EXPANDING THE SEARCH FOR GALAXIES AT z ∼ 7–10 WITH NEW NICMOS PARALLEL FIELDS^*

The data used here consist of NICMOS parallel observations taken during GO programs 10872 in GOODS and 11236 in COSMOS (PI: H. Teplitz), and 11188 in SSA22 (PI: B. Siana). For the GOODS fields, 15 parallel fields were observed in J₁₁₀ and H₁₆₀, and nine lie within the GOODS footprint where the Advanced Camera for Surveys (ACS) data are available. The positions of these fields within GOODS are shown in Figure 1 and coordinates are listed in Table 1. In total, this corresponds to 5.9 square arcminutes of new NICMOS imaging in GOODS. We note that three fields (CDFS-1,-2, and -3) are also included in the Bouwens et al. (2008) search, where they are found not to contain any z > 7 candidates. However, in light of the large discrepancy seen in the same NICMOS data by Richard et al. (2008) and Bouwens et al. (2009), we include these fields in our search as a consistency check. Typical exposures for these NICMOS parallels in GOODS were 8 ks in J₁₁₀ and 5 ks in H₁₆₀.

Zoom In Zoom Out Reset image size

The COSMOS parallels consist of 12 fields observed in J₁₁₀ and H₁₆₀, ⁷ 11 of which lie within the Subaru/SuprimeCam images in B, r', i', and z'. Seven of these 11 fields are also within the ACS I₈₁₄ footprint. A twelfth parallel field lies in the northeast corner of COSMOS, where the limited SuprimeCam coverage is not sensitive enough to discriminate between z > 7 galaxies and interlopers with typical galaxy colors. Therefore, we exclude this field from our survey. For the remaining 11 COSMOS fields, although the optical imaging is not as deep as in GOODS, it is adequate to remove interlopers, because, as we will show in Section 3.1, no z > 7 candidates are found in the COSMOS fields. In total, these 11 NICMOS parallel fields cover 7.2 arcmin². Their locations are shown in Figure 2 and coordinates are listed in Table 1. Typical exposures were 6–8 ks, divided between J₁₁₀ and H₁₆₀.

Zoom In Zoom Out Reset image size

Lastly, we include two parallel fields in SSA22, which comprise some of the deepest available NICMOS imaging. However, at these faint limits, optical data in SSA22 that are deep enough to be useful are limited. Ground-based optical images are not sensitive enough to detect the faintest sources in the NICMOS images, even if they have typical galaxy spectral energy distributions (SEDs). The only available observation that can adequately rule out interlopers is an ACS I₈₁₄ image (GO 10405; PI: S. Chapman), which covers only SSA22-2. Because all NICMOS sources are detected in this I₈₁₄ image, we know that no candidates are found in this parallel field (see Section 3) without considering z-band data, so we can include it in our survey volume. Although SSA22-1 cannot be used in the z ∼ 7 search, we are able to use both fields for the J₁₁₀-dropout LBG search, because there are no sources that are red enough in J₁₁₀–H₁₆₀ to meet the z ∼ 9 selection criterion in either SSA22 parallel field.

With these data, we select z > 7 candidates as z- and J₁₁₀-dropouts, using the deep optical images to reject interlopers. This will be discussed in detail in Section 3.

The NICMOS images were reduced and combined with a combination of custom IDL and Python scripts and available IRAF⁸ procedures. First, images were pedestal corrected, and the South Atlantic Anomaly (SAA) darks were subtracted for impacted orbits. Following the SAA correction, the pedestal correction was repeated to improve the subtraction. Next, the sky frames were made and subtracted using McLeod's NICRED (1997) code, and a static bad pixel mask that included the vignetted rows was created from these sky frames. To remove any remaining gradients in the images, we made sky images with each column replaced by its median. This image was smoothed by a three-pixel wide boxcar and subtracted from each NICMOS frame. Then, this process was repeated for each row of pixels to remove top-to-bottom gradients. Next, intermittent bad pixels were identified in each image using the IRAF package crutil. These masks were combined with the static bad pixel mask, and finally frames were drizzled (Fruchter & Hook 1997), using the parameters recommended in the dither handbook (pixfrac = 0.6 and scale = 0.5). Shifts were derived so that the final J₁₁₀ and H₁₆₀ images are drizzled onto the same frame and are therefore aligned. The resulting pixels are 01, and the zero points that we use are adjusted by −0.16 and −0.04 mag in J₁₁₀ and H₁₆₀, to correct for the nonlinearity reported by de Jong (2006).

Sensitivities were measured by randomly placing 06 diameter apertures in the images, rejecting apertures which contained light from objects.⁹ This procedure is repeated for each NICMOS image, as exposure times varied. The 5σ limits are 26.0–27.5 in J₁₁₀ and 25.9–27.0 in H₁₆₀, with the faintest limits reached in the small area in SSA22 (see Table 1). The point-spread function (PSF) for these NICMOS images was derived by stacking several isolated, unsaturated stars. The resulting PSF has an FWHM ∼03 in both bands. The point source aperture correction for a 06 diameter aperture is 0.31 mag.

GOODS. We use the publicly available v2.0 ACS GOODS images in B₄₃₅, V₆₀₆, i₇₇₅, and z₈₅₀ bands.¹⁰ Included in v2.0 are additional data used to search for Type Ia supernovae, which double the v1.0 exposure time in the z₈₅₀ band, and also increase the sensitivity in i₇₇₅. This significantly enhances the sensitivity to galaxies at z ≳ 6–7 and improves identification of faint interlopers.

As with the NICMOS images, a PSF is determined by stacking several point sources found in the ACS images. We find an FWHM of ∼ 01 in z₈₅₀. Typical 3σ limits are 28.7, 28.8, and 28.3, in B₄₃₅, V₆₀₆, and i₇₇₅, measured in 04 diameter apertures. As we will describe in Section 2.4, z₈₅₀ magnitudes are measured from 06 diameter apertures in images matched to the NICMOS resolution. For these, the 3σ sensitivity is ∼27–28 mag. Some parallel fields near the edge of the GOODS footprint have reduced sensitivity. We carefully measure the sensitivity in each of the fields, as our objective is to determine whether each source is detected in B₄₃₅, V₆₀₆, or i₇₇₅.

COSMOS. The COSMOS data that we use are less homogenous than the GOODS data, consisting of both Subaru/SuprimeCam images at B, r', i', and z', and where available, ACS I₈₁₄ images. The seeing is 08 in B, r', and i', and 12 in the z' images. Typical 3σ limits are 28.4, 27.8, 27.3, and 26.7 at B, r', i', and z' in 12 diameter apertures. The ACS I₈₁₄ images typically reach 27.7 in a 04 diameter aperture. Again, sensitivity varies within the COSMOS area, because some of the NICMOS parallels are in the less well-covered edges. As with the GOODS parallel fields, we measure the noise in each field so that we can accurately determine whether sources are detected in the B, r', i', or I₈₁₄ images.

SSA22. As described above, the only optical imaging that we use for the SSA22 parallels is an ACS I₈₁₄ image that covers SSA22-2. We use the "drz" image, directly from the archive, which has a 3σ sensitivity of 28.3 in a 04 diameter aperture.

To select z > 7 galaxies, we compare the above described z'or z₈₅₀ data to NICMOS images. As these data have widely differing resolution, different techniques are required to measure accurate z'–J₁₁₀ or z₈₅₀–J₁₁₀ colors. We describe these approaches below.

GOODS. To measure accurate colors of all the galaxies in the nine NICMOS fields in GOODS, we downgraded the resolution of the ACS z₈₅₀ images by matching the NICMOS PSF. To achieve this, we use the IRAF task PSFMATCH, which convolves the ACS images with a kernel made from the NICMOS and ACS PSFs. The convolved ACS images are then rebinned and aligned with the NICMOS images. Then, we use SExtractor (Bertin & Arnouts 1996) in dual-image mode, with an inverse-variance weighted average J₁₁₀ + H₁₆₀ image as the detection image. For detection, we require five contiguous pixels 1.3σ above the background. In addition, we use the gauss_3.0_5x5.conv filter, which is optimized for finding faint sources. Lastly, spurious detections, artifacts, and electronic ghosts are manually removed from the catalog. The z₈₅₀–J₁₁₀ and J₁₁₀–H₁₆₀ colors are measured in 06 diameter apertures, and the two-color plot is shown in Figure 3 (left).

Zoom In Zoom Out Reset image size

In order to test for nondetections at bands shorter than z₈₅₀, we measure the flux in 04 apertures in the original, unconvolved B₄₃₅, V₆₀₆, and i₇₇₅ images, at the positions predicted by our NICMOS detections.

COSMOS. The COSMOS data require a different approach, because downgrading the resolution of the NICMOS images to 12 seeing causes a significant loss of sensitivity. Instead, we resample the z' images to 01 per pixel (the same as NICMOS) and align them to match NICMOS. We then used SExtractor in the same manner as with GOODS, except we use 06 diameter apertures in J₁₁₀ and H₁₆₀, and 12 in z'. The aperture corrections for point sources in these apertures are 0.31 mag for NICMOS and 0.74 mag for COSMOS. Because galaxies at z ≳ 7 should be compact in NICMOS images (Bouwens et al. 2004a; Ferguson et al. 2004; Dow-Hygelund et al. 2007), this treatment is appropriate for the sources that we are interested in. For extended sources, we expect blueward scatter in z'–J₁₁₀ (away from the z > 7 selection), as more light will be missed from the higher resolution data. This trend is confirmed for simulated galaxies, using the IRAF artdata package. The two-color plot for the COSMOS fields is shown in Figure 3 (right).

As with the GOODS data, we measure the flux at the positions predicted by the NICMOS detections, using 08 apertures in B, r' and i', and 04 apertures in I₈₁₄.

SSA22 As we are not using any z-band data for SSA22, there is no need to properly account for z − J₁₁₀ colors measured with mismatched apertures and resolutions. Therefore, we simply follow the same procedures described above for the GOODS and COSMOS parallels—measuring J₁₁₀ and H₁₆₀ with SExtractor, and testing for I₈₁₄ detections in 04 diameters apertures in SSA22-2 where the ACS data are available.

Candidates for z > 7 galaxies are selected using the following criteria: first, we require that galaxies are detected in the J₁₁₀ + H₁₆₀ detection image at >5σ significance. In total, we find 696 sources that meet this criterion in GOODS, 701 in COSMOS, and 211 in SSA22. Next, z > 7 candidates must be undetected at the 2σ level in bands bluer than z' or z₈₅₀. This eliminates the vast majority of sources, with only two candidates remaining in the GOODS fields, two in COSMOS, and none in SSA22-2. We list these sources in Table 2, and we will proceed to show that all are interlopers.

We next use the colors of these sources to determine if any have SEDs consistent with z ∼ 7 galaxies. We adopt the color cut¹¹ of Bouwens et al. (2008, 2009), so that candidates must have z–J₁₁₀ > 0.8 and z–J₁₁₀ >0.8 + 0.4 (J₁₁₀–H₁₆₀), and J₁₁₀–H₁₆₀ <1.2 (where z refers to both z₈₅₀ and z'). All four of the "dropout" sources mentioned above lie outside this selection. One source (C5-zD1), has z'–J₁₁₀ ∼ 0.2, and the others (CDFS-3-JD1, CDFS-4-JD1, and C8-JD1) have J₁₁₀–H₁₆₀ >1.2. While Oesch et al. (2009) have suggested a stricter cut of z₈₅₀–J₁₁₀ > 1.3, adopting this cut would make no difference in our search, because we have not found any candidates with the most generous selection.

The red J₁₁₀–Hnic colors of these three sources could be an indication of the Lyman break in the J₁₁₀ band and redshifts z > 8 (Bouwens et al. 2005; Henry et al. 2007, 2008). However, J₁₁₀-dropouts must also be undetected at the 2σ level in the z' or z₈₅₀ bands. This restriction eliminates CDFS-3-JD1 and CDFS-4-JD1. The remaining source, C8-JD1, cannot be ruled out on the basis of optical/NIR data alone, but longer wavelength data from IRAC on the Spitzer Space Telescope show strong detections at 3.6 and 4.5 μm (H₁₆₀ − [3.6] = 2.4, H₁₆₀ − [4.5] = 2.7). For z ∼ 9, these colors correspond to a rest-frame UV slope which is much redder than LBGs, so this galaxy is more likely an interloper at z ∼ 1–3 with a dusty starburst or an old stellar population. In conclusion, none of the four optical "dropout" sources that we find can be described as a plausible z > 7 galaxy.

We have rejected as interlopers any sources which have 2σ detections in B₄₃₅, B, V₆₀₆, r', i', or I₈₁₄. While this approach is commonly taken in LBG surveys, it does not consider the possibility that a weak detection in any or all of these "veto" bands could arise from foreground contamination. In fact, as we will demonstrate, the probability of contamination is significant.

To estimate the influence of foreground contamination, we use the UDF ACS catalogs. The surface density of sources brighter than our typical 2σ detection limits in GOODS (B₄₃₅, V₆₀₆, and i₇₇₅ ∼ 29.1, 29.2, and 28.7) is ∼400 arcmin⁻². This corresponds to about a 10% probability of a foreground contaminant lying within 05 of an NICMOS-detected source. For the COSMOS fields, the B, r', and i' limits are shallower, so the surface density of possible contaminants is lower (∼250 arcmin⁻²). However, the seeing-limited resolution requires larger apertures. In this case, we find that the probability of a foreground source lying within 1'' of a z > 7 candidate is about 20%. We therefore estimate that 10%–20% of true z > 7 galaxies would be rejected by our survey because of faint foreground contaminants.

While we have not detected any candidate z > 7 galaxies, we can place limits on the LFs of z-dropouts at z ∼ 7 and J₁₁₀-dropouts at z ∼ 9. Furthermore, we can compare this limit to predictions from LFs at z ∼ 3–6 and place constraints on evolution from z ∼ 6 to 7.

First, we calculate the survey volume following Steidel et al. (1999):

$$\begin{equation} V_{\rm eff}(M) = \int _{z} {p(M, z) { { dV} \over {dz}} dz}. \end{equation} \tag{ 1 }$$

The quantity p(M,z) is the probability of both detecting a source of a given absolute magnitude and redshift, and selecting it as a z- or J₁₁₀-dropout based on the criteria that we established in Section 3. This probability can be expressed as p(M, z) = S(M, z) × C(m), with S(M, z) representing the selection function, and C(m) the photometric completeness. We use simulations to determine these quantities for both z ∼ 7 and z ∼ 9. To obtain S(M, z), we require only one assumption, namely, a distribution of galaxy spectra. We use a Gaussian distribution of UV power-law slopes estimated from z ∼ 6 galaxies (f_λ ∝ λ^β; β = −2.2 ± 0.2; Stanway et al. 2005). Then, for every M, z, and β, we predict the z'-, z₈₅₀-, J₁₁₀-, and H₁₆₀-band magnitudes, as well as the magnitude in the J₁₁₀ + H₁₆₀ image that we used for detection. Sources are required to be (1) bright enough to be detected at >5σ significance in the J₁₁₀ + H₁₆₀ image and (2) meet the color selection criteria discussed in Section 3. We also assume that 15% of all z > 7 galaxies are missed because of foreground contamination, as we showed in Section 3.2. We calculate S(M, z) for both the z' (COSMOS) and z₈₅₀ (GOODS) filter sets, and find that the difference is less than 2% (for a fixed J₁₁₀ + H₁₆₀ apparent magnitude limit). Therefore, the only difference in the GOODS and COSMOS portions of this survey is that the NICMOS images in COSMOS are slightly shallower.

We measure the photometric completeness, C(m), using the IRAF package, artdata, to add point sources to the J₁₁₀ + H₁₆₀ images. We then use SExtractor with the same configuration that we used for the photometry described in Section 2.4. We find a typical completeness of 70%–80% at the 5σ detection threshold for the aggressive SExtractor parameters that we have chosen. Finally, to evaluate Equation (1), we assume that all of the z-dropouts are at z = 7, and the J₁₁₀-dropouts are at z = 9, so that C(m) translates to C(M). The resulting effective survey volumes for z ∼ 7 and z ∼ 9 are shown in Figure 4. As mentioned in Section 2, due to limited optical data, we can only include SSA22-2 in the z ∼ 7 search, but both SSA22 fields are included in the z ∼ 9 upper limit, as they contain no candidates with J₁₁₀–H₁₆₀ >1.2.

Zoom In Zoom Out Reset image size

We next constrain the UV LF. Assuming a Schecter parameterization of the LF, we show the space allowed for ϕ* and M* in Figure 5. The shaded areas show the upper limits for 68% and 95% confidence for the z ∼ 7 survey, and the area below and to the right of the dotted lines indicate the same for the z ∼ 9 search.¹² We also plot measured LFs from Bouwens et al. (2007, 2008) at z∼ 4, 5, 6, 7, and the upper limit at z ∼ 9. The nondetections that we find in this survey are consistent with the Bouwens et al. measurements, which predict 0.7 z ∼ 7 candidates in our survey, although the error bars on their z ∼ 7 LF are large, due to the small sample. The dashed line shows the upper limit at z ∼ 7 from Mannucci et al. (2007). Their constraint on luminous M* is stronger than what we have measured here, due to their wide-area survey (∼130 arcmin²). Our result is also consistent with the constraint reported by Stanway et al. (2008), where a z ∼ 7 upper limit that is similar to the z ∼ 6 LF is found.

Zoom In Zoom Out Reset image size

This limit can be used to address the controversy over the numbers of strongly lensed z > 7 galaxies (Richard et al. 2008; Bouwens et al. 2009). These authors have found differing numbers of candidates behind the same lensing clusters, using the same NICMOS data. Richard et al. find a few times more candidates than are predicted from the small unlensed sample in the field (Bouwens et al. 2008). In fact, such a comparison is difficult to make, as the lensed and field surveys observe mostly different ranges of luminosities. While the unlensed 〈J + H〉 apparent magnitudes of the Richard et al. sources range from 27 to 30, the Bouwens et al. field survey finds sources down to H₁₆₀ ∼28. However, within this 1 mag of overlap, the Richard et al. density agrees more with the Bouwens et al. measurement at z ∼ 6 than at z ∼ 7. While our survey probes even brighter magnitudes, we can compare to the Bouwens et al. z ∼ 6 LF. Assuming no evolution, this LF predicts 3.2 z ∼ 7 galaxies in our survey volume—a scenario which we can exclude with 97% confidence (Poisson statistics). Our result is more consistent with the z ∼ 7 result from Bouwens et al. (2008), as shown in Figure 5.

We also constrain the amount of star formation at z ∼ 7 and 9. To do this, we fix ϕ* ∼ 10⁻³ Mpc⁻³ mag⁻¹. This is supported by LFs that have been measured by many authors (Bouwens et al. 2007, and references therein), from z ∼ 3–6. While some scatter is present at z ∼ 6, most LFs agree with this value of ϕ* to within a factor of 2, so that any evolution of this parameter must be small. For this choice of ϕ*, we find that M* ⩾ −20.0 at z ∼ 7 and ⩾−20.7 at z ∼ 9. Assuming a steep faint end slope of α = −1.74 (Bouwens et al. 2007), and integrating the LF to zero luminosity, we find a luminosity density of ρ_L ⩽ 1.5 × 10²⁶ erg s^-1 Hz⁻¹ Mpc⁻³ at z ∼ 7. This limit is 1.9 times higher at z ∼ 9. This corresponds to a star formation density of ρ_SFR ⩽ 0.019 M_☉ yr^-1 Mpc⁻³ at z ∼ 7, when the conversion from Madau et al. (1998) is used. It is important to note that this conversion assumes no extinction, solar metallicity, and a Salpeter IMF with dN/dM ∝ M^−2.3 from M = 0.1–100 M_☉. While a correction to a more likely metallicity of 0.2 Z_☉ is negligible less than 5%), a shallower IMF slope of −1.7 will decrease the star formation rate (SFR) by a factor of 3.2 (calculated from Starburst99; Leitherer et al. 1999).

An important question remains whether galaxies at z ∼ 6–7 are capable of reionizing the neutral hydrogen in the IGM. This question is difficult to address, as it depends on the duration of the reionization. A longer reionization will require more ionizing photons over the lifetime of the galaxies in order to account for recombination (Chary 2008). Nonetheless, it is interesting to compare our upper limit to the recombination rate at z ∼ 7, for a completely ionized IGM (consistent with the WMAP 5 yr electron scattering optical depth; Dunkley et al. 2009). Madau et al. (1999) report this rate in terms of the critical SFR required to maintain an ionized IGM:

$$\begin{eqnarray} \rho _{\rm SFR, {\rm crit}} &=& \frac{0.039\;M_{\odot }\;{\rm yr^{-1}\;Mpc^{-3} }}{f_{\rm esc}}\nonumber\\ &&\times \left(\frac{1+z}{8}\right)^3 \left(\frac{C}{30}\right) \left(\frac{\Omega _b h^2}{0.0227}\right)^2\!\!, \end{eqnarray} \tag{ 2 }$$

where, again, solar metallicity and a Salpeter IMF from 0.1–100 M_☉ are assumed. This critical SFR also depends on a number of other important, but uncertain parameters. The escape fraction of ionizing photons, f_esc, has been difficult to measure. While a number of authors have found that the escape fraction is small (less than 5%–10% relative to photons escaping at 1500 Å;¹³ Malkan et al. 2003; Siana et al. 2007, C. R. Bridge et al. 2009, in preparation), there remains some evidence that it could increase with redshift (Steidel et al. 2001; Shapley et al. 2006; Iwata et al. 2009). The H ii clumping factor, C = 〈n²_H ii〉/〈n_H ii〉², is also important, as this dictates the average recombination rate per hydrogen atom relative to an IGM of uniform density. While many authors have adopted an estimate of C = 30, based on simulations by Gnedin & Ostriker (1997), more recent work suggests that this estimate is much too high, and C ≲ 10 may be more appropriate (e.g., Bolton & Haehnelt 2007; Trac & Cen 2007).

In order to meet the requirement posed by our upper limit of 0.019 M_☉ yr^-1 Mpc⁻³ at z ∼ 7, we find that C/f_esc ⩽ 15. However, this number is strongly influenced by the faint end slope of the LF, because we have integrated our constraints to zero luminosity. We have assumed a faint end slope of α = −1.74, based on Bouwens et al. (2007), but Oesch et al. (2007) show that this slope is influenced by input assumptions such as dust extinction and IGM neutral hydrogen absorption (which alter the effective survey volume). For the shallower slope of α = −1.6, reported by Oesch et al., our upper limit is reduced by a factor of 1.7, and we then require C/f_esc ≲ 9. On the other hand, it has been predicted that α approaches −2 for a sample of young galaxies undergoing their first significant bursts of star formation (Overzier et al. 2008). In this case, constraints are more dependent on the true low-luminosity cutoff.

The effects of metallicity and IMF are also important in determining the ionizing output of galaxies. We use Starburst99 models (Leitherer et al. 1999) to calculate the ionizing photon rate for metal-poor stellar populations (Z = 0.2 Z_☉) and for a shallower IMF slope. For a Salpeter IMF and Z = 0.2 Z_☉, a stellar population will produce 1.4 times more ionizing photons than a solar metallicity population with the same UV luminosity. Consequently, the constraint from this survey becomes C/f_esc < 21. Likewise, with Z = 0.2 Z_☉ and a shallower IMF slope of dN/dM ∝ M^−1.7, this constraint is relaxed to C/f_esc < 36.

Lastly, it has also been noted that the electron temperature in the primordial H ii regions will play an important role (e.g., Tumlinson et al. 2001; Stiavelli et al. 2004). Because the recombination coefficient is proportional to T^−0.7, a factor of 2 increase in temperature decreases the critical SFR by a factor of ∼1.6.

In summary, we find that for reasonable models, C/f_esc ≲ 30–40 is required to maintain an ionized IGM at z ∼ 7. This echos constraints reported by Chary (2008), who finds that for C/f_esc ∼ 60 ("high-V" case) and a Salpeter IMF the number of ionizing photons produced is too low to reionize hydrogen, unless the reionization occurred rapidly between 6 < z < 7.

The i' observations consisted of a series of exposures, of length 300–500 s each (∼6.8 hr), taken on the nights of 2008 June 19–24 with Megacam at the 6.5 m MMT (Mcleod et al. 2006). The observations were carried out through thin cirrus, except for the nights of 2008 June 20 and 21, which were photometric. Seeing varied from as low as 08 to as high as 16, and averaged about 10. The data were reduced interactively using standard techniques: bias-subtracted and flattened exposures were treated to remove cosmic rays and bad pixels before calculating the coordinates using stars from the USNOB1.0 catalog and correcting the photometry for off-axis scattered light. The resulting exposures were spatially registered to a common coordinate system. All frames taken on the nights of 2008 June 19 and 22–24 were then flux calibrated using exposures from the photometric nights, and all the frames were then co-added to create an i' mosaic. The final image has a seeing FWHM of 12. We show a cutout image centered on JD2325+1433 in Figure 6, alongside our NICMOS and IRAC images that are described in Henry et al. (2008).

Zoom In Zoom Out Reset image size

We use the NICMOS images to predict the position of JD2325+1433 in the i' image, and measure the flux in a 13 diameter aperture at this position. The noise is measured by randomly placing apertures in blank parts of the image, as was done with the NICMOS and other optical images (see Section 2). We find a signal-to-noise ratio (S/N) of 2.6 and an aperture magnitude of 26.8 ± 0.4. The aperture correction measured for point sources in the field is 2.18 ± 0.04 in flux units, and so the result is i'= 26.0 ± 0.4, total. Although the detection is weak, it strongly suggests an intermediate-redshift interloper. The probability of the i' detection being the result of a foreground contaminant (as we described in Section 3.2) within 1'' of JD2325+1433 is low (∼5%), as the i' image is not as deep as the GOODS and COSMOS optical images.

We update the photometric redshift of JD2325+1433 by including the i' measurement, and repeating the fit that we performed in Henry et al. (2008). To do this, we use the photometric redshift code, Hyperz (Bolzonella et al. 2000), with Bruzual & Charlot (2003) stellar synthesis templates. We fitted for redshift, allowing age, extinction (using Calzetti et al. 2000), and metallicity to be free (Z = 0.02, 0.2, 0.4, and 1 × Z_☉), and using four star formation histories: an instantaneous burst, a constant SFR, and two exponentially declining star formation histories with e-folding times (τ_SFR) of 100 and 500 Myr. As in Henry et al. (2008), we do not include the upper limits at 5.8 and 8.0 μm, as they do not constrain the fit. The revised, best-fitting model is shown in Figure 7. It is described by a 250 Myr instantaneous burst at z ∼ 2.7, with solar metallicity, A_V = 0.2, and a stellar mass of 9.9 × 10⁹ M_☉. The absolute B-band magnitude is M_B = −21.0.

Zoom In Zoom Out Reset image size

We use Monte Carlo simulations to assess this underconstrained problem by constructing a five-dimensional (z, age, A_V, metallicity, and star formation history) probability density function. This is done by generating 10⁵ realizations of the photometry, with magnitudes simultaneously perturbed according to the uncertainties. We then repeat the fit described above. The probability distribution in redshift space is shown in Figure 7. Now, z ∼ 2–3 solutions are favored, with 74% of realizations having a best fit at z < 5. The fact that the z ∼ 8–10 interpretation still comprises a significant fraction of the simulated fits is guaranteed by the low-S/N i' detection, which frequently dips below 1σ when perturbed in the Monte Carlo simulation. For these cases, we do not include the i' observations and the fits strongly favor the z ∼ 9 solution. Regardless, the inclusion of this weak detection in our analysis adjusts the preferred redshift to z ∼ 2–3. This is more in line with an extrapolation of the Bouwens et al. (2006, 2007) LFs, which imply a low likelihood of a galaxy at z ∼ 9.

The additional constraints from our Monte Carlo simulation suggest, for z < 5: (1) a poorly constrained age with a median of 360 Myr, and a 68% confidence interval ranging from 100 Myr to 1 Gyr, and (2) little or no extinction, with 68% of realizations preferring A_V of 0.5 or less.

The discovery that JD2325+1433 is an interloper has important implications for future z ∼ 9 surveys, because similar sources will be readily discovered with new NIR instruments. In addition to JD2325+1433  in the NICMOS Pure Parallel Survey, we find 12 more galaxies down to H₁₆₀ ∼24 which have similarly red J₁₁₀–H₁₆₀ > 1.7. As this wide-area survey is complete for such red galaxies at this limit, the density of these objects is approximately 200 degree⁻². Longer wavelength IRAC observations of a few of these sources indicate rising SEDs that are indicative of interlopers, but not all of these unusually red galaxies have yet been observed with IRAC. So it is likely that more galaxies with extremely red J₁₁₀–H₁₆₀ and a flat spectrum at longer wavelengths have been detected in the NICMOS pure parallel imaging. These sources will be difficult, if not impossible to distinguish from z > 7–8 galaxies in future surveys, meaning that deep optical imaging or a high-S/N detection of the Balmer break will be crucial.

Using deep optical imaging to distinguish z > 7–8 galaxies from interlopers will be challenging. At the faint magnitudes where these galaxies are more likely to be confirmed (H > 28 AB), optical observations with the James Webb Space Telescope (JWST) will take at least 10 hr per pointing to reach ≳30 AB at the 2σ level required for nondetection. In total, this investment in telescope time simply to confirm nondetections could amount to hundreds of hours. In addition, as we showed in Section 3.2, foreground contamination from faint, lower redshift objects can be a substantial source of incompleteness. Extrapolating number counts from the UDF, we estimate the surface density of galaxies brighter than 30 AB (total magnitudes) in B₄₃₅, V₆₀₆, and i₇₇₅ is ∼900 arcmin⁻², or 0.25 arcsec⁻². Clearly, high angular resolution will be necessary to distinguish interlopers from z > 8 galaxies, as ground-based seeing-limited observations would suffer from severe confusion.