AEGIS-X: DEEP CHANDRA IMAGING OF THE CENTRAL GROTH STRIP

The new AEGIS-XD Chandra data were taken at three nominal pointing positions, which we have designated AEGIS-1, AEGIS-2, and AEGIS-3. Each field was approved to receive approximately 600 ks in exposure as part of Chandra Cycle 9, supplementing the ∼200 ks exposures acquired in Cycle 3 (Nandra et al. 2005) and Cycle 6 (L09). The new exposures were split up into smaller segments ranging in duration from ∼7 to 100 ks. These observations were all taken in the time period 2007 December 11 to 2009 June 26 using the ACIS-I instrument (Garmire et al. 2003) without any grating in place. All new AEGIS-XD data were taken in VFAINT mode. Full details of the new Chandra observations are given in Table 1. The three subfields AEGIS-1, 2, and 3 were defined as the spatial limits of ObsIDs 9450, 9454, and 9458, respectively, with a border of ±20 pixels in the X and Y directions. These subfields were analyzed separately but have considerable overlap; common sources were removed at a later stage.

The total exposure times for the Cycle 9 pointings before screening were 583 ks (AEGIS-1), 597 ks (AEGIS-2), and 596 ks (AEGIS-3). The centers of these fields correspond fairly closely to those of the EGS-3, EGS-4, and EGS-5 fields of L09. Here we combine the data from the 200 ks EGS 3–5 fields with our new data, to a nominal 800 ks depth in each field. For the AEGIS-1 sub-field there is some overlap with the EGS2 data of L09, but this is very small and at large off-axis angles, so these data were not included. For the AEGIS-3 sub-field, we also add three ObsIDs from the original "Groth-Westphal Strip" (GWS) survey of Nandra et al. (2005), which were taken in FAINT mode. These have significantly different pointing centers and roll angles to the L09 AEGIS-XW survey, but there is substantial overlap of the tiles with the AEGIS-3 field at one edge. A list of the ObsIDs that overlapped with each field is given in Table 2. Note that the final catalog presented here consists only of the sources in the region that is covered by the new X-ray imaging (i.e., covered by the ObsIDs listed in Table 1), otherwise they should be present in the catalog of L09.

The total (pre-screening) exposure times for each field were 779 ks (AEGIS-1), 787 ks (AEGIS-2), and 782 ks (AEGIS-3, excluding the GWS/EGS-8 exposure, which only covers a relatively small part of the field). The data and analysis of the 200 ks imaging of the AEGIS-XW survey is fully described in L09. Here we largely follow the same reduction analysis methodology described in that paper, which we now describe in brief, noting any changes.

The data reduction was performed using the CIAO data analysis software v4.1.2 (Fruscione et al. 2009) and follows the scheme described in L09, with some minor changes and improvements. Briefly, for each individual Obsid, we corrected the data for aspect offsets, applied the bad pixel removal and destreaking algorithms, removed cosmic ray afterglows using standard tools, and corrected for CTI and gain effects. We also applied the ACIS particle background cleaning algorithm to the VFAINT mode data. Analysis was restricted to ACIS chips 0–3 and ASCA-style event grades 0, 2, 4, and 6. To reject periods of high background, we used the procedure of Nandra et al. (2007), adopting a threshold of 1.4 times the quiescent background level, determined as the count rate at which the background shows zero excess variance over that expected from statistical fluctuations alone. As in L09 and Nandra et al. (2005) we relaxed this criterion in ObsID 4365, which contains a period of relatively high but stable background. As in L09, a flare was also manually removed from ObsID 5850.

Zoom In Zoom Out Reset image size

Following this basic reduction and screening, the astrometry of the individual image frames was corrected using a reference catalog. Specifically, we use the CFHTLS i-band selected catalog to register the AEGIS-XD images to the optical reference frame. We first ran the Chandra wavelet source detection task wavdetect on the 0.5–7 keV image, using a detection threshold of 10⁻⁶, and then used the CIAO task reproject_aspect to correct the astrometry compared to the reference image to create new aspect solution files. The new aspect solutions were then applied to the original event files. The parameters used for reproject_aspect were a source match radius of 120 and a residual limit of 0.50. Typically, around 100 sources were detected in the individual ObsIDs, with around 60% of these having a counterpart used in the reprojection. The absolute value of the offset was typically small ($\lt 0\buildrel{\prime\prime}\over{.} 5$). Following this step we created event files for the individual frames in the standard full (FB; 0.5–7 keV), soft (SB; 0.5–2 keV), hard (HB; 2–7 keV), and ultrahard (UB; 4–7 keV) bands; exposure maps for each were produced for each energy band using the merge_all task. The exposure maps were created at multiple energies with weights appropriate for a ${\rm{\Gamma }}=1.4$ power-law spectrum (see Table 3). The individual ObsID images and exposure maps were then stacked together. At the native resolution of ACIS (0492 pixels), images of the entire AEGIS-XD region would be too large to manipulate efficiently. We therefore created separate image stacks for the AEGIS-1, AEGIS-2, and AEGIS-3 fields. These overlap significantly, meaning that edge effects can be avoided, but in the final catalog we remove duplicated entries detected in more than one image (see below). Exposure maps for the stacked images were created by summing the exposure maps of the individual ObsIDs. Figure 2 shows the effective exposure time as a function of survey area for the AEGIS-XD data.

Zoom In Zoom Out Reset image size

Source detection was the same as described in L09 and the reader is referred to that paper for full details. Briefly, a "seed" source catalog was first created using the wavdetect task run at a low probability threshold (10⁻⁴) on the mosaic images. This low threshold is intended to capture all potential sources, but likely contains many spurious sources. Aside from this thresholding, the only information used from wavdetect in the final catalog is the source position. Using these positions for the candidate sources, counts were extracted from the mosaic images using a circular aperture with radius equal to the exposure-weighted 70% encircled energy fraction (EEF) of the Chandra point-spread function (PSF). The PSFs were taken from a lookup table calculated using the MARX simulation software as described in L09. Background was determined using an annulus with inner radius equal to 1.5 times the 95% EEF at the source position and outer radius 100 pixels larger than this, excluding detected sources. The background was then scaled to the size of the source region, and the Poisson false probability of observing the total counts given the expected background was calculated, masking out the 95% EEF of candidate sources. A significance threshold of $4\times {10}^{-6}$ was then applied, and a further detection iteration was performed, masking out only the sources more significant than this. This iteration ensures that the background is not underestimated due to the masking of random positive variations identified as candidate sources in the wavdetect seed list. Any source detected at this $4\times {10}^{-6}$ probability level in the second iteration in any individual band was included in the final catalog. The sources that were considered significant were band-merged using the matching criteria specified in Table 2 of L09. Photometry was then performed to estimate the fluxes in several energy ranges, even if the source was not considered a significant detection in that particular band. After performing the band merging, we visually inspected the images of the sources in each of the fields and checked that the correct cross-band counterparts were identified. Two sources in the catalog were detected in the soft and full bands, but not correctly matched with their hard (and in one case ultrahard) band counterparts. These were combined manually. Two single-band detected sources (one soft, one ultrahard) were removed from the catalog after visual inspection revealed strong contamination from nearby bright sources. These removed sources were most likely incorrectly identified in the initial wavdetect seed catalog.

Finally, the source catalogs for the individual subfields AEGIS 1–3 were merged to remove duplicate sources in the overlapping regions. A 2'' search radius was adopted, and as in L09, the source was chosen from the field with the smallest off-axis angle for the final catalog.

One significant change in the current paper compared to L09 is in the X-ray photometry. Specifically, here we have adopted elliptical apertures to extract the counts from the individual ObsIDs using the 90% EEF PSF appropriate for the ObsID in question (95% in the case of the soft band). This contrasts with the source detection described above, and the photometry in L09, both of which adopt circular apertures. Fluxes were estimated using the Bayesian methodology described in L09 using a ${\rm{\Gamma }}=1.4$ spectrum with Galactic ${N}_{{\rm{H}}}$ of $1.3\times {10}^{20}$ cm⁻² (Dickey & Lockman 1990). The count rates in the full, hard, and ultrahard bands were also extrapolated to fluxes in standard energy bands: 0.5–10, 2–10, and 5–10 keV, respectively. Hardness ratios were calculated using the Bayesian methodology BEHR (Park et al. 2006).

The IRAC Guaranteed Time Observations of the AEGIS field cover a region approximately $2^\circ$ by 10' (Barmby et al. 2006, 2008). The ACIS field of view is wider than this (see Figure 3), meaning that the GTO IRAC observations miss the edges of the deep Chandra imaging. As shown by L09, for example, and discussed below, Spitzer IRAC observations are critical for secure identifications of X-ray sources. For this reason, as part of the Chandra program we obtained additional Spitzer IRAC imaging of the edges of the strip. The IRAC coverage map is shown in Figure 1, and the data reduction was performed as described in Barro et al. (2011a), and incorporated into the Rainbow database.

Zoom In Zoom Out Reset image size

Sensitivity maps were computed according to the procedure described in Georgakakis et al. (2008b) as implemented by L09. This approach accounts for incompleteness and Eddington bias in the sensitivity calculation, which is performed in a manner that is also fully consistent with the source detection procedure. The flux limits for the new AEGIS-XD survey as a function of area are shown for various energy bands in Figure 5, compared to the deepest X-ray survey in existence, the Chandra Deep Field South, together with the sensitivity curves of the G12 source catalog of the entire EGS Chandra survey. Following L09, we define the limiting flux of our observations as the flux to which at least 1% of the survey area is sensitive. We find the limiting fluxes so defined to be $1.5\times {10}^{-16}$ (FB; 0.5–10 keV), $3.3\times {10}^{-17}$ (SB; 0.5–2 keV), $2.5\times {10}^{-16}$ (HB; 2–10 keV), and $3.2\times {10}^{-16}$ erg cm⁻² s⁻¹ (UB; 5–10 keV). We also show in Table 5 the 50% and 90% completeness limits of the survey.

Zoom In Zoom Out Reset image size

The source detection algorithm applied here (and in L09) is designed to provide an accurate estimate of the sensitivity of the X-ray observations at each position. The detection threshold can be altered depending on the number of likely spurious detections to be deemed acceptable in the catalog. Here we adopt a relatively conservative threshold, which should result in only a small number of spurious X-ray detections; L09, for example, estimated that $\lt 1.5$% of the AEGIS-XW 200 ks sources were likely to be spurious. To assess this for the deeper AEGIS-XD data, we performed tests of the source detection on simulated X-ray fields. The number of spurious sources expected in any given field may depend on the exposure map of the field: for example, with multiple overlapping pointings, as are present here, edge effects might introduce problems into the source detection.

The false source content of our catalog was estimated based on the field configuration of the AEGIS-2 field, overlapping the individual subframes just as in the real data. The simulations were initially run just using the background, subsequently applying the source detection procedure for the real observations to these simulated background-only images for the real observations. Correcting the area of the simulated observations to the total AEGIS-XD survey, we predict about 12 spurious sources in the catalog (i.e., around 1.3%), which is consistent with the estimates from L09. We also estimated the false source contamination using simulated sources, based on the log N–log S function of Georgakakis et al. (2008b), yielding similar results.

A further estimate of the spurious source content can be made using the ultrahard band (UB) images. Due to a combination of the strong energy-dependence of the effective area and PSF of the Chandra optics (which is inherent in the Wolter-1 design) and the nature of the spectra of the underlying source population, it is unlikely for real sources to be detected in only the UB, because it is the least sensitive of all the detection bands. As a result, UB-only sources must have heavy obscuration at just the right level to suppress the soft, full, and hard band detections below the threshold, but not so high that it also suppresses the UB flux. Furthermore, in practice, even heavily obscured sources are often detected in the soft X-ray band due to the presence of scattered X-ray light (e.g., Brightman & Nandra 2012). None of these considerations applies to spurious sources, which, if present in the ultrahard band, should not be detected in any other band. One source in our catalog was detected only in the UB, and is thus a candidate for being spurious in the UB catalog, which contains a total of 299 sources. This consideration suggests that the simulations might overestimate the number of spurious sources in our catalog.

With our deeper X-ray data, we can also make a post-hoc check of the number of spurious sources in the 200 ks L09 catalog. Naively, it would be expected that all real sources in the AEGIS-XW catalog also appear in our deeper observations in the overlapping area. This is not the case, however, if the sources are spurious because the significance would tend to go down (and eventually below the detection threshold) with deeper data. There are 17 sources in the AEGIS-XW catalog, which are covered in the deeper Chandra pointings but not listed as significant sources in our AEGIS-XD catalog. These are listed in Table 6, with postage-stamp images of the objects shown in Figure 6. The images show that the majority of the cases are low significance detections in the 200 ks catalog that are not confirmed in the deeper data. Two (egs_0511 and egs_0529) are sources that were detected in the wings of nearby bright (and possibly extended) sources. A relatively large number of the others are in the very central regions of the image where the PSF is small, and where the original detections by L09 are based only on a very small number of photons. In these and indeed other cases the objects identified by L09 may not be truly spurious but simply represent sources whose true flux lies below the detection limit even of the 800 ks data, but which generated a significant number of counts in the 200 ks observation due to Poisson statistics. In other words, they may be sources that were originally detected due to the Eddington bias. This assertion is supported by the fact that many have optical counterparts (see Table 6). This may also be explained if the X-rays are variable and the true flux has dropped by a large factor since the initial observations by L09.

Zoom In Zoom Out Reset image size

Table 6.  AEGIS-XW Sources from L09 Not Included in AEGIS-XD Catalog

L09	R.A. J2000a	Decl. J2000b	${p}_{\mathrm{min}}$ a,b	Det.a	Bayesian fluxa,c	$i{^{\prime} }_{\mathrm{AB}}$ d	Classical Flux Limite
cat ID	(J2000)	(J2000)		Bands	(10−16 erg s−1cm−2)	(mag)	(10−16 erg s−1cm−2)
egs_0379f	214.363649	52.525411	$1.0\times {10}^{-8}$	fs	${8.8}_{-2.7}^{+3.2}$	25.12	$\lt 2.6$
egs_0463	214.491077	52.632192	$3.6\times {10}^{-6}$	s	$\lt 2.8$	18.43	$\lt 2.3$
egs_0503	214.530413	52.508574	$3.3\times {10}^{-6}$	s	$\lt 3.9$	21.78	$\lt 2.2$
egs_0511f	214.548133	52.759199	$1.0\times {10}^{-8}$	f	${20.4}_{-5.3}^{+5.9}$	23.69	$\lt 3.7$
egs_0521	214.559843	52.568176	$2.1\times {10}^{-6}$	s	$\lt 8.2$	$\gt 27.0$	$\lt 4.2$
egs_0528	214.572498	52.446772	$3.2\times {10}^{-6}$	f	${5.8}_{-4.2}^{+3.2}$	23.79	$\lt 4.9$
egs_0529	214.573633	52.731312	$1.0\times {10}^{-8}$	fs	${9.2}_{-3.3}^{+3.9}$	24.89	$\lt 2.6$
egs_0533	214.575732	52.442809	$3.6\times {10}^{-6}$	f	${4.3}_{-2.8}^{+3.8}$	22.00	$\lt 4.9$
egs_0555	214.597394	52.488107	$1.6\times {10}^{-6}$	s	$\lt 1.5$	21.97	$\lt 7.1$
egs_0590	214.643180	52.698915	$4.8\times {10}^{-7}$	fs	${4.7}_{-2.5}^{+3.1}$	$\gt 27.0$	$\lt 1.8$
egs_0593	214.653435	52.627987	$4.8\times {10}^{-7}$	s	${3.6}_{-1.9}^{+2.5}$	22.23	$\lt 1.4$
egs_0602	214.662429	52.684361	$3.3\times {10}^{-6}$	s	$\lt 2.9$	25.39	$\lt 1.8$
egs_0687	214.813366	52.856652	$3.0\times {10}^{-6}$	h	${2.4}_{-1.3}^{+1.9}$	$\gt 27.0$	$\lt 1.4$
egs_0688	214.815105	52.793307	$2.7\times {10}^{-6}$	h	${2.4}_{-1.7}^{+1.9}$	$\gt 27.0$	$\lt 1.6$
egs_0711f	214.856439	52.745895	$4.8\times {10}^{-7}$	s	${5.9}_{-3.3}^{+3.5}$	24.28	$\lt 3.5$
egs_0728f	214.877813	53.007428	$1.0\times {10}^{-8}$	fs	${1.6}_{-0.7}^{+0.7}$	21.54	$\lt 4.6$
egs_0851f	215.076587	53.032611	$2.6\times {10}^{-6}$	s	${3.2}_{-1.7}^{+2.5}$	18.99	$\lt 1.4$

Notes.

^aValues from L09. ^bMinimum false detection probability found for the four analysis bands. ^cFull band flux or upper limit. ^dOptical identification from L09 or upper limit where no counterpart exists. ^eValues from this work. ^fSource also detected in G12.

Download table as:  ASCIITypeset image

G12 have already published a catalog of X-ray sources and optical associations in this field. The set of Chandra data used in their catalog is similar but not identical to ours, in that they analyze the entire AEGIS-X area, both deep and wide, whereas we use only the regions with nominal 800 ks exposure. Further, we use two more ObsIDs in these areas, 9876 and 9881, which are not analyzed by G12. The other 55 of our pointings are in common with G12, who nonetheless adopt a different procedure for data reduction, source detection, and identification of the X-ray sources. A comparison of the last can be found in Section 4.2.

We cross-correlated the X-ray source list of G12 with our catalog using a match radius of 3'', and restricting to the common area. From this we find 115 sources in G12 that have no match in our catalog. The vast majority (108) are faint sources close to the detection threshold (see Figure 7). In these cases, small differences in the analysis or source detection procedure (in particular the adopted detection threshold) can easily account for a detection in one catalog but not the other. In the photon-starved regime of Chandra the inclusion or exclusion of a single count in the source detection cell can easily change a "detection" to a "non-detection" or vice versa. Relatively minor differences in the determination of the background can have a similar effect. The other seven sources in G12 but not in our catalog are brighter and potentially a greater cause for concern. Visual examination shows three that are far off axis and have a counterpart in our catalog, but just outside the 3'' match radius. The remaining four are in crowded or bright source regions. Visual inspection suggests that these may indeed be distinct sources, but have not been separated or identified as such by our detection algorithm.

Zoom In Zoom Out Reset image size

Performing the opposite comparison, we find 73 sources in our catalog that have no counterpart within 3'' in G12. An examination of Figure 7 shows that a number of these sources are rather X-ray bright (${F}_{0.5-10}\gt {10}^{-15}$ erg cm⁻² s⁻¹). Examining the spatial distribution of these sources in our images, we find that the majority are at large off-axis angles, where the PSF is relatively broad.

We also compared the positions of our X-ray sources with those of G12, with the results of the comparison being shown in Figure 8. A systematic offset is found amounting to 037. This may be attributed in part to the different astrometric solutions adopted for the X-ray images, with ours being tied to the CFHTLS i-band, and the G12 positions registered to the DEEP2 reference frame.

Zoom In Zoom Out Reset image size

Following the cross-matching procedure, we can make a post-hoc estimate of the astrometric accuracy of the X-ray positions in our catalog. Figure 10 shows the offsets between the X-ray position and that of the multiband counterpart. Overall, we find that 84% of the of the counterparts lie within 1'' of the X-ray position, and 97% within 2''. As discussed by L09, the astrometric accuracy is a function of both off-axis position and source counts (see their Figure 7 and Table 8). Figure 10 demonstrates the latter effect, whereby the positions degrade somewhat for fainter sources. For relatively bright X-ray sources ($\gt 100$ net counts) we find 92% of counterparts within 1'' and 99% within 2''. With $\lt 100$ net counts, we find 78% of the counterparts within 1'', but still 96% within 2''. The rms error for all sources is 083 (062 for $\gt 100$ net counts and 093 for $\lt 100$ counts).

Zoom In Zoom Out Reset image size

Table 8.  Summary of Likelihood Ratio Matching Results for AEGIS-XD

Catalog	Area/deg2	N0	${\sigma }_{0}$	Lth	R	C	NX	NID	NNoID	NMulti	NPri
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
RAINBOW [3.6 μm]	0.26	65675	0.20	0.80	0.98	0.94	902	860	42	451	860
Subaru R	0.28	126709	0.20	1.18	0.95	0.80	936	788	148	435	37
ACS I	0.14	61983	0.10	1.00	0.96	0.79	617	509	108	279	0
Subaru K	0.08	10908	0.20	3.04	0.97	0.86	408	363	45	165	2
CFHTLS i	0.26	116686	0.30	1.10	0.95	0.77	828	670	158	401	1
MMT i	0.29	93412	0.30	1.03	0.94	0.75	937	752	185	370	1
DEEP R	0.28	45968	0.30	4.06	0.97	0.45	922	430	492	312	0
Palomar Ks	0.25	19086	0.30	1.58	0.96	0.76	881	693	188	202	6

Note. Column 1: catalog name and detection band used to match counterparts. Column 2: area covered by both the multiwavelength data and deep X-ray data. Column 3: number of multiwavelength sources in the X-ray area. Column 4: 1 σ positional accuracy of multiwavelength catalog in arcseconds. Column 5: likelihood ratio threshold determined by the iterative procedure described in Section 4. Column 6: sample reliability, the mean of the individual reliabilities of each secure counterpart, as in Luo et al. (2010). Column 7: sample completeness, the sum of the reliabilities for all counterparts divided by the total number of X-ray sources. Column 8: total number of X-ray sources in the area covered by the multiwavelength data. Column 9: number of secure counterparts to X-ray sources. Column 10: number of X-ray sources without secure counterparts. Column 11: number of X-ray sources with more than one candidate counterpart within the 35 search radius. Column 12: number of X-ray sources assigned a primary counterpart in this band.

Download table as:  ASCIITypeset image

For the 864 X-ray sources that are in common between our catalog and that of G12, we compared the counterpart identifications. There is a substantial difference in the multiwavelength datasets used: G12 used only the DEEP2 optical photometry, whereas we employ a much wider variety of data including (deblended) IRAC photometry, which is known to yield more efficient counterpart identification for X-ray sources (Cardamone et al. 2008; L09). The AGN are bright in the IRAC band and the number of background sources decreases, making it easier to obtain a secure association. Of the 864 common sources, G12 assign a counterpart to 606. By comparison, our method yields reliable associations for 830 sources, including 595 of the G12 counterparts. Of the 606 G12 counterparts, our methodology indicates an unreliable association in 11 cases.

Matching the counterpart positions, we found a systematic offset between the counterpart positions of about 058 (see Figure 11). This is presumably due to the different astrometric systems adopted by DEEP2 and the Rainbow database. After correction for this offset, we found 572 of the 595 sources with high confidence counterparts in both works to be within 05 of each other, and hence are presumably the same counterpart (see Figure 11). For the remaining 23 sources we seem to identify a different counterpart to G12. There can be a number of reasons for this, including differences in X-ray positions, deblending or non-deblending of photometry in crowded regions, and the fact that we match to a wider range of catalogs. For example, in several mismatched cases, we identify a single point-like counterpart in the HST Advanced Camera for Surveys (ACS) images, which in the ground-based images is merged with two or more very nearby objects into an apparent single source, identified as the counterpart by G12. A further check of the consistency between our associations those of G12 can be made by comparing the spectroscopic redshifts of the counterparts, when available. The G12 catalog contains 180 sources with reliable redshift from DEEP2. For 155 of these, we find the same redshift. For a further two (aegis_704 and aegis_762 in this paper) we assign the same counterpart but the redshift is different because we adopted a revised redshift value using DEEP3. Of the remaining 23 sources, there are 14 with spectroscopy in G12 that are not detected as significant X-ray sources in our reduction. For the final nine, we assign a different counterpart to G12, for the reasons discussed above, and hence have a different spectroscopic redshift.

Zoom In Zoom Out Reset image size

A total of 353 sources in our AEGIS-XD X-ray sample have a reliable spectroscopic data, obtained from a variety of sources, which are listed in priority order in Table 9. The largest number (167) of these were derived from the DEEP2 redshift survey (Newman et al. 2013). DEEP2 was a magnitude-limited redshift survey (to ${R}_{\mathrm{AB}}$ = 24.1) covering approximately 2.8 square degrees over four fields. Among these fields, the AEGIS area features the most extensive multiwavelength coverage (Davis et al. 2007). Furthermore, the target selection in the AEGIS field was largely based on optical magnitude, unlike the other fields in DEEP2 where color selections were also applied to isolate high redshift galaxies. The redshift success for the AEGIS-XD sources targeted in DEEP2 is very high ($\gt 77$%). Note that we count only the spectra with redshift quality 3 and 4 as defined by Newman et al. (2013) as secure redshifts, ignoring lower qualities. On the other hand, the sampling rate of DEEP2 combined with selection against presumed stellar objects in the survey means that not all X-ray source counterparts brighter than the magnitude limit were covered. In addition, because the X-ray sources were not known at the time the DEEP2 survey was designed, they could not be targeted explicitly.

The DEEP3 survey (Cooper et al. 2011), an extension of DEEP2, addressed this issue by specifically targeting the counterparts of X-ray sources in the field regardless of their optical properties. DEEP3 provides an additional 89 spectroscopic redshifts for the AEGIS-XD survey. The success rate in DEEP3 (∼51%) was lower than in DEEP2 because targets fainter than the DEEP2 magnitude limit were included. These spectra often failed to yield a secure redshift. A further important spectroscopic dataset was provided using the MMT/Hectospec instrument (Coil et al. 2009), again explicitly targeting X-ray sources that were not already covered by DEEP2. This campaign provided a total of 81 secure redshifts for AEGIS-XD sources. Finally, a handful of spectroscopic redshifts of X-ray source counterparts have been obtained by other campaigns, for example the Canada–France Redshift Survey (Lilly et al. 1995), the Keck Lyman Break Galaxy (LBG) surveys of Steidel et al. (2003, 2004), and the SDSS (Ahn et al. 2012). Of these, SDSS provides six additional redshifts, and the LBG survey provides additional six. The five redshifts from CFRS were all duplicated in DEEP2/3 or the MMT surveys, so we use those in preference.

The grand total of 353 spectra implies a total spectroscopic completeness of the sample of ∼38%. This is, however, a very strong function of optical magnitude, as can be seen from Figure 12. Only 20 spectroscopic redshifts have been obtained at magnitudes fainter than ${R}_{C}=24$, despite this being the peak of the magnitude distribution of the X-ray counterparts. This accounts for the relatively low spectroscopic completeness of the whole sample, despite major efforts in terms of spectroscopy in this field. A total of 111 X-ray source counterparts were targeted in the course of the various surveys without a reliable redshift being obtained.

Zoom In Zoom Out Reset image size

Despite the intensive spectroscopy in this field, a relatively large fraction of the AEGIS-XD sources do not have spectroscopic redshifts. Photometric redshifts were therefore computed for the remaining sources with multiwavelength counterparts using SED fitting. The photometric redshifts were computed following the analysis procedure described in Salvato et al. (2011), which takes into account knowledge of the optical morphology, optical variability, and X-ray emission to determine the most appropriate library and priors to be used. This method has been successfully applied in the COSMOS field (Salvato et al. 2009), the Lockman Hole (Fotopoulou et al. 2012), and the Extended Chandra Deep Field South (Hsu et al. 2014). More details regarding the adopted libraries can be found in Hsu et al. (2014).

Morphological classifications for the optical counterparts were taken from the Rainbow survey. There, the Hubble Space Telescope-ACS images were primarily used for the separation of the sources into point-like and extended. When the ACS data were not available, the analysis was performed on the Subaru R_c band image (FWHM ∼ 07). The motivation for this is that optically extended sources (i.e., resolved galaxies) are likely to be better modeled with a galaxy dominated template. Optical point sources and/or variable sources are candidate type 1 QSOs, which are better fit with an AGN template. We classified 536 sources as optically extended (EXTNV) and the remaining 401 as point-like or unresolved (QSO variable or QSOV).²⁰ The EXTNV group was also split on the basis of the soft X-ray flux, with 530 sources fainter than ${{\rm{F}}}_{(0.5-2\mathrm{keV})}=8\times {10}^{-15}$ erg cm⁻² s⁻¹ and the remaining six brighter than that.

Source variability can be a major issue in determining the photometric redshifts for type I AGN because the multiband data were usually taken at different epochs (i.e., there can be significant flux-offsets worsening the SED fits or causing the wrong template to be selected). In addition, in many cases the photometry in a given band was produced by adding the results from different observation runs separated in time, which means that a multi-epoch variability analysis was not possible. However, we do find that 59 sources have a clear offset (typically 0.5 mag and up to ∼1) when comparing photometry from the same or similar filters, suggesting variability. Visual inspection shows that 38 of these sources are point-like, and while the morphological classification is a strong function of magnitude and image resolution, this suggests that the variability is likely real and due to the QSO or stellar nature of the source. For the remainder, which are classified as optically extended, half are located close to nearby stars and the variability is likely more related to variation in the flux of the stars or the background in their vicinity. For lack of further information, all 59 apparently variable sources have been flagged, as the variability might have a significant effect on the redshift determination. As in Salvato et al. (2011), we used the LePhare code (Ilbert et al. 2006) to compute the photometric redshifts. The first step was to correct the photometry for Galactic extinction using a median $E(B-V)$ = 0.04 (see second-to-last column of Table 7, as in Barro et al. 2011a). Then we searched for possible zeropoint offsets that could affect the accuracy of the photometric redshift. To do this we computed the photometric redshift of a sample of normal galaxies (i.e., non X-ray detected) with the reliable spectroscopic redshift available. For each source, we kept the redshift fixed and searched for the best fitting template in the library of normal galaxies used in Ilbert et al. (2009). Then, for each photometric band, we computed the average difference between the photometry of all the sources and the photometry of the template. In this way, we can correct for second-order problems in the photometric calibration. We adopted these zeropoints (reported in the last column of Table 7) when computing the photometric redshifts for the X-ray selected sources. We note that the zeropoint corrections depend on the templates used, and different libraries could provide slightly different values for the correction. For this reason we do not apply these corrections to the photometric catalog released with this paper. Furthermore, the same procedure could not be applied directly to the X-ray sources because (a) variability could affect the results and (b) the relative host/AGN contribution to the SED in a given band is unknown. Ignoring these facts can potentially introduce a greater uncertainty in the zeropoints when applying a relatively limited number of templates.

We compute the photometric redshift for the EXTNV sources using the new hybrid templates of Hsu et al. (2014). These templates are tuned for sources dominated by galaxies, with special care given in reproducing the emission lines. For the QSOV sample we used the AGN-dominated hybrids of Salvato et al. (2009). Different absolute magnitude priors were considered for EXTNV ($-24\lt {M}_{B}\lt -8$) and QSOV ($-30\lt {M}_{B}\lt -20$) sources, respectively. In order to search for stars in the sample, a stellar library was also used to fit the data. Whenever a source in the QSOV sample was better fit with the stellar template it was classified as a star. There are 21 sources that satisfy this criterion (∼3.1% of the entire sample), with 12 of these also spectroscopically confirmed. Turning to the extragalactic sources, the overall accuracy for the spectroscopic sample, measured by the normalized mean absolute deviation, is ${\sigma }_{\mathrm{NMAD}}$ = 0.040, with an outlier fraction of η = 5.1% (Figure 13). The accuracy is slightly better (${\sigma }_{\mathrm{NMAD}}$ = 0.030, with an outlier fraction of η = 3.8%, where deeper and better resolved NIR photometry from CANDELS is available). A close look at the outliers revealed that most of those for which the photometric redshift solution was not unique (empty circles in the figure) had the correct solution in the second higher peak. The use of a redshift probability distribution function (made publicly available with this work) is recommended. The remaining outliers can be explained by blending with nearby sources for which the low resolution of the ground-based imaged misplace the sources in the "EXTNV" group, rather than in the "QSOV," thus adopting the wrong templates or priors. The effect of the resolution of the images used for the classification has been addressed in Hsu et al. (2014). These authors demonstrate that the fraction of outliers increases significantly when using the morphological classification from the ground-based images rather than from the space-based ones. We break down the results in redshift, magnitude, and type in Table 10. The outlier fraction and uncertainties in the QSOV sample are larger, partly due to the degeneracies associated with the typical power-law SED of this subsample. This is a general problem for photometric redshifts for AGN, which can be mitigated when narrower filters sensitive to strong emission lines are used in the photo-z determination (e.g., Salvato et al. 2009; Cardamone et al. 2010; Hsu et al. 2014).

Zoom In Zoom Out Reset image size

Table 10.  Accuracy of Photometric Redshifts

Sub Sample	QSOV			EXTNV
	N. of Sources	η (%)	${\sigma }_{\mathrm{NMAD}}$	N. of Sources	η (%)	${\sigma }_{\mathrm{NMAD}}$
	(1)	(2)	(3)	(1)	(2)	(3)
z < 1.2	53	11.3	0.067	201	3.5	0.034
z > 1.2	72	5.6	0.059	24	4.2	0.027
R < 22 mag	41	7.3	0.067	111	2.7	0.040
R > 22 mag	84	8.3	0.061	114	4.4	0.028
Combined	125	8.0	0.063	225	3.6	0.033

Note. (1) Number of sources, (2) outlier fraction, and (3) normalized mean absolute deviation. QSOV refers to optically point-like and/or variable sources. EXTNV refers to optically extended and non-variable sources.

Download table as:  ASCIITypeset image

The full probability distribution function for the photometric redshifts is made available for all sources in our catalog (see Appendix). We caution that the errors associated with the photometry can easily be underestimated, with the result that the 1–3σ errors associated with the photometric redshift can also be underestimated. This is true for photometric redshifts in general (see Dahlen et al. 2013 for a recent test with a sample of normal galaxies in the CANDELS fields). We verified that the situation is similar for our sample where ${z}_{\mathrm{phot}}-1\sigma \lt {z}_{\mathrm{spec}}\lt {z}_{\mathrm{phot}}+1\sigma$ only for 57% of the sources, and only 79% of the sample the spectroscopic sample is within the 3σ error of the photometric redshift. Thus, while using the 3σ errors associated with photometric redshifts provides a reasonably accurate estimate of the uncertainty for most of the sources, we recommend using the entire redshift probability distribution function for a more complete analysis.

The redshift distribution of our X-ray sources (after excluding the stars) is shown in Figure 14, distinguishing between spectroscopic and photometric redshifts. There is a peak in the spectroscopic redshift distribution around $z\;\sim 0.7$, which is also seen in the photometric redshift distribution, presumably due to a large-scale structure at this redshift in the field. The redshift distribution shows clearly that the spectroscopic completeness is a very strong function of redshift, with the sample being highly spectroscopically complete at $z\lt 1$ (210 of 303), and highly incomplete above this redshift (140 of 613). This shows the great importance of computing accurate photo-z when considering AGN evolution and/or host galaxy properties at high redshift.

Zoom In Zoom Out Reset image size

Figure 14 also shows in blue the sources that are characterized by at least one further peak in the redshift solution (200 sources, 23% of the sample). Sources shown in red are those for which a single peak in the redshift solution is found, but for which the peak P(z) is lower than 50% (34 sources, 6%), with a distribution over a broad redshift range. The majority of the high redshift sources (e.g., at z > 3.5) are among these poorly defined sources. We find 14 X-ray sources at $z=3.5-4.5$ in the AEGIS field (two spectroscopically confirmed), and 15 at redshift $z\gt 4.5$, although all of the latter show a secondary peak at a lower redshift. In this regime the photometric redshifts are strongly dependent on the upper limits adopted, making the results unstable. The high redshift nature of these sources thus needs to be assessed carefully, ideally via spectroscopy or possibly with deeper photometry (e.g., Venemans et al. 2007). For the time being the photometric redshifts should be considered with the associated errors and preferably with the full P(z).

Figure 15 shows our sample in the ${L}_{{\rm{X}}}-z$ plane. The left panel distinguishes between sources with spectroscopic redshifts and those with photometric redshifts. This clearly shows that that vast majority of sources at $z\gt 1.5$ do not have spectroscopic redshift, making the photo-z essential. Obtaining accurate photo-z, as in our work, is thus crucial for the investigation of AGN properties at high redshift. The AEGIS survey alone provides reasonably good sampling of the population in the luminosity range 10^42–44 up to around $z-3$, after which it becomes incomplete at the faint end. This can be improved by combining it with deeper surveys such as the CDF-N and, particularly, the CDF-S. Similarly, the luminosity range above 10⁴⁴ erg s⁻¹, above L_*, the knee in the XLF, is relatively sparsely sampled by our data, and larger area surveys are required to determine the bright end of the XLF with good accuracy. The right panel of Figure 15 shows the objects color coded by the best-fit template fitted to the multiwavelength photometry during the photo-z determination. There is a considerable mix of template types depending on whether the light at longer wavelengths is dominated by the AGN, the galaxy, or a mixture. Previous works using the same photometric redshift methods used here have shown good agreement between the SED type and the spectroscopic classification, where available (Salvato et al. 2009; Lusso et al. 2010, 2012; Civano et al. 2012)

Zoom In Zoom Out Reset image size

A subset of the basic X-ray properties of the AEGIS-XD sources are given in Tables 11 and in 12. In the first we provide coordinates of the sources and detection properties in the various X-ray bands. In the second table we provide X-ray properties as fluxes and hardness ratios. Each table is clearly described on the corresponding notes.

Table 12.  Chandra AEGIS-X Source Catalog: Source Fluxes and HRs

	Bayesian Flux				Bayesian	Phot.
XID	${f}_{0.5-10}$	${f}_{0.5-2}$	${f}_{2-10}$	${f}_{5-10}$	HR	Flag
(1)	(2)	(3)	(4)	(5)	(6)	(7)
aegis_001	${2.752}_{-0.628}^{+0.663}$	${0.814}_{-0.179}^{+0.195}$	${0.000}_{-0.000}^{+1.392}$	${0.000}_{-0.000}^{+1.483}$	−0.414${}_{-0.241}^{+0.247}$	0
aegis_002	${8.664}_{-0.796}^{+0.831}$	${1.853}_{-0.225}^{+0.240}$	${7.349}_{-1.067}^{+1.130}$	${3.517}_{-1.182}^{+1.265}$	−0.037${}_{-0.096}^{+0.097}$	0
aegis_003	${5.388}_{-0.719}^{+0.763}$	${1.255}_{-0.207}^{+0.227}$	${4.254}_{-0.968}^{+1.045}$	${2.070}_{-1.183}^{+1.110}$	−0.095${}_{-0.135}^{+0.139}$	0
aegis_004	${2.688}_{-0.616}^{+0.647}$	${0.285}_{-0.152}^{+0.149}$	${3.388}_{-0.960}^{+1.012}$	${0.000}_{-0.000}^{+1.978}$	${0.370}_{-0.179}^{+0.205}$	0
aegis_005	${1.571}_{-0.466}^{+0.482}$	${0.252}_{-0.115}^{+0.118}$	${1.348}_{-0.774}^{+0.659}$	${0.000}_{-0.000}^{+1.129}$	${0.107}_{-0.195}^{+0.299}$	0
aegis_006	${5.946}_{-0.607}^{+0.638}$	${1.006}_{-0.155}^{+0.169}$	${6.520}_{-0.894}^{+0.951}$	${2.550}_{-0.940}^{+1.010}$	${0.198}_{-0.096}^{+0.103}$	0
aegis_007	${2.087}_{-0.452}^{+0.470}$	${0.406}_{-0.114}^{+0.122}$	${1.761}_{-0.725}^{+0.723}$	${0.000}_{-0.000}^{+1.326}$	${0.023}_{-0.182}^{+0.224}$	0
aegis_008	${2.347}_{-0.447}^{+0.464}$	${1.074}_{-0.141}^{+0.150}$	${0.000}_{-0.000}^{+0.444}$	${0.000}_{-0.000}^{+0.634}$	−0.884${}_{-0.116}^{+0.031}$	0
aegis_009	${2.412}_{-0.524}^{+0.549}$	${0.220}_{-0.129}^{+0.116}$	${3.344}_{-0.827}^{+0.871}$	${1.765}_{-1.147}^{+0.856}$	${0.452}_{-0.165}^{+0.190}$	0
aegis_010	${3.095}_{-0.423}^{+0.443}$	${0.818}_{-0.120}^{+0.129}$	${1.658}_{-0.610}^{+0.627}$	${0.000}_{-0.000}^{+0.938}$	−0.308${}_{-0.144}^{+0.160}$	0

Notes. (1)Unique source name. (2) Flux in the 0.5–10 keV band in units of ${10}^{-15}\;\mathrm{erg}\;{{\rm{s}}}^{-1}\;{\mathrm{cm}}^{-2}$ estimated using the Bayesian methodology of L09. (3) Flux in the 0.5–2 keV band in units of ${10}^{-15}\;\mathrm{erg}\;{{\rm{s}}}^{-1}\;{\mathrm{cm}}^{-2}$ estimated using the Bayesian methodology of L09. (4) Flux in the 2–10 keV band in units of ${10}^{-15}\;\mathrm{erg}\;{{\rm{s}}}^{-1}\;{\mathrm{cm}}^{-2}$ estimated using the Bayesian methodology of L09. (5) Flux in the 5–10 keV band in units of ${10}^{-15}\;\mathrm{erg}\;{{\rm{s}}}^{-1}\;{\mathrm{cm}}^{-2}$ estimated using the Bayesian methodology of L09. (6) Hardness ratio determined by BEHR (Park et al. 2006) using the counts in the 0.5–2 and 2–7 keV spectral bands. (7) Quality of the X-ray photometry. A flag of "1" indicates the presence of a nearby source that may be contaminating the photometry. A flag of "2" indicates that another source was detected with in the 90% EEF and that the photometry is likely heavily contaminated and the source position uncertain. All other sources have a flag of "0."

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image

The multiwavelength properties of the counterparts are provided in Tables 13 and 14. In the first table we provide the coordinates of the counterparts in the optical and mid-infrared catalogs, together with their ID and offset from the X-ray coordinates. In the second table, for each counterpart we provide the magnitude in the AB system for all bands listed in Table 7. The photometry is corrected for Galactic extinction but not for zeropoint offset because this value is not independent of the SEDs used in its computation.

The redshift catalog is shown in Table 15. For each of the sources we list the spectroscopic redshift if available, the redshift reliability and the source in the literature from which it originates. In addition, we provide the photometric redshift values and the 68% and 90% confidence limits, as well as a measure of the width of the peak of the photo-z probability distribution function P(z), and the adopted template. If a significant second peak is seen in the P(z) solution, details are also provided for this secondary solution. In general, we recommend adopting the full P(z) rather than a single value for the photometric redshift. This is particularly important when the value of P(z_p1) is low, which is often the case when the number of available photometric points is low or where a secondary solution exists. The full P(z) and the templates used for fitting are available from the public web site http://www.mpe.mpg.de/XraySurveys.

Table 15.  Extract from the AEGIS-XD Redshift Catalog

Rainbow	Index No.	zs	zconf	${z}_{\mathrm{ref}.}$	Nb	zp1	${z}_{p1}$ l68	${z}_{p1}$ U68	${z}_{p1}$ l90	${z}_{p1}$ U90	P(zp1)	Mod1	${z}_{p2}$	P(zp2)	Mod2
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)	(15)	(16)
aegis_001	1	−1.000	0	0	23	1.992	1.907	2.068	1.818	2.148	95.066	30	−99.00	0.00	−999
aegis_002_3	4	−1.000	0	0	23	1.765	1.677	1.865	1.554	1.909	96.528	5	−99.00	0.00	−999
aegis_003	5	−1.000	0	0	23	1.485	1.396	1.954	1.334	1.996	54.259	7	−99.00	0.00	−999
aegis_004_1	6	−1.000	0	0	20	2.718	2.633	2.903	2.256	2.958	81.099	139	−99.00	0.00	−999
aegis_005_3	10	−1.000	0	0	23	1.472	1.436	1.766	1.412	1.822	75.805	139	−99.00	0.00	−999
aegis_006_1	11	−1.000	0	0	22	1.341	1.266	1.393	1.232	1.433	99.676	3	−99.00	0.00	−999
aegis_007_2	14	−1.000	0	0	15	2.567	2.098	3.147	1.472	3.383	50.924	4	−99.00	0.00	−999
aegis_008	15	0.281	3	4	22	0.177	0.162	0.279	0.125	0.344	83.952	104	−99.00	0.00	−999
aegis_009_1	16	0.769	2	2	22	0.747	0.713	0.784	0.692	0.873	92.742	121	−99.00	0.00	−999
aegis_010	18	−1.000	0	0	24	0.806	0.800	0.882	0.800	0.938	99.116	4	−99.00	0.00	−999

Note. Excerpt from the redshift catalog for AEGIS-X, which is published in its entirety in the HTML edition of the journal. Column 1: optical identifier number from the Rainbow catalog. Column 2: index number used to identify files in online database. Column 3: spectroscopic redshift, when available, otherwise −1. Column 4: redshift confidence; we consider the spectroscopic redshift to be reliable when this value is 3 or greater. Column 5: spectroscopic redshift reference, 0 = no z_spec, 1 = DEEP2+3 (Cooper et al. 2012; Newman et al. 2013), 2  = MMT (Coil et al. 2009; note, however, that this is the first time the redshifts themselves are published), 3 = CFRS (Lilly et al. 1995), 4 = SDSS (DR9, Ahn et al. 2012), 5 = LBG (Steidel et al. 2003). Column 6: number of photometric points available for the fit. Column 7: photometric redshift defined as the primary peak of the P(z) distribution. Columns 8 and 9: 68% confidence lower and upper value of photometric redshift. Columns 10 and 11: 90% confidence lower and upper value of photometric redshift. Column 12: area under the curve P(z) computed in the range ${z}_{p1}$ ± 0.1(1 + z_p1), normalized to a percentage of the entire P(z). Column 13: best fitting template; from 1 to 31 the templates are from S09, and templates from 100 + (1...30) are from the I09 library. Column 14: second solution for the photometric redshift, when P(z_p2) exceeds 5%, otherwise −99. Column 15: same as column 12, but for z_p2 if quoted in column 14, otherwise 0. Column 16: same as column 13, but for z_p2 if quoted in column 14, otherwise −999.

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image