∼1.5σ in J₁₂₅, but it nevertheless has a significantly higher χ² value (13.7 compared to χ²_best = 7.0). Deeper HST data shortward of the break would be extremely useful in order to constrain the possible non-detection of the source shortward of 1.4 μm. As with other high-redshift catalogs, the added shorter-wavelength optical/near-IR data would play a key role in helping to further tighten its photometric redshift measurement.

The properties of the 16 intermediate brightness sources that did not pass the IRAC non-detection criteria are discussed in the Appendix. From fitting their SEDs, these sources are found to be mostly massive galaxies (M > 5 × 10¹⁰ M_☉) with obscured but evolved stellar populations at z ∼ 2–4.

Interestingly, all these galaxies are essentially limited to H₁₆₀ ∼ 24–26 mag. This is ∼1 mag brighter than the detection limits of our bright survey fields (see Figure 11 in the Appendix) and thus suggests that such very red galaxies have a somewhat peaked LF. This will have to be confirmed and quantified with future wide-area data. However, at face value, it would appear that contamination from such sources with J₁₂₅ − H₁₆₀ > 1.2 and with H₁₆₀ − [3.6] ≳ 2 is less problematic at fainter magnitudes. We note, however, that contamination from similar galaxies with less extreme colors (which may be more abundant also at H₁₆₀ > 26 mag) may still be non-negligible due to photometric scatter (see also Section 3.5).

Furthermore, it is interesting to note that it is very difficult to construct clean z ∼ 10 galaxy samples purely based on HST data alone. Even if we were to increase the color criteria to J₁₂₅ − H₁₆₀ > 2.0, there would still be two contaminating galaxies in the sample together with our only viable z ∼ 10 candidate. Thus, also for future constraints on the bright end of the z ∼ 10 LF, it will be important to perform additional follow-up studies to validate the candidates, e.g., with Spitzer. This will be less of a concern for future z ∼ 9 galaxy samples, which can be obtained, e.g., based on new F140W filter data. In such a data set intrinsically red galaxies can be sorted out by requiring a blue continuum across F140W and H₁₆₀.

Here, we only give a brief summary of the possible sample contamination. For a thorough discussion, we refer the reader to Bouwens et al. (2011a).

Essentially, the only probable chance for contamination is due to photometric scatter of a red lower redshift source. We estimate this to be a 10% chance based on our photometric scatter experiments described in Section 3.2, including the χ²_opt cuts.

Other typical contaminants to Lyman break galaxy (LBG) selections such as very cool dwarf stars and supernovae can essentially be excluded based on the relatively blue J₁₂₅ − H₁₆₀ (<1.1) colors of stellar SEDs, on the fact that the source is detected in both the first and the second year of the HUDF09 WFC3/IR data, and due to the fact that the candidate shows clear signs of an extended morphology.

Additionally, it is very unlikely that this source is spurious, since the flux distribution in circular apertures randomly distributed over empty regions of the HUDF H₁₆₀ image are nearly exactly Gaussian, and the source is well detected at 6.3σ.

In order to estimate the number of sources we expect in our data from a given LF, we have to estimate the completeness as a function of magnitude C(m) and selection function as a function of redshift and magnitude S(z,m). Following Oesch et al. (2007, 2009), this is done by inserting artificial galaxies in the observational data and rerunning the source detection with the exact same setup as for the original catalogs. This is done for each of our fields individually.

Two sets of simulations were run. In the first set, we follow Bouwens et al. (2003), where the artificial galaxies are "cloned" from the z ∼ 4 dropout sample of the GOODS and HUDF fields. The images of these sources are adjusted for surface brightness dimming, the difference in angular diameter distance, as well as a size scaling of (1 + z)⁻¹ as observed for the LBG population across z ∼ 3–7 (see, e.g., Ferguson et al. 2004; Bouwens et al. 2004; Oesch et al. 2010a). These are then inserted in the observed images with galaxy colors as expected for star-forming galaxies between z = 8 and z = 12. When computing the galaxy colors we assume a distribution of UV continuum slopes with β = −2.5 ± 0.4 (see, e.g., Bouwens et al. 2009; Bouwens et al. 2010b; Finkelstein et al. 2010; Stanway et al. 2005).

From these simulations, we compute the completeness as a function of observed H_{160, AB} magnitude for each field, taking into account the scatter and offsets between input and output magnitudes. Additionally, we compute the selection probabilities as a function of redshift and magnitude by measuring the fraction of sources that meet our selection criteria. By construction, galaxies are selected at >50% at redshifts z ≳ 9.5.

In the second set of simulations, we repeat the above procedure. However, instead of using observed galaxy images, we use theoretical galaxy profiles from a mix of exponential (Sersic n = 1) and de Vaucouleur (Sersic n = 4) profiles. The size distribution is chosen to be lognormal, again with the same size scaling as a function of redshift. The completeness and selection functions of our two procedures are in excellent agreement. This demonstrates the reliability of our approach, which appears to be essentially independent of the adopted galaxy profiles (unlike what has been claimed elsewhere, e.g., Grazian et al. 2010, but see also Bouwens et al. 2010b).

The expected number of sources in a given magnitude bin m_i can be estimated for any given LF, ϕ(M), through

$$N^\mathrm{exp}_i = \int _{\Delta m} dm \int dz \frac{dV}{dz} S(m,z)C(m)\phi (M[m,z]).$$

The result of this calculation is shown in Figure 6 for each individual field using the observed LFs at z ∼ 8 (Bouwens et al. 2011b) as well as that expected at z ∼ 10. The latter is based on extrapolating the evolution of the Schechter function parameters presented in Bouwens et al. (2011b), which is based on fitting the likelihood contours of LBG LFs across z ∼ 4–8. In particular, significant evolution is only seen in the characteristic luminosity M_*, which is found to dim by 0.33 mag per unit redshift. The other two Schechter function parameters were kept constant, in agreement with the small evolutionary trends that were found at low significance in Bouwens et al. (2011b). For reference, the final parameter evolution we use in this work is

$$\begin{eqnarray*} &&M_*(z) = -21.02 \, \mathrm{mag} + (z-3.8) \times 0.33 \, \mathrm{mag}\\ &&\phi _* = 1.14\times 10^{-3} \, \mathrm{Mpc}^{-3}\,\mathrm{mag}^{-1} =\mathrm{const}\\ &&\alpha = -1.73 =\mathrm{const}. \end{eqnarray*}$$

Zoom In Zoom Out Reset image size

As can be seen in Figure 6, the wide-area, shallow data of CANDELS and ERS are very useful for constraining the evolution in the bright end of the LF. In particular, if there was no evolution in the LF from z ∼ 10 to z ∼ 8, the wide-area data should contain 8.5 z ∼ 10 sources. In total, we would expect to see ∼27 sources if the LF was unchanged over the 170 Myr from z ∼ 8 to z ∼ 10. This implies that the wide-area data provide essentially 30% of the total search power in the case of an LF evolution in ϕ_* only.

Given that we only detect one probable candidate, the LF appears to drop faster than expected from the empirical lower redshift extrapolation. In particular, we do not detect the ∼6 galaxies that we would have expected to find at z ∼ 10 if the lower redshift trends remained valid. From these extrapolations, we predicted to find three sources in the HUDF (consistent with the expectations from Bouwens et al. 2011a) and about one in each of the two HUDF09 parallel fields.

The Poissonian probability to find ⩽1 source, given that six are expected is <2%. This remains significant, even after including cosmic variance, which adds an additional uncertainty on these low number counts of about 45%–50% for an individual WFC3/IR pointing (see, e.g., Trenti & Stiavelli 2008; Robertson 2010a). We derive an upper limit of 6% to the probability of finding ⩽1 source in our search area, based on the cosmic variance calculator of Trenti & Stiavelli (2008) and combining the number counts uncertainty in the different fields assuming the final distribution is Gaussian (justified by the central limit theorem). Therefore, the detection of an accelerated evolution relative to the low-redshift extrapolation is significant at ⩾94%.

The accelerated evolution in the UV LF can also be seen from our constraints on the z ∼ 10 LF. The stepwise LF is computed using an approximation of the effective selection volume as a function of observed magnitude $V_{\rm eff}(m) = \int _0^\infty dz \frac{dV}{dz} S(z,m)C(m)$. The LF in bins of absolute magnitude is then given by ϕ(M_i)dM = N^obs_i/V_eff(m_i). This is shown in Figure 7, where the LF was evaluated in bins of 0.5 mag, and non-detections correspond to 1σ upper limits including the effects of 50% cosmic variance per pointing.

Zoom In Zoom Out Reset image size

As can be seen in Figure 7, the current constraints on the z ∼ 10 LF are significant even at bright magnitudes where the wide-area data are particularly valuable. These data sets reduce the upper limits by more than an order of magnitude relative to using only the HUDF09 fields, therefore indicating that the z ∼ 10 LF at M₁₄₀₀ < −20 drops by a factor of ∼4–5 with respect to the observed LF at z ∼ 8. However, from these shallow data sets, no constraints can be obtained on any accelerated evolution of the UV LF around z ∼ 8–10. This only becomes apparent at M₁₄₀₀ > −20 (corresponding to H_{160, AB} ∼ 27.5 mag). Such faint limits are only probed by the HUDF09 data set. In particular, at M₁₄₀₀ = −19, the upper limit on the LF is a factor of ∼3 below the expectation (Figure 7). Thus, it is clear that data reaching to deeper than H_{160, AB} = 28.5 mag will be necessary to further constrain the drop in the LF in the future (which is beyond the reach of the current MCT programs).

In order to quantify the change in the LF from z ∼ 10 to z ∼ 8 more robustly, we consider two possible scenarios. First, we assume that the accelerated LF evolution occurs only in M_*, at a constant rate since z ∼ 6, and we fit the Poissonian likelihood for the observed number of sources. This can be written as $\cal {L} = \prod _{j} \prod _i P(N^{\rm obs}_{j,i},N^{\rm exp}_{j,i})$, where j runs over all fields, and i runs over the different magnitude bins, and P is the Poissonian probability.

Our extrapolation of the UV LF is a modification of the fitting formulae of Bouwens et al. (2011b). We thus use ϕ_*(z) = 1.14 × 10⁻³ Mpc⁻³ = const. and α(z) = −1.73 = const., and assume M_*(z) = −20.29 + ζ(z − 6). We then fit for ζ, finding ζ = 0.58^+0.14_−0.11, which results in an estimate for M_*(z = 10) = −18.0 ± 0.5 mag.

Alternatively, we assume M_*(z) = −19.63 = const. (as derived for z ∼ 8 from our empirical extrapolation), α = −1.73, and we fit only for an evolution in the normalization with redshift relative to the z ∼ 8 LF. This results in ϕ_* = 1.14 × 10⁻³ 10^{−ϒ(z − 8)} Mpc⁻³, with best-fit ϒ = 0.54^+0.36_−0.19. Thus, using this extrapolation, the normalization of the UV LF from z ∼ 10 to z ∼ 8 is expected to increase by a factor of 12. These results are summarized in Table 3.

The quantity most easily comparable between observations and simulations is the observed LD (ρ_L) above a given limiting magnitude, which is shown in Figure 8. The LD from the z ∼ 10 candidate alone amounts to log₁₀ρ_L = 24.1^+0.5_−0.7 erg s⁻¹ Hz⁻¹ Mpc⁻³. However, a more realistic estimate of the LD can be obtained from the two possible extrapolations of the UV LF we derived in the previous section, which include the contribution from galaxies at M₁₄₀₀ < −19 mag that are currently undetected. Assuming that the LF evolution only occurs in the characteristic luminosity, we find for the z ∼ 10 LD log₁₀ρ_L = 24.2 ± 0.5 erg s⁻¹ Hz⁻¹ Mpc⁻³, while assuming the evolution to be driven by a normalization of the Schechter function only, we find log₁₀ρ_L = 24.4 ± 0.5 erg s⁻¹ Hz⁻¹ Mpc⁻³. These different estimates are also summarized in Table 3.

Zoom In Zoom Out Reset image size

Given that the observed LD at z ∼ 8 is log₁₀ρ_L(z = 8) = 25.6 erg s⁻¹ Hz⁻¹ Mpc⁻³ (Bouwens et al. 2011b), the inferred increase in LD in the 170 Myr from z ∼ 10 to z ∼ 8 amounts to more than an order of magnitude. This is a factor ⩾4 higher than what would have been inferred from the empirical relation for the UV LF evolution, which predicts an increase by only a factor of ∼3.

Note, however, that such a rapid increase in LD is actually predicted by many theoretical models. In Figure 8, we also show the luminosity densities at M₁₄₀₀ < −18 mag as derived from the semi-analytical model (SAM) of Lacey et al. (2011), and from the theoretical model of Trenti et al. (2010).

Although there is still some discrepancy on the exact shape of the z > 6 UV LF between the Lacey et al. (2011) SAM and the observations, the integrated LD and its evolution are remarkably well reproduced across the full redshift range z ∼ 4–8 (see also discussion in Raičević et al. 2011). The SAM predicts a constant growth in ρ_L with cosmic time at somewhat faster pace than observed, leading to some discrepancy between the observations and the model at z ∼ 7 and z ∼ 8, where the observational data show higher luminosity densities. However, the model reproduces our estimates of the z ∼ 10 LD remarkably well.

Similar conclusions are reached for the purely theoretical model of Trenti et al. (2010). In particular, this model predicts the UV LD to evolve at an accelerated rate at z > 8, being only log₁₀ρ_L = 24.4 erg s⁻¹ Hz⁻¹ Mpc⁻³ at z ∼ 10, in excellent agreement with our observed estimates. Since the model is only based on the evolution of the underlying dark matter mass function, this indicates that an accelerated evolution in the galaxy population can be explained even without the need for a change in the physical mechanisms of galaxy formation.

For further theoretical model predictions, see also, e.g., Mao et al. (2007), Salvaterra et al. (2011), or Muñoz (2011).

It is also interesting to note that the star formation rate (SFR) densities (see Table 3) inferred from our data are more than an order of magnitude too low to account for the stellar mass densities observed at z ∼ 7 in systems of similar brightness. With constant star formation over ∼300 Myr from z ∼ 10 to z ∼ 7 the observed galaxy population would only produce a stellar mass density of log₁₀ρ_M = 4.7–5.3 M_☉ Mpc⁻³, compared to log₁₀ρ_M(z = 7) = 6.6 M_☉ Mpc⁻³ as estimated by, e.g., Labbé et al. (2010a, 2010b); Gonzalez et al. (2010). If the inferred mass densities and SFRs are correct, this suggests that the majority of the stars found in z ∼ 7 galaxies down to M_UV < −18 mag have to be formed in systems below our detection limit at z ∼ 10 or are younger than 300 Myr.

Here, we briefly summarize the properties of the galaxies that formally met the J₁₂₅-dropout criteria based on the HST data alone, but which were found to exhibit very red H₁₆₀ to IRAC colors. This makes it extremely unlikely that these galaxies are at z ≳ 9. A summary of these sources is listed in Table 4.

Note that all these galaxies are obviously clearly detected in the IRAC images, with the exception of one source (jD-2080646581), which is close to a quintet of IRAC bright galaxies. However, after subtraction of the contaminating flux from its neighbors based on a careful convolution of the H₁₆₀ image to the IRAC PSFs (see, e.g., Labbé et al. 2010a, 2010b; Gonzalez et al. 2010), it also shows a clear detection (see Figure 10). The best estimate for its color is H₁₆₀ − [3.6] = 1.6 ± 0.5, which is too red for a likely z > 9 source.

Zoom In Zoom Out Reset image size

Interestingly, six of these 16 sources are present in catalogs of potential passive high-redshift galaxies of Rodighiero et al. (2007) and Wiklind et al. (2008). One of these is also detected in X-ray emission and classified as a so-called EXO source hosting a potentially highly obscured active galactic nucleus, particularly if it is an evolved galaxy at intermediate redshift, as pointed out by Koekemoer et al. (2004).

We have derived photometric redshifts and mass estimates for these galaxies by complementing our HST photometry with the IRAC fluxes from the GOODS MUSIC catalog (Santini et al. 2009) or from the SIMPLE images. Based on SED fits with Bruzual & Charlot (2003) models, these galaxies are confirmed to be evolved, reddened systems with stellar masses around 10¹¹ M_☉ at redshifts z ∼ 2–3.5. Note that with the exception of four sources these are all detected individually in MIPS 24 μm data.

In Figure 11, we additionally show the surface density of these evolved intermediate-redshift sources. Even at their peak surface density, they are only found at 0.06 sources magnitude^-1 arcmin^-2. Therefore, only ∼0.4 sources are expected per WFC3/IR pointing, providing an explanation for why none of these bright sources are found in the three deep HUDF09 fields. The dearth of such galaxies faintward of H_{160, AB} = 26, at ∼1 mag brighter than the completeness limit of our bright surveys, suggests that such extremely red sources become even less frequent (i.e., that their LF may have peaked at these magnitudes) and so they could be less problematic for contamination of fainter z ≳ 9 dropout searches. This will have to be confirmed, however, with future wide-area WFC3/IR data, searching explicitly for these types of galaxies.

Zoom In Zoom Out Reset image size

Note that our finding of a disappearance of passive intermediate-redshift galaxies to fainter magnitudes is in agreement with Stutz et al. (2008), who found an absence of passive, red galaxies at z ∼ 1.5–3 at masses below ∼10¹⁰ M_☉.