z ∼ 7–10 GALAXIES BEHIND LENSING CLUSTERS: CONTRAST WITH FIELD SEARCH RESULTS^*

Because of the great distances and extreme faintness of galaxies at z ≳ 7, as well as the high sky backgrounds, detection of galaxies at such high redshifts remains extremely challenging. It is not surprising that the number of robust z ≳ 7 candidates is still very small (see e.g., Bouwens & Illingworth 2006; Bouwens et al. 2008; Oesch et al. 2008; Bradley et al. 2008). Gravitational lensing by galaxy clusters has been highlighted as an efficient way of improving this situation, due to the significant areas on the sky behind these clusters with sizeable magnification factors to amplify light from faint sources. However, this advantage is offset by the greatly reduced source plane volume in the highly magnified regions.

As a result of the trade-off between depth and area, the utility of clusters for these searches depends strongly on the slope of the luminosity function (e.g., Broadhurst et al. 1995; Santos et al. 2004). If the effective slope of the LF (−d(log dΦ)/dlog L) is greater than 1, the gain in depth more than compensates for the loss in area, increasing the overall number of sources (e.g., Broadhurst et al. 1995) over that found in the field. Such steep slopes are found at magnitudes brightward of L* (corresponding to H ≲ 27 AB mag for z ∼ 7 galaxies). By contrast, at fainter magnitudes (H ≳ 27 for z ∼ 7 galaxies) the effective slope of the LF is not quite so steep (e.g., the faint-end slope for the Bouwens et al. 2007 z ∼ 6 LF corresponds to −d(log Φ)/d(log L) ∼ 0.7 < 1). This trade-off between depth and area is such that the surface density of dropouts is lower behind clusters than in the field at faint magnitudes. Overall these considerations suggest that the most significant advantages will be achieved at bright magnitudes (>L*) where the LF is very steep. Shallow searches over many clusters, in particular, would seem to be the most rewarding.

Here we assess the promise of clusters for studying z ≳ 7 galaxies by conducting a careful search for high-redshift galaxies in all the currently available Hubble Space Telescope (HST) NICMOS imaging data over 11 massive low-redshift galaxy clusters. Bradley et al. (2008) have already conducted such a search around Abell 1689 and reported one highly robust z ≳ 7 galaxy. Richard et al. (2006) examined two clusters and reported 13 z ≳ 6 candidates, while Richard et al. (2008) have examined six clusters and reported 12 z ≳ 6 candidates. The present paper represents an independent assessment of the prevalence of these sources behind massive low-redshift clusters. We take advantage of ∼21 arcmin² of very deep NICMOS data behind 11 lensing clusters with optical ACS+WFPC2 coverage, seven of which were already considered in the Bradley et al. (2008) and Richard et al. (2008) papers. For our z ≳ 7 search, we will utilize many of the same photometric techniques we have employed over the past few years to identify large samples of z ∼ 4–8 BViz dropouts in the field (Bouwens et al. 2004, 2006, 2007, 2008).

We begin this paper with a summary of the observational data available to search for z ≳ 7 galaxies behind galaxy clusters (Section 2). In Section 3, we describe our techniques for constructing source catalogs and doing the photometry (Section 3). In Section 4, we summarize the results of our selection and discuss sources of contamination. In Section 5, we compare our results with (1) what we would expect based upon the z ∼ 7 LF derived in the field and (2) the search results from other teams. We then conclude by discussing the implications of this study (Section 6). In the Appendix, we provide an assessment of the z ≳ 7 candidates reported by Richard et al. 2008. In particular, we assess the merits of searching for z ≳ 7 galaxies behind clusters versus searching for these galaxies in the field (Section 6). We assume Ω₀ = 0.3, Ω_Λ = 0.7, H₀ = 70 kms^-1Mpc⁻¹ throughout. Although these parameters are slightly different from those determined from the WMAP five-year results (Dunkley et al. 2008), they allow for convenient comparison with other recent results expressed in a similar manner. The HST filters F555W, F606W, F625W, F702W, F775W, F814W, F850LP, F110W, and F160W will be denoted as V₅₅₅, V₆₀₆, r₆₂₅, R₇₀₂, i₇₇₅, I₈₁₄, z₈₅₀, J₁₁₀, and H₁₆₀, respectively. We will express all magnitudes in the AB system (Oke & Gunn 1983).

We conduct searches for dropout galaxies over the 11 low-redshift galaxy clusters with deep near-IR NICMOS and optical ACS coverage (Table 1). The near-IR data here were taken with the goal of finding z ≳ 7 galaxies. The NICMOS coverage of the first six clusters considered here (CL0024, MS1358, Abell 2218, Abell 2219, Abell 2390, and Abell 2667) extends over 8.4 arcmin² (12 NIC3 pointings) and includes very deep imaging in both the J₁₁₀ and H₁₆₀ bands (Richard et al. 2008). The NICMOS coverage of three other clusters considered here (i.e., Abell 1689, Abell 1703, and 1E0657-56) is somewhat shallower in general, extends over ∼10 arcmin² (17 NIC3 pointings), and is mainly in the J₁₁₀ band. The final two clusters considered here (Abell 1835 and AC114) only have deep near-IR NICMOS data in the H₁₆₀ band and over one ∼0.8 arcmin² NIC3 pointing per cluster. The layout of our NICMOS search fields is illustrated graphically in Figure 1.

Zoom In Zoom Out Reset image size

Deep optical imaging data are necessary for the selection of z ≳ 7 galaxies and can be a significant limitation, if not available. MS1358, CL0024, Abell 1689, Abell 1703, and Abell 2218 all possess useful optical data, with ⩾ two orbits in each of the ACS g₄₇₅, r₆₂₅, and i₇₇₅ bands and ⩾ six orbits in the ACS z₈₅₀ band. The optical coverage of Abell 2219, Abell 2390, Abell 2667, Abell 1835, and AC114 is generally shallower in depth and primarily with WFPC2, though ∼ four orbits of ACS z-band coverage are available for each.

All the available ACS and NICMOS data over these clusters are processed into image mosaics using the ACS GTO pipeline "apsis" (Blakeslee et al. 2003) and NICMOS pipeline "nicred.py" (Magee et al. 2007). Reductions of the WFPC2 data are obtained from the Canadian Astronomy Data Centre. All reductions are registered onto the same frame as the NICMOS data.

1. Catalog Generation. Our procedure for generating source catalogs and doing photometry is identical to that performed in Bouwens et al. (2008). Briefly, we run SExtractor (Bertin & Arnouts 1996) in double-image mode to do object detection and photometry. For the detection image, we use the square root of the χ² image (Szalay et al. 1999), which we construct from the NICMOS J₁₁₀ and H₁₆₀ images for our z₈₅₀ dropout selection and the NICMOS H₁₆₀ band image for our J₁₁₀ dropout selection. Our photometry is then conducted on our ACS, WFPC2, and NICMOS images, which are point-spread functions (PSF), matched to the NICMOS H₁₆₀ image. Colors are measured in small-scalable (Kron 1980) apertures, assuming a Kron (1980) factor of 1.2. These fluxes are then corrected to total magnitudes using the light within a larger Kron (1980) aperture (adopting a Kron factor of 2.5). These latter corrections are made using the square root of the χ² image to improve the S/N. Figure 5 of Coe et al. (2006) provides a graphical description of a similar multistage procedure for measuring colors and total magnitudes. Typical aperture diameters are 03 and 06 for color and total magnitude measurements, respectively.

2. Selection Criteria. We use the same selection criteria for identifying star-forming galaxies at z ≳ 7 that we used in our previous work on the identification of such galaxies in field data sets such as GOODS or the HUDF (e.g., Bouwens et al. 2008). Specifically, we require a z-dropout candidate to satisfy the criterion ((z₈₅₀ − J₁₁₀)_AB)>0.8) ∧ ((z₈₅₀ − J₁₁₀)_AB > 0.8 + 0.4(J₁₁₀ − H₁₆₀)_AB) where ∧ represents the logical AND symbol. J-dropout candidates are expected to satisfy the criterion (J₁₁₀ − H₁₆₀)_AB > 1.3. In cases of a nondetection in the dropout band, we set the flux in the dropout band to be equal to the 1σ upper limit. We require each candidate to be completely undetected (<2σ) in all passbands blueward of the dropout band. We also demand that each candidate be detected at 5.5σ in the H₁₆₀ band in a 05 diameter aperture to eliminate spurious sources.

Our selection of z ≳ 7 galaxy candidates behind clusters not possessing NICMOS imaging in both the J and H bands is slightly more involved. Our approach was to first identify possible z ≳ 7 candidates in the available single band NICMOS data (be it either J or H band data). We then obtained follow-up observations on these candidates in the other NICMOS band to thus permit the use of the above full z and J dropout criteria. In detail, for clusters with only J-band imaging (Abell 1689, Abell 1703, and 1E0657-56), the initial selection is based on a (z₈₅₀ − J₁₁₀)_AB > 1.0 criterion while for clusters with only H-band imaging (Abell 1835 and AC114), a (z₈₅₀ − H₁₆₀)_AB > 1.2 criterion is applied. Follow-up NICMOS observations are then obtained for all z ≳ 7 candidates identified behind these clusters. As only one such candidate is found, this required that NICMOS H-band observations be obtained for just the one source A1689-zD1 (Bradley et al. 2008).

After careful application of our selection criteria to all 11 clusters under study here, we identify four z-dropout candidates and one J-dropout candidate. We also uncover a small number of candidates that appeared to be z ≳ 7 candidates (e.g., the candidates at 00:26:37.90, 17:09:10.4 or 00:26:35.11, 17:10:10.3 candidate behind CL0024), but which show modest (∼2σ) detections in passbands blueward of z₈₅₀ and therefore were excluded.

Postage stamps of the four candidate star-forming galaxies at z ≳ 7 are provided in Figure 2. Other properties are given in Table 2. Our candidate in Abell 1689 has already been reported upon before by Bradley et al. (2008) and is by far the strongest candidate. The spectral energy distribution (SED) derived from the very deep optical, NICMOS, and IRAC data can only be successfully fitted by the SED of a star-forming galaxy at z ∼ 7.6 (Bradley et al. 2008). We estimate, using the techniques in the following section (i.e., by adding noise to the colors of intermediate magnitude galaxies from the HUDF), that the probability is very small (<0.2%) that this source is a low-redshift galaxy.

Zoom In Zoom Out Reset image size

Possible Contamination of our Selection by Lower-Redshift Galaxies. The four new sources in our selection are much less robust z ≳ 7 candidates. All are in clusters for which the optical data are much shallower. The V₅₅₅, V₆₀₆, or I₈₁₄ WFPC2 coverage over Abell 2390 and Abell 2667 reaches to only ∼26.6 and ∼26.5 (5σ, 03 diameter aperture) over Abell 2390, and 2667, respectively. These WFPC2 data are ⩾1 mag shallower than the ACS data available over five other clusters studied here (Table 1). This will result in a much higher contamination rate from low-redshift galaxies scattering into our color selection due to noise.

To estimate the expected contamination level, we start with a selection of intermediate magnitude galaxies in the HUDF NICMOS field (Thompson et al. 2005) and add noise to the fluxes to match the errors for sources in our search fields. On average, we find approximately one low-redshift contaminant over Abell 2390 and approximately two such sources over Abell 2667 in these simulations. This is comparable to the observed numbers and suggests that A2390-zD1, A2667-zD1, A2667-zD2, and A2390-JD1 are likely to be low-redshift interlopers. As an alternate estimate of the contamination level, we start with the catalog of sources in the three clusters with the deepest ACS and NICMOS data (Abell 2218, MS1358, CL0024) and add noise (again to match the flux errors for sources behind Abell 2390 and Abell 2667). On average, we predict approximately two low-redshift contaminants over Abell 2390 and ∼3.5 low-redshift contaminations over Abell 2667. Again these numbers are comparable to the observed number of candidates over these clusters and suggest that all four candidates behind these clusters may be interlopers. The modest differences in the predicted number of contaminants for the two methods is consistent with what one would expect from small number statistics.

Obviously, there is a clear difference in quality between all four of our weaker z ≳ 7 candidates—which are quite consistent with being low-redshift interlopers—and our strongest candidate where this is unlikely (the probability it is not a low-redshift interloper is > 99.8% (>3σ)). We therefore report A1689-zD1 as our only robust z ∼ 7 candidate.

Surface Density of Robust z ≳ 7 Candidates Behind Massive Galaxy Clusters. We tabulate the surface density of robust z ≳ 7 candidates found behind clusters as a function of magnitudes in Table 3. Also included in this table are the surface densities found in the field (Bouwens et al. 2008; R. J. Bouwens et al. 2008, in preparation). The surface density of z ≳ 7 candidates behind clusters appears to be substantially larger than in the field at bright magnitudes (i.e., H < 25.5 or >5L*). This is exactly what we expect as a result of the slope of the LF at bright magnitudes. Interestingly, current observations do not provide any evidence for an enhancement in the surface density of z ≳ 7 sources between H ∼ 25.5 and H ∼ 27. This is despite the expected steep slope of the LF at such magnitudes. This could reflect the small number of sources involved here and point to the need for imaging more clusters to improve the overall statistics (see also Section 6 and R. J. Bouwens et al. 2008, in preparation).

At H ≳ 27 AB magnitude we are probing faintward of L* and likely reaching the regime where the effective slope of the LF is only moderately steep (i.e., −d(log Φ)/d(log L) ∼ 0.7 < 1; Bouwens et al. 2007). For such slopes, the trade-off between depth and area is such that the surface density of z ≳ 7 dropouts behind clusters will be lower than that observed in the field.

5.1. Comparison with Model Expectations

5.2. Comparison with Previous Results

The purpose of the present study is to increase the sample of candidate z ∼ 7 galaxies and to assess the potential of gravitational lensing by galaxy clusters to identify and quantify the properties of galaxies at z ≳ 7. Lensing will increase the depth of the survey by the magnification factors, but decrease the search area by the same factor. If the effective slope of the LF (−d(log dΦ)/dlog L) is greater than 1, the gain in depth more than compensates for the loss in area, increasing the overall number of sources (e.g., Broadhurst et al. 1995; Santos et al. 2004; Richard et al. 2008). We would expect this effect to increase the numbers at bright magnitudes where the effective slope of the LF is very steep due to an apparent cut-off at the bright end (i.e., H ≲ 27), but to decrease the numbers at fainter magnitudes (H ≳ 27) where the faint-end slope is shallower than this (i.e., −d(log dΦ)/dlog L ∼ 0.7 ≲ 1; e.g., Bouwens et al. 2007).

Our simulations (Section 5.1) show the expected gains at the bright end of the LF for searches behind clusters (see Table 4). We expect three z ≳ 7 galaxies over the present set of cluster data (144 NICMOS orbits) using the observed LF of Bouwens et al. (2008). The 48 NICMOS orbits/galaxy in the clusters for bright sources contrasts with the ∼120 NICMOS orbits/galaxy needed in the field (nine z ≳ 7 galaxies are found in ∼1050 orbits of NICMOS data over the GOODS fields; R. J. Bouwens et al. 2008, in preparation). Nearly identical search procedures and selection criteria are used in both the cluster and field searches. Interestingly, the expected gains at bright magnitudes in clusters are not reflected in the observational results above, if we only include our most robust candidate (for ∼144 NICMOS orbits/galaxy). This could easily arise because of small number statistics. Any gains in using clusters are likely to disappear at lower luminosities faintward of the LF knee, since the slope is not steep enough, as noted above.

Of course, lensing clusters can be used to potentially detect objects much fainter than in a field sample, thereby possibly extending the LF function to fainter limits. Unfortunately, both small number statistics and the challenges of modeling clusters make this very difficult, and likely not to be a very practical approach. This is because a substantial sample of objects is needed to faint limits to accurately determine the LF, as well as extremely accurate models of both lensing by the foreground cluster and incompleteness suffered by the lensed high-redshift population. Determining either of these latter two quantities well (and without any systematics) is a great challenge. As a result, it can be difficult to even measure quantities such as the faint-end slope of the UV LF at z ∼ 4, z ∼ 5, and z ∼ 6 from current samples of g, r, and i dropouts behind lensing clusters (where the samples are much larger: see R. J. Bouwens et al. 2008, in preparation).

Our findings also underline the importance of having very deep optical data below the Lyman break for identifying high-z dropout galaxies. Without such data, it is essentially impossible to distinguish bona fide z ≳ 7 galaxies from the large number of low-redshift galaxy interlopers that may scatter into z ≳ 7 samples as a result of noise—as we found for the four weaker z ≳ 7 dropout candidates in our selection (Section 4). Unfortunately, obtaining sufficiently deep optical data can be expensive and often requires ≳2 − 3× as much time as is spent obtaining the near-IR data.

We thank Richard Ellis, Marusa Bradač, Johan Richard, Daniel Schaerer, and Dan Stark for useful discussions. We thank Johan Richard for helping us to more precisely locate some of his team's lower S/N z ≳ 7 candidates in our own reductions of the available NICMOS data. We thank our referee for important feedback that was valuable for improving the overall clarity of our manuscript. This research used the facilities of the Canadian Astronomy Data Centre operated by the National Research Council of Canada with the support of the Canadian Space Agency. We acknowledge support from NASA grants HST-GO09803.05-A, HST-GO10874.04-A, and NAG5-7697.

Richard et al. (2008) also conducted a search for z ≳ 7 galaxy candidates behind 6 of the 11 galaxy clusters considered in this study (i.e., CL0024, CL1358, Abell 2218, Abell 2219, Abell 2390, Abell 2667). They reported finding 10 z ∼ 7 z-dropout candidates and 2 z ∼ 9 J-dropout candidates. Surprisingly, none of the 12 candidates in the Richard et al. (2008) z ≳ 7 sample make it into our own z ≳ 7 dropout selections (Section 4) —nor do our z ≳ 7 candidates appear in their selection—even though our color selection criteria are essentially identical (and their search data a subset of ours).

To understand the reason for these differences, we performed photometry on the sources using the same 03 diameter (for ACS+WFPC2) and 06 diameter (for NICMOS) apertures employed by Richard et al. (2008) for their photometry. We then corrected the magnitudes we measured from the ACS+WFPC2 or NICMOS data by 0.3 mag and 0.6 mag, respectively (again following the methodology of Richard et al. 2008). For the four Richard et al. (2008) candidates where light from the central cluster galaxy would contaminate the photometry (i.e., A2218-z2, A2390-z2, A2667-z2, A2667-j1), we explicitly fit the isophotes to the central galaxy and subtracted them off. A summary of our photometry of the 12 Richard et al. (2008) candidates is presented in Table 5. We present images of the 12 z ≳ 7 candidates from the Richard et al. (2008) selection in Figure 3.

Zoom In Zoom Out Reset image size

Of the 12 candidates, 2 (i.e., A2219-z1 and A2667-z2) seem like plausible z ≳ 7 galaxies (A2219-z1 being the stronger of the 2). Both candidates satisfy our selection criteria (but were blended with foreground galaxies in our catalogs and therefore not included in our candidate lists). The other 10 candidates appear unlikely to correspond to high-redshift sources, given our photometry (although in the case of A2667-z1 it is difficult to rule out the source being at z ≳ 7). Ascertaining the precise redshift of A2667-z2 is difficult given the limited depth of the optical data around Abell 2667. Without deeper optical data, it is essentially impossible to know which z ≳ 7 candidates correspond to high-redshift galaxies and which are lower redshift contaminants. Note that we encountered similar difficulties in ascertaining the nature of the 4 weaker z ≳  7 candidates found in our selection behind Abell 2390 and Abell 2667 (neither of which has particularly deep optical coverage; see Table 1). Our simulations and tests (Section 4) suggested that substantial contamination by low-redshift galaxies is quite likely.

Richard et al. (2008) were also concerned about contamination in their z ≳ 7 selection and ran numerous simulations in an attempt to correct for it. They argue that only five of the candidates (or 50%) are likely at z ≳ 7. However, from our photometry, it would appear that only two of their candidates (and perhaps three) are plausible z ≳ 7 galaxies, and so we would argue that the actual number of z ≳ 7 sources in their selection is much fewer than five and likely ≲2 (i.e., fewer than 25% of their candidates correspond to z ∼ 7 galaxies).

In addition to our above comments about specific z ≳ 7 candidates in the Richard et al. (2008) selection, we also have several general concerns about the collective properties of their z₈₅₀-dropout sample. Our first concern regards the J₁₁₀ − H₁₆₀ color distribution of this sample. The colors are ∼0.5 mag bluer on average than what we find in the field for our z-dropout sample (Bouwens et al. 2008; Figure A1). They are also ∼0.5 mag bluer than what one would expect for a young β = −2 star-forming galaxy population (as observed at z ∼ 5 − 6: e.g., Lehnert & Bremer 2003; Stanway et al. 2005; Bouwens et al. 2006). It is hard to understand why z-dropouts from the Richard et al. (2008) selection would be so much bluer than these expectations unless the candidates had strong Lyα emission in the J₁₁₀ band. However, such an explanation would appear to be ruled out by the follow-up spectroscopy that Richard et al. (2008) conduct that find no such emission for the 7 z-dropout candidates they observe. Richard et al. (2008) also note this concern.

Second, we find it worrisome that almost all (8 out of 10) of their z₈₅₀-dropout candidates only satisfy their z₈₅₀ − J₁₁₀ color criterion by a small margin (<0.4 mag) whereas in the Bouwens et al. (2008) field sample, almost half of the z-dropouts satisfy the z₈₅₀ − J₁₁₀ color criterion by a wide margin (i.e., >1 mag). One would expect the situation to be quite opposite here, as a result of the much flatter number counts expected behind galaxy clusters (the number of bright sources in cluster fields would increase due to magnification by the cluster and the number of faint sources would decrease because of the loss of area). This should result in a larger fraction of sources satisfying the selection criterion by a wide margin, not a smaller fraction.