A survey of NB921 dropouts in the Subaru Deep Field

In order to search for high-redshift galaxies beyond $z = 6.6$ in the Subaru Deep Field, we have investigated NB921-dropout galaxies where NB921 is the narrowband filter centered at 919.6 nm with FWHM of 13.2 nm for the Suprime-Cam on the Subaru Telescope. There are no secure NB921-dropout candidates brighter than $z^\prime = 25.5$. Based on this result, we discuss the UV luminosity function of star-forming galaxies at $z>6.6$.


Introduction
In the last decade, surveys for high-redshift galaxies have been very successful thanks to the great observational capabilities of 8-10m class telescopes. Both the Lyman break method and the Lyman α emitter search technique have provided us a large number of high-redshift galaxies at z > 3 (e.g., Steidel et al. 1996Steidel et al. , 1999Ouchi et al. 2004aOuchi et al. , 2004bHu et al. 2002;Ajiki et al. 2003;Kodaira et al. 2003;Taniguchi et al. 2005). In particular, these surveys tell us the evolution of the star formation rate density up to z ∼ 6 (e.g., Taniguchi et al. 2005; see also for review, Taniguchi et al. 2003;Spinrad 2004) and the presence of large scale structure in the very early universe (e.g., Ouchi et al. 2005 and references therein). Now, the frontier of high-redshift galaxy surveys have been shifted to z > 6. It becomes important to clarify the star formation rate density and the clustering properties at z > 6. Spectroscopic observations of the highest known quasars suggest that the reionization of the intergalactic medium (IGM) could have just completed at z ≥ 6 (Djorgovski et al. 2001;Becker et al. 2001; see for a review Loeb & Barkana 2001). On the other hand, a large amplitude signal in the temperature-polarization maps of the cosmic microwave background by The Wilkinson Microwave Anisotropy Probe (WMAP) is interpreted as that the universe became reionized at z ∼ 17 (Spergel et al. 2003;Kogut et al. 2003). This also reinforces the importance of investigation of galaxies beyond z ∼ 6 from various observational points of view.
Recently, Bouwens et al. (2004bBouwens et al. ( , 2005 tried to search galaxies at z ∼ 7-8 as z 850dropouts in the Hubble Ultra Deep Field (HUDF) and those at z ∼ 10 as J 110 -dropouts in the fields with deep NICMOS image. They found four and less than three candidates, respectively, in their surveys. These pioneering works may lead to a conclusion that the star formation rate density is declining toward higher redshift beyond z ∼ 6. Although the HUDF surveys are unprecedentedly deep, their survey field is not wide enough to obtain a general conclusion because the field of view of the NICMOS is very small. We therefore tried to search galaxies at z > 6.6 using the deep optical imaging data set for the Subaru Deep Field (SDF: Kashikawa et al. 2004). Applying the NB921-dropout method, we investigate how many such very highz galaxies are present in this field. We describe the NB921 dropout method in section 2. In section 3, we then apply the method to the SDF data We discuss the star formation rate density at z > 6.6 in section 4.

NB921 dropouts
The dropout method is a powerful tool to find high redshift galaxies. The basic idea of the NB921-dropout method used in this study is almost the same as that used in previous dropout galaxy surveys (e.g., Steidel et al. 1996Steidel et al. , 1999Stanway et al. 2004). Since the UV continuum shortward Lyα of high-z galaxy is absorbed by the intergalactic neutral hydrogen gas clouds, we can use this property to select high-z galaxies. We call galaxies with z > 6.6 as (genuine) NB921-dropout galaxies and galaxies with NB921 − z ′ > 0 simply NB921-depression galaxies. Note that a similar narrowband dropout method has been already applied to the SDF data; the NB816 dropout method to select galaxies with 5.7 < z < 6.5 (Shioya et al. 2005b; see also Shioya et al. 2005a).
In fig.1, we show the transmission curves of the two filters of z ′ and NB921. We also show SEDs of an LBG at z = 6.1, 6.57, and 6.8. The stellar continuum of an LBG is produced by GALAXEV (Bruzual & Charlot 2003). Here we adopt τ model, i.e., SF R(t) ∝ exp(−t/τ ) with τ = 1 Gyr and age of t = 1 Gyr. We then add the Lyman α emission line with EW 0 (Lyα) = 65 Ato the above stellar continuum. We also use the cosmic transmission by Madau (1996) to evaluate the absorption by intergalactic medium. Note, however, that we ignore the extinction by dust in this figure.
Since the bandpass of the NB921 filter is completely included in that of the z ′ filter, the relation between NB921 − z ′ and z is slightly complex. For example, as we mentioned in Taniguchi et al. (2005), if the UV continuum depression shortward of Lyman α (121.6 nm) due to the absorption of the intergalactic medium is redshifted not into the NB921 window but into the z ′ window, corresponding to a case that galaxies are located at 5.9 < z < 6.5, the NB921 − z ′ color becomes to be bluer (i.e., NB921-excess). However, as we mentioned in Nagao et al. (2004), even if this is the case, when a galaxy has a strong Lyα emission line, the NB921 − z ′ color may become redder (i.e., NB921-depression) although the degree of this depression depends on the strength of Lyα emission.
Adopting both the cosmic transmission derived by Madau et al. (1996) and the extinction curve derived by Calzetti et al. (2000), we calculate the NB921 − z ′ color as a function of z for the case of EW 0 (Lyα) = 0 (stellar continuum only), 65, 130, and 260Å, respectively. The results are shown in figure 2. It is clearly shown that Lyman break galaxies with strong Lyα emission are identified as NB921-depression objects even if their redshifts are smaller than 6.6. Here we note that i ′ -dropout galaxies with NB921-depression found by Nagao et al. (2004Nagao et al. ( , 2005 are Lymanα emitters at z ∼ 6 and their EW 0 (Lyα) ranges from 110Åto 280Å. We therefore adopt a color criterion of NB921 − z ′ > 1.5 for unambiguous NB921-dropout galaxies.

Application to the Subaru Deep Field Data
We apply the NB921-dropout method to the SDF data (Kashikawa et al. 2004). Its photometric catalog (Version 1) is in public 1 . In the following analysis, we use 2 ′′ φ aperture magnitudes and photometric errors in a 2 ′′ φ aperture magnitude in the z ′ -band selected catalog. We also calculate errors of colors using the above photometric errors.
We adopt the following selection criteria for NB921-dropout galaxies, and since we calculate the lower limit of NB921 − z ′ using the 2σ limiting magnitude for galaxies fainter than the above z ′ limiting magnitude; the 2σ limiting magnitude of NB921 is 26.98 mag. Figure 3 shows a diagram between NB921 − z ′ and z ′ . It is found that 14 galaxies satisfy the above criteria (open crosses in figure 3). In order to reduce any contaminations from foreground objects that are free from absorption by the intergalactic neutral hydrogen, we also adopt the following additional criteria and V > 27.74 (3σ).
By using the above three criteria, we located no NB921-dropout object. We therefore conclude that there is no galaxy at z > 6.6 with z ′ < 25.48 in the SDF. All the objects with NB921 − z ′ > 1.5 are detected both in B and in V bands and considered to be low redshift galaxies. Galaxies at z > 6.6 could be selected as a subsample of i ′ -dropout galaxies. We now investigate the NB921 − z ′ color of z ′ -dropouts. We select i ′ -dropout galaxies adopting the following criteria: i ′ −z ′ > 1.5, z ′ < 26.07 (5σ), B > 28.45 (3σ), V > 27.74 (3σ), and R > 27.80 (3σ). The number of i ′ -dropout candidates is 49. We plot them on a diagram between NB921 − z ′ and z ′ (figure 3) and on another diagram between i ′ − z ′ and NB921 − z ′ (figure 5). All the i ′dropouts except three are detected in NB921. For the three galaxies undetected in NB921 (open circles in figure 3), we cannot judge if they are genuine NB921-dropout galaxies unless future spectroscopic confirmation is available. However, we conclude that there are three possible genuine NB921-dropout candidates in the SDF if we lower the z ′ limiting magnitude down to 26.07.
Finally, we comment on another three galaxies whose NB921 − z ′ color is larger than 0.9 together with above 3σ(NB921 − z ′ ) (open squares in figure 3). As we discussed in section 2, strong Lyman α emitters at 6.0 < z < 6.5 have a color of NB921 − z ′ ≃ 1. We, therefore, do not select them as NB921-dropout galaxies in this paper, although they may be genuine NB921-dropout galaxies at z ∼ 6.6. It is noted that all these galaxies are detected in NB921 and some of z ′ -excess objects are spectroscopically confirmed as Lyman α emitters at z < 6.6 (Nagao et al. 2004(Nagao et al. , 2005.

DISCUSSION
4.1. UV Luminosity Function of Galaxies at z > 6.6 We discuss how our new result gives a constraint on the UV luminosity function of starforming galaxies at z > 6.6. In the present study, we find no genuine NB921-dropout galaxy in the SDF from a sample of galaxies with z ′ < 25.48 and three possible candidates from a sample of z ′ < 26.07. The observed z ′ -magnitude of a galaxy depends not only on the UV luminosity but also on both the redshift of the galaxy and the strength of the Lyα emission line. Therefore, it seems convenient to show how our upper limits are consistent with predictions from previous luminosity functions for high-z galaxies.
In figure 6, we show expected cumulative numbers of galaxies at z > 6.6 if the UV luminosity function of galaxies at z > 6.6 is the same as that at z ∼ 4.0 (Ouchi et al. 2004a) or at z ∼ 6 (Bouwens et al. 2004a) for the following two cases; no Lymanα emission or Lymanα emission whose rest-frame equivalent width of 65Å. Our upper limits are also shown in this figure. Here we incorporate the completeness for the SDF sample as a function of z ′ magnitude (Kashikawa et al. 2004). We expect that the number of galaxies brighter than z ′ = 25.48 ranges from 4 × 10 −4 to 0.5 and that brighter than z ′ = 26.07 ranges from 3 × 10 −2 to 4 (figure 6).
We may conclude that the number of UV bright galaxies at z > 6.6 is smaller than that at z ∼ 6 even if all the star-forming galaxies emit the strong Lyman α emission. Bouwens et al. (2004bBouwens et al. ( , 2005 surveyed for the star-forming galaxies at z ∼ 7-8 and z ∼ 10 in the Hubble Ultra Deep Field and other deepest HST NICMOS field and demonstrate that the star formation rate density at z > 6 is decreasing with increasing redshift. Our result is consistent with their demonstration.

Further Comments
In this paper, we used a combination of the two filters, NB921 and z ′ . In general, when the filter band-passes overlap, interpretations may sometimes get convoluted. However, in particular, when we study either LBGs or Lyman α emitters or both at high redshift, there is little ambiguity in selecting such galaxies as demonstrated in figure 4.
Our study as well as previous works (e.g., Bouwens et al. 2005) have suggested possible evidence for the decreasing star formation activity with increasing redshift. In future, it seems desirable to obtain near infrared imaging data of the SDF. In particular, if we would have a deep imaging data with a filter that starts at λ = 1008 nm, we could make a comparative study of the star formation activity between z ∼ 6.6 and z ∼ 7.5 unambiguously. The NICMOS camera on board the HST is the most efficient camera for this kind of works. However, its field of view is too small to cover the entire filed of the SDF. The new near infrared camera and spectrograph, MOIRCS (Multi-Object Infrared Camera and Spectrograph;Tokoku et al. 2003), will be available on the Subaru Telescope; its FOV is 4 ′ × 7 ′ . This camera will be useful in exploring the star formation activity at z > 7 in the SDF.
We would like to thank the staff of the Subaru Telescope. We would also like to thank the anonymous referee for the useful comments and suggestions. This work was financially supported in part by the Ministry of Education, Culture, Sports, Science, and Technology (Nos. 10044052, and 10304013) and JSPS (No. 15340059 and 17253001). MA, SSS, and TN are JSPS fellows. Wavelength (angstrom) m AB +constant z= 6.57 6.8 6.1 Fig. 1. The upper panel shows the band response curves of the filters, z ′ and NB921. These are convolved with the CCD sensitivities, instrument and atmospheric transmission. The lower panel shows the SED of Lyman break galaxies at z ∼ 6.1, 6.57, and 6.8. i ′ -dropout galaxies. Open circles show the galaxies those are not detected in NB921-band (< 2σ). Open squares show the galaxies whose N B921 − z ′ colors are larger than 3σ(N B921 − z ′ ) but smaller than 1.5. . Color-color diagram between i ′ − z ′ and N B921 − z ′ . The meanings of lines are the same as these given in Fig.2. All the i ′ -dropouts are plotted. Filled circles with the error bars show galaxies detected in both i ′ and NB921 (> 2σ), triangles show galaxies those are not detected in i ′ -band (< 2σ), and inverted triangles show those are not detected both in i ′ and NB921 (< 2σ). Lines show the the expected number of galaxies with z > 6.6 for different assumptions. Thin dotted (solid) lines show the case for the SED with (without) Lyα emission together with the UV luminosity function at z ∼ 4.0 (Ouchi et al. 2004). Thick dotted (solid) lines show the case for the SED with (without) Lyα emission together with the UV luminosity function at z ∼ 6.0 (Bouwense et al. 2004).