SEARCH FOR z ∼ 7 Lyα EMITTERS WITH THE SUPRIME-CAM AT THE SUBARU TELESCOPE

The data were taken with the Suprime-Cam instrument (the Subaru Prime Focus Camera), installed at the 8 m Subaru telescope at the National Astronomy Observatory of Japan. This instrument delivers a mosaic of 10 2048 × 4096 CCDs, which covers a 34' × 27' field of view with a pixel scale of 020. A new CCD, with a quantum efficiency two times better than the preceding one at the red end (λ = 8000–10400 Å), was installed.

By taking exposures lasting 20 minutes, we ensured background-limited performance. We did offset the telescope between each exposure. The offset sequence for each field and night was chosen randomly.

We targeted two fields (hereafter called D33 and D41) covered by the Canada–France–Hawaii Telescope Legacy Survey (CFHT-LS) and the WIRCam Deep Survey (WIRDS; PIs: C. Willott and J.-P. Kneib).⁶ The total area of the survey, the two fields considered, is 2340 arcmin².

We observed using the custom-made NB filter centered at 9755 Å  with a wavelength range of Δλ = 200 Å  (NB973) during two nights in 2009 July. During these observations, the conditions were partially good: we obtained 15 and 19–20 minute exposures during these nights with a seeing varying between 05 and 085.

The total exposure time for each epoch of data is given in Table 1.

Using the Data Reduction Software developed for the Subaru Suprime-Cam instrument (Ouchi et al. 2004; Yagi et al. 2002), we perform the following steps: a bias subtraction, a flat fielding, a distortion and atmospheric dispersion correction, a sky subtraction, a bad region masking of elements such as satellite trails and an acquisition and guiding probe, the alignment of the individual exposures, and the final co-adding step resulting in a stacked image. After eliminating the low signal-to-noise regions at the edges of the field of view, we obtain an effective area of 1118 arcmin² for the D33 field and 1202 arcmin² for the D41 field.

We realize the astrometric calibration on the individual images before the final stacking using the UCAC2 catalog (Zacharias et al. 2004).

We then need to adjust the World Coordinate System parameters to obtain a more precise alignment. For this purpose we use the IRAF task msctpeak to interactively align the catalog stars in our images and update their headers. We finally obtain an astrometry calibration for each individual image with a precision of rms ∼ ± 0.1 arcsec in both directions.

The photometric calibration of the u*, g', r', i', z' broadband CFHT-LS data (described in the next section) is based on the Sloan Digital Sky Survey (SDSS) data for stars with 17 < i'  < 21 and the MegaCam–SDSS color transformation equations of Regnault et al. (2009). The precision obtained in u*, g', r', i', z' is between 0.03 and 0.02 mag. As the NB973 filter is included in the z'-band filter, we calibrate our NB973-stacked images using the AUTO magnitude from the z'-band SExtractor catalog. We perform the calibration on 1500 non-saturated stars within 16 < z'  < 20. Considering the photometric error on the broadband calibration, we obtain a photometric calibration precisely at 0.1 mag in NB973.

To estimate the limiting magnitude for the different bands, we use the script limitmag.cl available in the SDFRED package. This task constructs a count distribution from random photometric apertures on the image. Then it fits a Gaussian profile into this distribution, obtains the sigma number of this profile, and calculates the limiting magnitude of the image. We report in Table 1 the limiting magnitude for each band in each field.

Very deep optical imaging data of our observed fields are available through the CFHT-LS. For the purpose of this study, we make use of the T0006 release. These data products are available from the CADC archive and take the form of image stacks in the u*, g', r', i', z' filters and of ancillary data such as weight maps, catalogs, etc. The spectral curves of the filters u*, g', r', i', z' are similar to those of SDSS filters. We also have in hand the Near-Infrared Deep Imaging data from the WIRDS (PIs: C. Willott and J.-P. Kneib; R. Bielby et al. in preparation).

These optical data have been calibrated photometrically using the SDSS photometry, and the NIR data have been calibrated using the 2MASS photometry (McCracken et al. 2010). Considering internal and external photometric error sources, the uncertainty on the optical and the NIR data photometry is ∼0.03 mag and ∼0.02 mag, respectively.

As the broadband data and the NB data do not have the same pixel scale, we resample the broadband data using the software SWARP to obtain optical and NIR images with 0.2 arcsec pixel⁻¹. The alignment in pixels is then verified using IRAF geomap/geotran tasks.

Our complete set of data is therefore scaled at 0.2 arcsec pixel⁻¹ and covers an effective area of 1118 and 1202 $\textnormal {arcminutes}^{2}$, for D33 and D41 fields, respectively.

A summary of the observational data for both fields, D33 and D41, used in this paper is provided in Table 1. Figure 1 shows the transmission curves of the filters corresponding to the multi-band data used in this study.

Zoom In Zoom Out Reset image size

We generate the catalogs using the software SExtractor of Bertin & Arnouts (1996). We use the dual-image mode: the first image, the detection image, has been set as the combined NB973 image; the second image, the measurement image, corresponds to the resampled images from the optical and NIR bands.

We choose to detect objects in 7 pixels above a threshold of 1.2σ. The aperture used for the photometry is 1''.

Criterion 1: SNR(NB973_combined) > 5σ. We selected objects with a 5σ detection on the combined NB973 images.

Criterion 2: SNR(u*, g', r', i', χ²) < 3σ. Due to the Gunn–Peterson trough, i.e., the nearly complete absorption of the flux shortward of Lyα as a result of the large neutral hydrogen column density in the IGM, we observe spectral discontinuity at redshifts greater than about 6. We are therefore searching for objects that are not detectable in optical (u*, g', r', i') bands. A possible method, therefore, is to select objects with less than a 3σ detection in filters blueward of the expected Lyα emission: u*, g', r', i'.

The color break between the optical and NB973 filters is high and covers a wide spectral range. Moreover, for the CFHT-LS, the Terapix data center generated a deep χ² image combining the g', r', and i' images. We consider therefore this χ² image as well as the other blueward optical filters, and we select objects with less than a 3σ detection in the χ² image.

In summary, we select objects with less than a 3σ detection in the u*, g', r', i', and χ² bands.

Criterion 3: a color criterion between z' and NB973 data. The NB973 filter used for this study is included in the broad z-band filter. Although we expect an excess of flux in NB973, it is still possible to observe part of the continuum in the z-band filter. From Criterion 1 and Criterion 2, we can define the following color criterion: z' − NB973 > 0.65 mag. We use the same NB973 filter as Ota et al. (2008) have used for the search of robust z = 7 LAEs in the Subaru Deep Field. Their search was successful, as it resulted in the spectroscopic confirmation of the only z ∼ 6.96 spectroscopically confirmed LAE. Following the simulation they performed and plot in their Figure 3, we also check the color z' − NB973 of our candidates with a second color criterion: $z^{\prime }-{\it {\it NB}}973>1.72\, {\rm mag}$. We show our candidates in Figure 2, representing z' − NB973 versus NB973 for both fields, and the two color criteria. The points answering z' − NB973 > 0.65 mag but not z' − NB973 > 1.72 mag are the points with no z'-band detection. We then use the 3σ limit of the z'-band data for these objects.

Zoom In Zoom Out Reset image size

Criterion 4. NB973 − J < 0, NB973 − H < 0, and NB973 − Ks < 0.

To avoid some contamination by low-redshift objects, we choose to set NIR criteria. More explanations are given in Section 3.3.

Criterion 5. High-redshift objects are identified as compact point-like sources. After carefully applying all the criteria presented above, we visually inspect each candidate to confirm their point-like source aspects. Objects with a different aspect are more likely artifacts. For D33 7 out of 9 objects and for D41 7 out of 10 objects are selected as serious candidates.

3.3.1. Transient Objects

3.3.2. L–T Dwarfs

3.3.3. Foreground Emitters

We estimate that the lower value on equivalent width that a line emitter, with a flat continuum in f_ν, will have if selected with our criteria using the formula from Rhoads & Malhotra (2001) will require:

$$\begin{equation} {\rm EW}_{{\rm min}} \sim \left(\frac{f_{{\it NB}}}{f_{{\it BB}}}\right) \Delta \lambda _{{\it NB}} = \left[\frac{5\sigma _{{\it NB}}}{3\sigma {{\it BB}}}-1\right] \Delta \lambda _{{\it NB}}, \end{equation} \tag{ 1 }$$

with f_NB and f_BB being the flux in NB973 and the g' band, respectively, Δλ_NB the width of the NB973 filter, and σ_NB and σ_BB the flux uncertainties in NB973 and the g' band, respectively. We therefore obtained EW^D33_min ∼ 840 Å and EW^D41_min ∼ 714.2 Å. Foreground line emitters would then require an equivalent width EW^D33_min ⩾ 840 Å and EW^D41_min ⩾ 714.2 Å to contaminate our Lyα selection sample.

• 1.  
  Hα at z ∼ 0.47. Salzer et al. (2005) presented the typical equivalent width of low-redshift galaxies. They found that Hα emitters at z ∼ 0.4 have EW < 200 Å. From Equation (1), the lower limit on the Hα emitter equivalent width in our survey is 840 Å in the D33 field and 714.2 Å in the D41 field. Hα emitters at z ∼ 0.47 can therefore not contaminate our candidate selection.
• 2.  
  <?CDATA[? >O iii] at z ∼ 0.95. Xia et al. (2011) studied the line emitters in the PEARS survey and observed that [O iii] emitters at z ∼ 1 generally have an equivalent width lower than 500 Å. Moreover, looking at the typical spectral energy distributions of [O iii] emitters at z ∼ 0.9, we remark that these foreground emitters are brighter in the Ks band than in the J band and with typical J band magnitude ∼ 23 (AB). We will have therefore observed these objects in our very deep NIR data. Considering the lower limit on equivalent width that line emitters will have to contaminate our candidate sample and the NIR magnitudes of typical z ∼ 0.9–1 [O iii] emitters, the contamination by these foreground emitters is therefore ruled out.Cardamone et al. (2009) found low-redshift [O iii] emitters with very high observed EWs, reaching values up to 1500 Å. However, these emitters have very bright magnitudes in optical bands: 18 ⩽ g', r', i', z' ⩽ 21 and blue spectra. They would therefore be easily detected by the CFHT-LS data. These emitters are identified as Pea galaxies: luminous blue compact galaxies.
• 3.  
  [O ii] at z ∼ 1.6. Hayashi et al. (2011) studied [O ii] emitters at z ∼ 1.46. Figure 8 of Hayashi et al. (2011) shows these emitters in a color–magnitude diagram z' − K_s versus K_s. z ∼ 1.46 [O ii] emitters have K_s[20; 23], and z' − K_s > 0. As our NB filter is included in the z'-band filter, as seen in Figure 1, we can assume that z' − K_s > 0. This color does not agree with our Criterion 4. Figure 10 of Hayashi et al. (2011) shows the color–color diagram z' − K_s versus J − K_s. Most of their [O ii] emitters have not only a z' − K_s > 0 but also a J − K_s > 0. In the case of our sample, a deeper J-band image would help us eliminate the [O ii] emitters as a possible contaminant to verify that our candidates have J − K_s < 0, as deduced from Criterion 4.However, our K_s-band data do not entirely cover both fields. Although the contamination by z ∼ 1.6 [O ii] emitters is very unlikely due to the K_s magnitude range, we cannot completely rule it out.

3.3.4. Red Continuum Galaxies

3.3.5. False Detections

After this analysis, we can conclude that it is unlikely that our z ∼ 7 LAE candidates are contaminated by low-redshift interlopers.

A more detailed study of the foreground emitters (Hα at ∼0.47, [O iii] at ∼0.95, and [O ii] at ∼1.6) will be presented in a forthcoming paper.

We obtain a final sample in each observed field. The D33 final sample contains seven z ∼ 7 LAE candidates over the range of NB973 = 24.3–24.8 and SNR(NB973) = 5.2–7.7, and the D41 final sample contains seven z ∼ 7 LAE candidates in a range of NB973 = 24.6–25.0 and SNR(NB973) = 5.6–7.9, as described in Tables 2 and 3 and presented in Figures 3 and 4. Using the equation given below, we estimate the lower limits of the rest-frame equivalent widths derived from the photometric data and report them in Tables 2 and 3. It is interesting to observe that the candidates from the D41 field, which is deeper than the D33 field (see Table 1), are automatically fainter than the D33 ones. The observational conditions for both fields show a systematic difference of at least 0.1 arcsec in seeing:

$$\begin{eqnarray} &&{\rm EW}_\mathrm{rest} = \left(\frac{f_{{\it NB}973} \Delta \lambda _{z^{\prime }}-f_{z^{\prime }} \Delta \lambda _{{\it NB}973}}{f_{z^{\prime }} - f_{{\it NB}973}} \right) \times \frac{1}{1 + z}, \end{eqnarray} \tag{ 2 }$$

where f_NB973 is the observed flux in the NB combined image, $f_{z^{\prime }}$ is the observed flux in the z'-broadband image, and Δλ_NB973 and $\Delta \lambda _{z^{\prime }}$ are the widths of the NB973 filter (200 Å) and the z'-band filter (928 Å), respectively.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Following Ota et al. (2010), we assume that 77% of the NB flux comes from the Lyα line: f_Lyα ∼ 0.77f_NB973. We use the detection limit in the z' band to derive in turn lower EW limits.

In the D33 field, LAE 1 has an NB973 magnitude corresponding to 94% completeness, and LAE 7 has an NB973 magnitude corresponding to 89% completeness. In the D41 field, LAE 1 has an NB973 magnitude corresponding to 96% completeness, and LAE 7 has an NB973 magnitude corresponding to ∼89% completeness. To estimate the completeness, we add 200 artificial star-like objects per bin of 0.1 mag in the blank regions of the stacked images. We then run SExtractor on the image with the same parameters as those previously used for object detection. This procedure is repeated 20 times. The average count on 20 times of the number of artificial stars retrieved in each magnitude bin provides a direct measure of the completeness limit.

Two sources of variance can affect a high-redshift study: the Poisson variance and the cosmic variance; in other words, the fluctuations in the large-scale distribution of the galaxies. In order to estimate the value of this cosmic variance we use the online calculator⁷ from the model of Trenti & Stiavelli (2008). We then obtain values of 37% and 36% of cosmic variance in the D33 and D41 fields, respectively.

Considering the limited number of objects in our sample and the large comoving volume of our survey, our results are as limited by Poisson noise (∼38% for the seven objects of the D33 field and for the seven objects of the D41 field) as by clustering. Therefore, in the case in which the full sample in each observed field (seven candidates in each field) is taken into account (see Section 4.2.1), the error bars represent the Poisson variance and the cosmic variance. In the two other cases (see Sections 4.2.2 and 4.2.3), our results are more limited by the Poisson noise than by clustering. The error bars therefore represent the Poisson noise only.

For the D33 field, we obtain a total fractional error on number counts of 0.65. Since we obtain seven objects in D33, we should expect between 4.5 and 9.4 objects in D41, if the field-to-field variation is within 1σ. Since we found seven objects in the D41 field, we can therefore confirm that the field-to-field variation is within 1σ. This result was expected, as none of our fields are known to be located in overdense or underdense large-scale structure at this redshift.

Following Ota et al. (2010), we assume that on average 77% of the NB flux comes from the Lyα line. We therefore apply the same correction factor during the conversion from NB973 to Lyα fluxes.

To the Lyα LF of these z ∼ 7 LAE samples, we fit a Schechter function, Φ(L), given by

$$\begin{equation} \Phi (L){{{d}}}L=\Phi ^{*}\left(\frac{L}{L^{*}}\right)^{\alpha }\mathrm{exp}\left(-\frac{L}{L^{*}}\right)\frac{{{{d}}}L}{L^{*}}, \end{equation} \tag{ 3 }$$

in order to compare with previous high-redshift works (Hibon et al. 2010; Ouchi et al. 2010, 2008; Ota et al. 2008). The error bars shown in Figure 5 represent the total errors (cumulative Poisson errors and cosmic variance) as explained in Section 4.1. In the case of Figures 6 and 7, the results are more limited by the Poisson noise than by the cosmic variance. The error bars then represent only the cumulative Poisson errors. Considering the low number of candidates in our sample, we choose to fit two out of three of the Schechter function parameters. Following Hibon et al. (2010) and Ouchi et al. (2008), we set the faint end slope of the luminosity, α, to α = −1.5, and determine Φ* and L* by χ² minimization. We correct the data from the completeness by number weighting.

We decide to study the following different cases:

• 1.  
  The full sample of each field is real.
• 2.  
  The samples are contaminated by 50%. We then consider only the four brightest objects of each field as real (later referred to as bright samples). This is justified by the fact that the brightest candidates are the most robust ones, and we are focusing on building the bright end of the z ∼ 7 Lyα LF.
• 3.  
  We choose to derive a common Lyα LF for the D33 and D41 LAE brightest candidates. We therefore have a sample of eight bright objects in a total area of 2320 arcmin², corresponding to the sum of the D33 and D41 effective areas.

The best-fit Schechter LF parameters for each case are summarized in Table 4.

4.2.1. All the Candidates are Real

From the z ∼ 7 Lyα LF derived from the full sample of the D41 field, shown in Figure 5 as the blue line, and the corresponding candidates represented as filled squares, we observe a possible but weak evolution in luminosity from the z ∼ 6.5 Lyα LF from Ouchi et al. (2010) (dot-long-dashed line in Figure 5), but no possible evolution from the z ∼ 6.5 Lyα LF from Kashikawa et al. (2011) (dot-short-dashed line in Figure 5). The z ∼ 7 Lyα LF from the complete sample of the D33 field, shown as the red line and the filled triangles in Figure 5, shows weaker evolution in luminosity from the z ∼ 6.5 Lyα LF. However, looking at the error bars from the D33 and D41 field candidates, both weak evolutions in luminosity become very discussable. From these possible z ∼ 7 Lyα LFs, if all the candidates are real for one of the observed fields, we cannot therefore make a conclusion about possible evolution in luminosity from z ∼ 6.5 to z ∼ 7.

Zoom In Zoom Out Reset image size

4.2.2. Only the Brightest Candidates are Real

We derive Lyα LFs for the two different z ∼ 7 bright LAE candidate samples. The best-fit Schechter functions are shown in Figure 6. In both cases, we do not observe a significant evolution in either density or in luminosity from z ∼ 6.5, from Ouchi et al. (2010) or Kashikawa et al. (2011), to z ∼ 7.

Zoom In Zoom Out Reset image size

4.2.3. One Common LF for both Fields Bright Sample

By fitting a unique Schechter LF for both samples (as seen in Figure 7), we assume that the brightest objects from both samples are real z ∼ 7 LAEs. Looking only at the filled triangles in Figure 7, representing the four brightest candidates from D41 and the ones from D33, we remark that they fit the z ∼ 6.5 Lyα LF from Ouchi et al. (2010) but not the z ∼ 6.5 Lyα LF from Kashikawa et al. (2011). Depending on the z ∼ 6.5 Lyα LF we considered, the best-fit Schechter-derived LF, shown as the red line in Figure 7, could agree with the observed evolution between z ∼ 5.7 and z ∼ 6.5 but would not be in favor of a strong evolution between z ∼ 6.5 and z ∼ 7.

Zoom In Zoom Out Reset image size

4.2.4. Summary