The Subaru–XMM-Newton Deep Survey (SXDS). VIII. Multi-wavelength identification, optical/NIR spectroscopic properties, and photometric redshifts of X-ray sources^†

Abstract

We report on the multi-wavelength identification of the X-ray sources found in the Subaru–XMM-Newton Deep Survey (SXDS) using deep imaging data covering the wavelength range between the far-UV and mid-IR (MIR). We select a primary counterpart of each X-ray source by applying the likelihood ratio method to R-band, 3.6 μm, near-UV, and 24 μm source catalogs as well as matching catalogs of active galactic nucleus (AGN) candidates selected in 1.4 GHz radio and i^′-band variability surveys. Once candidates for Galactic stars, ultra-luminous X-ray sources in a nearby galaxy, and clusters of galaxies are removed there are 896 AGN candidates in the sample. We conduct spectroscopic observations of the primary counterparts with multi-object spectrographs in the optical and NIR; 65% of the X-ray AGN candidates are spectroscopically identified. For the remaining X-ray AGN candidates, we evaluate their photometric redshift with photometric data in 15 bands. Utilizing the multi-wavelength photometric data of the large sample of X-ray-selected AGNs, we evaluate the stellar masses, M_*, of the host galaxies of the narrow-line AGNs. The distribution of the stellar mass is remarkably constant from z = 0.1 to 4.0. The relation between M_* and 2–10 keV luminosity can be explained with strong cosmological evolution of the relationship between the black hole mass and M_*. We also evaluate the scatter of the UV–MIR spectral energy distribution (SED) of the X-ray AGNs as a function of X-ray luminosity and absorption by the nucleus. The scatter is compared with galaxies which have redshift and stellar mass distribution matched with the X-ray AGN. The UV–NIR (near-IR) SEDs of obscured X-ray AGNs are similar to those of the galaxies in the matched sample. In the NIR–MIR range, the median SEDs of X-ray AGNs are redder, but the scatter of the SEDs of the X-ray AGN broadly overlaps that of the galaxies in the matched sample.

1 Introduction

After the discovery of super-massive black holes (SMBHs) at the center of every massive galaxy in the local Universe (e.g., Kormendy & Richstone 1995), the issue of how these SMBHs formed and evolved over cosmic history has become one of the major unanswered questions in observational cosmology. Luminosities of active galactic nuclei (AGNs) reflect the mass accretion rates of their SMBHs, therefore the luminosity function of AGNs and their cosmological evolution reflect the growth history of SMBHs through accretion. Hard X-ray selection is an effective selection method for AGNs, it can detect obscured AGNs without serious bias, except for Compton-thick AGNs, and can efficiently sample AGNs without contamination by a large number of star-forming galaxies. The hard X-ray luminosity function of AGNs and its evolution as a function of redshift have been examined by combining X-ray-selected AGN samples with various survey depths and areas (Ueda et al. 2003, 2014; Barger et al. 2005; La Franca et al. 2005; Hasinger et al. 2005; Silverman et al. 2008; Aird et al. 2008, 2010; Yencho et al. 2009; Ebrero et al. 2009). Based on the cosmological evolution of the hard X-ray luminosity function, the average accretion growth curves of SMBHs are inferred (Marconi et al. 2004; Ueda et al. 2014).

In order to construct a large sample of X-ray-selected AGNs, we have conducted deep multi-wavelength imaging observations covering the far-UV (FUV) to mid-IR (MIR) and deep optical and near-IR (NIR) multi-object spectroscopic observations of the X-ray sources found in the Subaru–XMM-Newton Deep Survey (SXDS: Sekiguchi et al. 2005). The X-ray images obtained with XMM-Newton cover a 1.3 deg² area down to 1 × 10⁻¹⁵ erg s⁻¹ cm⁻² in the 0.5–2 keV band and 3 × 10⁻¹⁵ erg s⁻¹ cm⁻² in the 2–10 keV band (Ueda et al. 2008). The log N − log S relation of the X-ray sources has a knee at a hard X-ray flux of 1 × 10⁻¹⁴ erg s⁻¹ cm⁻² and the contribution to the cosmic X-ray background is maximum in this flux range (Cowie et al. 2002). Several hard X-ray surveys covering a ∼ 1 deg² area in this flux range have been conducted (Fiore et al. 2003; Kim et al. 2004; Yang et al. 2004; Chiappetti et al. 2005; Murray et al. 2005; Laird et al. 2009; Elvis et al. 2009; Cappelluti et al. 2009; Civano et al. 2012). Among the X-ray surveys with similar depth and area, SXDS is notable in having deep multi-wavelength data in the wavelength range from far-UV to mid-IR.

In this paper, we describe the X-ray sample in section 2, and the multi-wavelength imaging data in section 3. The 3 σ detection limits of the imaging data in the i^′-, K-, and 3.6 μm bands reach 27.4 mag, 25.3 mag, and 24.5 mag in the AB magnitude system, respectively. They are expected to be deep enough to detect counterparts of almost all of the X-ray sources at the flux limit: 90% of the X-ray sources found in a wide-area Chandra survey at a flux limit of 1 × 10⁻¹⁵ erg s⁻¹ cm⁻² in the 0.5–2.0 keV band have a counterpart brighter than 26.4 mag (i^′-band), 23.6 mag (K-band), and 22.8 mag (3.6 μm-band: Civano et al. 2012). The positional uncertainties of the XMM-Newton X-ray sources are larger than those of Chandra sources, and they are not small enough to securely identify the faintest optical counterparts (Loaring et al. 2005; Chiappetti et al. 2005). In order to quantitatively select the counterparts, we apply the likelihood ratio method (de Ruiter 1977; Wolstencroft et al. 1986; Sutherland & Saunders 1992). We apply the method not only to the deep optical source catalog but also to the deep 3.6 μm, near-UV (NUV), and 24 μm source catalogs to improve the reliability of the counterpart selection. The identification procedure with the multi-wavelength source catalogs is described in section 4.

In order to reveal the nature of the X-ray sources, we have been conducting a large campaign of spectroscopic observations in the NIR as well as in the optical wavelength range. Because obscured AGNs at z = 1–2 can be significant contributors to the X-ray source population, NIR spectroscopy is effective in revealing their nature by detecting strong rest-frame optical emission lines. The spectroscopic follow-up observations are described in section 5. Even with the deep optical and NIR spectroscopic observations, a significant fraction of the X-ray sources are fainter than the spectroscopic limit. In order to overcome the limitation of the spectroscopic identification, photometric redshift estimation with a deep imaging dataset covering a wide wavelength range is important (Zheng et al. 2004). The multi-wavelength photometry and photometric redshift evaluation are described in section 6. In section 7, we discuss the X-ray properties and host galaxy stellar mass of the X-ray-selected AGNs. Utilizing the multi-wavelength photometry, we examine the scatter of the spectral energy distributions (SEDs) of the X-ray AGNs, and compare the scatter with that of non-X-ray galaxies with similar redshifts and stellar masses. Throughout the paper, the cosmological parameters H₀ = 70 km s⁻¹ Mpc⁻¹, Ω_M = 0.3, and Ω_Λ = 0.7 are used. The magnitudes are based on the AB magnitude system (Oke, 1974).

Using the large sample of X-ray selected AGNs, Hiroi et al. (2012) derive the comoving space density and obscured fraction of AGNs at 3 < z < 5. Furthermore, Ueda et al. (2014) examine the cosmological evolution of the X-ray AGN luminosity function including our sample. The black hole mass and Eddington ratio distribution functions of the X-ray-selected AGNs at z ∼ 1.4 are discussed in Nobuta et al. (2012), using the optical and NIR spectra of the X-ray-selected AGNs.

2 X-ray sample definition

The SXDS field is centered on 02^h18^m and − 05°00′. The field is selected to be observable with 8–10 m-class optical telescopes in both of the northern and the southern hemispheres. The field was observed with XMM-Newton with one central 30′-diameter field with a 100 ks exposure and six flanking fields with 50 ks exposure time each (Ueda et al. 2008). These seven pointings cover a 1.3 deg² field. Details of the observations and data reduction procedures are described in Ueda et al. (2008).

In total, 866 and 645 sources are detected above a likelihood of 7 in the 0.5–2 keV (soft) and 2–10 keV (hard) bands, respectively, through source detection utilizing a maximum likelihood fit method. The spatial distribution of the sources is shown in figure 1 (small dots). A likelihood (L) of 7 corresponds to the probability of being random background noise, P_random of 9.1 × 10⁻⁴ (L = −ln P_random). The 0.5–2 keV and 2–10 keV band X-ray flux distributions of the X-ray sources are shown in figure 2. The 0.5–2 keV (2–10 keV) flux is calculated by multiplying the 0.5–2 keV (2–10 keV) count rate with 0.167 (0.838) × 10⁻¹⁴ [erg s⁻¹ cm⁻²]/[cts ks⁻¹] (Ueda et al. 2008). These are the conversion factors for a power-law spectrum with Γ = 1.8, which is the typical apparent X-ray spectral slope of a non-obscured AGN.

Fig. 1.

Areas covered by u-band, optical, NIR, and MIR imaging observations are shown overlaid on a map of the detected X-ray sources from the seven XMM-Newton images (Ueda et al. 2008). The five thin solid rectangles indicate the area covered by the deep Suprime-Cam imaging observations. The thin dashed line shows the area covered by the deep u-band observations with the Mosaic-II camera. The thick solid square indicates the region covered by the deep J-, H-, K-band imaging observations of UKIDSS/UDS. The gray region indicates the region covered by SpUDS IRAC observations. The lower (upper) light gray region is only covered by 3.6 μm and 5.8 μm (4.5 μm and 8.0 μm) observations. The thick dashed line represents the region covered by the SpUDS MIPS observations. The black dots show the positions of the X-ray sources with likelihood larger than 7 either in the soft or hard band from Ueda et al. (2008). The blue squares indicate X-ray sources included in the current sample which are covered by the Suprime-Cam observations and not affected by halos of very bright stars. Red filled symbols indicate spectroscopically identified sources. (Color online)

Fig. 2.

Left: 0.5–2.0 keV band flux distribution of the X-ray sources in the SXDS. The solid histogram represents the number of the entire soft-band sample of Ueda et al. (2008), and the dotted histogram shows the number of the sample considered in this paper. Right: 2.0–10.0 keV band flux distribution with details as for left panel.

In this paper, only X-ray sources in the region covered with the deep optical imaging data described in the next section are considered. The region is shown as five rectangles with thin solid lines. Some of the X-ray sources whose optical images are affected by halo or electron leakage from very bright objects are removed as well. A total of 781 (584) soft-(hard-)band X-ray sources with likelihood above 7 are covered. Hereafter, these sources are referred to as the soft-(hard-)band sample. Considering sources common to both samples, there are 945 unique X-ray sources; we refer to the combined sample as the total sample. The flux distributions of the samples are shown with dotted histograms in figure 2. The corresponding survey area for the soft-(hard-)band sample is shown in the left (right) panel of figure 3. The solid line represents the survey area of the entire sample of Ueda et al. (2008), and the dotted line shows the survey area of the sample considered in this paper.

Fig. 3.

Left: Survey area as a function of 0.5–2.0 keV X-ray flux for the soft-band sample. The solid line represents the survey area of the entire soft-band sample of Ueda et al. (2008), and the dotted line shows the survey area of the sample considered in this paper. The flux is determined by multiplying the soft-band count rate by 0.167 × 10⁻¹¹ (erg s⁻¹ cm⁻²) / (cts ks⁻¹) assuming a power-law spectrum with Γ = 1.8. Right: Survey area for the hard-band sample. The flux is determined by multiplying the hard-band countrate by 0.838 × 10⁻¹¹ (erg s⁻¹ cm⁻²)/(cts ks⁻¹) assuming the same power law.

3 Multi-wavelength imaging dataset

3.1 Deep optical imaging data

Deep optical imaging observation in the field was conducted with the optical wide-field 34′ × 27′ mosaic CCD camera, Suprime-Cam, attached to the prime focus of the 8.2 m Subaru telescope. The observation was conducted as a Subaru observatory project from 2002 September to 2004 January. Imaging data in five overlapping fields of view (FoVs) in the B, V, R, i^′, and z′ bands were obtained. The thin solid rectangles in figure 1 indicate the area covered. The total exposure times for each field range were 5.5–5.8 hr (B), 4.9–5.3 hr (V), 3.9–4.1 hr (R), 5.2–10.8 hr (i^′), 3.1–5.2 hr (z′). For details of the deep imaging data, see Furusawa et al. (2008). In these images the full width at half maximum (FWHM) of stellar objects is 0|${^{\prime\prime}_{.}}$|7–0|${^{\prime\prime}_{.}}$|9. We apply a Gaussian convolution to make the resulting point spread functions (PSFs) in the five bands constant for each region. We use the photometric zero point provided in Furusawa et al. (2008). The typical 2 σ detection limits in 2″-diameter apertures are determined by measuring the standard deviation of fluxes within a fixed aperture at positions where no object is detected. The limits are 28.7 mag (B), 28.1 mag (V), 28.1 mag (R), 28.0 mag (i^′), and 26.9 mag (z′). Correcting for the aperture loss for point sources, the typical magnitude limits are 28.1 mag (B), 27.6 mag (V), 27.6 mag (R), 27.6 mag (i^′), and 26.6 mag (z′) for the total magnitudes of point sources. The limiting magnitudes in the aperture and for the total magnitudes are summarized in table 1. For other bands, we derive the limiting magnitudes for aperture and total magnitudes in the same way.

Table 1.

Multiwavelength dataset.*

Obs./Inst.	Band	Detection limits		Correction	Note
		Aperture	Total
		(mag)	(mag)	(mag)
(1)	(2)	(3)	(4)	(5)	(6)
GALEX	FUV	26.64	25.52 (4 σ,0.36)	0.18	Deep Imaging Survey
GALEX	NUV	26.18	25.15 (3 σ,0.59)	0.17	Deep Imaging Survey
Subaru/Suprime	U	27.39	26.46 (3 σ,0.49)	0.11	South FoV only
CTIO/Mosaic-II	U	26.86	26.29 (3 σ,0.13)	0.11
CFHT/Mega-cam	U	26.65	26.11 (3 σ,0.10)	0.11	CFHTLS wide
INT/WFCam	U			0.11	Only used for bright objects
Subaru/Suprime	B	28.50–28.83	27.91–28.24 (3 σ,0.15)	0.09
INT/WFCam	G				Only used for bright objects
Subaru/Suprime	V	28.09–28.22	27.55–27.68 (3 σ,0.10)	0.07
Subaru/Suprime	R	27.95–28.18	27.38–27.61(3 σ,0.13)	0.07
INT/WFCam	R				Only used for bright objects.
Subaru/Suprime	i   ′	27.91–28.10	27.30–27.49 (3 σ,0.17)	0.06
INT/WFCam	i   ′				Only used for bright objects.
Subaru/Suprime	z′	26.83–27.08	26.25–26.50 (3 σ,0.14)	0.04
INT/WFCam	z′				Only used for bright objects.
UKIRT/WFCAM	J	26.06	25.40(3 σ,0.22)	0.02	UKIDSS/UDS-DR8
UKIRT/WFCAM	H	25.48	24.82(3 σ,0.22)	0.02	UKIDSS/UDS-DR8
UKIRT/WFCAM	K	25.96	25.30(3 σ,0.22)	0.01	UKIDSS/UDS-DR8
Spitzer/IRAC	3.6 μm	23.54	22.77(3 σ,0.33)	0.01	SWIRE
Spitzer/IRAC	3.6 μm	25.26	24.49(3 σ,0.33)	0.01	SpUDS
Spitzer/IRAC	4.5 μm	22.87	22.07(3 σ,0.36)	0.00	SWIRE
Spitzer/IRAC	4.5 μm	24.83	24.03(3 σ,0.36)	0.00	SpUDS
Spitzer/IRAC	5.8 μm	21.27	19.97(4 σ,0.54)	0.00	SWIRE
Spitzer/IRAC	5.8 μm	22.99	21.69(4 σ,0.54)	0.00	SpUDS
Spitzer/IRAC	8.0 μm	21.17	19.91(4 σ,0.66)	0.00	SWIRE
Spitzer/IRAC	8.0 μm	23.19	21.77(4 σ,0.66)	0.00	SpUDS
Spitzer/MIPS	24 μm	20.21	18.22(4 σ,1.23)	0.00	SWIRE
Spitzer/MIPS	24 μm	22.06	20.07(4 σ,1.23)	0.00	SpUDS

*Column (3) is the 2 σ aperture magnitude limit measured in a fixed aperture. The diameters of the apertures are 7|${^{\prime\prime}_{.}}$|6 for GALEX FUV and NUV images, 2|${^{\prime\prime}_{.}}$|0 for U-band, optical, and NIR images, 3|${^{\prime\prime}_{.}}$|8 for IRAC images, and 6|${^{\prime\prime}_{.}}$|0 for MIPS images. Column (4) gives the 3 σ limits for the total magnitudes, for GALEX FUV and for Sptitzer MIPS data the 4 σ magnitude limits. They are determined from the aperture magnitude limit and the aperture correction for stellar objects. These upper limits are used in the photometric redshift determination and in table 8. Applied aperture corrections are shown on the right-hand side in parentheses. Column (5) is the correction for Galactic extinction.

We use the Suprime-Cam R-band astrometry as the astrometric reference frame in this paper. Astrometric calibration is carried out in the R-band images using the 2MASS point source catalog. The rms residual of the fit is typically 0|${^{\prime\prime}_{.}}$|03. The multi-wavelength data described below are aligned to the R-band image using commonly detected stellar objects.

3.2 U-band imaging data

The entire survey region is covered by the Canada France Hawaii Telescope Legacy Survey (CFHTLS).¹ We use the U-band data from the CFHTLS archive. The typical 3.0 σ total magnitude limit for a point source is 26.11 mag. Part of the survey region, the southern FoV out of the five Suprime-Cam FoVs, is covered by a deeper U-band imaging observation with Suprime-Cam with 3.7 hours integration. Details of these data are described in Yoshida et al. (2008). The 3.0 σ limiting total magnitude of a point source is 26.46 mag. The central 37′ × 37′ area of the survey region is also covered by deep imaging data obtained with the Mosaic-II camera attached to the CTIO Blanco 4 m telescope (Díaz Tello et al. 2013; PI. N. Padilla). In this observation, the Sloan Digital Sky Survey (SDSS) U-band filter is used. The region covered is shown with a thin dashed line in figure 1. The total integration time is 10 hr and the 3.0 σ limiting total magnitude of a point source is 26.29 mag. Because the filter systems used in the three U-band images are different, we adjust the Suprime-Cam and Mosaic-II magnitudes to the photometric system used in the MegaCam. Using stars in the overlapping regions, we fit the magnitude difference between the Suprime-Cam (Mosaic-II) and MegaCam photometry with a linear function of the U − B color, and apply the function to the Suprime-Cam (Mosaic-II) photometry. We check the photometric zero-point of the MegaCam data using the B and R photometry from the Suprime-Cam data and the U − B and B − R color–color diagram of stellar objects. The distribution of the stellar objects on the diagram is consistent with the colors derived with the SEDs of stars from Gunn and Stryker (1983).

3.3 Shallow optical and U-band imaging data

For bright objects which are saturated in the deep optical images, their optical magnitudes are determined with shallow optical images taken with the Wide Field Camera attached to the Isaac Newton Telescope. U-, G-, R-, i-, and z-bands data are available. The G, R, i, and z optical magnitudes are adjusted to the Suprime-Cam B, R, i^′, and z′ system with magnitude offset and its color dependence derived with stars in the overlapping region.

3.4 NIR imaging data

A large part of the SXDS region is covered by the Ultra Deep Survey of the UKIRT Infrared Deep Sky Survey (UKIDSS UDS: Lawrence et al. 2009). The coverage of this survey is shown with the thick solid rectangle in figure 1. Archival image data from data release 8 (UKIDSS DR8) is used in this paper.² The stellar image size is 0|${^{\prime\prime}_{.}}$|7. We use the photometric zero-point provided in the data release. The 3 σ detection limit of the image for a point source is measured to be 25.40 mag, 24.82 mag, and 25.30 mag for the J-, H-, and K-bands, respectively, with 2″-diameter aperture photometry and aperture correction for point sources. For X-ray sources outside of the UKIDSS UDS region, we do not have NIR photometric data.

3.5 MIR imaging data

3.5.1 Spitzer IRAC data

Almost the entire FoV of the SXDS is covered by the Spitzer observation with the Infra-Red Array Camera (IRAC) instrument obtained in the course of the Spitzer Wide-area Infra-Red Extragalactic (SWIRE) legacy survey project (Lonsdale et al. 2003, 2004). Seven sources in the total sample are outside of the field covered. We use publicly available mosaiced images of IRAC (DR2 2005-06-03).³

The FWHM of the PSFs in the IRAC images are determined to be 1|${^{\prime\prime}_{.}}$|9, 2|${^{\prime\prime}_{.}}$|0, 1|${^{\prime\prime}_{.}}$|9, and 2|${^{\prime\prime}_{.}}$|2 by combining images of > 10 stars in the FoV, and the corresponding aperture corrections for a 3|${^{\prime\prime}_{.}}$|8-diameter aperture are 0.33 mag, 0.36 mag, 0.54 mag, and 0.66 mag for the 3.6 μm, 4.5 μm, 5.8 μm, and 8.0 μm bands, respectively.³ We determine the photometric zero points of the mosaiced IRAC images using K − 3.6 μm, K − 4.5 μm, K − 5.8 μm, and K − 8.0 μm colors of bright blue stars following Lacy et al. (2005); values of 0.04 mag, 0.02 mag, 0.00 mag, and 0.00 mag are used for stars with J − K colors bluer than 0.3 mag based on a theoretical stellar SED model (table 5 of Lacy et al. 2005). Near-infrared photometry of stellar objects in the SXDS field is taken from the Two Micron All-Sky Survey Point Source Catalog (2MASS PSC: Skrutskie et al. 2006). K-band magnitudes in the 2MASS PSC are determined with a profile fitting method. We select blue stars with J − K < 0.3 mag and examine their average K − 3.6 μm, K − 4.5 μm, K − 5.8 μm, and K − 8.0 μm colors. Photometry in the IRAC bands are determined with a 3|${^{\prime\prime}_{.}}$|8-diameter aperture with an aperture correction derived from the PSF. We determine the photometric zero points to match the expected colors from the theoretical stellar SED model.

The 3 σ detection limit of the image for point sources is measured to be 22.77 mag, 22.07 mag, 19.97 mag, and 19.91 mag for the 3.6 μm, 4.5 μm, 5.8 μm, and 8.0 μm bands, respectively, with a 3|${^{\prime\prime}_{.}}$|8 aperture photometry diameter and assumption for the aperture correction for point sources.

Comparing the coordinates of objects detected in both the R-band and 3.6 μm-band, we found a 0|${^{\prime\prime}_{.}}$|36 offset in the right ascension direction between the deep Suprime-Cam R-band and the IRAC 3.6 μm-band images. We shift the IRAC image to match the coordinates of the Suprime-Cam image. After the correction, the difference between the R-band and 3.6 μm-band coordinates is 0|${^{\prime\prime}_{.}}$|4 rms, which is thought to be dominated by the positional uncertainty of the 3.6 μm source.

A significant part of the SXDS region is covered by one of the Spitzer legacy programs, Spitzer UKIDSS Ultra Deep Survey (SpUDS). IRAC observations cover the region shown in gray in figure 1. The photometric zero point of the SpUDS data is adjusted to SWIRE photometry with overlapping objects. The typical 2 σ detection limits for a 3|${^{\prime\prime}_{.}}$|8 aperture are 25.26 mag, 24.83 mag, 22.99 mag, and 23.19 mag, and the detection limits for the point source are estimated to be 24.49 mag (3.0σ), 24.03 mag (3.0 σ), 21.69 mag (3.0σ), and 21.77 mag (4.0 σ) for 3.6 μm, 4.5 μm, 5.8 μm, and 8.0 μm bands, respectively.

3.5.2 Spitzer MIPS data

The Spitzer observation with the Multiband Imaging Photometer (MIPS) instrument in the SWIRE legacy survey (Lonsdale et al. 2003, 2004) also covers almost the entire area of the SXDS. We use publicly available mosaiced images of MIPS data in the 24 μm band (DR5 2006-12-23).

The FWHM of the PSF of the MIPS image is 5|${^{\prime\prime}_{.}}$|9, and aperture correction for a 6|${^{\prime\prime}_{.}}$|0-diameter aperture is 1.23 mag.⁴ We use the photometric zero point provided in the data release. The 2 σ detection limit of the data in the aperture is 20.21 mag, and the 4 σ detection limit for the total magnitude of stellar objects is 18.22 mag.

A large part of the SXDS region is also covered by the MIPS observation of SpUDS. The coverage is shown with the thick dashed line in figure 1. The 2 σ detection limit in a 6|${^{\prime\prime}_{.}}$|0-diameter aperture is 22.06 mag, and the 4 σ detection limit in total magnitude of stellar objects is 20.07 mag.

3.6 UV imaging data

The Deep Imaging Survey (DIS) of the Galaxy Evolution Explorer (GALEX: Morrissey et al. 2007) covers almost the entire area of the SXDS in the FUV (effective wavelength of 1516 Å) and NUV (effective wavelength of 2267 Å) bands. We use the reduced data obtained from the GALEX archive.⁵ The XMMLSS_03 (29189 s, 30289 s), XMMLSS_04 (27306 s, 28561 s), XMMLSS_07 (27440 s, 27520 s), and XMMLSS_09 (30606 s, 32354 s) tiles overlap with the SXDS. The numbers in parentheses show the total exposure times in FUV and NUV bands, respectively, for each tile.

The FWHM of the reduced images are 5|${^{\prime\prime}_{.}}$|0 and 5|${^{\prime\prime}_{.}}$|1 in the FUV and NUV images, respectively. We use the photometric zero point provided in Morrissey et al. (2007): 18.82 mag and 20.08 mag for FUV and NUV bands. The aperture corrections for point sources are 0.36 mag and 0.59 mag for FUV and NUV bands for a 7|${^{\prime\prime}_{.}}$|6-diameter aperture (Morrissey et al. 2007). The typical point source limiting magnitudes estimated with the 7|${^{\prime\prime}_{.}}$|6 aperture photometry reach down to 25.52 mag (4 σ) and 25.15 mag (3 σ) in the FUV and NUV bands, respectively.

4 Multi-wavelength identification of the X-ray sources

4.1 Identification with R-band objects and its reliability

For each X-ray source the statistical error in the position estimated through the maximum likelihood fit, σ_Xi, is given in Ueda et al. (2008). When we match the X-ray and R-band sources, we use an uncertainty radius, σ_X:Ri, which is determined by the square of the σ_Xi and 1|${^{\prime\prime}_{.}}$|1 added in quadrature. The latter term represents any positional uncertainties other than σ_Xi, i.e., any systematic uncertainty in the positions of the R-band sources and X-ray sources. The term is estimated through the scatter of positional differences of X-ray sources whose σ_Xi is less than 0|${^{\prime\prime}_{.}}$|5 and their candidate R-band counterparts.

Sample images of three X-ray sources in the R-band are shown in the second column of figure 4. Almost all of the X-ray sources have at least an R-band source in the 3 × σ_X:Ri radius above the detection limit of the R-band images. Some X-ray sources have multiple R-band sources within their error radii. The upper left panel of figure 5 shows the radial distribution of R-band sources down to 27.0 mag from the X-ray positions. The dotted line shows the average number density of sources outside the X-ray uncertainty area. The excess sources above the line statistically represent the counterparts of the X-ray sources. The average number density corresponds to the expected number of contaminating sources. If we consider the region within 2 × σ_X:Ri down to R = 27.0 mag, the ratio between the number of excess and contaminating sources is 0.30. It is thus important to select the most probable counterpart for each X-ray source.

Fig. 4.

Sample identification images. From left to right, GALEX NUV, Suprime-Cam R-band, Spitzer 3.6 μm-band, and Spitzer 24 μm-band images with 15″ × 15″ FoV. Top row for SXDS0061, whose primary candidates in the four bands match, middle and bottom rows for SXDS0285 and SXDS0524, respectively, whose primary counterpart is based on the Spitzer 3.6 μm identification. Green circles indicate the X-ray source position; the inner and outer circles have radii of 1.0 × σ_Xi and 3.0 × σ_Xi. The position of the primary counterpart is indicated with the large open square. The small open square indicates the object selected in an AGN variability search. (Color online)

Fig. 5.

Radial distribution of sources from the centers of the X-ray sources. Radial distance is normalized by the uncertainty radius, which is an X-ray positional uncertainty convolved with a positional uncertainty in each band. The number per unit area is divided by the number of X-ray sources covered in each dataset. Dotted lines indicate the average source density outside of the X-ray uncertainty area determined at 3.5–4.0 σ_X:band. Upper left: R-band sources down to R = 27.0 mag. Upper middle: 3.6 μm sources down to 22.7 mag determined with SWIRE coverage. Upper right: 3.6 μm sources down to 24.4 mag determined with SpUDS coverage. Lower left: NUV source down to 25.1 mag. Lower middle: 24 μm sources down to 18.1 mag determined with SWIRE coverage. Lower right: 24 μm sources down to 20.0 mag determined with SpUDS coverage.

At first, we identify the X-ray sources with sources detected in the R-band. We construct a catalog of R-band sources by applying Sextractor (Bertin & Arnouts 1996) to the deep Suprime-Cam R-band images. In order to quantitatively select the most probable identification, we utilize the likelihood ratio defined as the probability of an X-ray source i with X-ray flux f_Xi having a counterpart j with magnitude m_j and distance r_ij divided by the probability of it being a chance match (de Ruiter 1977; Wolstencroft et al. 1986; Sutherland & Saunders 1992). The likelihood ratio is calculated as

\begin{equation} L_{ij} \equiv \frac{Q(<m_{j}, f_{{\rm X}i}) \exp (-r_{ij}^{2}/2\sigma _{{\rm X:R}i}^{2})}{2\pi \sigma _{{\rm X:R}i}^{2} N(<m_{j})}, \end{equation}

(1)

assuming the spatial distribution of the counterparts in the X-ray error circles follows a Gaussian distribution. N(<m_j) is a cumulative number density of optical objects brighter than m_j. Q(<m_j, f_Xi) is the probability that the i-th X-ray source has its counterpart brighter than m_j. If the probability of being a counterpart for the X-ray source exceeds the probability of being a chance match, the likelihood ratio is greater than 1. The number density of optical objects is calculated from the Suprime-Cam R-band image directly without correction for detection incompleteness. We determine Q(<m_j, f_Xi) by examining the distribution of log f_X/f_R for the X-ray sources, and assume that the distribution is constant in the SXDS X-ray flux range. We consider R-band sources within 5.0 × σ_X:Ri of each X-ray source.

First, we calculate L_ij with Q(<m_j, f_Xi) = 1.0 and select the R-band object with the highest likelihood ratio for each X-ray source as the initial guess for the R primary candidate. Based on the R-band magnitudes of the identified objects, we determine Q(<m_j, f_Xi). The log f_Xi/f_Rj value is calculated with log f_Xi/f_Rj = log f_Xi + 5.53 + 0.4Rj in the AB magnitude system. The half-maximum wavelength range of the R-band filter response (5950 Å–7085 Å) of the Suprime-Cam is used in the calculation. We use the 0.5–10 keV flux, which is derived by summing the flux in the 0.5–2 keV and 2–10 keV bands, as the X-ray flux f_Xi. We call the R-band object with the highest likelihood ratio for a given X-ray source the R primary candidate for the source.

The reliability of the identification is evaluated as the probability that an R-band object j is the counterpart of an X-ray source i. The reliability is given by

\begin{equation} R_{ij} = \frac{L_{ij}}{\sum _{j} L_{ij} + [1-Q(<\!m_{\rm limit}, f_{{\rm X}i})]} \end{equation}

(2)

(Sutherland & Saunders 1992). Q(<m_limit, f_Xi) represents the probability of finding an R-band counterpart above the detection limit m_limit. We use a value of Q(<m_limit, f_Xi) of 0.8 for all X-ray sources for simplicity. The cumulative distribution of the reliability of the R primary candidate is shown in figure 6 with thick and thin black solid lines for the soft- and hard-band samples, respectively. A significant fraction of the X-ray sources have low reliability; 30% of the X-ray sources have primary optical counterpart with reliability smaller than 0.6. X-ray sources that have the R primary candidate with low reliability have a few R-band objects with similar R-band magnitudes and distances from the X-ray position. These are the problematic cases and the fraction of such cases in the total sample is similar to those in XMM-Newton survey fields with similar depth (Brusa et al. 2010).

Fig. 6.

Distribution of reliability R_ij for the primary candidate identified in each band. Black solid curves are for the R-band identification. Red dotted lines are for the 3.6 μm identification. Blue dot-dashed lines are for 24 μm identification. Green dashed lines are for NUV identification. The thick (thin) lines are for the hard-band (soft-band) sample. (Color online)

In order to quantitatively examine the reliability of the optical identification process, we conduct Monte Carlo simulations of the optical identification procedure. We use the large catalog of the optical objects in the fields made with the deep optical imaging. We regard each optical object as an X-ray source and add pseudo X-ray flux information and simulate the X-ray coordinates by adding positional uncertainty to the optical coordinates. We use 1 × 10⁻¹⁴ erg s⁻¹ cm⁻² in the total band for the X-ray flux. We assume that the positional error follows a Gaussian distribution. We simulated a positional error σ_X:Ri of 0|${^{\prime\prime}_{.}}$|0 to 3|${^{\prime\prime}_{.}}$|0 with a 0|${^{\prime\prime}_{.}}$|2 step. We apply the same optical identification process described above to the pseudo-X-ray source catalog. The identification process recovers more than 90% of the counterparts for the pseudo-X-ray sources with σ_X:Ri < 0|${^{\prime\prime}_{.}}$|8, and the fraction decreases down to 40% for the pseudo-X-ray sources with σ_X:Ri = 2|${^{\prime\prime}_{.}}$|0 and counterparts of R = 26.0 mag.

It is important to evaluate the fraction of correct identifications among the R-band primary candidates. For pseudo-X-ray sources with σ_X:Ri < 0|${^{\prime\prime}_{.}}$|8 and primary candidate of R < 25.5 mag, 95% of the primary candidates are the correct identification. On the other hand, only 50% of the primary candidates with R ∼ 25.0 mag are correct for pseudo-X-ray sources with σ_X:Ri ∼ 2|${^{\prime\prime}_{.}}$|0–3|${^{\prime\prime}_{.}}$|0. The results also show that even among the optical identifications with a bright optical counterpart, R ∼ 20 mag, there can be a non-negligible fraction of incorrect identifications.

Based on the simulations, we can estimate the probability of mis-identification for each X-ray source with σ_X:Ri and the R-band magnitude of the primary candidate. Summing the probabilities of mis-identification for all of the X-ray sources, we calculate the expected number of mis-identifications to be 102 among the 945 sources in the total sample. The identifications only with R-band information are clearly affected by mis-identification.

4.2 Identification with 3.6 μm objects and its reliability

Most of the expected mis-identifications are thought to be associated with X-ray sources with large σ_Xi and faint optical counterparts. Because optically faint X-ray sources have on average a red optical–MIR color (Yan et al. 2003; Koekemoer et al. 2004; Mignoli et al. 2004), identification in the MIR-band can be more reliable (Brusa et al. 2010; Civano et al. 2012). The upper middle and upper right panels of figure 5 show the radial distribution of 3.6 μm sources from the centers of the X-ray sources. For the uncertainty radius, |$\sigma _{{\rm X:3.6{\,\mu{\rm m}}}i}$|⁠, we use the root sum squared of the σ_Xi and 1|${^{\prime\prime}_{.}}$|1. The latter term is derived in the same way as for the R-band sources. If we take the ratio between the numbers of excess and contaminating sources within 2 × σ_X:3.6i, the ratios are 2.2 and 0.9 at the depth of the SWIRE and SpUDS coverages, respectively. The ratios are higher than the ratio in the R-band, and it is expected that more reliable identification can be made in the 3.6 μm-band. Feruglio et al. (2008) show that 3.6 μm identification provides a better counterpart assignment than optical-only identification for XMM-Newton X-ray sources. They checked the R and 3.6 μm identifications of XMM-Newton X-ray sources with the source positions measured with Chandra, and found 20% and 4% different identifications from Chandra for the R and 3.6 μm primary candidates, respectively.

In order to improve the reliability of the identification, we apply the 3.6 μm identification utilizing the SWIRE and SpUDS 3.6 μm images. First, we construct a 3.6 μm object catalog applying Sextractor (Bertin & Arnouts 1996) to the 3.6 μm images with detection criteria of 3 connected pixels above 2 σ of the background. We use MAG_AUTO as the 3.6 μm magnitude. Then, we identify the X-ray sources with the 3.6 μm objects in the same way for the R-band objects using L_ij and R_ij determined with the 3.6 μm object list. We consider 3.6 μm sources within |$5 \times \sigma _{{\rm X:3.6{\,\mu{\rm m}}}i}$| of each X-ray source. The number count of the 3.6 μm objects are derived from the source number density in the 3.6 μm source catalog. We select 3.6 μm primary candidates with |$Q(<m_{{\rm 3.6\,\mu{\rm m}}j}, f_{{\rm X}i})=1.0$|⁠, then we derive |$Q(<m_{{\rm 3.6\,\mu{\rm m}}j}, f_{{\rm X}i})$| with the 3.6 μm magnitudes of the primary candidates. Finally, we conduct the 3.6 μm identification with the derived |$Q( < m_{{\rm 3.6\,\mu{\rm m}}j}, f_{{\rm X}i})$|⁠.

The distribution of the reliability of the 3.6 μm primary candidate is shown in figure 6 with the red dotted lines. The fraction of high reliability identifications among the 3.6 μm identifications is higher than that among the R-band identifications: 80% of the sources have a primary candidate with R_ij > 0.8. We evaluate the expected number of mis-identifications among the 3.6 μm primary candidates using a Monte Carlo simulation as for the R-band identification. The mis-identification rate for all of the X-ray sources is 44 among the 945 X-ray sources. The fraction of expected mis-identifications is consistent with that reported in Feruglio et al. (2008). The expected number of mis-identifications is significantly reduced from that of the R-band identification.

The PSF size of the 3.6 μm-band IRAC image is larger than that of the R-band image, but as can be seen in figure 4, sufficiently small to pinpoint which is the more reliable counterpart among the multiple R-band sources that are in the X-ray uncertainty area. We use 3.6 μm identification to pinpoint the optical counterpart, and we refer the position of the selected optical counterpart as the coordinate of the counterpart.

Similar to the 3.6 μm identification, identifications in the K-band can be effective in selecting optically faint and red counterparts. Additionally, the PSF size of the UKIDSS UDS K-band image is significantly smaller than that of the IRAC 3.6 μm image, therefore, it is possible that true counterparts that are blended with contaminating sources in the 3.6 μm image can be resolved in the K-band image. Using the UKIDSS UDS K-band image, we check the difference of the K-band identifications from those in the R and 3.6 μm bands. The K-band identifications basically follow the R and 3.6 μm identifications; among 762 X-ray sources covered in the K-band image, 636 and 650 sources have a K-band primary candidate consistent with the primary candidate in the R and 3.6 μm identifications, respectively. If we remove K-band primary candidates which are located far outside the X-ray error circle, there are 11 K-band primary candidates that are different from both the R-band and 3.6 μm identifications. All of the 11 K-band primary candidates have reliability lower than 0.5, i.e., there are other K-band sources with similar likelihood, and the identifications are marginal. Considering the similarity to the R and 3.6 μm identifications and the limited coverage of the K-band image, we do not refer to the K-band identifications in the final selection of the primary counterparts.

4.3 Primary counterpart selected with multi-wavelength identification

Utilizing the multi-wavelength dataset of the SXDS, we can further improve the reliability of the identification. Luminous blue broad-line AGNs at low to intermediate redshifts are reliably identified with strong NUV emission. On the other hand, obscured narrow-line AGNs show strong 24 μm emission. In the bottom panels of figure 5, radial distributions of NUV and 24 μm sources are shown. The 1 σ uncertainty radii, σ_X:NUVi and |$\sigma _{{\rm X:24{\,\mu{\rm m}}}i}$|⁠, are calculated with root sum squares of σ_Xi and the additional term associated with the systematic X-ray source and NUV or 24 μm position uncertainty. We use 2|${^{\prime\prime}_{.}}$|3 and 1|${^{\prime\prime}_{.}}$|4 as the latter term for NUV and 24 μm identifications, respectively. In the same way as for the R and 3.6 μm-bands, if we consider within 2 × σ_X:NUVi or 2 × σ_X:24 μmi, the ratios between the number of excess and contaminating sources are 2.5, 22.0, and 8.3 for NUV down to 25.1 mag, 24 μm down to 18.1 mag (SWIRE depth), and 24 μm down to 20.0 mag (SpUDS depth), respectively. Thus, most of the associated 24 μm sources are expected to be counterparts of the X-ray sources. Some contaminations are expected for NUV sources; see the discussions on the distributions of reliability below. Therefore, we utilize NUV and 24 μm identifications as well as the R and 3.6 μm identifications to determine the primary counterpart of each X-ray source. We refer to a primary object selected in each identification as the primary candidate, and the best candidate in the combined identifications as the primary counterpart.

We perform NUV and 24 μm identifications in the same way as for the R and 3.6 μm identifications. First, we construct object catalogs from the NUV and 24 μm images with Sextractor. We use the NUV and 24 μm magnitude determined in 7|${^{\prime\prime}_{.}}$|6 and 6|${^{\prime\prime}_{.}}$|0 diameter apertures with aperture corrections of 0.59 mag and 1.23 mag determined with average PSF, respectively. The number densities of NUV and 24 μm sources are directly derived from the source list made by applying Sextractor to the SXDS dataset. For the NUV and 24 μm identifications, we do not consider |$Q(<m_{j}, f_{{\rm X}_i})$| and set |$Q(<m_{j}, f_{{\rm X}_i})=1.0$| for simplicity. We calculate R_ij in the same way for R and 3.6 μm identification. As shown in figure 4, the PSF sizes of the NUV and 24 μm-band images are significantly larger than for the R-band images. The PSF sizes are still sufficiently small to pinpoint one R-band counterpart in most of the cases. We apply a visual check for the R-band images of the NUV and 24 μm sources. As for the 3.6 μm identification, even if we use NUV or 24 μm identification, we refer to the position of the selected optical counterpart as the coordinate of the counterpart.

The distribution of the R_ij for NUV and 24 μm identifications are shown with green dashed and blue dot-dashed lines, respectively, in figure 6. The NUV identification is good for blue luminous broad-line AGNs, and higher reliability identification can be achieved than R identification for more than 40% of the X-ray sources. However, another 40% of the sources have multiple NUV sources with similar magnitude in the uncertainty area, giving a reliability less than 0.5. In the 24 μm band we achieve highly reliable identification, R_ij > 0.8, for 80% of the sources. For the rest of the X-ray sources, the 24 μm data is not deep enough to detect the counterparts of the X-ray sources and the fraction is saturated at 80%. The higher reliability of the NUV and 24 μm identifications can partly be caused by the blending effect due to the large PSF size in the two bands. The counterparts of the X-ray sources are bright in the NUV and 24 μm bands, and fainter NUV or 24 μm band sources that are contaminating and are located at the X-ray source position by chance can be hidden due to blending with the brighter true counterpart. Such a blending effect is discussed for 3.6 μm sources in Cardamone et al. (2008).

Additionally, AGN candidates are selected in radio and variability surveys in SXDS, and we utilise this information in the final identification. A large part (0.8 deg²) of the SXDS field is covered by the Very Large Array (VLA) radio survey observation at 1.4 GHz down to 100 μJy (Simpson et al. 2006, 2012). In total, 505 radio sources are detected, and 480 sources are within the X-ray coverage region. Among the 945 X-ray sources in the total sample, 789 of them are covered in the 100 μJy sample. If we consider 1 σ positional matching uncertainty, σ_X:radioi with the root sum of squares of σ_Xi and 1|${^{\prime\prime}_{.}}$|0, there are 62 radio sources associated with the X-ray sources within 4.0 × σ_X:radioi. Most of the broad-line AGNs detected in the radio survey are also detected in X-ray; among 25 broad-line AGNs detected in the VLA survey (Simpson et al. 2012), 21 of them are in the X-ray survey region, and 17 of them are associated with X-ray sources. Three radio broad-line AGNs without X-ray detection are fainter than i^′ = 23.5 mag, and are likely to be below the X-ray detection limit. The remaining radio broad-line AGNs without X-ray detection are located in a gap of the X-ray detectors, and the AGN can be missed due to bad sensitivity in the region.

Variable AGN candidates are selected with i^′-band Suprime-Cam imaging data (Morokuma et al. 2008a, 2008b). There are 936 variable objects detected down to the magnitude of the varying component of |$i^{\prime }_{\rm vari}=25.5$| in the overlapping region with the X-ray images (Morokuma et al. 2008b), and 193 of them are matched with X-ray sources.

Finally, the primary counterpart is determined utilizing all of the R, 3.6 μm, 24 μm, and NUV identifications. Radio and variability AGN candidates are also considered during the selection of the primary counterpart.

Among the 945 X-ray sources, 213 sources have a matched primary candidate considering all four identifications, and we select this object as the primary counterpart. 404 X-ray sources have only one object with a likelihood ratio larger than 1 in the four identifications within the X-ray uncertainty area, and we select the object as the primary counterpart. For the remaining X-ray sources, there are more than two objects with L_ij > 1 within the X-ray uncertainty area. We multiply the likelihood ratios of the four identifications for each source, and if there is an object with a likelihood ratio product larger than 1000, we select this object as the primary counterpart. There are 112 X-ray sources that meet the criteria. Of the remaining 183 X-ray sources, 90 X-ray sources have a primary candidate selected both in the R and 3.6 μm identifications. Among them, we select 77 of them as the primary counterpart. For 13 X-ray sources, there is another primary candidate selected at 24 μm or in the NUV, and the candidate is selected as the primary counterpart. The other 93 X-ray sources have different primary candidates in the R and 3.6 μm identifications. For such source, we determine the primary counterpart comparing the likelihood ratios in the four identifications. We select 52, 24, 20, and 6 primary candidates with the 3.6 μm, R, 24 μm, and NUV identifications, respectively.

The distributions of the reliability of the primary counterparts in the R, 3.6 μm, 24 μm, and NUV identifications are shown in figure 7. Almost all of the primary counterparts have high reliability in one of the four identifications. The list of the primary counterparts is shown in table 2. In the list, the X-ray identification number (column 1), X-ray RA and Dec position (columns 2 and 3), X-ray 1 σ uncertainty radius, σ_Xi (column 4), and likelihood in the 0.5–2 keV and 2–10 keV bands (columns 5 and 6) are shown. Columns 7 to 10 provide the information from the multi-wavelength identification process in the R (7), NUV (8), 3.6 μm (9), and 24 μm (10) bands. Flag 1 means the primary counterpart is the primary candidate in the band, flag −1 means the counterpart is the primary candidate but the likelihood is less than 1.0, and flag 0 means the counterpart is not the primary candidate in the band. Columns 11 and 12 show the matching with the 1.4 GHz radio source and variability AGN candidates. The RA and Dec of the primary counterpart is given in columns 13 and 14. The multi-wavelenth images of the X-ray sources are available online.⁶

Fig. 7.

Reliability R_ij determined in the 3.6 μm (top), NUV (middle), and 24 μm (bottom) band identification vs. R-band identification. Red filled circles represent objects where all four identifications are the same. Red open circles have matched identifications in the R and 3.6 μm bands. Blue filled squares and blue open squares are identified either in R or 3.6 μm bands. Blue crosses are identified in the NUV or 24 μm bands. Black open squares and black open circles indicate objects matched with radio and variability AGN candidates, respectively. (Color online)

Table 2.

Identification of the X-ray sources.*

ID	X-ray		ERR	ML2	ML34	Primary flag				Matching		Counterpart
	RA	Dec	(″)			O	U	I	M	R	V	RA	Dec
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)
SXDS0001	02:15:24.9	−04:55:43.3	2.40	19.9	5.1	1	1	1	1	0	0	02:15:24.9	−04:55:41.2
SXDS0002	02:15:24.9	−04:54:07.9	2.20	33.7	6.2	1	1	1	0	0	0	02:15:24.9	−04:54:07.8
SXDS0003	02:15:27.2	−05:02:00.7	2.90	0.0	7.5	0	1	0	0	0	0	02:15:26.9	−05:02:04.3
SXDS0004	02:15:29.0	−04:57:46.5	3.00	14.1	0.0	1	1	1	0	0	0	02:15:28.9	−04:57:47.1
SXDS0005	02:15:29.4	−04:53:27.5	1.90	11.8	8.6	1	0	1	1	0	1	02:15:29.3	−04:53:25.2
SXDS0008	02:15:37.1	−04:56:56.5	1.20	41.6	5.9	1	1	1	0	0	1	02:15:37.1	−04:56:57.2
SXDS0010	02:15:38.8	−05:00:52.2	1.30	72.1	11.0	1	1	1	1	0	0	02:15:38.7	−05:00:50.3
SXDS0012	02:15:41.6	−05:04:13.5	2.40	7.4	8.3	−1	0	−1	0	0	0	02:15:41.6	−05:04:11.0
SXDS0015	02:15:43.9	−05:07:15.8	0.30	830.0	150.0	1	1	0	0	0	0	02:15:43.9	−05:07:16.3
SXDS0016	02:15:44.0	−04:55:25.8	2.30	11.7	4.4	−1	0	−1	0	0	0	02:15:44.0	−04:55:22.2

*Column (1) is the X-ray name. Columns (2) and (3) are the X-ray coordinates. Column (4) is σ_Xi from Ueda et al. (2008). Columns (5) and (6) are the maximum likelihoods in the 0.5–2 keV and 2–10 keV bands. Columns (7)–(10) show whether the primary counterpart is selected as the primary candidate in R-band [column (7)], GALEX NUV band [column (8)], Spitzer 3.6 μm [column (9)], and Spitzer 24 μm [column (10)] identifications. 1 means the object is selected as the primary candidate of the identification in the band. −1 means the object is selected as the primary candidate, but the likelihood is less than 1.0. Columns (11) and (12) indicate the primary counterpart is selected as counterpart of a 1.4 GHz radio source and variability AGN candidate, respectively. Columns (13) and (14) are coordinates of the primary counterpart. (This table is available in its entirety in a machine-readable form in the online journal. A portion is shown here for guidance regarding its form and content.)

4.4 Candidates of clusters of galaxies

The identifications discussed above assume the X-ray source is a single object. This is not the case for clusters of galaxies, which are a non-negligible population among extra-galactic X-ray sources. Some of the identifications with low reliability can be caused by the multiple galaxies associated with an X-ray source that is associated with a cluster of galaxies.

Finoguenov et al. (2010) constructed catalogs of groups and clusters of galaxies in the SXDS field through wavelet detection of extended X-ray emission for the XMM-Newton imaging dataset. They cross-match their candidates with the extended sources in Ueda et al. (2008), and 14 sources in Ueda et al. (2008) are matched. Among them, we regard 10 extended X-ray sources as candidate clusters of galaxies as listed in table 3. Among the remaining four sources, we regard three as candidates for AGNs. SXDS0051 is cross-identified with SXDF47XGG at z = 0.348, but SXDS0051 is significantly detected in the 2.0–10 keV band and the observed 0.5–2 keV and 2–4.5 keV hardness ratio, HR2, is relatively large (−0.50). Thus we regard the X-ray source as an AGN candidate. SXDS0396 is cross-identified with SXDF64XGG at z = 1.030, but SXDS0396 is significantly detected in the 2.0–10 keV band and HR2 (−0.36) implies a power law with photon index of 1.6. Therefore, we regard the X-ray source as an AGN candidate. The source is also close to SXDS0386. SXDS0934 is cross-identified with SXDF06XGG at z = 0.451, but the primary optical counterpart is identified with a broad-line AGN at z = 1.669, therefore we regard the AGN as the X-ray source. SXDS0784 matched with SXDF07XGG has a detection likelihood less than 7 in both the soft and hard X-ray bands, and it is not included in the total sample.

Table 3.

Candidate clusters of galaxies.*

ID	HR2	zspec	zphot	Note
(1)	(2)	(3)	(4)	(5)
SXDS0047	−0.79	9.99	1.00
SXDS0280	−0.95	9.99	0.45	SXDF79XGG
SXDS0295	−0.53	9.99	0.60
SXDS0434	−0.44	9.99	0.62
SXDS0453	−1.00	0.20	0.20	SXDF16XGG
SXDS0514	−0.54	0.65	0.65	SXDF69XGG
SXDS0552	−0.82	0.52	0.52	SXDF34XGG, Radio
SXDS0621	−0.98	0.69	0.69	SXDF04XGG, SXDF85XGG, Radio
SXDS0647	−0.64	9.99	0.46	SXDF35XGG
SXDS0829	−0.82	0.87	0.88	SXDF46XGG
SXDS0876	−0.53	0.38	0.38	SXDF01XCG
SXDS0889	−0.70	0.20	0.38	SXDF01XCG
SXDS0913	−0.88	9.99	0.47
SXDS1176	−0.74	0.33	0.33	SXDF36XGG

*Column (1) is the X-ray source name. Column (2) is HR2. Sources only detected in the soft-band are indicated with HR2 of −1.00. Column (3) is spectroscopic redshift: 9.99 means no spectroscopic redshift is available. Column (4) is photometric redshift. Column (5) describes the corresponding name in Finoguenov et al. (2010).

We also found four additional candidates of clusters of galaxies with an excess of objects with similar B − z′ and z′ − K colors within r = 15″ from the X-ray source position. These are also listed in table 3. The median photometric redshift of objects within r = 30″ from the X-ray position is used as the photometric redshift for the candidates.

In the table, the X-ray source ID (column 1), HR2 (2), spectroscopic redshift (3), and photometric redshift (4) are summarized.

4.5 Ultra-luminous X-ray source candidates

Four X-ray sources, SXDS0416, SXDS0390, SXDS0404, and SXDS0402, are associated with a bright nearby galaxy KUG 0214−057 at z = 0.018 (Watson et al. 2005). They are strong ultra-luminous X-ray (ULX) source candidates.

5 Spectroscopic identification

5.1 Optical spectroscopic observations

5.1.1 FOCAS/Subaru observation in selected fields

We conducted a spectroscopic campaign observation with a multi-slit optical spectrograph, FOCAS on the Subaru telescope, from 2001 January to 2005 November. We targeted the primary counterparts with the highest priority, and in addition we observed secondary counterparts if possible. We put the slits at the positions of the primary counterparts determined in the optical images, even for primary counterparts selected by the identifications in the 3.6 μm, 24 μm, or NUV band. The situation is the same for the following spectroscopic observations. The FoV of the instrument is a 6′-diameter circle. In total, 50 fields were observed. About three quarters of the fields were selected to cover as many as X-ray sources as possible. The remaining fields were chosen to cover as many as z ∼ 1 red sequence galaxies as possible (Yamada et al. 2005) and to cover a candidate z = 5.7 cluster of galaxies found by narrow-band surveys (Ouchi et al. 2005, 2008). In such fields, X-ray sources are observed as filler targets for the masks.

In the observations, we use a 150R grating with SY47 order-cut filter to cover 4700 Å to 9400 Å. A slit width of 0|${^{\prime\prime}_{.}}$|8 is used and the resulting spectral resolution is 20 Å. Hereafter, we use the spectral resolution measured with night-sky emission lines obtained simultaneously with the target spectrum. The spectral resolution for the target can be higher because the slit illumination by a stellar object can be more concentrated than the smooth illumination by night-sky emission. In later observations, we use a 300B grating with order-cut filter SY47 or without the filter. With the order-cut filter the spectral coverage is 4700 Å to 9200 Å. Without the filter the spectral coverage in the blue extends down to 3800 Å, but second-order light can contaminate the wavelength range above 7600 Å. A slit width of 0|${^{\prime\prime}_{.}}$|8 is used and the resulting spectral resolution is 13 Å. The central coordinate, observing mode, and the exposure time of each field are summarized in table 4. During the observation, a few additional objects were observed in long-slit mode with the 300B grating with SY47 order-cut filter.

Table 4.

Journal of FOCAS/Subaru observations.*

Mask name	FoV Center		Period	Exptime	Grism	Filter	λ Coverage	Note
	RA	Dec		(s)			(Å)
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
150A_20	02:17:57.60	−05:02:25.1	2001/01	5400	150	SY47	4700–9400
150A_21	02:17:39.00	−05:01:35.4	2001/11	7200	150	SY47	4700–9400
150A_22	02:18:28.80	−05:00:32.4	2001/11	5400	150	SY47	4700–9400
150A_23	02:18:21.10	−05:05:27.0	2001/11	1800	150	SY47	4700–9400
150A_01	02:17:09.67	−05:13:17.9	2003/10	7200	150	SY47	4700–9400
150A_02	02:19:40.20	−04:50:54.5	2003/12	3600	150	SY47	4700–9400
150A_03	02:17:16.41	−04:58:49.6	2003/10	7200	150	SY47	4700–9400
150A_04	02:19:23.44	−05:05:26.8	2003/12	3600	150	SY47	4700–9400
150A_05	02:16:54.54	−04:48:08.3	2003/12	3600	150	SY47	4700–9400
150A_06	02:17:23.92	−05:28:55.2	2003/12	3600	150	SY47	4700–9400
150A_07	02:19:31.70	−04:58:42.6	2003/12	3600	150	SY47	4700–9400
150A_08	02:20:06.97	−04:56:13.6	2003/12	4800	150	SY47	4700–9400
150A_09	02:19:23.09	−04:50:37.3	2003/12	3600	150	SY47	4700–9400
150A_10	02:18:11.78	−05:02:23.2	2003/12	3600	150	SY47	4700–9400
150A_11	02:17:34.52	−05:17:15.6	2003/12	3600	150	SY47	4700–9400
150A_12	02:17:59.68	−04:54:43.2	2003/12	4800	150	SY47	4700–9400
150A_13	02:20:22.53	−05:02:12.1	2003/12	3600	150	SY47	4700–9400
150A_14	02:19:33.89	−05:14:18.7	2003/12	3600	150	SY47	4700–9400
150A_15	02:16:18.00	−05:10:32.7	2003/10	5400	150	SY47	4700–9400
150A_16	02:18:26.74	−04:46:35.3	2003/12	3600	150	SY47	4700–9400
150A_17	02:17:07.19	−04:44:21.3	2003/12	4800	150	SY47	4700–9400
150A_18	02:18:38.77	−05:05:14.4	2003/12	5400	150	SY47	4700–9400
150A_19	02:17:35.15	−05:03:11.8	2003/12	3600	150	SY47	4700–9400
150B_01	02:18:08.28	−05:01:32.3	2003/12	3600	150	SY47	4700–9400
150C_01	02:19:40.70	−05:06:41.3	2003/10	3600	150	SY47	4700–9400
150C_05	02:17:44.60	−04:51:05.9	2003/10	4800	150	SY47	4700–9400
150C_07	02:17:20.50	−05:06:19.8	2003/10	6000	150	SY47	4700–9400
150C_08	02:17:20.80	−05:12:07.8	2003/10	4800	150	SY47	4700–9400
150C_10	02:16:53.50	−04:57:08.5	2003/10	3600	150	SY47	4700–9400
300A_10	02:19:01.33	−04:31:07.2	2003/10	7200	300B	SY47	4700–9200
300C_05	02:17:51.50	−05:29:30.0	2003/12	7200	300B	SY47	4700–9200
300C_06	02:18:25.92	−05:32:47.3	2003/12	7200	300B	SY47	4700–9200
300C_07	02:20:22.32	−04:50:23.9	2003/12	7200	300B	SY47	4700–9200
300E_01	02:18:09.35	−05:03:03.2	2004/10	14400	300B	NONE	3800–9200
			2004/12	7200	150	SY47	4700–9400
300E_02	02:17:24.23	−04:53:28.5	2004/10	9000	300B	NONE	3800–9200
300E_03	02:17:34.08	−05:11:11.7	2004/11	5400	150	SY47	4700–9400
300E_04	02:17:43.84	−05:08:31.6	2004/11	10800	300B	NONE	3800–9200
300E_05	02:17:17.19	−05:00:15.3	2004/11	10800	300B	NONE	3800–9200
300E_06	02:18:12.63	−04:53:22.8	2004/11	10800	300B	NONE	3800–9200
300E_07	02:18:44.59	−05:00:22.7	2004/11	10800	300B	NONE	3800–9200
300E_08	02:17:42.10	−04:49:54.6	2004/11	18000	300B	NONE	3800–9200
300E_09	02:17:57.53	−04:56:47.0	2004/11	9000	300B	NONE	3800–9200
300G_01	02:17:18.07	−05:08:26.3	2005/11	9000	300B	NONE	3800–9200
300G_02	02:18:33.27	−05:18:31.1	2005/11	7200	300B	NONE	3800–9200
300G_04	02:16:28.60	−05:09:35.9	2005/11	10800	300B	NONE	3800–9200
300G_05	02:16:11.08	−05:10:14.0	2005/11	7200	300B	NONE	3800–9200
300G_06	02:16:51.87	−04:45:07.1	2005/11	12600	300B	NONE	3800–9200
300G_08	02:16:51.21	−04:50:36.5	2005/11	16200	300B	NONE	3800–9200
300G_10	02:19:10.45	−05:12:59.1	2005/11	7200	300B	NONE	3800–9200
300G_12	02:17:47.25	−04:33:39.3	2005/11	7200	300B	NONE	3800–9200

*Column (1) is the name of the FoV, and columns (2) and (3) are coordinates of the FoV center. Column (4) is the observation period. Column (5) is the integration time. Columns (6) and (7) indicate the name of the grism and filter used in the observation. Column (8) shows the typical wavelength coverage.

The data are reduced with a script based on IRAF. The procedure follows the usual analysis of long-slit observations. The bias counts are subtracted using the overscan area of the CCD, and a flat-fielding correction is applied with a dome flat frame taken after the observation. Wavelength calibration is performed using a ThAr lamp spectrum. We did not dither the object position along the slit during the observation, and the sky subtraction is performed by using the sky spectra obtained on either side of an object. The flux calibration of the reduced spectra is carried out utilizing the spectra of spectroscopic standard stars taken with a 2|${^{\prime\prime}_{.}}$|0-width slit. Feige 110 and G191 B2B are used as the standard stars. The standard stars are obtained with and without the SY47 order-cut filter, in order to evaluate the contamination of the second-order light at the red end of the wavelength coverage. The sensitivity curve for each night is determined with the standard stars. Furthermore, we adjust the normalization of the spectra of the targets using their photometric data. After the sensitivity correction, we also applied correction for atmospheric absorption features, A-band and B-band, using the absorption profile of the bands derived with a bright object in the same mask.

5.1.2 2dF/AAT observation for bright optical counterparts

Spectroscopic follow-up observations for bright optical counterparts with R < 21.0 mag were obtained with the 2dF 400 multi-fibre optical spectrograph on the Anglo-Australian 3.9 m Telescope on 2002 November 3–6 in director's discretionary time. The integration time ranges from 6 min for the brightest objects to 3 hr for the faintest objects. The 2dF has two spectrographs, each covering 200 fibres. We used a 270R grating with wavelength range of 4400–9100 Å for the first spectrograph, and a 316R grating with wavelength range of 4700–8500 Å for the second spectrograph. The spectral resolutions are 14 and 12 Å and the spectral samplings are 4.8 and 4.1 Å pixel⁻¹ for the former and latter spectrographs. We observed a relatively bright star in each fibre configuration in order to correct for wavelength dependence of the sensitivity. The data are reduced in the standard way for the 2dF data using the 2dfdr software.

5.1.3 Additional spectroscopic observations for faint optical counterparts

Some of the radio and X-ray sources in the SXDS field were observed with the Visible Multi-Object Spectrograph (VIMOS) on the Very Large Telescope (VLT) (P074.A-0333; Simpson et al. 2012). The MR-Orange grating and the GG475 order-cut filter were used and a spectral resolution of 10 Å in the wavelength range between 4800 Å and 10000 Å was achieved. Twenty-seven slit masks were prepared to cover ∼ 80% of the radio and X-ray sources in the SXDS. ∼ 75% of the 27 masks were observed with one 2700 s and two 1350 s exposures.

Additionally, two FoVs of VIMOS were observed in the course of spectroscopic observations of extended Lyα emitters by Saito et al. (2008) (P074.A-0524). The HR-Orange grism and the GG435 order-cut filter were used. The wavelength range between 5000 and 6800 Å was covered with 2.8 Å resolution.

A deep spectroscopic survey for K-band-selected galaxies in the UKIDSS UDS region was conducted with the VLT VIMOS and VLT FOcal Reducer and low dispersion Spectrograph 2 (FORS2) instruments in the UDSz ESO Large Programme (180.A-0776, PI: Almaini; Bradshaw et al. 2013; McLure et al. 2013). In the VIMOS observations, the LR-Blue and LR-Red grisms were used for each mask and wavelength range between 3700 and 9500 Å was covered. A spectral resolution of 35 Å was achieved at 7000 Å. The FORS2 observations were conducted with a 300I grism covering 6000–10000 Å and with spectral resolution of 12 Å.

In the course of the optical spectroscopic survey of Lyman Alpha Emitters and Lyman Break Galaxies in the SXDS field, the primary counterparts of the X-ray sources were observed as filler objects in the masks. A multi-object spectroscopic observation with the DEep Imaging Multi-Object Spectrograph (DEIMOS) instrument attached to the Keck II telescope was also carried out (Ouchi et al. 2009). The observations used the 830G grating and the GG495 filter. The 5700–9500 Å wavelength range was covered with a spectral resolution of 2.6 Å.

Further spectroscopic survey observations were made with the Inamori-Magellan Areal Camera and Spectrograph (IMACS) on the Magellan 6.5 m Baade telescope (Ouchi et al. 2009). In the IMACS observations, the F/2 camera with 27|${^{\prime}_{.}}$|4 diameter FoV and 300 lines mm⁻¹ grism with central wavelength of either 6700 Å or 8000 Å were used. Wavelength ranges of from 4700 Å to 8000 Å (from 6200 Å to 9600 Å) were covered with 7 Å (6 Å) resolution and 1.34 Å pixel⁻¹ (1.25 Å pixel⁻¹) by the former (latter) setup.

X-ray sources are also observed as filler targets during spectroscopic observations of the Balmer-break galaxies in the field with IMACS on the Magellan Baade telescope (PI. N. Padilla; Díaz Tello et al. 2013). In the observation, the F/2 camera with 27|${^{\prime}_{.}}$|4 diameter FoV and 200 lines mm⁻¹ grism with central wavelength of 6600 Å was used. The wavelength range from 4500 Å to 9000 Å is covered with 10 Å spectral resolution and 2.0 Å pixel⁻¹ sampling.

Some quasi-stellar object (QSO) candidates identified with the KX selection technique were spectroscopically observed with the AAOmega spectrograph on the AAT 3.9 m telescope, and X-ray AGNs were also observed as filler targets (Smail et al. 2008). The 580V and 385R gratings were used in the dual-beam spectrograph of AAOmega and the wavelength range from 3700 Å to 8600 Å was covered simultaneously. Spectral resolution is 5.0 Å with spectral sampling of 3.4 Å pixel⁻¹.

5.2 NIR spectroscopic observations with FMOS

In addition to the optical spectroscopic observations, we conducted intensive NIR spectroscopic observations with the Fiber Multi Object Spectrograph (FMOS) on the Subaru telescope (Kimura et al. 2010). This instrument can observe up to 200 objects simultaneously over a 30′-diameter FoV in the cross-beam switching mode with two spectrographs, IRS1 and IRS2. The fibers have an aperture of 1|${^{\prime\prime}_{.}}$|2 diameter on the sky. The spectrographs cover the wavelength range between 9000 Å and 18000 Å with spectral resolution of R ∼ 800 at λ ∼ 1.55 μm in the low-resolution mode. A total of 851 sources were observed in observations of 22 FoVs with this setup during guaranteed, engineering, and open-use (S11B-098 Silverman et al. and S11B-048 Yabe et al.) time observations. The central coordinate, observation period, integration time, and observing mode of each FoV are summarized in table 5.

Table 5.

Journal of the FMOS/Subaru observations.*

FoV name	FoV center		Period	Exptime	MODE	Note
	RA	Dec		(s)
(1)	(2)	(3)	(4)	(5)	(6)	(7)
SXDS00	02:17:32.6	−05:00:34	2009/10	9000	LR	IRS1 only
SXDS01	02:17:18.0	−05:04:19	2009/12	5400	LR	IRS1 only
SXDS02	02:17:32.5	−05:10:50	2009/12	15300	LR
SXDS03	02:17:45.4	−05:05:23	2010/09	11700	LR
SXDS05	02:17:32.9	−04:39:00	2010/11	6300	LR
SXDS06	02:17:43.3	−04:38:20	2010/11	3600	LR
SXDS07	02:17:40.4	−05:18:17	2010/11	35100	LR
SXDS08	02:19:26.8	−04:59:37	2010/11	16200	LR
SXDS09	02:16:35.6	−05:01:44	2010/11	18900	LR
SXDS10	02:17:39.2	−04:38:25	2010/11	11700	LR
SXDS11	02:17:41.1	−05:02:24	2010/11	14400	LR
SXDS12	02:17:59.2	−04:51:00	2010/12	13500	LR	IRS1 only
SXDS13	02:18:26.9	−05:18:03	2010/12	1800	LR
			2011/02	2700	LR
			2011/02	1800	LR
SXDS16	02:16:59.7	−05:00:19	2011/10	10800	LR
SXDS17	02:17:48.7	−05:15:38	2011/10	19800	LR
SXDS18	02:18:42.2	−04:38:29	2011/10	18000	LR
SXDS19	02:17:19.4	−05:21:00	2011/10	18000	LR
SXDS20	02:19:11.8	−05:07:26	2011/10	18900	LR
SXDS21	02:17:40.7	−05:01:30	2011/10	17100	LR
			2011/10	9000	LR
			2011/11	3600	LR	IRS1 only
			2011/11	5400	LR
SXDS22	02:16:19.3	−04:58:58	2011/12	9000	LR
			2011/12	14400	LR	IRS1 only
SXDS23	02:18:52.5	−05:02:11	2011/12	3600	LR
			2011/12	14400	LR	IRS1 only
SXDS24	02:17:34.0	−05:18:25	2011/12	11700	LR	IRS1 only

*Column (1) is the name of the FoV, columns (2) and (3) the central coordinates of the FoV, column (4) the observation period, column (5) the total integration time, column (6) the observation mode, and column (7) notes.

The NIR spectra are reduced with the pipeline data reduction software FIBRE-pac (Iwamuro et al. 2012). The resulting one-dimensional spectra are corrected for atmospheric absorption and sensitivity dependence on wavelength using relatively faint F–G-type stars observed simultaneously with the targets.

5.3 Spectroscopic identification summary

The primary counterparts of 31 X-ray sources with a bright stellar object show optical spectra consistent with Galactic stars. We identify them as Galactic stars. The ID number and HR2 of the X-ray sources are given in table 6. Almost all of them have a small HR2 value, which is consistent with the thermal X-ray emission from Galactic stars. Hereafter, we exclude X-ray sources that are selected as candidates for clusters of galaxies (subsection 4.4), ULXs (subsection 4.5), and Galactic stars, and refer to the remaining X-ray sources as AGN candidates. There are 896 AGN candidates in the total sample as summarized in table 7.

Table 6.

Candidate galactic stars.*

ID	HR2		ID	HR2
(1)	(2)		(1)	(2)
SXDS0020	−1.00		SXDS0686	−0.63
SXDS0039	−1.00		SXDS0692	−1.00
SXDS0120	−0.98		SXDS0693	−0.96
SXDS0125	−0.93		SXDS0731	−1.00
SXDS0191	−1.00		SXDS0783	−1.00
SXDS0245	−0.90		SXDS0816	−0.92
SXDS0248	−0.82		SXDS0822	−0.98
SXDS0297	−0.99		SXDS0832	−0.80
SXDS0302	−1.00		SXDS0844	−0.97
SXDS0365	+0.32		SXDS0923	−0.92
SXDS0448	−0.67		SXDS0974	−0.91
SXDS0469	−0.74		SXDS0985	−1.00
SXDS0532	−0.94		SXDS0995	−0.99
SXDS0545	−0.90		SXDS1062	−1.00
SXDS0601	−0.93		SXDS1181	−0.98
SXDS0633	−1.00

*Column (1) is the X-ray source name. Column (2) shows HR2. Sources only detected in the soft-band are indicated with HR2 of −1.00.

Table 7.

X-ray source content of the sample.

	N  total	N  soft	N  hard	Comment
X-ray sources	1045	866	645	Listed in Ueda et al. (2008)
X-ray sample	945	781	584	Within deep Suprime-Cam image coverage
Covered by 1.4 GHz radio survey	789	655	497	Covered in Simpson et al. (2006)
Clusters of galaxies candidates	14	14	1	See table 3
Galactic star candidates	31	30	4	See table 6
ULX candidates	4	4	3	See subsection 4.5
AGN candidates	896	733	576
Observation summary for the 896 AGN candidates
Optical spec. observed	597	523	399
NIR spec. observed	851	704	548
Spec. identified	586	514	397
Photz flag = 1	1	1	0
Photz flag = 2	10	10	7
Photz flag = 3	12	6	7

Among the primary counterparts of the 896 AGN candidates, 597 counterparts are observed in the optical spectroscopic observations discussed above, as summarized in table 7. As a first step, we determine the redshift of each object by examining its optical spectrum and fitting Gaussian profiles to the observed emission and absorption lines.

A large fraction of objects show multiple features. If a narrow emission line is detected, we determine the redshift with that line. For example, if the [O ii]λ3727 emission line is detected, we determine the redshift of the object with that emission line. For low-redshift objects, the optical spectrum covers the Hα and Hβ wavelength range and shows a strong [O iii]λ5007 emission line. We determine their redshifts with the [O iii]λ5007 line. For high-redshift objects, the optical spectrum covers the rest-frame wavelength range below the Mg iiλ2800 emission line. If that emission line is detected, it is used to determine the redshifts, otherwise, C iii]λ1909 or C ivλ1550 emission lines are used. For objects only showing absorption lines, we determine their redshift with the Ca ii HK doublet absorption lines at 3969 Å and 3934 Å, if these absorption lines are detected.

We regard a permitted emission line with velocity width larger than 1000 km s⁻¹ (after deconvolving the instrumental profile) as a broad-emission line. This threshold is narrower than the typical criteria for broad-line AGNs (1500 or 2000 km s⁻¹). We set the threshold considering the distribution of the FWHM of the broad-line AGNs in the local Universe (Hao et al. 2005; Stern & Laor 2012). Although the threshold is narrower, most of the broad emission lines are broader than 1500 km s⁻¹ and the distribution of the FWHM of the broad-emission lines, especially the lack of broad-emission lines below 1500 km s⁻¹, is similar to that of optically selected broad-line AGNs in the SDSS as discussed in Nobuta et al. (2012).

Some objects only show one either broad or narrow emission line in the observed optical wavelength range. For such objects, we identify most of the single narrow-emission lines with [O ii]λ3727, considering the absence of any other emission line. If the identification is not conclusive due to the narrow wavelength coverage of a spectrum, we identify the single narrow-emission line, utilizing the photometric redshift described in the next section. For objects with one prominent broad-emission line, we mostly identify it with Mg iiλ2800, consistent with the absence of another broad emission line in the observed wavelength range. The identifications with a single broad-emission line in the optical are checked with the NIR spectra covering the rest-frame optical wavelength range.

Intensive spectroscopic observations in the NIR wavelength range cover 851 of the total X-ray sources as summarized in table 7. Spectroscopic identification in the NIR spectra is conducted in the same manner as for the optical spectra. In the NIR spectra, if one narrow or broad emission line is detected, almost all of them are identified with the Hα emission line. We identify some single narrow-emission lines with [O iii]λ5007, considering the photometric redshift. In such cases, [O iii]λ4959 is presumed to be weaker than the detection limit. Among the objects that are observed but not identified in the optical spectroscopy, 23 objects are identified with NIR spectroscopic observations with FMOS. They are AGNs at z ∼ 0.9–2.5, and 15 of them are narrow-line AGNs, which are mostly identified through their strong Hα or [O iii]λ5007 lines.

Objects that show broad emission in at least one of the observed permitted emission lines are referred to as “broad-line” AGNs, hereafter. It needs to be noted that the criteria is not homogeneous and depends on the rest-frame wavelength coverage of the optical and NIR spectroscopic observations. For example, some objects are identified as “broad-line” AGNs with one prominent broad Hα emission line, but the same objects can be identified as “narrow-line” AGNs if spectra covering the Hα line are not available. Furthermore, an object with broad Hα emission lines can be type 1, 1.5, 1.8, or 1.9 Seyferts. On the other hand, objects without broad-emission lines are referred to as “narrow-line” AGNs. Considering the X-ray luminosity distribution of the sample, their X-ray emission is thought to be dominated by the AGN component, and we ignore contributions to X-ray sources from star-forming galaxies in the AGN candidates.

Utilizing the optical and NIR spectra, 586 out of the 896 AGN candidates (65%) are spectroscopically identified. The fraction of spectroscopically identified objects in the entire AGN candidate sample is shown in figure 8 as a function of radius from the center of the survey field. If we limit the sample to primary counterparts brighter than i^′ = 23 mag, the identification rate reaches 86%. The spectroscopic identification of each AGN candidate is summarised in table 10. Column 22 shows the spectroscopic redshift, and column 23 describes the classification. BLA and NLA represent broad-line and narrow-line AGNs, respectively. NID indicates the object is observed in the optical but cannot be identified with the optical spectrum. NSP indicates the object is not observed in the optical. NID and NSP do not refer to the availability of NIR spectrum.

Fig. 8.

Spectroscopic identification rate of the AGN candidates as a function of radius from the center of the survey field. From top to bottom, the lines are for X-ray sources with primary counterparts brighter than i^′ = 23 mag, for all X-ray sources, and for X-ray sources with i^′ > 23 mag.

6 Multi-wavelength photometry and photometric redshift determination

6.1 Multi-wavelength photometry

For X-ray sources without spectroscopic identification, we determine their photometric redshifts. In order to determine photometric redshifts reliably, we conduct multi-wavelength photometry of each primary counterpart paying careful attention to the difference in the PSF size and shape.

In the photometry, the i^′-band image taken with Suprime-Cam is used as the reference frame for the object detection. Sextractor is used with a detection threshold of three connected pixels above 2.0 σ. The aperture magnitude determined with a variable elliptical aperture (MAG_AUTO) is used as the i^′-band total magnitude of an object. We use PHOT_AUTOPARAMS of (2.5, 3.5) to determine the variable elliptical aperture. With the parameters, 94% of the total flux will be covered within the variable elliptical aperture (Bertin & Arnouts 1996). Using the i^′-band-detected catalog, optical colors of the object are measured in the 2|${^{\prime\prime}_{.}}$|0-diameter aperture using the multi-image mode of Sextractor. The total magnitude in each optical band is determined with a fixed aperture color and the aperture correction determined in the i^′-band, i.e., the difference between i^′-band MAG_AUTO and i^′-band MAG_APP.

For photometry in wavelengths other than the optical bands, we measure a fixed aperture color between a selected band and i^′-band using the PSF-matched i^′-band image whose PSF is matched to the image in the selected band. The fixed aperture color is converted into the total magnitude in the selected band using the difference between the i^′-band MAG_AUTO in the original i^′-band image and the aperture magnitude measured in the PSF-matched image. We try to conserve the fixed aperture color, and assume the light profile in the selected band follows that in the i^′-band. The details are described as follows.

At first we subtracted the background component of each image using the object detection software, Sextractor. The background is determined locally with median filtering in a 48″ box. For very bright and extended objects with sizes larger than 30″, i.e., nearby galaxies, the background removal subtracts the extended halo component, thus the photometry is not reliable for such objects. Background-subtracted images are magnified and aligned to the Suprime-Cam i^′-band imaging data. Coordinate conversions are determined with positions of common bright stars using the geomap command in IRAF with a second-order polynomial, and the images are registered with the gregister command. Fluxes of objects in the original images are conserved.

The PSF size of the Suprime-Cam i^′-band image is the smallest amongst the multi-wavelength data. We make the PSF-matched i^′-band images by convolving with a kernel which represents the PSF difference between the i^′-band and a given band. The determination of the kernel and the convolution is performed with the psfmatch command in IRAF.

After matching the PSF size and shape, we measure colors of objects with the multi-image mode of Sextractor. Colors in the i^′-band and a given band are measured from the images in the band and the PSF-matched i^′-band at the position in the original i^′-band image. The total magnitude in the band is derived from the aperture photometry and the aperture correction derived from the difference between the MAG_AUTO in the original i^′-band image and the MAG_APP in the PSF-matched i^′-band image. Different aperture diameters are used for different datasets: 7|${^{\prime\prime}_{.}}$|6 diameter for GALEX FUV and NUV, 2|${^{\prime\prime}_{.}}$|0 diameter for U-band images and UKIDSS J-, H-, and K-bands, 3|${^{\prime\prime}_{.}}$|8 diameter for the IRAC images.

The uncertainties in the photometric measurement are determined as the 1 σ uncertainty of the flux measurement in the aperture magnitude assuming a background-limited regime. For bright objects, the systematic uncertainty dominates the measurement uncertainty, and constant values, 0.01–0.05 mag depending on wavelength, are added to represent such systematic uncertainties.

In order to obtain photometric data for primary counterparts that are not detected in the i^′-band-selected catalog, we also construct B-, R-, V-, z′-, and 3.6 μm band selected catalogs using the images in each band as the reference frame. We apply the same procedures as for the i^′-band-selected catalog.

The results of the multi-wavelength photometry are summarized in tables 8, 9, and 10. The total magnitude and associated error for each band are given. An error of 99.99 indicates the associated magnitude is an upper limit. A total magnitude of 99.99 indicates the object is not covered in the imaging data of that band. The values in the tables are corrected for Galactic extinction. The applied corrections are summarized in table 1.

Table 8.

Multi-wavelength properties of the X-ray AGN candidates I.*

ID	FUV		NUV		U		B		V		R		i   ′		z′
	(mag)		(mag)		(mag)		(mag)		(mag)		(mag)		(mag)		(mag)
(1)	(2)		(3)		(4)		(5)		(6)		(7)		(8)		(9)
SXDS0001	23.40	0.07	21.66	0.05	21.13	0.05	20.62	0.01	19.99	0.01	19.30	0.01	19.00	0.01	18.72	0.01
SXDS0002	25.34	99.99	23.36	0.09	22.99	0.06	22.61	0.01	22.17	0.01	21.55	0.01	21.07	0.01	20.83	0.01
SXDS0003	24.39	0.10	23.73	0.10	23.84	0.10	23.39	0.01	22.82	0.01	22.60	0.02	22.66	0.01	22.51	0.02
SXDS0004	23.04	0.05	22.18	0.05	18.98	0.05	20.93	0.01	99.99	99.99	19.21	0.00	17.94	0.00	17.55	0.00
SXDS0005	25.34	99.99	24.98	99.99	25.36	0.26	24.76	0.02	23.91	0.02	22.86	0.02	22.00	0.01	21.20	0.01
SXDS0008	23.09	0.06	22.13	0.06	22.28	0.05	22.43	0.01	21.96	0.01	22.02	0.01	22.04	0.01	21.97	0.01
SXDS0010	22.13	0.05	21.16	0.05	20.94	0.05	20.85	0.01	20.64	0.01	20.56	0.01	20.60	0.01	20.61	0.01
SXDS0012	25.34	99.99	24.98	99.99	24.42	0.12	24.12	0.01	24.01	0.02	23.70	0.02	23.54	0.01	23.11	0.02
SXDS0015	21.10	0.05	20.67	0.05	18.33	0.05	20.95	0.01	99.99	99.99	19.48	0.00	18.31	0.00	17.86	0.00
SXDS0016	25.34	99.99	24.98	99.99	23.52	0.07	22.01	0.01	21.01	0.01	20.87	0.01	20.87	0.01	20.82	0.01

*Column (1) is the X-ray source name. Columns (2)–(9) show the measured total magnitude and associated error from FUV to z′-band. 99.99 in the error column means the corresponding magnitude is an upper limit. 99.99 in the total magnitude means the object is not covered in that band. (This table is available in its entirety in a machine-readable form in the online journal. A portion is shown here for guidance regarding its form and content.)

Table 9.

Multi-wavelength properties of the X-ray AGN candidates II.*

ID	J		H		K		3.6 μm		4.5 μm		5.8 μm		8.0 μm		24 μm
	(mag)		(mag)		(mag)		(mag)		(mag)		(mag)		(mag)		(mag)
(10)	(11)		(12)		(13)		(14)		(15)		(16)		(17)		(18)
SXDS0001	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	15.91	0.05
SXDS0002	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	17.32	0.05
SXDS0003	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99
SXDS0004	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	17.25	0.01	99.99	99.99	18.13	0.01	18.22	99.99
SXDS0005	99.99	99.99	99.99	99.99	99.99	99.99	19.34	0.02	19.63	0.04	20.39	0.36	19.96	0.31	17.92	0.05
SXDS0008	99.99	99.99	99.99	99.99	99.99	99.99	20.47	0.04	20.18	0.06	19.97	99.99	19.47	0.19	18.22	99.99
SXDS0010	99.99	99.99	99.99	99.99	99.99	99.99	18.91	0.01	18.64	0.01	18.47	0.02	18.13	0.01	17.43	0.05
SXDS0012	99.99	99.99	99.99	99.99	99.99	99.99	21.34	0.03	21.09	0.03	21.69	99.99	21.77	99.99	18.22	99.99
SXDS0015	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	99.99	15.56	0.05
SXDS0016	99.99	99.99	99.99	99.99	99.99	99.99	20.40	0.01	20.04	0.01	19.55	0.04	18.70	0.02	17.41	0.05

*Column (10) is the X-ray source name. Columns (11)–(18) show measured total magnitude and associated error from J- to 24 μm-band. 99.99 in the error column means the corresponding magnitude is the upper limit. 99.99 in the total magnitude column means the object is not covered in that band. (This table is available in its entirety in a machine-readable form in the online journal. A portion is shown here for guidance regarding its form and content.)

Table 10.

Multi-wavelength properties of the X-ray AGN candidates III.*

				Spec.z		Phot.z
ID	VLA		Si
	(μJy)			z  spec	flag	z  phot	type	flag
(19)	(20)		(21)	(22)	(23)	(24)	(25)	(26)
SXDS0001	0.0	0	0	0.505	NLA	0.480	GAL	0
SXDS0002	0.0	0	0	0.700	NLA	0.644	GAL	0
SXDS0003	0.0	0	0	9.999	NSP	0.622	GAL	0
SXDS0004	0.0	0	0	9.999	NSP	0.496	GAL	2
SXDS0005	0.0	0	0	1.050	NLA	0.782	GAL	0
SXDS0008	0.0	0	1	1.033	NLA	1.010	QSO	0
SXDS0010	0.0	0	1	1.224	BLA	0.800	QSO	0
SXDS0012	0.0	0	0	9.999	NSP	1.344	GAL	0
SXDS0015	0.0	0	0	0.098	NLA	0.598	GAL	2
SXDS0016	0.0	0	1	3.512	BLA	3.298	QSO	0

*Column (19) is the X-ray name. Column (20) is the 1.4 GHz flux measured with VLA observations from Simpson et al. (2012); 0 in the first column means the object is outside of the VLA coverage. 1 (0) in the second column indicates the object is (not) detected. Column (21) is the stellarity index determined in the i^′-band image; 1 means stellar object, and 0 means extended object. Columns (22) and (23) are the spectroscopic redshift and spectroscopic type; BLA and NLA represent broad-line and narrow-line AGNs, respectively. NID indicates the object is observed in the optical but cannot be identified with the optical spectrum. NSP indicates the object is not observed in the optical. Columns (24)–(26) are photometric redshift, photometric type, and flag for photometric redshift. GAL (QSO) is the photometric type showing whether the SED of the object is fitted better with the galaxy (QSO) templates. (This table is available in its entirety in a machine-readable form in the online journal. A portion is shown here for guidance regarding its form and content.)

In the photometric redshift determination, we consider the stellarity of the primary counterpart. The stellarity is determined in the deep i^′-band image. For objects brighter than i^′ = 20.0 mag, the CLS_STAR parameter from Sextractor is used. An object with CLS_STAR larger than 0.90 is classified as stellar. For fainter objects, we use the FWHM in the i^′-band image to define stellar objects. If the i^′-band FWHM is smaller than 0|${^{\prime\prime}_{.}}$|9, an object is classified as stellar. For bright objects with saturation in their peak, the stellarity measurements are not reliable. In such cases we check the optical images by eye, and assign the corrected stellarity in the catalog. The resulting stellarity is also listed in column 21 of table 10.

6.2 Photometric redshift determination with galaxy and QSO SED templates

The photometric redshift of each primary counterpart is determined with the hyperz code (Bolzonella et al. 2000), which determines photometric redshifts via template fitting with χ² minimization of the multi-wavelength photometry ranging from 1500 Å to 8.0 μm. We use two sets of templates: one represents galaxy SEDs and the other one represents QSO SEDs.

For each object, we determine photometric redshifts from the galaxy and QSO templates separately. For the galaxy templates, we use the spectral evolutionary models of Bruzual and Charlot (1993) with the Miller and Scalo Initial Mass Function (IMF: Miller & Scalo 1979) and evolving metallicity. It needs to be noted that we assume a Salpeter IMF (Salpeter 1955) in the determination of the stellar mass of host galaxies of narrow-line AGNs in subsection 7.2 in order to compare with results in the literature. The difference in the estimated photometric redshift caused by the difference in the applied IMF is expected to be negligible. In the hyperz code, Burst, constant, and τ models with τ = 1 Gyr to 30 Gyr are available. The dust extinction curve of star-forming galaxies (Calzetti et al. 2000) is utilized with A_V = 0.0 mag to A_V = 2.5 mag in steps of δA_V = 0.5.

The galaxy templates only represent the stellar continuum component, and neither the excess emission of an AGN hot dust component nor polycyclic aromatic hydrocarbon (PAH) emission are included, although such emission can affect the rest-frame wavelength range above 2 μm. Instead of including such a component in addition to the galaxy SED model, we increase the IRAC photometric errors to compensate.

For QSO templates, we construct our own QSO template by combining average UV–optical average SEDs from SDSS and NIR–MIR SEDs of QSOs from Infrared Space Observatory (ISO) observations. We use the average QSO SEDs in Richards et al. (2003) in the wavelength range between 1200 Å and 7000 Å. A total of six average QSO SEDs are constructed depending on the UV–optical color. We extend the SED to the shorter wavelength range by extrapolating the continuum component in the above range with a power law. In the longer wavelength range above 1 μm, we used three QSO IR SEDs from Leipski et al. (2005). The SEDs are constructed from the ISO observations of broad-line QSOs. By adjusting the normalization, we connect the six optical and the three IR SEDs smoothly and make 18 QSO templates covering the UV to MIR wavelength range. Because the observed slopes in the UV–optical and IR wavelength ranges are independent, we consider SEDs with blue UV–optical SED and red IR SED, and vice versa. The templates are compared with the observed SEDs of spectroscopically identified broad-line AGNs in figure 9. All of the SEDs are normalized at 1 μm in the rest frame. The red and blue dots represent SEDs of X-ray AGNs with absorption characterized by a hydrogen column density, N_H, larger than and smaller than 10²² cm⁻², respectively. The derivation of N_H is described in subsection 7.1. The templates enclose the entire range of SEDs of broad-line AGNs, although some AGNs with log N_H > 22.0 show redder SED in the UV to optical wavelength range. Their SEDs are similar to the galaxy SEDs (see subsection 7.3). The templates also cover the variety of the SEDs observed in the X-ray-selected broad-line AGNs in the COSMOS survey (Elvis et al. 2012).

Fig. 9.

Template SEDs for QSOs (green solid lines) compared with UV to MIR SEDs of spectroscopically identified broad-line AGNs. Blue and red symbols represent broad-line AGNs with X-ray absorption smaller and larger than log N_H = 22.0, respectively. Small dots indicate upper limits. (Color online)

In the template fitting, two additional constraints are applied in addition to the χ² minimization. The first one is that an object with stellar morphology in the i^′-band image is assumed to be at z > 0.8. In the left panel of figure 10, the i^′-band magnitudes of the spectroscopically identified objects are shown as a function of redshift. Stellar objects are shown with asterisks and crosses in this figure. Out of 261 stellar objects, 209 are spectroscopically identified. A total of 182 and 27 of them are identified with broad-line and narrow-line AGNs at z > 0.8, respectively. Stellar broad-line AGNs show bright nuclei, and their observed optical light is dominated by the nuclear component with stellar morphology. It should be noted that this constraint is only applicable to the X-ray samples with similar depth and area to SXDS. As shown in subsection 7.1, in the limited area of the SXDS, there is a correlation between redshifts and luminosities, therefore luminous (log L_2–10 keV > 44) broad-line AGNs, i.e., QSOs, whose optical morphologies are dominated by the nuclear component, statistically appear only above z > 0.8. Such correlation is typical of flux-limited surveys. Less luminous broad-line AGNs are found at lower redshifts and they show extended morphologies due to their host galaxies. Stellar narrow-line AGNs are not resolved, possibly due to the fact that they are at high redshifts.

Fig. 10.

Left: Stellarity distribution of spectroscopically identified AGNs plotted on the redshift vs i^′-band magnitude plane. Blue asterisks and red crosses represent AGNs with stellar morphology in the i^′-band image. Blue (square and asterisk) and red (circle and cross) symbols show broad-line and narrow-line AGNs, respectively. The vertical dotted line marks z = 0.8. The green lines indicate the redshift evolution of the i^′-band magnitude for a 10 Gyr-old (solid) and 2 Gyr-old (dashed) single burst stellar population with stellar mass of log M_* = 10.5 (upper) and log M_* = 12.0 (lower). Right: Same for AGNs without spectroscopic identification. In this panel, AGNs are classified photometrically as described in subsection 6.2. (Color online)

An additional consideration is the absolute magnitude range of the galaxy and QSO templates. Taking into account the absolute magnitude ranges of spectroscopically identified AGNs, we limit the z-band absolute magnitude range of the galaxy (QSO) template to lie between M_z = −20.0 and −25.0 mag (M_z = −22.0 and −26.5 mag). In subsection 7.3, it will be shown that the rest-frame optical to NIR continuum of a narrow-line AGN is dominated by the stellar continuum of its host galaxy. Moreover, the range of the estimated stellar mass of the host galaxies of spectroscopically identified narrow-line AGNs is limited (see subsection 7.2). Such a limited mass range of X-ray AGN host galaxies has already been reported (e.g., Yamada et al. 2009). The absolute magnitude range of the galaxy template corresponds to a stellar mass range of 10¹⁰–10¹² M_⊙ with the galaxy SED templates used in subsection 7.2. We introduce the faint limit of the QSO absolute magnitude range, because the host galaxy component will dominate the total magnitudes if the nuclear component is fainter than the limit. These considerations are necessary to reduce the number of outliers in the photometric redshift estimation.

Once we determine photometric redshifts with both the QSO and galaxy templates for each AGNs, we compare the minimum |$\chi ^2_{\nu }$| with QSO templates, |$\chi ^2_{\nu \ {\rm QSO}}$|⁠, and that with galaxy templates, |$\chi ^2_{\nu \ {\rm GAL}}$|⁠, to choose the best photometric redshift for each AGN. In the left panel of figure 11, |$\chi ^2_{\nu \ {\rm QSO}}$| and |$\chi ^2_{\nu \ {\rm GAL}}$| are compared for spectroscopically identified AGNs. The redshift is not fixed to their spectroscopic redshift during the fitting process, in order to include the effect of redshift uncertainty. Narrow-line AGNs (red circles and crosses) on average fitted better with galaxy templates, and broad-line AGNs (blue squares and asterisks) on average fitted better with QSO templates. Crosses and asterisks represent AGNs with stellar morphology. Because the number of galaxy templates is larger than the number of QSO templates, and the average |$\chi ^2_{\nu }$| value for extended narrow-line AGNs with galaxy templates is smaller than that for stellar broad-line AGNs with QSO templates, we do not choose the best photometric redshift just by comparing |$\chi ^2_{\nu \ {\rm QSO}}$| and |$\chi ^2_{\nu \ {\rm GAL}}$|⁠, but consider the stellarity as well as the |$\chi ^2_{\nu }$| values. Considering the distribution in the panel, we regard the QSO templates as better if |$\chi ^2_{\nu \ {\rm QSO}}$| is smaller than |$2.5 \times \chi ^2_{\nu \ {\rm GAL}}$| for stellar objects. On the contrary, for extended objects, we regard the GAL template to be better if |$\chi ^2_{\nu \ {\rm GAL}}$| is smaller than |$1.5 \times \chi ^2_{\nu \ {\rm QSO}}$|⁠. These selection criteria are shown with the dashed line in figure 11.

Fig. 11.

Left: Reduced χ² values of fitting results with QSO and galaxy templates for the spectroscopically identified sample. Blue (square and asterisk) and red (circle and cross) symbols represent broad-line and narrow-line AGNs. Asterisks and crosses are for stellar objects. In order to evaluate the uncertainties in the type determination due to the photometric redshift uncertainties, the redshift is NOT fixed to the spectroscopic redshift during the fitting process. Right: Same figure for objects without spectroscopic identification. Circles and asterisks represent non-stellar and stellar objects. In both panels, the solid line shows the equal |$\chi ^2_\nu$| line and the upper and lower dashed lines represent the type determination criteria for stellar and non-stellar objects, respectively. (Color online)

Based on the comparison, we constrain not only their redshifts but also their AGN types. We classify an object whose photometric redshift is determined with the galaxy (QSO) templates as a narrow-line (broad-line) AGN, hereafter. We can examine the reliability of the photometric classification with the template-fitting results for the AGNs with spectroscopic classification. As described above, we do not fix their redshifts with spectroscopic redshifts during the fitting process. Among the 302 narrow-line (284 broad-line) AGNs with spectroscopic identification, 267 (206) of them are correctly classified as a narrow-line (broad-line) AGN in the photometric classification. Because less luminous broad-line AGNs have extended morphology, a fraction of them are fitted better with the galaxy templates and photometrically classified as narrow-line AGNs. Their optical continuum can be dominated by the host galaxy component—see subsection 7.3. Those mis-classifications lower the rate of correct classification for the broad-line AGNs. If we concentrate on the 180 broad-line AGNs with stellar morphology, 170 are correctly classified as a broad-line AGN in the photometric classification.

The derived photometric redshifts and classification are summarized in columns 24 and 25 of table 10, respectively. Column 26 describes flags for the photometric redshift determination: flag = 1 means the reduced χ² is larger than 10, flag = 2 means photometric redshift is determined with unreliable photometric data for a bright and extended galaxy, and flag   3 means photometric redshift is determined with less than four bands. The numbers of X-ray AGNs with each flag are summarized in table 7.

6.3 Photometric redshift results

The reliability of the photometric redshifts is examined by comparing them with the spectroscopic values. A comparison of the photometric and spectroscopic redshifts of spectroscopically identified AGNs is shown in figure 12. We separate the sample into four categories using the stellarity and the best-fit template. The difference between the spectroscopic and photometric redshifts is defined as dz ≡ (z_phot − z_spec), and the bottom part of the panels show dz/(1 + z_spec). Photometric redshifts for objects that are fitted better with galaxy templates have a small scatter (shown in the left panels), but the scatter is larger for objects fitted better with QSO templates (shown in the right panels). The distribution of dz/(1 + z_spec) is shown in figure 13. There is no systematic difference between the spectroscopic and photometric redshifts; the median of of dz/(1 + z_spec) is −0.026 for all the AGNs in the total sample, and −0.038 and −0.022 for broad-line and narrow-line AGNs, respectively. The scatter of the difference is examined with the normalized median absolute deviation (NMAD; σ_z) following Brammer, van Dokkum, and Coppi (2008),

\begin{equation} \sigma _{\rm NMAD} = 1.48 \times {\rm median} \left[ \left| \frac{ {\rm d}z - {\rm median}({\rm d}z) }{1+z_{\rm spec}} \right| \right]. \end{equation}

(3)

σ_NMAD corresponds to the standard deviation of a Gaussian distribution, and is less sensitive to the outliers than the usual definition of the standard deviation (Brammer et al. 2008). For the total sample, σ_NMAD is 0.098, which is larger than that of the photometric redshift estimations for X-ray-selected AGNs with medium band filters (Salvato et al. 2009; Cardamone et al. 2010; Luo et al. 2010). For example, Luo et al. (2010) report σ_NMAD of 0.059 for a blind test sample. Because we do not train the photometric redshift code with SEDs obtained in the spectroscopic redshift sample, our σ_NMAD should be compared with their blind test results. The larger σ_NMAD can be explained by the lack of photometric data with medium band filters. The σ_NMAD for broad-line AGNs (0.109) is larger than that for narrow-line AGNs (0.072), as expected from figure 13. This is because there is no strong feature in the SEDs of the broad-line AGNs except for the break below Lyα. Since the break can only be covered by deep optical observations for AGNs at z > 3, the discrepancy between z_spec and z_phot is smaller for AGNs at z > 3.

Fig. 12.

Spectroscopic redshift vs photometric redshift for objects with spectroscopic identification. Blue squares and crosses and red circles and asterisks represent broad-line and narrow-line AGNs. The top panels show the results for extended objects (squares and circles) and the left (right) panel is for objects fitted better with the GAL (QSO) template. The bottom panels are the same as the upper panels for stellar objects (crosses and asterisks). The upper part of each panel shows spectroscopic redshift vs photometric redshift and the bottom part is dz/(1 + z_spec) as a function of z_spec. (Color online)

Fig. 13.

Distribution of dz/(1 + z_spec) for all AGNs (top panel), broad-line AGNs (middle), and narrow-line AGNs (bottom).

Photometric redshift determination is crucial to understand the nature of the X-ray-selected AGNs. The redshift distribution of the AGN sample is summarized in figure 14. For narrow-line AGNs, the redshift distribution of spectroscopically identified objects is different from that of objects without spectroscopic identification; they are located at higher redshifts on average. As shown in the right panel of figure 10, among the 310 primary candidates without spectroscopic identification, 258 objects have extended morphology. Photometric redshift estimation suggests that they are narrow-line AGNs at z ≥ 1. The green lines in the figure indicate the redshift evolution of the i^′-band magnitude of 10 Gyr (solid) and 2 Gyr (dashed) single burst stellar population with stellar mass of log M_* = 10.5 (upper) and log M_* = 12.0 (lower). The i^′-band magnitudes of most of the spectroscopically identified narrow-line AGNs are enclosed within the green lines, and the i^′-band magnitudes of narrow-line AGNs with photometric redshifts are also enclosed within the lines.

Fig. 14.

Redshift distributions of the AGNs in the soft-band (hard-band) sample in the left (right) row. The thin solid histogram indicates the redshift distribution of spectroscopic plus photometric redshift samples, the thick solid histogram shows spectroscopic samples only. Shown from top to bottom are the redshift distributions of all X-ray sources, broad-line AGNs, and narrow-line AGNs.

The photometric redshift results suggest that there are 45 broad-line AGN candidates that have not been spectroscopically identified, and 86% of the broad-line AGNs are thought to be spectroscopically identified. Among broad-line AGN candidates without spectroscopic identification, seven objects have a photometric redshift above 3, where photometric redshift estimation is reliable even for broad-line AGNs.

7 Discussion

7.1 X-ray luminosity and absorption

In order to examine the physical properties of the X-ray AGNs, at first, we evaluate the intrinsic 2–10 keV luminosity, L_2–10 keV (erg s⁻¹), and the column density of photoelectric absorption to the nucleus, N_H (cm⁻²), of each AGN. In order to evaluate L_2–10 keV, it is necessary to correct for the effect of photoelectric absorption using the N_H value. The number of X-ray photons of each AGN in the current sample is not sufficient to evaluate N_H for an individual AGN accurately with spectral fitting. Thus we estimate N_H for each AGN from its HR2 value. It needs to be noted that the relation between HR2 and N_H is non-linear (Ueda et al. 2008), and the uncertainty associated with N_H can be large, especially around an HR2 of ±1. Furthermore, the relation varies with redshift: at higher redshifts N_H is a steeper function of HR2, and thus the uncertainty is larger than the AGNs at low redshifts with same uncertainty in HR2. Due to these limitations, a statistical analysis, like that applied in Hiroi et al. (2012) and Ueda et al. (2008), is necessary to derive the distribution function of N_H and the hard X-ray luminosity function. Therefore, in this paper, we use L_2–10 keV and N_H to evaluate average properties of a population of AGNs (subsection 7.2) and to construct a statistical sample of AGNs sub-divided by their X-ray properties (subsection 7.3).

In order to estimate N_H from HR2, we assume an intrinsic shape of the X-ray spectrum of AGNs following Ueda et al. (2003). The intrinsic X-ray spectrum of AGNs is modeled by a combination of a power-law component with high-energy cutoff E^−Γ × exp (−E/E_c) and a reflection component. A photon index Γ of 1.9 and cutoff energy E_c of 300 keV are assumed. We calculate the reflection component with the “pexrav” (Magdziarz & Zdziarski 1995) model in the XSPEC package, applying a solid angle of 2π, an inclination angle of cos i = 0.5, and solar abundance of all elements. The strength of the reflection component is about 10% of the direct component just below 7.1 keV. For the assumed intrinsic X-ray spectrum, we ignore any scattered light component and the luminosity dependence of the strength of the reflection component for simplicity. We only consider log N_H up to 24.0. For Compton-thick AGNs whose X-ray spectrum is affected by a scattered component, N_H can be underestimated due to the soft X-ray spectrum of that component. However, the number of such AGNs should be small in the X-ray flux range covered by the current sample (Ueda et al. 2014). If HR2 is smaller than that expected for Γ = 1.9, we determine the corresponding photon index by assuming no absorption. If an object is only detected in the soft-band, we assign Γ = 1.9 and we neglect their log N_H. For objects only detected in the hard-band, we assign log N_H = 24.0 and Γ = 1.9.

L  _2–10 keV is derived from the observed 2–10 keV count rate assuming the intrinsic spectrum discussed above and correcting for the effect of the photoelectric absorption with the best estimated N_H. For objects only detected in the 0.5–2 keV band, the count rate in that band is used instead of the 2–10 keV count rate to derive L_2–10 keV. The best-estimated L_2–10 keV are compared with L_2–10 keV estimated simply assuming a power law of Γ = 1.8, typical apparent spectra of non-absorbed AGNs, in figure 15. We use the count rate in the 2–10 keV band in the estimation, but for objects only detected in the 0.5–2 keV band, we use the count rate in the 0.5–2 keV band and show them with open symbols. Because the observed 2–10 keV energy range covers a higher energy range in the object rest frame, for most of the X-ray sources, the amounts of absorption correction are not large. Therefore, the statistical conclusions derived using the best-estimated L_2–10 keV in later sections do not significantly change even if we use the L_2–10 keV derived simply with Γ = 1.8.

Fig. 15.

Best-estimated L_2–10 keV vs L_2–10 keV estimated simply assuming a power law with Γ = 1.8. Blue squares and red circles represent broad-line and narrow-line AGNs, respectively. Open symbols represent objects that are not detected in the 2–10 keV band and L_2–10 keV is estimated from the 0.5–2 keV count rate. (Color online)

The best-estimated L_2–10 keV and N_H are shown as a function of redshift in the left and right panels of figure 16. In the right panel, we show typical uncertainties in log N_H for objects with log N_H ≃ 22.0. They are determined by the median uncertainties of objects with the best-fit log N_H = 21.5–22.5 at z = 0.5–1.0, 1.0–1.5, 1.5–2.0, and 2.0–2.5. The uncertainty of each object is determined from the 1 σ statistical error in HR2.

Fig. 16.

Left: Redshift vs luminosity of the broad-line (blue squares) and narrow-line AGNs (red circles). Open symbols represent objects with photometric redshift. Right: Redshift vs log N_H (cm⁻²). Same symbols as in the left panel. log N_H is derived from the HR2 value. AGNs only detected in the soft- (hard-)band are plotted below log N_H = 20.0 (above log N_H = 24.0) with an arbitrary shift for clarity. Error bars at log N_H = 22.0 represent medians of uncertainties for objects with log N_H = 21.5–22.5 at z = 0.5–1.0, 1.0–1.5, 1.5–2.0, and 2.0–2.5. (Color online)

The SXDS AGN sample covers a wide range of the hard X-ray luminosity, from 10⁴² to 10⁴⁶ erg s⁻¹, and encloses the knee of the hard X-ray luminosity function at z = 0.4–4.0 (log L_2–10 keV ∼ 44 at z ≤ 1.0 and log L_2–10 keV ∼ 44.6 at z ≥ 2.0; Ueda et al. 2014). There is an apparent relation between the redshift and luminosity due to the flux limit of the sample.

The distribution of N_H is broadly consistent with the optical classification on the basis of the existence of broad-emission lines or by SED modeling: broad-line AGNs have smaller N_H on average. In the redshift range below 0.6, almost all of the AGNs with log N_H < 22 lack broad-emission lines. For these AGNs, it is likely that the broad-emission line component may be diluted by the host galaxy component. In table 11, columns 2 and 3 show the redshift and HR2 values assumed to derive the best-estimated log N_H, Γ, and log L_2–10 keV. Derived values are shown in columns 4–6. “—” in column 4 means log N_H is not given because HR2 is smaller than expected for a power law of Γ = 1.9, or the object is only detected in the soft-band.

Table 11.

X-ray properties of AGN candidates.

ID	z	HR2	log NH	Photon index	log L2–10 keV
			(cm−2)		(erg s−1)
(1)	(2)	(3)	(4)	(5)	(6)
SXDS0001	0.505	−0.42	21.9	1.9	43.2
SXDS0002	0.700	−0.64	20.8	1.9	43.5
SXDS0003	0.622	−0.34	22.1	1.9	43.6
SXDS0004	0.496	−1.00	—	1.9	42.7
SXDS0005	1.050	−0.24	22.5	1.9	44.1
SXDS0008	1.033	−0.62	21.3	1.9	43.7
SXDS0010	1.221	−0.72	—	2.1	44.1
SXDS0012	1.344	−0.36	22.5	1.9	44.0
SXDS0015	0.098	−0.62	20.8	1.9	42.3
SXDS0016	3.512	−0.60	21.9	1.9	44.5

*Column (1) is the X-ray name. Columns (2) and (3) are the redshift and HR2 used to calculate the best-estimated X-ray properties. HR2 of −1.00 (1.00) means the source is only detected in the soft- (hard-)band. Column (4) is log N_H, column (5) is Γ, and column (6) is log L_2–10 keV. “—” in column (4) means log N_H is not given, because HR2 is smaller than expected for a power law of Γ = 1.9. Γ of 1.9 means the value is fixed. Details of determination of the X-ray properties are described in subsection 7.1. (This table is available in its entirety in a machine-readable form in the online journal. A portion is shown here for guidance regarding its form and content.)

7.2 Stellar mass of narrow-line AGNs and the M_*–L_X relation

We derive the stellar mass of host galaxies of narrow-line AGNs, M_*, by assuming that their UV to NIR SEDs are dominated by the host galaxy component. The similarity of the SEDs of narrow-line AGNs to those of non-X-ray galaxies (e.g., Kiuchi et al. 2009) supports this assumption. The similarity will be discussed further in subsection 7.3. We apply SED template fitting with a fixed redshift which is determined either spectroscopically or photometrically, and we use a larger number of SED templates than those used in the photometric redshift estimation in order to cover the variety of SEDs of galaxies accurately. Furthermore, the effect of the hot dust component above the 1 μm wavelength range is considered in the templates as described below.

We construct the SED templates using PEGASE.2 stellar population synthesis code (Fioc & Rocca-Volmerange, 1997). A Salpeter IMF with mass range of 0.1–120 M_⊙ is assumed. SED templates with burst and continuous star formation histories are constructed with a simple exponentially declining star formation rate proportional to exp (−t/τ). t represents the age of the stellar system and τ represents the decay time scale of the star formation. We use τ of 0.5 Gyr and 2.0 Gyr for the burst and continuous models. We use ages of 1.0, 2.0, 2.5, 3.0, 3.5, 4.0, 4.5, 5.0, 6.0, 8.0, and 10.0 Gyr for the burst model and 0.5, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0 , 8.0, and 10.0 Gyr for the continuous model. We considered three metallicities, 0.5, 1.0, and 2.0 times the solar metallicity. We apply reddening with the Calzetti extinction curve (Calzetti et al. 2000) and consider color excess E(B − V) with 0.1 mag steps up to 0.9 mag. In addition to the simple exponentially declining and continuous models, we construct templates by combining the two models with a stellar mass ratio of 0.0, 1.0, and 50.0, representing the mixture of the bulge and disk systems of galaxies. In the wavelength range above 1 μm, a significant fraction of narrow-line AGNs show an excess compared to the galaxy templates due to the hot dust component. In order to reproduce the excess, we construct galaxy templates with an additional power-law component representing the excess above 1 μm. Figure 17 shows samples of the SED fitting results for narrow-line AGNs at z ∼ 1 in order of the strength of the excess component from the top-left panel with no excess to the bottom-right panel with the strongest excess. The excess components with various strengths that can be seen above the observed wavelength of 2 μm are well reproduced with the best-fit templates.

Fig. 17.

Sample SED fitting results for narrow-line AGNs at z∼1 in order of the strength of the excess component above 1 μm from top-left to bottom-right panels. Filled squares indicate observed magnitudes at each observed wavelength and arrows show upper limits. Open squares represent magnitudes of the best-fit SED model.

Furthermore, for galaxies at z < 0.6, the 8.0 μm photometry is affected by the 8 μm PAH feature, which is not included in the galaxy templates. We do not use 8.0 μm photometry for galaxies at z < 0.6. We do not consider the constraints on the absolute magnitude in the fit.

The resulting stellar masses of host galaxies of narrow-line AGNs are shown in figure 18. Although the SXDS sample covers a wide range in luminosity and redshift, the estimated M_* values are remarkably constant. If we divide the sample into three redshift bins with 0.1 < z < 1.0 (low), 1.0 < z < 2.0 (mid), and 2.0 < z (high), the median M_* increases by 0.18 dex from the low to the high redshift bins, although the median L_2–10 keV increases by 1.4 dex. The distributions of the M_* in the three redshift bins are compared with each other in figure 19. The probabilities that a pair of distributions is drawn from the same sample are examined with the Kolmogorov–Smirnov test, and they are 1.0%, 4.7%, and 28% with the low–mid, low–high, and mid–high redshift samples. The mass range of the host galaxies is broadly consistent with those for various X-ray-selected samples of narrow-line (e.g., Akiyama 2005; Alonso-Herrero et al. 2008; Kiuchi et al. 2009; Brusa et al. 2009; Yamada et al. 2009; Xue et al. 2010; Mainieri et al. 2011; Rosario et al. 2013) and broad-line AGNs (e.g., Jahnke et al. 2009; Merloni et al. 2010; Bongiorno et al. 2014) at redshifts up to 4. It needs to be noted that we consider the absolute magnitude range of the galaxy template in the photometric redshift determination. The absolute magnitude range can affect the distribution of M_* of narrow-line AGNs without spectroscopic redshifts, however, only 15 out of the 259 narrow-line AGNs without spectroscopic redshifts are significantly affected by the faint limit of the range and have absolute magnitude close to the limit. Some narrow-line AGNs have M_* less than 10¹⁰ M_⊙. Their stellar mass estimation would be affected by scattered nuclear light in the UV. Such contamination with a blue continuum possibly makes the SED appear similar to a younger stellar population with a smaller mass-to-light ratio, and the estimated stellar mass can thus appear smaller even with the rest-frame NIR photometry constraints.

Fig. 18.

Estimated stellar mass vs 2–10 keV luminosity for narrow-line AGNs. Symbols are coded with redshift: blue triangles are for z = 0.1–1.0, green squares are for z = 1.0–2.0, and red circles are for z > 2.0. Spectroscopically identified narrow-line AGNs are plotted with large open symbols, and photometric narrow-line AGN candidates are plotted with small filled symbols. Photometric candidates with phot-z flags of 0 or 1 are plotted. Big crosses indicate the median values in the three redshift ranges. Solid lines indicate the stellar mass and 2–10 keV luminosity relation assuming the M_BH–M_* relation from Bennert et al. (2011) and bolometric correction of Marconi et al. (2004). From bottom to top, an Eddington ratio of 1, 0.1, 0.01, and 0.001 are assumed. Dotted and dashed lines indicate the relations derived with the M_BH–M_*,bulge relations of Sani et al. (2011) and Marconi and Hunt (2003), respectively, for an Eddington ratio of 1. (Color online)

Fig. 19.

Distribution of M_* for narrow-line AGNs. The solid histogram is for the entire sample, and blue dotted, green dashed, and red dot-dashed histograms represent the distribution of 0.1 < z < 1.0, 1.0 < z < 2.0, and 2.0 < z narrow-line AGNs, respectively. The numbers for the entire sample are divided by 2 for clarity. (Color online)

The solid lines in figure 18 indicate the relationship between the stellar mass and X-ray luminosity if we assume the local black hole and host galaxy stellar mass (M_BH–M_*) relation from Bennert et al. (2011) and the AGN bolometric correction for 2–10 keV X-ray luminosity from Marconi et al. (2004). The lines from bottom to top correspond to an Eddington ratio of 1, 0.1, 0.01, and 0.001. Bennert et al. (2011) derive the relationship for broad-line AGNs following Häring and Rix (2004) with a Chabrier IMF (Chabrier 2002). We convert their relation to that with the Salpeter IMF, considering that M_* is 0.255 dex heavier with the Salpeter IMF. We also plot the M_* and L_2–10 keV relation based on the black hole and bulge stellar mass, M_BH–M_*,bulge, and relations from Sani et al. (2011) and Marconi and Hunt (2003) with dotted and dashed lines, respectively. They are consistent with the relation derived with the M_BH–M_* relation of Bennert et al. (2011) around M_* = 10¹² M_⊙, and they show steeper dependence on L_2–10 keV than the relation. The difference below M_* = 10¹² M_⊙ can be explained with the difference between M_* and M_*,bulge. The constant M_* of narrow-line AGNs with wide luminosity and redshift ranges implies higher λ_Edd in the higher redshift bins; a significant fraction of narrow-line AGNs at z > 2 possibly have L_2–10 keV close to or above the Eddington limit.

On the other hand, the distribution of narrow-line AGNs on the M_*–L_2–10 keV plane can be explained with the cosmological evolution of the M_BH–M_* relationship with higher M_BH/M_* at higher redshifts (e.g., McLure et al. 2006; Peng et al. 2006; Decarli et al. 2010; Bennert et al. 2011; Schramm & Silverman 2013), and λ_Edd for the narrow-line AGNs at high redshifts may be overestimated. Assuming that the λ_Edd and L_bol relation that is derived with broad-line AGNs at 0.5 < z < 2.3 in the same SXDS X-ray sample [equation (16) of Nobuta et al. 2012] is applicable to the narrow-line AGNs, we estimate M_{BH, L} of the narrow-line AGNs from L_2–10 keV and the bolometric correction of Marconi et al. (2004). Figure 20 shows the difference between the estimated M_{BH, L} and the |$M_{{\rm BH},M_{*}}$|⁠, which is calculated from M_*, and the local M_BH–M_* relationship from Bennert et al. (2011). The difference implies the cosmological evolution of the M_BH–M_* relationship with larger M_BH compared to M_* at higher redshifts can explain the observed distribution on the M_*–L_2–10 keV plane; the 0.1 < z < 1.0 narrow-line AGNs have −0.23 dex smaller M_BH than that expected from M_*, and the z > 2.0 narrow-line AGNs have 0.63 dex larger M_BH. If we simply fit the distribution, cosmological evolution of the M_BH–M_* relationship with (1 + z)^2.6 is required. If we take it at face value, the derived cosmological evolution of the M_BH–M_* relation is stronger than that in Bennert et al. (2011), shown with a dotted line in the figure. In order to derive the cosmological evolution of the M_BH–M_* relationship reliably, we need to consider the bias associated with the luminosity-limited sample (Schulze & Wisotzki 2014). Such detailed analysis is beyond the scope of this paper, and will be discussed elsewhere.

Fig. 20.

Difference between M_BH derived from L_2–10 keV and the λ_Edd–L_bol relation (Nobuta et al. 2012) and M_BH derived from M_* and the local M_BH–M_* relation (Bennert et al. 2011). Symbols are coded with redshift: blue triangles are for z = 0.1–1.0, green squares are for z = 1.0–2.0, and red circles are for z > 2.0. Spectroscopically identified narrow-line AGNs are plotted with large open symbols, and photometric narrow-line AGN candidates are plotted with small filled symbols. Photometric candidates with phot-z flags of 0 or 1 are plotted. Big crosses indicate the median values in the three redshift ranges. The dotted line indicates the redshift dependence of the difference with cosmological evolution of the M_BH–M_* relation with (1 + z)^1.15 (Bennert et al. 2011). (Color online)

7.3 SEDs of X-ray-selected AGNs

Utilizing the multi-wavelength photometric data of the large sample of X-ray-selected AGNs, we examine the scatter of their UV–MIR SEDs as a function of L_2–10 keV and N_H, and compare the range of the scatter with that of galaxies without X-ray AGN activity. We choose galaxies which have the same mass and redshift distributions as the X-ray AGNs. Because X-ray AGNs only appear among massive galaxies, it is important to compare the properties of AGNs with those of galaxies without X-ray AGN activity in the same mass range (Silverman et al. 2009; Xue et al. 2010).

We construct a sample of galaxies without X-ray AGN activity in the overlapping region of the deep Suprime-Cam optical and UDS NIR imaging data. As a first step the photometric redshift is determined with the FUV to MIR photometric data for each object in the same way as for the X-ray AGNs as described in section 6. We do not apply a QSO template to the galaxies and do not consider the constraints on the absolute magnitude. Once photometric redshifts are determined, we evaluate the stellar masses of the galaxies, applying the model described in subsection 7.2. In order to construct the sample of galaxies without X-ray activity, we select galaxies that have similar stellar mass and redshift distribution of the X-ray AGNs. For each X-ray AGN, we select galaxies without X-ray detection within dz = ±0.2 of the redshift of the AGN. We assume broad-line AGNs follow the same host galaxy stellar mass distribution shown in figure 19. Within the redshift range, we randomly select 14 galaxies for each AGN following the mass distribution of 1 in the range 9.5 < log M_* < 10.0, 2 in the range 10.0 < log M_* < 10.5, 5 in the range 10.5 < log M_* < 11.0, 5 in the range 11.0 < log M_* < 11.5, and 1 in the range 11.5 < log M_* < 12.0. The mass distribution roughly reproduces that of the total sample of the narrow-line AGNs. In the redshift range between 1.0 and 2.0, there are 12248 (9.5 < log M_* < 10.0), 9140 (10.0 < log M_* < 10.5), 5899 (10.5 < log M_* < 11.0), 2557 (11.0 < log M_* < 11.5), and 310 (11.5 < log M_* < 12.0) galaxies that are not detected in X-ray in the mass range shown in parentheses. We allow overlap in the selection, and a galaxy can be selected multiple times. We refer to this sample of galaxies as the matched sample, hereafter.

The SEDs of X-ray AGNs are shown with dots in figure 21. From top to bottom, the sample is subdivided by X-ray luminosity: 44.0 < log L_2–10 keV, 43.0 < log L_2–10 keV < 44.0, and log L_2–10 keV < 43.0, and from left to right, the sample is subdivided by X-ray absorption to the nucleus: log N_H < 22.0 and 22.0 < log N_H. Blue and red symbols represent broad- and narrow-line AGNs, respectively. All of the SEDs are normalized at 1 μm, and each photometric point is located at its rest-frame effective wavelength. It is clear that the luminous and less-absorbed AGNs show SEDs within the range of the SEDs of the optically selected QSOs (green dashed lines).

Fig. 21.

Rest-frame UV to MIR SEDs of X-ray AGNs. All SEDs are normalized at 1 μm in the rest frame. Broad-line and narrow-line AGNs are plotted with blue and red symbols, respectively. Small dots indicate upper limits. The thick black solid and dotted lines represent the median (50th percentile) SED of galaxies in the matched sample for each luminosity bin. Dotted parts represent upper limits, i.e., less than 50% of the galaxies are detected in the wavelength range. Thin black solid and dotted lines show the 10th percentile (upper) and 90th percentile (lower) SED of the galaxies, and dotted lines represent upper limits. Green dashed lines indicate the range of the template QSO SEDs used in the photometric redshift determination. The X-ray AGN sample is subdivided by X-ray luminosity and log N_H: from top to bottom, the panels are for AGNs with 44.0 < log L_2–10 keV, 43.0 < log L_2–10 keV < 44.0, and log L_2–10 keV < 43.0, and from left to right, log N_H < 22.0 and 22.0 < log N_H. The typical redshift range of each AGN sample is indicated in parentheses. (Color online)

Less luminous or absorbed AGNs show redder UV–optical SEDs, which are similar to those of galaxies in the matched sample. The thick black solid and dotted lines in the figure show the median (50th percentile) value of the 1 μm-normalized SEDs of the galaxies in the matched sample. We construct the matched sample for each luminosity bin. In order to determine the median value at each wavelength, we linearly interpolate the photometric data points of each galaxy in the log νFν–log λ plane. In the percentile calculation, we include photometric data points which are upper limits. In the wavelength range where less than 50% of galaxies in the sample are detected, the derived median value is an upper limit, and we plot this with the dotted line. The thin black solid and dotted lines represent the 10th percentile and 90th percentile from top to bottom at each wavelength range.

In order to quantitatively compare the range of the SEDs, we also construct the percentile SEDs of X-ray AGNs and show them in figure 22. The SED of each AGN is interpolated in the same way as for the galaxies, and is normalized at 1.0 μm. At each wavelength we derive from brightest 10th percentile to the 90th percentile SEDs at 10% intervals. If the fraction of detected objects at a certain wavelength is less than the corresponding percentile, we regard the percentile SED as an upper limit and plot the SED with dotted lines.

Fig. 22.

Rest-frame UV to MIR percentile SEDs of X-ray AGNs (red lines). All SEDs are normalized at 1 μm in the rest frame. From top to bottom, 10th percentile to 90th percentile SEDs are shown with 10% interval. Thick solid and dotted lines indicate the median (50th percentile) SED. Dotted lines represent upper limits. The thick black solid and dotted lines represent median SEDs of the galaxies in the matched sample for each luminosity bin. The thin black solid and dotted lines show 10th percentile and 90th percentile SEDs. Dotted parts indicate upper limits. Green dashed lines indicate the range of the template QSO SEDs used in the photometric redshift determination. The X-ray AGN sample is subdivided in the same way as in figure 21. (Color online)

At wavelengths below 1 μm where the SEDs of broad-line AGNs are dominated by a blue power-law component, most of the broad-line AGNs have bluer SEDs than the galaxies in the matched sample, especially in the upper-left highest luminosity and lowest absorption bin: 80% of the X-ray AGNs have bluer SEDs than the median SED of the galaxies in the matched sample. The almost constant median SED of the most luminous less-absorbed AGNs is consistent with the median pre-host subtraction SEDs of broad-line AGNs derived with the COSMOS dataset (Elvis et al. 2012; Hao et al. 2014). If we compare the 90th percentile range of Elvis et al. (2012), the upper envelope is consistent, but the current sample has larger scatter in the lower envelope due to the inclusion of narrow-line AGNs. The SEDs of X-ray AGNs become redder with lower luminosity or higher N_H. In the lowest luminosity bin, even with the low N_H, the median SED of the AGNs is similar to that of galaxies in the matched sample. As can be seen in figure 21, almost all of them are narrow-line AGNs, and they do not show strong broad-emission lines in their observed spectra.

The median SEDs of the absorbed (log N_H > 22.0) X-ray AGNs have similar SEDs to those of the galaxies in the matched sample. For example, the rest-frame U − V color of the median SED of AGNs with log N_H > 22.0 and 43.0 < log L_2–10 keV < 44.0 is only 0.05 mag bluer than the galaxies in the matched sample. The similarity of the U − V colors of the AGN host galaxies and the galaxies in the matched sample is consistent with the results obtained in the Chandra Deep Fields (Xue et al. 2010), and does not support the distribution of AGN host galaxies in a distinct region of color–magnitude space (e.g., Nandra et al. 2007).

In the wavelength range above 1 μm, the SEDs of the AGNs are thought to be dominated by ≤ 1000 K hot thermal emission from a dusty torus, and are roughly constant on the ν–L_ν plane. On the other hand, the SEDs of galaxies without X-ray AGN activity are dominated by the stellar continuum, which shows a 1.6 μm bump and continuous decrease at longer wavelengths. PAH 8 μm and cold dust emissions from star-forming regions contribute in the wavelength range longward of 5 μm. As can be seen in figure 21, AGNs with 44 < log L_2–10 keV are distributed around a constant SED above 1 μm for both the log N_H < 22.0 and >22.0 samples, and the SEDs are consistent with the observed range of broad-line AGNs shown with green dashed lines. For AGNs with lower luminosity, the contribution from the hot dust component decreases and the SEDs are bluer and closer to those of galaxies peaked at around 1.6 μm.

The galaxies in the matched sample do not show an excess component above 1 μm in the lowest luminosity bin, which samples galaxies in the low redshift range, shown in the bottom panels. On the contrary, in the highest luminosity bin, which samples galaxies in the higher redshift range, shown in the top panels, the median SED of the galaxies in the matched sample show a constant SED in this wavelength range. As a result, even though in the middle and high luminosity ranges more than 70%–80% of X-ray AGNs show excess component compared to the median SED of the galaxies in the matched sample, the 10th percentile SEDs of X-ray AGNs are similar to those of the galaxies in the matched sample.

Several AGN selection methods based on the detection of hot thermal emission by multi-band photometric data are proposed with the Spitzer (Lacy et al. 2004; Ivison et al. 2004; Stern et al. 2005; Donley et al. 2007, 2012; Daddi et al. 2007; Fiore et al. 2008), AKARI (Takagi et al. 2010; Hanami et al. 2012), and Wide-field Infrared Survey Explorer (WISE: Mateos et al. 2012) datasets. The similarity of the SEDs of AGNs with log L_2–10 keV > 43 to those of the galaxies in the matched sample suggests that AGN selection based on detecting >1.6 μm excess emission above the stellar continuum can be contaminated by galaxies without X-ray AGN activity at z > 1 (e.g., Georgakakis et al. 2010; Alexander et al. 2011; Donley et al. 2012). It needs to be noted that the SEDs above rest-frame 3 μm in the current study are based on broad-band photometry only at 8 and 24 μm, and detailed continuum structure can be missed.

The constant SEDs of galaxies without X-ray AGN activity at z > 1 can be caused by high star-formation activity associated with massive galaxies whose star-formation activity is strongly affected by dust. The cosmological evolution of the M_*–star-formation rate relation suggests galaxies with log M_* = 10.5 at z = 1.0 and z = 2.0 have 7 and 24 times higher star-formation rate than those at z = 0.0 on average (Elbaz et al. 2011). Furthermore, star-formation activity in massive galaxies is strongly affected by dust (Pannella et al. 2015). Donley et al. (2007) show SEDs of IR-excess galaxies can be reproduced with the SED of a local star-forming galaxy with additional reddening with A_V = 0.8 mag. A larger contribution from thermal dust emission than for local massive galaxies can explain the constant SEDs of the non-X-ray galaxies at z > 1 (Georgakakis et al. 2010). Such evolution of SEDs of galaxies may affect the discussion of MIR SEDs of torus components of the broad-line AGNs (Lusso et al. 2013).

8 Summary

We report the multi-wavelength identification of the X-ray sources found in the SXDS field using deep imaging data covering the wavelength range between far-UV and mid-IR. Among the 1045 X-ray sources detected in the seven FoVs of XMM-Newton (Ueda et al. 2008), we consider 945 X-ray sources that are detected above likelihood 7 in either soft- or hard-band and covered by the deep Suprime-Cam imaging observations. We select a primary counterpart for each X-ray source applying the likelihood ratio method to R-band, 3.6 μm, near-UV, and 24 μm source catalogs. We also consider the matching catalogs with AGN candidates selected in 1.4 GHz radio and i^′-band variability surveys.

Once candidates of Galactic stars, ULXs in a nearby galaxy, and clusters of galaxies are removed there are 896 AGN candidates in the sample. We conduct spectroscopic observations of 596 and 851 primary counterparts in the optical and NIR wavelengths, respectively, and spectroscopically identify 65%, 597, of the total sample. For the remaining X-ray AGN candidates, we evaluate their photometric redshift with photometric data in 15 bands. In the photometric redshift determination, we apply SED templates of QSOs and galaxies separately to each primary counterpart, and select the best-fit photometric redshift considering the i^′-band stellarity.

Utilizing the multi-wavelength photometric data of the large sample of X-ray-selected AGNs, we examine M_* of the host galaxies of the narrow-line AGNs. The median M_* in the three redshift bins, 0.1 < z < 1.0, 1.0 < z < 2.0, and 2.0 < z, only increases 0.18 dex from the low to the high redshift bins, although the median L_2–10 keV increases 1.4 dex in the redshift range. If we consider the M_BH–M_* relation in the local Universe, the constant M_* implies higher λ_Edd for luminous AGNs at high redshifts. On the other hand, if we assume the λ_Edd and L_bol relation derived for broad-line AGNs, we can estimate M_BH of the narrow-line AGNs. The constant M_* can also be explained with strong redshift evolution of the M_BH–M_* relation from that determined in the local Universe.

Finally, the scatter of the rest-frame UV–MIR SEDs of X-ray AGNs is examined as a function of L_2–10 keV and N_H. The scatter is compared with the scatter of the SEDs of the galaxies without X-ray detection. The comparison sample of galaxies are selected to have redshift and M_* distributions matched to the X-ray AGNs. The UV–NIR SEDs of X-ray AGNs with log N_H > 22.0 are remarkably similar to those of the galaxies in the matched sample. In the NIR–MIR wavelength range, the median SEDs of X-ray AGNs are redder than those of the galaxies in the matched sample, and luminous AGNs show constant median SED in the wavelength range, suggesting the existence of the hot dust component. However, the scatter of the SEDs of the luminous X-ray AGNs broadly overlaps with that of the galaxies in the matched sample. The galaxies in the matched sample of the luminous AGNs also show a more constant SED than the galaxies in the matched sample of the less luminous AGNs, which are at lower redshifts than the luminous AGNs on average. A higher star-formation rate in massive galaxies at z = 1–3 can explain the constant SED of the galaxies in the matched sample. Due to the similarity of the SEDs of X-ray AGNs to those of the galaxies in the matched sample, it is expected that AGN selection criteria looking for excess in the NIR–MIR wavelength range can be significantly contaminated by non-AGN galaxies.

We warmly thank the staff members of the Subaru telescope for their support during the observations. We would like to thank the anonymous referee for valuable comments that improved this paper. MA is supported by JSPS Grant-in-Aid for Young Scientist (B) (18740118), and MA and HH are supported by Grant-in-Aid for Scientific Research (B) (21340042). HH acknowledges the Grant-in-Aid for Challenging Exploratory Research (24650145) from JSPS, which partially supported this research.

This publication makes use of data products from the Two Micron All Sky Survey, which is a joint project of the University of Massachusetts and the Infrared Processing and Analysis Center/California Institute of Technology, funded by the National Aeronautics and Space Administration and the National Science Foundation.

Based on observations obtained with MegaPrime/MegaCam, a joint project of CFHT and CEA/IRFU, at the Canada–France–Hawaii Telescope (CFHT) which is operated by the National Research Council (NRC) of Canada, the Institut National des Science de l'Univers of the Centre National de la Recherche Scientifique (CNRS) of France, and the University of Hawaii. This work is based in part on data products produced at Terapix available at the Canadian Astronomy Data Centre as part of the Canada–France–Hawaii Telescope Legacy Survey, a collaborative project of NRC and CNRS.

Supporting information

Additional supporting information can be found in the online version of this article: Tables 2, 8, 9, 10, 11.

Supplementary data is available at PASJ Journal online.

†

Based in part on data collected at Subaru Telescope, which is operated by the National Astronomical Observatory of Japan.

1

See 〈http://www.cfht.hawaii.edu/Science/CFHTLS〉.

2

See 〈http://surveys.roe.ac.uk/wsa〉.

3

〈http://irsa.ipac.caltech.edu/data/SPITZER/SWIRE/docs/delivery_doc_r2_v2.pdf〉.

4

Taken from Multiband Imaging Photometer for Spitzer Data Handbook Version 3.2.

5

See 〈http://galex.stsci.edu/GR6/〉.

6

See 〈http://www.astr.tohoku.ac.jp/~akiyama/SXDS/post_stamp.html〉.