Submitted to AJ The Extended Chandra Deep Field-South Survey: X-ray Point-Source Catalog Shanil N. Virani, Ezequiel Treister1,2, C. Megan Urry1, and Eric Gawiser1,3 6 Department of Astronomy, Yale University, P.O. Box 208101, New Haven, CT 06520 0 0 2 [email protected] n a ABSTRACT J 0 3 The Extended Chandra Deep Field-South (ECDFS) survey consists of 4 2 Chandra ACIS-I pointings and covers 1100 square arcminutes ( 0.3 deg2) v ≈ ≈ centered on the original CDF-S field to a depth of approximately 228 ks. This is 1 5 the largest Chandra survey ever conducted at such depth, and only one XMM- 5 6 Newtonsurvey reaches a lower fluxlimit in thehard2.0–8.0keV band. Wedetect 0 651 unique sources — 587 using a conservative source detection threshold and 5 0 64 using a lower source detection threshold. These are presented as two separate / h catalogs. Of the 651 total sources, 561 are detected in the full 0.5–8.0 keV band, p 529 in the soft 0.5–2.0 keV band, and 335 in the hard 2.0–8.0 keV band. For - o point sources neartheaimpoint, thelimiting fluxes areapproximately 1.7 10−16 r × st erg cm−2 s−1 and 3.9 10−16 erg cm−2 s−1 in the 0.5–2.0 keV and 2.0–8.0 keV a × : bands, respectively. Using simulations, we determine the catalog completeness as v i a function of flux and assess uncertainties in the derived fluxes due to incomplete X spectral information. We present the differential and cumulative flux distribu- r a tions, which are in good agreement with the number counts from previous deep X-ray surveys and with the predictions from an AGN population synthesis model that can explain the X-ray background. In general, fainter sources have harder X-ray spectra, consistent with the hypothesis that these sources are mainly ob- scured AGN. Subject headings: diffuse radiation — surveys — cosmology: observations — galaxies: active — X-rays: galaxies — X-rays: general. 1Yale Center for Astronomy and Astrophysics, Yale University, P.O. Box 208121,New Haven, CT 06520 2Departamento de Astronomia, Universidad de Chile, Casilla 36-D, Santiago, Chile. 3 NSF Astronomy and Astrophysics PostdoctoralFellow – 2 – 1. Introduction Wide-area X-ray surveys have played a fundamental role in understanding the nature of the sources that populate the X-ray universe. Early surveys like the Einstein Medium Sen- sitivity Survey (Gioia et al. 1990), the ROSAT International X-ray/Optical Survey (Ciliegi et al. 1997), and the ASCA Large Sky Survey (Akiyama et al. 2000) showed that the vast majority of the bright X-ray sources are active galactic nuclei (AGNs). More specifically, shallow wide-area surveys in the soft (0.5–2.0 keV) X-ray band yield mostly unobscured, broad-line AGNs, which are characterized by a soft X-ray spectrum with a photon index of Γ 1.9 (Nandra & Pounds 1994). In contrast, deep X-ray surveys — particularly surveys ≃ that make use of the unprecedented, sub-arsecond spatial resolution of the Chandra X-ray Observatory — find AGN with harder X-ray spectra (Γ 1.4) at fainter fluxes, more like ∼ the hard spectrum of the X-ray background. Deep Chandra surveys have thus opened a new vista on resolving the X-ray background and identifying the role and evolution of accretion power in all galaxies. The cosmic X- ray background is now almost completely resolved ( 70–90%) into discrete sources in the ∼ deep, pencil beam surveys like the Chandra Deep Fields (CDF-N, Brandt et al. 2001; CDF- S, Giacconi et al. 2002). To understand the composition of the sources that make up the X-ray background, population synthesis models have been constructed (Madau et al. 1994; Comastri et al. 1995; Gilli et al. 1999;Gilli et al. 2001;Treister & Urry 2005) which typically require approximately 3 times as many obscured AGN as traditional Type 1 (unobscured) AGN. While the deep fields provide the deepest view of the X-ray universe and have generated plentiful AGN samples at lower luminosities, the small area covered by pencil-beam surveys means luminous sources are poorly sampled. In an attempt to determine the luminosity function of X-ray emitting AGN up to z 5, as well as to leverage existing deep multiwave- ∼ ′ ′ length data in the extended 30 30 field centered on the CDF-S, the region surrounding the × CDF-Swas recently observed by Chandra. Covering 1100 square arcminutes ( 0.3 deg2), ≈ ≈ theExtended Chandra DeepField-South(ECDFS) survey isthelargestChandra survey field at this depth ( 230 ks), and is the second deepest and widest survey ever conducted in the ≈ X-rays (the XMM-Newton survey of the Lockman Hole is deeper in the hard band and has 30% more area; Hasinger 2004). ∼ In this paper, we present the X-ray catalog for the ECDFS and the number counts in two energy bands. In subsequent papers, we will present the optical and near-IR properties of these X-ray sources, including first results from our deep optical spectroscopy campaign obtained as part of the one-square-degree MUltiwavelength Survey by Yale/Chile (MUSYC) (Gawiser et al. 2005). – 3 – In Section 2, we describe our data reduction procedure. In Section 3, we describe the point source detection and astrometry. The X-ray source catalog and basic survey results are presented in Section 4 and the conclusions are given in Section 5. The average Galactic column density along this line of sight for the four pointings is 9.0 1019 cm−2 (Stark et al. × 1992). H0 = 70 km s−1 Mpc−1, ΩM = 0.3, and ΩΛ = 0.7 are adopted throughout this paper which is consistent with the cosmological parameters reported by Spergel et al. (2003). All coordinates throughout this paper are J2000. 2. Observations and Data Reduction 2.1. Instrumentation and Diary of Observations All nine observations of the ECDFS survey field were conducted with the Advanced CCD Imaging Spectrometer (ACIS) on-board the Chandra X-ray Observatory1 as part of theapprovedguestobserver programinCycle5(proposalnumber05900218–PINielBrandt; Lehmer et al. 2005). ACIS consists of 10 CCDs, distributed in a 2 2 array (ACIS-I) and a × 1 6 array (ACIS-S). All 4 of the ACIS-I CCDs are front-illuminated (FI) CCDs; 2 of the 6 × ACIS-S CCDs are back-illuminated CCDs (S1 and S3). Of these 10 CCDs, at most 6 can be operated at any one time. Table 1 presents a journal of the Chandra observations of the ECDFS. All nine observations were conducted in VERY FAINT mode (See the Chandra Proposers’ Guide, pg. 95) so that the pixel values of the 5 5 event island are telemetered × rather than just the 3 3 event island as in FAINT mode. This telemetry format offers × the advantage of further reducing the instrument background after ground processing (see Section 2.2). Observation ids (ObsIds) 5019–5022 and 6164 also had the ACIS-S2 CCD powered on (see Table 1). However, due to the large off-axis angle of the S2 CCD during these observations, it has a much broader point spread function (PSF) and hence lower sensitivity so we exclude data from this CCD and focus only on data collected from the ACIS-I CCDs. The on-axis CCD for each ACIS-I observation is I3 and the ACIS-I field of view is 16.′9 16.′9. × 1For additional information on the ACIS and Chandra see the Chandra Proposers’ Guide at http://cxc.harvard.edu/proposer/POG/html. – 4 – 2.2. Data Reduction Alldatawere re-processedusing thelatestversion oftheChandra Interactive Analysisof 2 Observations (CIAO; Version 3.2.1; released 10 February 2005) software as well as version 3.0.0 of the calibration database (CALDB; released 12 December 2004). We chose to re- reduce all nine datasets rather than simply use the standard data processing (SDP) level 2 event files because we wanted the datasets to be reduced in a consistent manner, and more importantly, we wanted to take advantageof betterChandra X-rayCenter (CXC) algorithms to reduce the ACIS particle background by using the information located in the outer 16 3 pixels of the 5 5 event island , as well as executing a new script that is much more efficient × 4 at identifying ACIS “hot pixels” and cosmic ray afterglow events . This new hot pixel and cosmic ray afterglow tool is now implemented in the standard data processing pipeline at the CXC but was not applied in the SDP pipeline of the present observations. The following procedure was used to arrive at a new level 2 event file. Before applying the new CIAO tool to identify ACIS hot pixels and cosmic ray afterglow events, the pixels identified in the CXC-provided level 1 event file as being due to a cosmic ray afterglow event were reset. An afterglow is the residual charge from the interaction of a cosmic ray in a front-side illuminated CCD frame. Some of the excess charge is captured by charge traps created by the radiation damage suffered early in the mission (see Townsley et al. 2000 and references therein) and released over the next few to a few dozen frames. If these afterglow events are not removed from the data, they can result in the spurious detection of faint sources. To better account for such events, a new, more precise method for identifying afterglow events was developed by the CXC and has now been introduced into the standard data processing pipeline. This CIAO tool, “acis run hotpix,” was then run on the reset level 1 event file to identify and flag hot pixels and afterglow events in all 9 ACIS observations of the ECDFS. The last step in producing the new level 2 event file was to run the CIAO tool “acis process events.” In addition to applying the newest gain map supplied in the latest release of the CALDB, this tool also applies the pixel randomization and the ACIS charge transfer inefficiency (CTI) correction. The latter corrects the data for radiation damage sustained by the CCDs early in the mission. All of these corrections are part of the standard data processing and are on by default in acis process events. The time-dependent gain correction is also applied to the event list to correct the pulse-invariant (PI) energy 2See http://cxc.harvard.edu/ciao/. 3See “Reducing ACIS Quiescent Background Using Very Faint Mode”, http://cxc.harvard.edu/cal/Acis/Calprods/vfbkgrnd/index.html. 4See http://cxc.harvard.edu/ciao/threads/acishotpixels/. – 5 – channel for the secular drift in the average pulse-height amplitude (PHA) values. This drift is caused primarily by gradual degradation of the CTI of the ACIS CCDs (e.g., Schwartz & Virani 2004). Finally, the observation-specific bad pixel map created by “acis run hotpix” was supplied, and the option to clean the ACIS particle background by making use of the additional pixels telemetered in VERY FAINT mode was turned on. Once a new level 2 event file was produced, we applied the standard grade filtering to each observation, choosing only event grades 0, 2, 3, 4, and 6 (the standard ASCA grade set), and the standard Good Time Intervals supplied by the SDP pipeline. We also restrict the energy range to 0.5–8.0 keV, as the background rises steeply below and above those 5 limits . Lastly, we examined the background light curves for all 9 observations as the ACIS background is known to vary significantly. For example, Plucinsky & Virani (2000) found −1 thatthefront-illuminatedCCDscanshowtypicalincreasesof1-5ctss abovethequiescent −1 level, while the back-illuminated CCDs can show large excursions — as high as 100 cts s above the quiescent level — during background flares. The durations of these intervals of 4 enhanced background are highly variable, ranging from 500 s to 10 s. The cause of these background flares is currently not known (see Grant, Bautz, & Virani 2002); however, they may be caused by low-energy protons (<100 keV; e.g., Plucinsky & Virani 2000; Struder et al. 2001). The time periods corresponding to these background flares are generally excised from the data before proceeding with further analysis although not always (see Brandt et al. 2001;Nandraetal.2005). Infact, Kimetal.(2004)findthatthesourcedetectionprobability depends strongly on the background rate. To examine our observations for such periods, we used the CIAO script ANALYSE LTCRV.SL which identifies periods where the background is 3σ above the mean. All 9 observations were filtered according to this prescription (see ± Table 1 for a comparison of raw exposure time vs. filtered exposure time), resulting in only 40.6 ks ( 4%) being lost due to background flares (954.2 ks vs. 913.6 ks). Of this, 20.3 ks ∼ ∼ were excluded from the end of ObsID 5017 due to a flare in which the count rate increased by a factor of 2. Table 2 lists the net exposure time for each of the 4 pointings used to ∼ image the ECDFS region. The net exposure time for each of the 4 pointings varies from a low of 205 ks to a high of 239 ks, with the mean net exposure time for the entire survey field of 228 ks. These extra steps in processing help remove spurious sources and result in fewer catalog sources than if the standard processing or pipeline products were used. 5See http://cxc.harvard.edu/contrib/maxim/stowed/. – 6 – 3. Data Analysis In this paper, we report on the sources detected in three standard X-ray bands (see Table 3): 0.5–8.0 keV (full band), 0.5–2.0 keV (soft band), and 2.0–8.0 keV (hard band). −1 The raw ACIS resolution is 0.492 arcsec pixel , however, source detection and flux deter- minations were performed on the block 4 images, i.e., 1.964 arcsec pixel−1, as the source detect tool and exposure map generation require significant computer resources for full size images; for greater accuracy, source positions were determined from the block 1 images. 3.1. Image and Exposure Map Creation ObservationsateachofthefourpointingswerecombinedviatheCIAOscript“merge all.” This script was executed using CIAO version 2.3 because of a known bug in the “asphist” tool under CIAO version 3.2.1; this bug results in incorrect exposure maps for the merged 6 image . At each pointing, the observation with the longest integration time was used for co-ordinate registration. For example, when merging ObsIds 5015 and 5016, the merged event list was registered to ObsId 5015 as it has approximately twice the integration time as 5016. Table 2 lists the ObsIds for each pointing, as well as the raw and the net integration time. For each pointing, we constructed images in the three standard bands: 0.5–8.0 keV (full band), 0.5–2.0 keV (soft band), and 2.0–8.0 keV (hard band); see Table 3. The full 7 band exposure-corrected image for the entire survey field is presented in Figure 1. We constructed exposure maps in these three energy bands for each pointing and for 8 the entire survey field . These exposure maps were created in the standard way and are normalized to the effective exposure of a source location at the aim point. The procedure used to create these exposure maps accounts for the effects of vignetting, gaps between the CCDs, bad column filtering, and bad pixel filtering. However, it should be noted that charge blooms caused by cosmic rays can reduce the detector efficiency by as much as few 9 percent . There is currently no way to account for such charge cascades; however, when a toolbecomes available, we will correct for this effect asnecessary and make the new exposure 6 See the usage warning at http://cxc.harvard.edu/ciao/threads/mergeall/. 7Raw and smoothed ASCA -grade images for all three standard bands (See Table 3) are available from http://www.astro.yale.edu/svirani/ecdfs/. 8Exposure maps for all three standard bands (see Table 3) are available from http://www.astro.yale.edu/svirani/ecdfs/. 9See http://cxc.harvard.edu/ciao/caveats/aciscaveats 050620.html. – 7 – maps publicly available at the World Wide site listed in Footnote 7. The exposure maps were binned by 4 so that they were congruent to the final reduced images. A photon index of Γ = 1.4, the slope of the X-ray background in the 0.5–8.0 keV band (e.g. Marshall et al. 1980; Gendreau et al. 1995; Kushino et al. 2002) was used in creating these exposure maps. In order to calculate the survey area as a function of the X-ray flux in the soft and hard bands, we used the exposure maps generated for each band and assumed a fixed detection threshold of 5 counts in the soft band and 2.5 in the hard band ( 2σ). Dividing these ∼ counts by the exposure map, we obtain the flux limit at each pixel for each band. The pixel area is then converted into a solid angle and the cumulative histogram of the flux limit is 2 2 constructed (Figure 2). The total survey area is 1100 arcmin ( 0.3 deg ). A more ≈ ≈ precise method of determining the survey area as a function of the X-ray flux is described by Kenter & Murray (2003); however, this would affect only the faint tail of the sample and would not significantly alter the present results. Therefore, a more sophisticated treatment is deferred to a later paper. 3.2. Point Source Detection To perform X-ray source detection, we applied the CIAO wavelet detection algorithm wavdetect (Freeman et al. 2002). Although several other methods have been used in other survey fields to find sources in Chandra observations (e.g., Giacconi et al. 2002; Nandra et al. 2005), we chose wavdetect to allow a straightforward comparison between sources found in our catalog with those found in the CDF-S (Giacconi et al. 2002; Alexander et al. 2003). Moreover, wavdetect is more robust in detecting individual sources in crowded fields and in identifying extended sources than the other CIAO detection algorithm, celldetect. Point- source detection was performed in each standard band (see Table 3) using a “√2 sequence” of wavelet scales; scales of 1, √2, 2, 2√2, 4, 4√2, and 8 pixels were used. Brandt et al. (2001), for example, showed that using larger scales can detect a few additional sources at large off-axis angles but found that this “√2 sequence” gave the best overall performance across the CDF-N field. Moreover, as Alexander et al. (2003) point out, sources found with larger scales tend to have source properties and positions too poorly defined to give useful results. Our criterion for source detection is that a source must be found with a false-positive probability threshold (p ) of 1 10−7 in at least one of the three standard bands. This thresh × false-positive probability threshold is typical for point-source catalogs (e.g., Alexander et al. 2003; Wang et al. 2004), although Kim et al. (2004) found that a significance threshold parameter of 1 10−6 gave the most efficient results in the Chandra Multiwavelength Project × – 8 – (ChaMP) survey. We ran wavdetect using both probability thresholds and found that using the lower significance threshold (i.e., 1 10−6) results in only an additional 64 unique × sources. Visual inspection of each of these sources suggest they are bona fide X-ray sources. However, because these are sources found with the lower significance threshold, we present them in a separate table (the secondary catalog; Table 5). The primary catalog (Table 4) is a compilation of 587 unique sources found using the higher significance threshold in at least one of the three energy bands. For the remaining source detection parameters, we used the default values specified in CIAO which included requiring that a minimum of 10% of the on-axis exposure was needed in a pixel before proceeding to analyze it. We also applied the exposure maps generated for each pointing (see Section 3.1) to mitigate finding spurious sources which are most often located at the edge of the field of view. The number of spurious sources per pointing is approximately p N , where N thresh pix pix × is the total number of pixels in the image, according to the wavdetect documentation. Since 6 there are approximately 2 10 pixels in each image for each pointing, we expect 0.2 × ∼ −7 spurious sources per pointing per band for a probability threshold of 1 10 . Therefore, × treating the 12 images searched as independent, we expect 2-3 false sources in our primary ∼ catalog (Table 4) for the case of a uniform background. Of course the background is neither perfectlyuniformnorstaticastheleveldecreasesinthegapsbetween theCCDsandincreases slightly near bright point sources. As mentioned by Brandt et al. (2001) and Alexander et al. (2003), one might expect the number of false sources to be increased by a factor of ∼ 2–3 due to the large variation in effective exposure time across the field and the increase in background near bright sources due to the point-spread function (PSF) wings. But our false-sourceestimate islikely tobeconservative byasimilarfactorsince wavdetect suppresses fluctuationsonscales smaller thanthePSF.Thatis, asinglepixelisunlikely tobeconsidered a source detection cell — particularly at large off-axis angles (Alexander et al. 2003). The source lists generated by the procedure above foreach ofthe standard bands ineach of the pointings of the ECDFS were merged to create the point-source catalogs presented in Tables 4 and 5. The source positions listed in each catalog are the full band wavdetect- determined positions except when the source was detected only in the soft or hard bands. To ′′ identify the same source in the different energy bands, a matching radius of 2. 5 or twice the PSF size of each detect cell, whichever was the largest, was used. For comparison, Alexander ′′ ′ et al. (2003) and Nandra et al. (2005) used a matching radius of 2. 5 for sources within 6 of ′′ the aimpoint, and 4. 0 for sources with larger off-axis angles. With our method, 9 and 3 soft- and hard-band sources, respectively, have more than 1 counterpart, so we took the closest one. Note that both Tables 4 and 5 excludes sources found by wavdetect in which one or both of the axes of the “source ellipse” collapsed to zero. Over the survey field, 70 such sources are found; in general, these are unusual sources and although the formal probability – 9 – of being spurious is low, there may be problems with these detections. Hornschemeier et al. (2001) found that using the wavdetect-determined counts for such objects as we do results in a gross underestimate of the number of counts even though the source was detected with −7 a probability threshold of 1 10 . Since these sources would appear in catalogs that do × circular aperture photometry instead, we present this list in a separate catalog (Table 6) for completeness. Below we define the columns in Tables 4 and 5, our primary and secondary source catalogs for the ECDFS survey. Column 1 gives the ID number of the source in our catalog. • Column 2 indicates the International Astronomical Union approved names for the • sources in this catalog. All sources begin with the acronym “CXOYECDF” (for “Yale 10 E-CDF”) . Columns 3 and 4 give the right ascension and declination, respectively. These are • wavdetect-determined positions for the unbinned images. If a source is detected in multiple bands, then we quote the position determined in the full band; when a source is not detected in the full band, we quote the soft-band position or the hard-band position. Column 5 gives the PSF cell size, in units of arcseconds, as determined by wavdetect. • The farther off-axis a source lies, the larger the PSF size. −1 Columns 6, 7, and 8 give the count rates (in units of cts s ) in the full band and • the corresponding upper and lower errors estimated according to the prescription of Gehrels (1986). If a source is undetected in this band, no count rate is tabulated. −1 Columns 9, 10, and 11 give the count rates (in units of cts s ) in the soft band and • the corresponding upper and lower errors estimated according to the prescription of Gehrels (1986). If a source is undetected in this band, no count rate is tabulated. −1 Columns 12, 13, and 14 give the count rates (in units of cts s ) in the hard band and • the corresponding upper and lower errors estimated according to the prescription of Gehrels (1986). If a source is undetected in this band, no count rate is tabulated. −2 −1 Column 15 lists the full band flux (in units of erg cm s ) calculated using a photon • slope of Γ =1.4 and corrected for Galactic absorption. If a source was undetected in 10 Name registration submitted to http://cdsweb.u-strasbg.fr/viz-bin/DicForm. – 10 – the full band but was detected in the hard or soft band, the hard- or soft- band flux (in that order of priority) was used to extrapolate to the full band assuming a photon slope of 1.4. −2 −1 Column 16 lists the soft band flux (in units of erg cm s ) calculated using a photon • slope of Γ =1.4 and corrected for Galactic absorption. If a source was undetected in the soft band but was detected in the full or hard band, the full- or hard- band flux (in that order of priority) was used to extrapolate to the soft band assuming a photon slope of 1.4. −2 −1 Column 17lists the hardbandflux (inunits of erg cm s ) calculated using a photon • slope of Γ =1.4 and corrected for Galactic absorption. If a source was undetected in the hard band but was detected in the full or soft band, the full- or soft- band flux (in that order of priority) was used to extrapolate to the hard band assuming a photon slope of 1.4. Column 18 provides individual notes for each source. Examples include the catalog ID • (c#) if detected in the CDF-S by Alexander et al. (2003), or if the source was selected from a band other the full band (’h’ or ’s’) or only detected in the full band (’f’). To determine source counts for each of our sources, we extracted counts in the standard bands from each of the images using the geometry of the wavdetect source cell and the wavdetect-determined source position. For example, to determine the counts in the soft band, we used the position and geometry determined by wavdetect in the soft band image to extract soft band counts. Some studies use circular aperature photometry to extract sources counts. However, as both Hornschemeier et al. (2001) and Yang et al. (2004; see their Figure 5) demonstrate, both techniques generally return the same result. Net count rates were then calculated using the effective exposure (which includes vignetting) for each pointing(exposuremapsgeneratedasdescribedinSection3.1). Errorswerederivedfollowing Gehrels (1986), assuming an 84% confidence level. Note that the exposure maps do account forthedegradationofthesoft X-rayresponse ofACIS due tothebuild-up ofacontamination layer on the ACIS optical blocking filter (Marshall et al. 2004; see Section 3.4). Therefore, the count rates reported in Table 4 are exposure- and contamination-corrected. In Table 7 we summarize the source detections in the three standard bands, and in Table 8 we summarize the number of sources detected in one band but not in another. To convert the count rates to flux, we determined the conversion factor for each band assuming a photon slope of Γ = 1.4 and the mean Galactic N absorption along the line-of-sight for H 19 −2 each of the 4 pointings (N = 9 10 cm ; Stark et al. 1992). H ×