X-RAY AND MULTIWAVELENGTH STUDIES OF ACTIVE GALACTIC NUCLEI IN THE CHANDRA DEEP FIELDS

Transcription

1 The Pennsylvania State University The Graduate School Department of Astronomy and Astrophysics X-RAY AND MULTIWAVELENGTH STUDIES OF ACTIVE GALACTIC NUCLEI IN THE CHANDRA DEEP FIELDS A Dissertation in Astronomy and Astrophysics by Bin Luo c 2010 Bin Luo Submitted in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy December 2010

2 The dissertation of Bin Luo was reviewed and approved 1 by the following: W. Nielsen Brandt Professor of Astronomy and Astrophysics Dissertation Advisor Chair of Committee Jane Charlton Professor of Astronomy and Astrophysics George Pavlov Professor of Astronomy and Astrophysics Donald P. Schneider Professor of Astronomy and Astrophysics Tyce DeYoung Assistant Professor of Physics Lawrence W. Ramsey Professor of Astronomy and Astrophysics Head of the Department of Astronomy and Astrophysics 1 Signatures on file in the Graduate School.

3 iii Abstract With the advent of the newest generation of X-ray space observatories, Chandra and XMM-Newton, X-ray surveys have become the most effective tool to detect Active Galactic Nuclei (AGNs) and explore their physics. The 2 Ms Chandra Deep Field- North (CDF-N) and Chandra Deep Field-South (CDF-S) surveys (jointly the Chandra Deep Fields) are the two deepest X-ray surveys ever performed. We utilized these unprecedented X-ray data along with the superb multiwavelength coverage in these fields to study AGN properties in the distant universe. (1) We constrained X-ray outbursts from galactic nuclei with harder spectra, higher redshifts, and lower luminosities than have been studied previously. We performed a systematic survey of optical galaxies in the Chandra Deep Fields to search for X-ray outbursts. No outbursts were found, and thus we set tight upper limits on the rate of such events in the Universe. For an outburst with X-ray luminosity erg s 1 and adurationof6months, theupperlimit onits event rateis 10 4 galaxy 1 yr 1, roughly consistent with theoretical predictions. Our results also suggest that the X-ray luminosity function for moderate-luminosity active galactic nuclei is not primarily due to stellar tidal disruptions (Chapter 2). (2) We discovered the most-distant double-peaked emitter, CXOECDFS J , at z = A Keck/DEIMOS spectrum shows a clearly double-peaked broad Mg ii λ2799 emission line, with FWHM km s 1 for the line complex. This is one of a handful of double-peaked emitters known to be a luminous quasar, with excellent multiwavelength coverage and a high-quality X-ray spectrum. The local viscous energy released from the line-emitting region of the accretion disk is probably insufficient to power the observed line flux, and external illumination of the disk appears to be required. The illumination cannot arise from a radiatively inefficient accretion flow as suggested for prototype double-peaked emitters (Chapter 3). (3) We presented point-source catalogs for the 2 Ms exposure of the CDF-S, including 462 main catalog sources and 116 supplementary catalog sources. The 2 Ms CDF-S achieves on-axis sensitivity limits of and erg cm 2 s 1 for the and 2 8 kev bands, respectively. We performed detailed classification of the main catalog sources. Optical to radio multiwavelength identifications were carried out using the likelihood-ratio method, resulting in reliable counterparts for 442 (95.7%) of the X-ray sources, with an expected false-match probability of 6.2%. High-quality photometric redshifts were calculated, which are the best obtained so far for faint X-ray sources. The median redshift is 1.3 for the CDF-S X-ray sources, and we have discovered 10 high-redshift (z > 4) AGN candidates. About 80% of the X-ray sources are AGNs, among which 72% are obscured (Chapters 4 and 5). (4) We utilized and improved the relative infrared star formation rate excess (ISX) selection method to search for heavily obscured and Compton-thick (CT) AGN candidates in the CDF-S at z We have discovered 242 ISX sources; an X-ray stacking analysis of 23 of the objects resulted in a very hard X-ray signal with an effective photon index of , indicating a significant contribution from obscured AGNs. Based on Monte Carlo simulations,we conclude that 74±25% of the galaxies selected host obscured AGNs, within which 94%

4 are heavily obscured and 79% are CT. The space density of these heavily obscured AGNs is (2.0 ± 0.7) 10 4 Mpc 3, much higher than those from previous discoveries. These heavily obscured objects contribute 3.1% of the total cosmic X-ray background (XRB) flux in the kev band, which accounts for 10 30% of the unresolved XRB in this energy band according to population synthesis models (Chapter 6). iv

5 v Table of Contents List of Tables List of Figures Acknowledgments vii viii x Chapter 1. Introduction AGNs in Deep Extragalactic X-ray Surveys The Chandra Deep Fields Overview of this Thesis Chapter 2. Deep-Survey Constraints on X-ray Outbursts from Galactic Nuclei Introduction The Survey Optical Galaxy Sample X-ray Data Source Characterization and Outburst Searching Results and Discussions Upper Limits on the Outburst Event Rate Comparison with Previous Results Conclusions and Future Work Chapter 3. Discovery of the Most-Distant Double-Peaked Emitter at z = Introduction Keck Observations and Disk-Model Fit Multiwavelength Properties Discussion Chapter 4. The Chandra Deep Field-South Survey: 2 Ms Source Catalogs Introduction Observations and Data Reduction Observations and Observing Conditions Data Reduction Production of the Point-Source Catalogs Image and Exposure Map Creation Point-Source Detection Point-Source Catalogs Main Chandra Source Catalog Supplementary CDF-S plus E-CDF-S Chandra Source Catalog

6 Supplementary Optically Bright Chandra Source Catalog Background and Sensitivity Analysis Number Counts for the Main Chandra Catalog Summary Chapter 5. Identifications and Photometric Redshifts of the 2 Ms Chandra Deep Field-South Sources Introduction Multiwavelength Identifications of the 2 Ms CDF-S X-ray Sources X-ray and Optical to Radio Data Matching Method Matching Results Comparison with the Error-Circle Matching Method Photometric Redshifts Multiwavelength Photometric Data Photometric Redshift Fitting Photometric Redshift Results Photometric Redshift Accuracy Comparison with Other AGN Photometric-Redshift Accuracies AGN Classification and Best-Fit SED Templates Discussion Nature of the Unidentified X-ray Sources Future Prospects for Improved Source Identification Summary of Results Chapter 6. Revealing a Population of Heavily Obscured Active Galactic Nuclei at z in the Chandra Deep Field-South Introduction The ISX Sample Multiwavelength Data and Source Properties Sample Selection X-ray Stacked Properties and CT AGN Fraction X-ray Stacking Analysis Heavily Obscured AGN Fraction Discussion Space Density and Contribution to the XRB Additional Samples and Subsamples Observational Prospects for Distant Heavily Obscured AGNs Conclusions and Future Work Chapter 7. Summary Bibliography vi

7 vii List of Tables 2.1 Observation Epochs for the Chandra Deep Fields SED Data for J Journal of Chandra Deep Field-South Observations Main Chandra Catalog Summary of Chandra Source Detections Sources Detected in One Band but not Another Supplementary CDF-S plus E-CDF-S Chandra Catalog Supplementary Optically Bright Chandra Catalog Background Parameters Summary of the Likelihood-Ratio Matching Parameters and Results List of the Primary Counterparts List of the Secondary Counterparts Photometric Catalog Photometric Redshift Catalog List of ISX and ISN Sources in the X-ray Stacking Analysis List of ISX and ISN Sources in the X-ray Stacking Analysis Stacked X-ray Properties Comparison of Stacked and Simulated Counts for the ISX Sample

8 viii List of Figures 1.1 The comoving space density of X-ray selected AGNs Deepest X-ray surveys and false-color image of the CDF-S Spatial distributions the galaxy samples Redshift distributions of the galaxy samples Distributions of the variability factors SMBH mass functions for the galaxy samples X-ray Luminosity distributions for the galaxy samples The total rest-frame sensitive time as a function of outburst duration for Γ = The total rest-frame sensitive time as a function of outburst duration for Γ = Upper limits on the observed outburst rates as functions of the lower limit on X-ray luminosity Upper limits on the observed and expected outburst rates as functions of the lower limit on the SB X-ray luminosity Upper limits on the observed and expected outburst rates as functions of the lower limit on the HB X-ray luminosity Upper limits on the observed and expected outburst rates as functions of the lower limit on the X-ray luminosity, with a modified theoretical prediction Upper limits on the observed and expected outburst rates as functions of the lower limit on the X-ray luminosity, without the constraint of Eddington limit Rest-frame NUV spectrum of J showing the double-peaked Mg ii line Multiwavelength images of J Radio through X-ray SED of J X-ray spectrum of J overlaid with the best-fit model Redshift-luminosity distribution of double-peaked emitters Full-band raw image of the 2 Ms CDF-S Full-band exposure map of the 2 Ms CDF-S Amount of survey solid angle having at least a given amount of full-band effective exposure for the CDF-S Chandra false-color image of the 2 Ms CDF-S Positional offset vs. off-axis angle for sources in the main Chandra catalog Histograms showing the distributions of positional offset for sources in the main Chandra catalog Histograms showing the distributions of detected source counts for sources in the main Chandra catalog

9 4.8 Histograms showing the distributions of X-ray fluxes for sources in the main Chandra catalog WFI R-band postage-stamp images for the sources in the main Chandra catalog Positions of the sources in the main and supplementary Chandra catalogs Band ratio vs. full-band count rate for sources in the main Chandra catalog WFI R-band magnitude vs. soft-band flux for X-ray sources in the main and supplementary catalogs Full-band sensitivity map of the 2 Ms CDF-S Survey solid angle as a function of the flux limit Number of sources, N(> S), brighter than a given flux, S, for the soft band and hard band The SED of source 312 along with the selected best-fit SED template The expected magnitude distributions of the background sources and counterparts for the GEMS z-band catalog Positional offset vs. off-axis angle for sources in the main X-ray source catalog that were matched to the VLA radio sources Positions of the 446 identified and 16 unidentified X-ray sources Distributions of the positional offset between an X-ray source and its primary counterpart Cumulative false-match probability as a function of the R-band limiting magnitude The observed median SED for the CDF-S X-ray sources and its best-fit SED template Distributions of the number of filters and the number of filters vs. the WFI/MUSYC R-band magnitude SEDs and best-fit templates for z > 5 sources and the median SEDs of z = 0 4 AGNs in the CDF-S Distribution of photo-z s and the redshift vs. luminosity distribution Comparison of the photo-z s and spec-z s The distribution of the photo-z accuracy The photo-z accuracy vs. WFI R-band magnitude The distributions of the photo-z errors X-ray band ratio vs. redshift for the AGNs R SFR vs. IR luminosity for sources in the parent sample The adaptively smoothed stacked images for the ISX and ISN samples The average SB and HB counts of the ISX sample as a function of the AGN fraction The effective photon index of the stacked X-ray signal vs. the average R SFR of the sample ix

10 x Acknowledgments I would like to thank my parents, Mingbao Luo and Shendi Huang, for their continuing support. No matter what happens, I always have you by my side. I am very grateful to my PhD advisor Niel Brandt for his invaluable guidance through my PhD study. Niel has taught me the proper way to carry out a scientific study, and has provided assistance in every step of the way. Every piece of the work presented here has been reviewed by Niel word by word for several times. Niel has also helped me in many other aspects that are crucial for a successful professional career in Astronomy. I am fortunate to have Niel as my advisor. I would also like to express my gratitude to my key collaborators, Dave Alexander, Franz Bauer, Marcella Brusa, Bret Lehmer, Don Schneider, Aaron Steffen, and Yongquan Xue, for their helpful roles in the work presented in this thesis. I acknowledge the suggestions and feedback from members of the CDFs team that have been very helpful for the research presented here. I also thank my thesis committee for their useful advice and suggestions during the progress of the thesis. I thank the faculty, staff, and students of the Astronomy Department for their assistance during my research, teaching, and outreach in the graduate school. I thank all the friends I have met during these five years in Penn State. In particular, I would like to thank my friend Junfeng Wang and his wife Lei Lin, who gave me a warm welcome to the US and have provided significant help since then.

11 1 Chapter 1 Introduction In this thesis, I utilize deep extragalactic X-ray surveys along with multiwavelength photometric and spectroscopic data to study the properties of active galactic nuclei (AGNs) in the Chandra Deep Fields, the 2 Ms Chandra Deep Field-North (CDF-N; Alexander et al. 2003) and 2 Ms Chandra Deep Field-South (CDF-S; Luo etal.2008). TheChandra DeepFieldshavefoundthehighestdensityofAGNsonthesky, providing a unique opportunity to explore the physics of typical, moderate-luminosity AGNs in the high-redshift universe. 1.1 AGNs in Deep Extragalactic X-ray Surveys An AGN is a compact region at the center of a galaxy emitting at almost all wavelengths of the electromagnetic spectrum. The radiation from the AGN outshines the hostgalaxy inmanycases, andisbelievedtobepoweredbyaccretion ontoasupermassive black hole (SMBH). The X-ray continuum emission is produced mainly from the hot corona of the accretion disk via inverse Compton scattering, and can be broadly described as a power law from 1 kev to a cut-off energy of a few hundred kev. X-ray emission is an ubiquitous property of AGNs. Historically, AGNs were mostly identified through optical spectroscopy, including luminous optical AGNs (e.g., quasars) and local moderate-luminosity AGNs (e.g., Seyfert galaxies). Optical identification of moderate-luminosity AGNs in the distant universe is challenging due to frequent obscuration and host-galaxy contamination; on the other hand, X-ray emission is less affected by these factors. With the advent of superb X-ray observatories, Chandra and XMM-Newton, X-ray surveys have become the most effective technique to detect AGNs. Compared to previous X-ray missions, Chandra and XMM- Newton can provide imaging spectroscopy surveys with much higher sensitivities and source positional accuracies, allowing construction of large samples of faint extragalactic X-ray sources. Deep X-ray surveys have identified reliable and fairly complete samples of AGNs out to z 5 (e.g., see Brandt & Hasinger 2005 for a review). The observed AGN sky density reaches 7200 deg 2 in the deepest X-ray surveys, the Chandra Deep Fields (e.g., Bauer et al. 2004), which is 10 times higher than that found in deep optical photometric and spectroscopic surveys (e.g., Wolf et al. 2003; Hunt et al. 2004). Compared to wide-field shallower surveys, deep surveys such as the Chandra Deep Fields are able to probe less luminous and more typical objects living in earlier epochs of the universe. Figure 1.1 displays the comoving space density of X-ray selected AGNs as a function of redshift and luminosity (Ueda et al. 2003; Brandt & Hasinger 2005; Hasinger et al. 2005). Moderate-luminosity AGNs are more common than bright quasars, and they reach their peak space density at lower redshifts.

12 2 Fig. 1.1 The comoving space density of AGNs selected in the (a) kev band and (b) 2 10 kev band as a function of redshift, adapted from Brandt & Hasinger (2005). Results for five X-ray luminosity ranges are shown in different colors and are labeled with logarithmic luminosity values. X-ray detected AGNs are largely responsible for the observed cosmic X-ray background (XRB). The XRB was discovered in the 1960s through a rocket flight (Giacconi et al. 1962), and it was the first cosmic background radiation detected. The near isotropy of the 2 kev XRB strongly suggested an extragalactic origin (e.g., Schwartz et al. 1976). With the development of X-ray telescopes (e.g., ROSAT), the XRB became resolved into discrete sources (e.g., Shanks et al. 1991). In the era of Chandra and XMM-Newton, the resolved fraction has increased to 70 90% in the kev range (e.g., Moretti et al. 2002; Bauer et al. 2004; Hickox & Markevitch 2006). Most of the discrete sources contributing to the XRB are obscured (N H cm 2 ) but Compton-thin (N H cm 2 ) AGNs (e.g., Bauer et al. 2004; Tozzi et al. 2006). The resolved fraction of the XRB decreases toward higher energies, being 60% in the 6 8 kev band and 50% in the 8 12 kev band. (Worsley et al. 2004, 2005). Population synthesis models suggest that Compton-thin AGNs discovered at energies below 10 kev cannot account for the XRB flux at kev entirely, and an additional population of Compton-thick (N H > cm 2 ) AGNs is required. The number density of these Compton-thick AGNs is estimated to be of the same order as that of moderately obscured AGN (e.g., Gilli et al. 2007). However, as the X-ray emission (below at least 10 kev) of such sources is significantly suppressed, only a few Compton-thick AGNs have been clearly identified (e.g., Tozzi et al. 2006; Alexander et al. 2008), and thus a significant fraction of the AGN population remains undetected even in the deepest X-ray surveys. Several future hard X-ray missions such as the Nuclear Spectroscopic Telescope Array (NuSTAR) and ASTRO-H are aiming to detect directly Compton-thick AGNs in the distant universe. An X-ray stacking technique, which stacks X-ray cutouts at given source positions, can also be used to assess the average properties of a sample of Compton-thick AGN candidates.

13 3 1.2 The Chandra Deep Fields Being the newest generation of X-ray space observatories, Chandra and XMM- Newton have complementary capabilities for performing deep surveys. Chandra has a lower kev background than XMM-Newton and can localize sources with sub-arcsecond positional accuracy. Therefore, Chandra can achieve the highest possible kev sensitivity without suffering from significant source confusion (e.g., Alexander et al. 2003). In contrast, XMM-Newton has a larger photon collecting area than Chandra, which is a distinct advantage for X-ray spectroscopy. XMM-Newton also has a larger field of view, allowing surveys of larger solid angles in the same amount of time. Fig. 1.2 (a) On-axis kev flux limit vs. solid angle for the deepest surveys by Chandra (blue and red circles), XMM-Newton (green stars), and ROSAT (black squares). The two red circles indicate the Chandra Deep Fields and the Extended Chandra Deep Field-South. (b) False-color image of the 2 Ms CDF-S. This image is a color composite of the exposure-corrected adaptively smoothed images in the kev (red), 2 4 kev (green), and 4 8 kev (blue) bands. Figure 1.2a shows the sensitivities and areas of the deepest Chandra, XMM-Newton, androsat surveys. Thecurrenthighest0.5 2keVsensitivity( ergcm 2 s 1 ) was accomplished by the 2 Ms CDF-N and CDF-S, jointly the Chandra Deep Fields; this sensitivity corresponds to sources with Chandra count rates of 1 photon per 4 days. The CDF-N and CDF-S each cover an area of 450 arcmin 2 with 500 sources detected. The Chandra Deep Fields also have superb coverage at optical, infrared, submillimeter and radio wavelengths, including observations from space-borne observatories such as HST, GALEX, and Spitzer, as well as ground-based telescopes such as VLA, VLT, and MPG/ESO. A smoothed image of the 2 Ms CDF-S is shown in Figure 1.2b. The Chandra Deep Fields will continue to be premiere multiwavelength deep-survey fields

14 for the coming decades, and the CDF-S has been scheduled to have another 2 Ms exposure released later 2010, reaching a total exposure of 4 Ms. The CDF-S was flanked by four contiguous 250 ks Chandra observations, the Extended Chandra Deep Field-South (E-CDF-S; Lehmer et al. 2005), which cover a solid angle of 1100 arcmin 2. The E-CDF-S complements the CDF-S with relatively shallower X-ray coverage over a larger area (see Fig. 1.2a), and it also has superb multiwavelength supporting data. AGNs in the Chandra Deep Fields and the E-CDF-S are the main focus of this thesis. 1.3 Overview of this Thesis In this thesis, I study several aspects of AGN physical properties, taking advantage of the deep X-ray and multiwavelength data in the Chandra Deep Fields. Each chapter is a self-contained work with comprehensive introductions and motivations provided at the beginning. In Chapter 2, I present tight constraints on X-ray outbursts from galactic nuclei using the CDF-N and CDF-S data; X-ray outbursts are associated with transient fueling events of SMBHs, and they were suggested to be related to the luminosity function of moderate-luminosity AGNs. In Chapter 3, I report the discovery of the most-distant double-peaked emitter; double-peaked emitters are a special type of AGNs that emit broad and double-peaked low-ionization lines. In Chapter 4, I present point-source catalogs for the 2 Ms exposure of the CDF-S; sensitivities and source number counts are also calculated. In Chapter 5, I perform detailed multiwavelength identifications for the 2 Ms CDF-S X-ray sources, and derive high-quality photometric redshifts for these sources. In Chapter 6, I select a sample of infrared star-formation-rate excess objects in the CDF-S, which are likely candidates for heavily obscured or even Compton-thick AGNs; I study the nature of these sources using an X-ray stacking technique. I summarize in Chapter 7. Chapters 2 and 3 are individual research projects based on the data in the Chandra Deep Fields and the E-CDF-S, while Chapters 4, 5, and 6 are closely related, focusing on the 2 Ms CDF-S region. Several of the chapters present in this thesis have been published, and the references to the journal articles are listed below: 1. Chapter 2: Published as Deep-Survey Constraints on X-Ray Outbursts from Galactic Nuclei, Luo et al. (2008a; ApJ, 674, 122) 2. Chapter 3: Published as Discovery of the Most-Distant Double-Peaked Emitter at z = 1.369, Luo et al. (2009; ApJ, 695, 1227) 3. Chapter 4: Published as The Chandra Deep Field-South Survey: 2 Ms Source Catalogs, Luo et al. (2008b; ApJS, 179, 19) 4. Chapter 5: Published as Identifications and Photometric Redshifts of the 2 Ms Chandra Deep Field-South Sources, Luo et al. (2010; ApJS, 187, 560) 4

15 5 Chapter 2 Deep-Survey Constraints on X-ray Outbursts from Galactic Nuclei 2.1 Introduction X-ray observations, mainly with ROSAT, have found about six transient, largeamplitude X-ray outbursts from galactic nuclei (e.g., Donley et al. 2002; Komossa 2002; Komossa et al. 2004; Vaughan, Edelson, & Warwick 2004; and references therein). These events have variability factors of , peak kev X-ray luminosities comparableto those of local Seyfertgalaxies ( erg s 1 ), decay timescales of months, and soft X-ray spectra. They have been observed in both active and inactive galaxies(i.e., galaxies with and without a persistently accreting supermassive black hole, respectively). The estimated event rates of these outbursts suffer from large systematic and statistical uncertainties. In nearby inactive galaxies the event rate is 10 5 galaxy 1 yr 1, while in active galaxies the event rate appears to be 100 times higher (e.g., Donley et al. 2002). At least some of these outbursts induce accompanying optical nuclear variability (e.g., Brandt, Pounds, & Fink 1995; Grupe et al. 1995). Recently, five new candidate X-ray outbursts were found in the XMM-Newton Slew Survey (Esquej et al. 2007), and one outburst has been detected in the ultraviolet and is thought to have the same origin as X-ray outbursts (Gezari et al. 2006). The physical origin of nuclear X-ray outbursts remains mysterious. The most likely explanation is that they are caused by transient fueling events of nuclear supermassive black holes (SMBHs). Since SMBHs are thought to be ubiquitous in nucleated galaxies (e.g., Ferrarese & Ford 2005 and references therein), fueling events seem inevitable in crowded galactic centers. Fueling may occur when a star, planet, or gas cloud is tidally disrupted and partially accreted (e.g., Rees 1990; Ulmer 1999; Li, Narayan, & Menou 2002). The predicted event rate for stellar tidal disruptions (e.g., Magorrian & Tremaine 1999; Syer & Ulmer 1999) is roughly consistent with the poorly constrained rate of large-amplitude X-ray outbursts in inactive galaxies. However, Wang & Merritt (2004) predicted a rate that was a factor of 10 higher than earlier results, mainly due to downwardly revised black-hole masses. This predicted rate, apparently in excess of the observed rate of X-ray outbursts, may indicate that a large fraction of tidal-disruption events do not exhibit expected X-ray outburst characteristics, perhaps due to short durations of the events or X-ray obscuration. In some cases, transient fueling could be due to accretion-disk instabilities (e.g., Siemiginowska, Czerny, & Kostyunin 1996). Some outbursts could also perhaps be explained by X-ray afterglows of gamma-ray bursts (GRBs), though no simultaneous GRBs were reported for the known X-ray outbursts (e.g., Komossa & Bade 1999).

16 Aside from the innate scientific interest of nuclear X-ray outbursts, determining the rate and properties of such events is relevant to planning future missions such as the Black Hole Finder Probe (BHFP), Lobster, and erosita. The BHFP, for example, is likely to observe in the hard X-ray band, and the large sky coverage of this mission will allow intrinsically rare transient events to be studied(e.g., Grindlay 2005). Facilities such as the Large Synoptic Survey Telescope, the Joint Dark Energy Mission, and the Laser Interferometer Space Antenna should also allow the accompanying optical (e.g., Brandt 2005) and gravitational-wave (e.g., Kobayashi et al. 2004) outbursts to be studied. Previous studies of X-ray outbursts from galactic nuclei have delivered fascinating results but require extension so that this phenomenon can be better understood. For example, advances are required in the following directions: 1. Outbursts with harder X-ray spectra. As noted above, the outbursts discovered to date have generally had soft X-ray spectra with effective kev power-law photonindicesofγ 3 5. However, thismaypartiallybeaselectioneffectowingto the soft X-ray bandpass of the ROSAT satellite. Transient fueling events of SMBHs should be capable of generating harder X-ray spectra (Γ ) via Compton upscattering in the accretion flow; such spectra are observed from most active galactic nuclei (AGNs) as well as from transient Galactic black holes. Indeed, one X-ray outburst has shown evidence for spectral hardening as it declined (Komossa et al. 2004). In addition, obscuration can harden the observed X-ray spectra of an outburst, which would make it more difficult to detect in the current outburst surveys. A fairly small column density of N H = cm 2 reduces the expected kev flux from a soft-spectrum outburst by a factor of Higher redshift outbursts. The X-ray outbursts discovered to date are at low redshift with z = Since only a small fraction of cosmic time is spanned by this redshift range and the source statistics are limited, little is known about the redshift evolution of their frequency. It is plausible that the X-ray outburst rate could show evolution over cosmic time, considering that AGNs and galaxies both show strong evolution. The number densities of comparably X-ray luminous AGNs evolve upward by factors of out to z 1 (e.g., Brandt & Hasinger 2005 and references therein), and Milosavljević, Merritt, & Ho (2006) have suggested that a significant portion of the X-ray luminosity function of such AGNs may be comprised of sources powered by tidal disruptions. 3. Lower luminosity outbursts. Transient phenomena in complex natural systems often follow power-law distributions in frequency of occurrence, such that low-amplitude events are more common than high-amplitude events(e.g., earthquakes, avalanches, solar flares, and ecological extinctions). It is plausible that such a distribution applies for X-ray outbursts in galactic nuclei. Low-mass fueling events, such as partial tidal disruptions or the accretion of brown dwarfs, planets, or small gas clouds, would then be more common than high-mass fueling events. In this paper, we utilize data from the Chandra Deep Fields (see Brandt & Hasinger 2005 for a review) to constrain the rate of X-ray outbursts with harder X-ray spectra, 6

17 higher redshifts, and lower luminosities than those studied to date. Observations of the Chandra Deep Fields were made in several well-separated and individually sensitive epochs, allowing effective constraints to be placed upon X-ray outbursts that evolve on timescales of months in the rest frame. Some variability work has been performed on the Chandra Deep Field sources. For example, Paolillo et al. (2004) studied the X-ray variability of 350 sources detected in the Chandra Deep Field-South (CDF-S), but this work has not been optimized to detect and to constrain systematically X-ray outbursts of the type relevant here. WeadoptH 0 = 70km s 1 Mpc 1, Ω M =0.3, andω Λ =0.7throughoutthispaper. All coordinates are J2000. The relevant Galactic column densities are cm 2 for the Chandra Deep Field-North (CDF-N; Lockman 2004) and cm 2 for the CDF-S (Stark et al. 1992). All X-ray fluxes and luminosities quoted throughout this paper have been corrected for Galactic absorption using these column densities. 2.2 The Survey In this survey, we focused on optically detected galaxies in the CDF-N (Brandt et al. 2001b; Alexander et al. 2003) and the CDF-S (Giacconi et al. 2002; Alexander et al. 2003). An optical galaxy sample was compiled from several catalogs, with the redshifts collected from literature. We grouped the Chandra observations into several well-separated epochs. The five epochs for the CDF-N span 798 days, and the four epochs for the CDF-S span 1828 days. At the median redshift of 0.8 for the galaxies in our sample, the rest-frame coverage would be 443 days for the CDF-N and 1016 days for the CDF-S. X-ray count rates or upper limits for every optical galaxy were measured for each epoch in three standard Chandra bands: kev (full band; FB), kev (soft band; SB), and 2 8 kev (hard band; HB). These measurements were then analyzed to search for any X-ray outburst candidates. Here we define X-ray outbursts to be transient events in galaxy nuclei that cause the count rate to vary by a minimum factor of 20 in one of the three standard bands. A minimum variability factor of 20 was chosen to discriminate against normal AGN variability, which typically has variability factors of 2 5 but can be as large as (e.g., Paolillo et al. 2004) Optical Galaxy Sample Optical galaxies were selected from the catalog of Capak et al. (2004) for the CDF-N, and from the COMBO-17 catalog (Wolf et al. 2004) for the CDF-S. To ensure that these galaxies were covered by either the CDF-N or the CDF-S, we used the kev exposure maps of the 2 Ms CDF-N and 1 Ms CDF-S as filters. A small portion ( 30 arcmin 2 ) of the CDF-N was not covered by the Capak catalog, so the coverage of this survey is 418 arcmin 2 for the CDF-N and 391 arcmin 2 for the CDF-S. We only chose optical galaxies with AB magnitudes R < 25, to optimize our selection of a large galaxy sample with redshift measurements. We examined manually those sources with R < 22 and removed some false sources, which were generally around bright objects. Galaxies in the COMBO-17 catalog with values for photometry flags 8 were ignored, 7

18 as suggested by the creators of that catalog. In this way, we selected galaxies in the CDF-N and galaxies in the CDF-S. We searched the literature for redshift measurements for this optical galaxy sample. Spectroscopic redshifts were obtained from the spectroscopic surveys of Cowie et al. (2004), Reddy et al. (2006), and the Team Keck Treasury Redshift Survey (Wirth et al. 2004) in the CDF-N, and from the VIMOS VLT Deep Survey (Le Fèvre et al. 2004) and the VLT/FORS2 spectroscopic survey (Vanzella et al. 2005, 2006) in the CDF-S. Photometric redshifts were obtained from P. Capak et al. (2010, in preparation) in the CDF-N, and from Mobasher et al. (2004) and Wolf et al. (2004) in the CDF-S. A matching radius of 1 was used when cross correlating different optical catalogs. We did not find redshift measurements for 108 faint galaxies in the CDF-S. After removing these sources our final optical galaxy sample contained and galaxies in the CDF-N and CDF-S, respectively. The spatial distributions of these galaxies in the CDF-N and CDF-S are shown in Figure 2.1. We found spectroscopic redshifts for 2100 galaxies in the CDF-N and 1534 galaxies in the CDF-S, with the remaining sources having photometric redshifts. The redshift distributions are shown in Figure 2.2. To test the reliability of the photometric redshifts, we examined sources with both photometric and spectroscopic redshift measurements, and we studied the offset distribution of these two types of redshifts in the CDF-N and CDF-S separately. In both fields, the median offsets approach 0. The interquartile ranges are only 0.11 in the CDF-N and 0.20 in the CDF-S. The stellarity index (class star parameter in SExtractor) is a robust galaxy identifier, with 0 for confirmed galaxies, and 1 for confirmed stars (e.g., Groenewegen et al. 2002). Less than 2.5% of these galaxies had stellarity indices 0.9, so stars were well removed from this sample. This sample of galaxies contains a representative mix of field galaxies. Only 2% of the galaxies are X-ray detected, and thus the AGN fraction is small. Lehmer et al. (2008) constructed a similar galaxy sample in the Chandra Deep Fields, with z 850 < 23 and z = They studied the rest-frame optical colors as well as the Sérsic indices (e.g., Häussler et al. 2007) of galaxies in their sample, finding 2544 late-type galaxies and 727 early-type galaxies out of galaxies. Our sample should have a similar composition in terms of galaxy morphology. Recent close-pair studies have suggested that the merger rate of field galaxies remains fairly small out to z 1 (e.g., Bundy et al. 2004, 7% infrared-selected pairs). Elmegreen et al. (2007) examined the GEMS (Galaxy Evolution from Morphology and SEDs) and the southern GOODS (Great Observatories Origins Deep Survey) fields, which overlap with the CDF-S, and identified 300 interacting galaxies to z 1.4 out of over 8000 optical galaxies. Thus our sample should also be dominated by non-interacting galaxies X-ray Data X-ray data analyses were performed on the CDF-N, CDF-S, and the Extended Chandra Deep Field-South (E-CDF-S; Lehmer et al. 2005). There are 20 observations for the CDF-N, 11 observations for the CDF-S (Alexander et al. 2003), and 9 observations for the E-CDF-S(Lehmer et al. 2005). We grouped these into epochs according to their observation dates, five epochs for the CDF-N and four epochs for the CDF-S (see Table 8

19 9 Fig. 2.1 Spatial distributions of the galaxies in this survey in the CDF-N and CDF-S. Dots with different colors represent galaxies in different magnitude bins. Squares are galaxies with X-ray counterparts as listed in Table 2.1. The gap at the top of the CDF-N is due to the lack of coverage of the optical catalog. 2.1). The E-CDF-S observations took place 4 years after the last CDF-S observation and were counted as the fourth epoch for the CDF-S. The E-CDF-S covered the entire CDF-S and consisted of four distinct observational fields with different observation dates. We list these fields separately in Table 2.1. We created images of each epoch from the reduced and cleaned level 2 event files (Alexander et al. 2003; Lehmer et al. 2005), using the standard ASCA grade set (ASCA grades 0, 2, 3, 4, and 6) for the three standard bands. There were thus 15 CDF-N images, 9 CDF-S images, and 12 E-CDF-S images. The aim point for a given image was taken to be the average value over its observations weighted by exposure time. To construct X-ray source lists, we ran wavdetect (Freeman et al. 2002) on the 24 images in the CDF-N and CDF-S using a 2 sequence of wavelet scales (i.e., 1, 2, 2, 2 2, 4, 4 2, and 8 pixels). The false-positive probability threshold in each wavdetect run was set to Source lists for the E-CDF-S were taken directly from Lehmer et al. (2005) because their source-detection method is the same as that used here. To improve the accuracy of the X-ray source positions, we matched the optical and X-ray source lists with a matching radius of 2. 5 and centered the distributions of offsets in right ascension and declination between the optical and X-ray source positions. This resulted in small (< 1. 0) image-dependent astrometric shifts for all X-ray sources. Compared to the main Chandra X-ray source catalogs in Alexander et al. (2003), our X-ray source lists give 10 new sources for each image, due to the less-conservative false-positive probability threshold we were using or the variability of some sources. We are interested in the X-ray properties of the optical galaxies selected in 2.2.1, so we searched for X-ray counterparts of these galaxies by cross-correlating the optical

20 Fig. 2.2 Redshift distributions of the galaxies in this survey for the CDF-N and CDF-S. The median and mean redshifts are 0.78 and 0.91 for the CDF-N, and 0.84 and 1.06 for the CDF-S, respectively. The dashed lines show the median values. 10

21 and X-ray source lists with a matching radius of 1. 0 for sources within 6 of the average aim point and 1. 3 for larger off-axis angles. In 47 cases where there is more than one optical galaxy associated with an X-ray source, the closest one is selected. None of the galaxies is associated with more than one X-ray source. A summary of these findings is given in Table 2.1. Around one third of the X-ray sources do not have counterparts in the galaxy sample, either because their R magnitude is fainter than 25 or they are stars. We also estimated the expected number of false matches by artificially offsetting the X-ray source coordinates in right ascension and declination by 5. 0 (both positive and negative shifts) and re-correlating with the optical sources. On average, the number of false matches is only 6% of the total matches in our sample for both the CDF-N and CDF-S Source Characterization and Outburst Searching We measured the X-ray counts coincident with every optical galaxy in all 36 images using circular-aperture photometry. We have chosen the aperture radii based on the encircled-energy function of the Chandra PSF, which was calculated using MARX (Model of AXAF Response to X-rays 1 ) simulations. The 90% encircled-energy radius of the PSF was used. The circular aperture was centered at the position of every optical galaxy or its X-ray counterpart if it was X-ray detected. If a galaxy was not fully covered by a certain image, then it was marked as undetectable for this image. The local background was determined in an annulus outside of the source-extraction region using background maps with known X-ray sources carefully removed to avoid possible contamination from adjacent sources. To create a background map for a given image, we first merged our Chandra X-ray source list from this image to the main Chandra catalogs (Alexander et al. 2003) to get a combined source list, using a matching radius of 2. 5 for sources within 6 of the average aim point and 4. 0 for larger off-axis angles. Then we masked out these sources using apertures with radii twice that of the 90% PSF encircled-energy radius. We filled the masked regions for each source with a local background estimated by making a probability distribution of counts using an annulus with inner and outer radii of 2 and 4 times the 90% PSF encircled-energy radius, respectively. The local background for every optical galaxy was then determined using the same annulus on this background map. When an optical galaxy was X-ray detected (i.e., had an X-ray counterpart) in a given image, the net number of source counts was calculated by subtracting the expected number of background counts from the number of counts in the aperture. Poisson errors were calculated following Gehrels (1986) and were propagated through this calculation. When an optical galaxy was not detected, an upper limit was calculated. If the number of counts in the aperture was 10, the upper limit was calculated using the Bayesian method of Kraft, Burrows, & Nousek (1991) for 99.87% confidence. For a larger number of counts in the aperture, a 3σ upper limit was set by multiplying the square root of the number of background counts by three. The number of source counts was then divided by the effective exposure time, which was the average value within the aperture on the 11 1 See

22 Table 2.1. Observation Epochs for the Chandra Deep Fields Exposure Time Obs. Date a Number of Sources b Epochs Obs. IDs (ks) t obs FB SB HB CDF-N Epoch 1 580, 967, 966, Dec CDF-N Epoch , 1671, Nov CDF-N Epoch , 2233, 2423, 2234, Feb CDF-N Epoch , 3388, 3408, Nov CDF-N Epoch , 3294, 3390, Feb CDF-S Epoch , Nov CDF-S Epoch 2 441, Jun CDF-S Epoch , 2405, 2312, 1672, 2409, 2313, Dec CDF-S Epoch 4 F01 c 5015, Mar CDF-S Epoch 4 F02 c 5017, May CDF-S Epoch 4 F03 c 5019, Nov CDF-S Epoch 4 F04 c 5021, 5022, Nov a Observations within each epoch only lasted for a few days, so we neglected the length of the epochs and only show the average observation dates weighted by exposure time. b Number of galaxies in the sample that are X-ray detected in this epoch for the particular band: kev (FB), kev (SB), or 2 8 kev (HB). c Observational fields of the E-CDF-S. Note that a large fraction of each of these fields extends outside the region of our outburst survey. 12

23 exposure map, to get the X-ray count rate (this procedure corrects the count rate for vignetting and other effects). The E-CDF-S, which serves as the fourth epoch for the CDF-S, contains four observational fields, which overlap in a few areas over 50 arcmin 2. If an optical galaxy was detected in more than one field, we chose data from the field with greater source counts. If it was detected in only one field, we chose data from this field. If it was not detected in any field but was in the overlapping area, we chose data from the field with the longest effective exposure time. We searched for outbursts following these steps: 1. For every optical galaxy, we calculated its count rate or upper limit on count rate in each epoch and each energy band (i.e., each of the 36 images), unless it was not fully covered by a given image. A galaxy had coverage in up to five epochs if it was in the CDF-N and up to four epochs if it was in the CDF-S. 2. For each energy band, we selected galaxies that were detected in at least one epoch. For each of these galaxies, we compared its highest count rate with its lowest count rate or upper limit on count rate. Thus we got either the variability factor or its lower limit for each source. If there was a variation of more than a factor of 20, then we considered that there was a candidate for an X-ray outburst in this galaxy. 2.3 Results and Discussions After systematically analyzing the count-rate variations, we found no outbursts in either the CDF-N or in the CDF-S. The distribution of the variability factors in the SB and HB is shown in Figure 2.3. The median relative uncertainty of these factors is 30%. The count-rate variations of a few sources exceed a factor of 10. However, these are all off-axis sources and the significances of the variations are < 1σ Upper Limits on the Outburst Event Rate The nondetection of any X-ray outbursts in this survey can be used to constrain the rate of such outbursts in the Universe. Since the detectability of an outburst depends upon its luminosity, the constraints will have a luminosity dependence. Moreover, not all galaxies in this survey are capable of producing outbursts. Galaxies without a central SMBH (e.g., dwarf irregulars) cannot tidally disrupt stars, and a SMBH of mass M BH M will swallow stars whole (e.g., Frank & Rees 1976). SMBH candidates with masses of a few 10 5 M in the centers of galaxies have been found (e.g., Greene & Ho 2004; Peterson et al. 2005). Thus we consider that only galaxies with a SMBH mass greater than 10 5 M and less than M can produce outbursts. SMBH masses were roughly estimated for all galaxies in this survey with the relation between M BH and total galaxy luminosity in the K band, L K,total. The relation was derived from the data for 27 low-redshift galaxies (10 late-type and 17 early-type galaxies) in Marconi & Hunt (2003). The intrinsic dispersion of this relation is 0.5 dex in log M BH. Although thereis evidence showingcosmic evolution of the M BH -σ (SMBH mass andbulge velocity 13

24 14 Fig. 2.3 Histograms showing the distribution of variability factors in the SB (upper panels) or HB (lower panels) for galaxies in the CDF-N and CDF-S. The variability factor of a given galaxy is defined as the ratio between its highest count rate and its lowest count rate or upper limit on count rate. The shaded areas are for galaxies with lower limits on variability factors only. dispersion) relation and the bulge-to-smbh mass ratio (e.g., Woo et al. 2006), the coexistent luminosity evolution makes the SMBHs at z 1 coincidently fall on nearly the same M BH versus R-band magnitude (M R ) relation (to 0.3 mag) as low-redshift galaxies (Peng et al. 2006). Thus we can expect that the M BH -L K,total relation also holds approximately at z 0.8 for this galaxy sample. L K,total for the galaxies in this survey was extrapolated from HK -band magnitudes (10826 galaxies, Capak et al. 2004) or z -band magnitudes (2873 galaxies, Capak et al. 2004) for the CDF-N, and from z-band magnitudes (10167 galaxies, Caldwell et al. 2008) or R-band magnitudes (802 galaxies, Wolf et al. 2004) for the CDF-S, using the spectral energy distribution for Sbc galaxies (Coleman, Wu, & Weedman 1980). The resulting conversions from HK, z, z and R magnitudes (AB system) to the standard K magnitude were given by HK K = 1.65, z K = 2.07, z K = 2.01, and R K = Based on the dispersion of the M BH -L K,total relation and the uncertainties in the color conversions, the derived M BH is estimated to be good within an order of magnitude (as we show below, our main results are not sensitive to M BH ). A comparison between the SMBH mass functions derived for the low-redshift galaxies in this sample and the local SMBH mass function for all galaxy types (Marconi et al. 2004) is shown in Figure 2.4. There is basic agreement between the shapes of these mass functions indicating that our mass measurements are reasonable. There are galaxies in the CDF-N and 8268 galaxies in the CDF-S with M BH in the range of 10 5 M < M BH < M. There will also be another constraint set bythephysicsofoutburstproduction, i.e., asmbhwithagiven masscannot produce outbursts with arbitrarily high luminosities owing to, e.g., the Eddington limit. However, we ignore this constraint for now, as we would first like to present model-independent limits. To determine the allowed rate of outbursts of X-ray luminosity greater than or equal to a certain value, we used the following recipe :

25 15 Fig. 2.4 SMBH mass functions for galaxies with redshift less than (a) 0.5 or (b) 1.0, and the local SMBH mass function from Marconi et al. (2004, solid curve) scaled to compare with the mass functions for our sample. The dotted curves indicate mass functions in the CDF-N while the dashed curves are for the CDF-S. The deviation at low SMBH mass in the right panel is simply due to a selection effect; it is difficult to detect the galaxies hosting SMBHs with small masses at relatively high redshifts. 1. For a given energy band (FB, SB, or HB), we picked an X-ray luminosity, L X,burst, as the lower limit, and we derived the rate of outbursts of X-ray luminosities L X,burst. Count rates or upper limits on count rates for all the galaxies in the survey were converted to luminosities based on their redshifts, using the Portable, Interactive, Multi-Mission Simulator (PIMMS) for a power-law photon index of Γ. Galaxies generally have soft X-ray spectra during outburst and harder spectra in the their basal states. We thus considered a few choices of Γ ranging from 2 to 5. The luminosity distributions for Γ = 2 and Γ = 4 are shown in Figure 2.5. Most of the luminosities are upper limits. 2. We assumed that the typical duration of an outburst was T burst. Theoretically predicted light curves of outbursts show a characteristic fast rise and slow decay. After the bulk of the material is accreted, which could be on a timescale of the order of months, the debris starts to form a radiation-pressure supported torus and the luminosity declines slowly as t 5/3 (Rees 1988, 1990). This long-term decay could last for years and may have been observed (with large light-curve gaps) in a few outbursts (e.g., Komossa & Bade 1999). The detailed luminosity evolution of outbursts, especially for the first few months, remains unknown, and likely varies greatly from event-to-event. For simplicity, we assumed that the luminosity is constant during the outburst, and after that the X-ray luminosity would drop by a minimum factor of 20. We then determined the total rest-frame time over which we are sensitive to outbursts of luminosity L X,burst : T total (L X,burst,T burst ) = n T i,sens (L X,burst,T burst ), (2.1) i=1

26 wheren = 19607isthenumberofgalaxiesinthesurveywith10 5 M < M BH < M, and T i,sens is the rest-frame time over which we are sensitive to an outburst of luminosity L X,burst for galaxy i. T i,sens can be expressed as: T i,sens = Time,rf 16 W i (t)dt. (2.2) This integration is over all the rest-frame observation period, from time T burst before the first epoch to the last epoch, and W i (t) is our sensitivity window function to outbursts starting from time t and with luminosity L X,burst and duration T burst. If we could detect such an outburst based on our searching criteria (step 2 in 2.2.3), then W i (t) = 1; otherwise W i (t) = The 90% confidence upper limit on 0 events is from Gehrels (1986). Thus the 90% confidence upper limit on the event rate of outbursts is given by N CDF (L X,burst,T burst ) = T total outbursts galaxy yr. (2.3) 4. Repeat the above steps for various values of L X,burst and T burst as well as different assumptions about outburst spectral shape. The constraints set by Equation 2.3 depend significantly on the outburst duration T burst and the lower luminosity limit L X,burst, and they weakly depend on the spectral shape (we consider X-ray photon indices Γ =2, 3, 4, or 5) and energy band (the SB or HB), as shown in Figures 2.6, 2.7 and 2.8. Longer outburst durations lead to tighter upper limits on the event rate. Photon indices affect the derived X-ray luminosities through the conversion from count rate to flux (from PIMMS) and the K correction term (1+z) (Γ 2). As Γ increases, theconversion factor increases inthesbanddecreases in the HB, while the K correction term always increases and preferentially dominates over the conversion factor for high-luminosity sources since they generally have large redshifts. Thus the dependence of the upper limits on the spectral shape behaves differently in the SB and HB. Note that luminosities L X,burst in the SB and HB are different, so the upper limits on the event rate in these energy bands cannot be compared to each other directly. Generally, assuming an outburst duration of 6 months, which was also adopted in Donley et al. (2002), the upper limit on the event rate is 10 4 galaxy 1 yr 1, for an outburst with X-ray luminosity ergs s Comparison with Previous Results We compared these new constraints with the theoretical study by Wang & Merritt (2004) using a singular isothermal sphere, which analytically predicted a stellar tidaldisruption rate of N WM (M BH ) yr 1( σ ) 7/2 70 km s 1 ( MBH 10 6 M ) 1 ( ) 1/3 ( ) 1/4 m R A(z), M R (2.4)

27 Fig. 2.5 SB and HB X-ray luminosity distributions of galaxies in the CDF-N and CDF-S. Luminosities for all the epochs were collected, so each galaxy is plotted multiple times, corresponding to the number of epochs in which it was observed. Most of the luminosities are 3σ upper limits. Luminosities of X-ray detected galaxies are represented by the shaded area. The number of galaxies was plotted using a logarithmic scale to show the small fraction of X-ray detected galaxies. A photon index of Γ = 2 or Γ = 4 was adopted in these plots. 17

28 18 Fig. 2.6 Dependence of T total on outburst duration at different outburst luminosities. T total is the total rest-frame time over which we are sensitive to outbursts of (a) SB luminosity L SB,burst or (b) HB luminosity L HB,burst ; see Equation 2.1. A photon index of Γ = 2 was adopted when making these plots. Fig. 2.7 The same as Figure 2.6, but for a photon index of Γ = 4.

29 19 Fig % confidence upper limits on the event rates of outbursts derived from this survey as functions of the lower limit on (a) SB or (b) HB X-ray luminosity; see Equation 2.3. Solid, dotted, dashed, and dash-dotted lines represent photon indices Γ of 2, 3, 4, and 5, respectively. We show the constraints on two different outburst durations, 1 month and 6 months. We made no assumptions about the physical process causing the outburst except that the SMBH mass is in the range 10 5 M < M BH < M. with M BH being the SMBH mass, σ the velocity dispersion of the host galaxy, and m and R the mass and radius of the tidally disrupted stars. We added an amplification factor A(z) here to represent any redshift evolution of the rate. As the evolution of X-ray outbursts is unknown (see 2.1), we set A(z) = 1 for now. Utilizing the M BH -σ relation from Ferrarese & Ford (2005): Equation 2.4 becomes ( M BH = σ ) 4.86 M, (2.5) 200 km s 1 ( ) 0.28 ( ) 1/3 ( ) 1/4 N WM (M BH ) yr 1 MBH m R 10 6 A(z). (2.6) M M R As we are here comparing with a physical model, we modified our previous constraints on event rate by also considering the capability of a SMBH with a given mass to produce outbursts of high luminosities. We simply assumed that there was a narrow range for the outburst luminosity, which was around a fraction f Edd of the Eddington luminosity. f Edd = L bol /L Edd, where L bol is the outburst s bolometric luminosity and L Edd (M BH /M ) ergs s 1 is the Eddington luminosity. We also assumed a bolometric correction f bc, which was defined as f bc = L bol /L X,burst. The constraint on M BH is then M BH f bcl X,burst f Edd M. (2.7) Applying this SMBH-mass requirement in addition to the 10 5 M < M BH < M requirement in step 2 of 2.3.1, we got weaker constraints on the event rate for high

30 outburst luminosities, since the number of qualified galaxies n is smaller. The derived upper limits as a function of L X,burst for the SB and the HB are shown in Figures 2.9 and 2.10, respectively. Here we assume T burst = 6 months, f bc = 10, and f Edd = 1.0 or 0.1. The event rate shows a dependence on the Eddington ratio, f Edd, since smaller f Edd will limit the number of galaxies that are capable of making bright outbursts. The new requirement on the SMBH mass does not affect the outburst rate at small L X,burst, because all galaxies are capable of producing outbursts of such low X-ray luminosity. To employ the formula for the theoretical tidal-disruption rate of a single galaxy (Equation 2.6) in our survey, we used M BH calculated in As the dependence of N WM on M BH is weak, there should not be a large error if our SMBH-mass estimation is not highly accurate; an error of a factor of 5 in M BH only changes N WM by a factor of 1.5. As we are only considering tidal disruptions of stars here, the disruption rate is dominated by subsolar stars according to the stellar mass function (Milosavljević et al. 2006). The radius of these stars follows the relation R /R (m /M ) 0.8 (e.g., Kippenhahn & Weigert 1990). Thus the dependence on stellar mass and radius is weak, and we simply assumed solar mass and radius. For each L X,burst, we took the average value of Ṅ WM for all galaxies for which we are sensitive to an outburst of luminosity L X,burst, which required T i,sens > 0. The predicted event rate for this survey as a function of L X,burst is then given by 20 n N i=1 th (L X,burst ) = k iṅwm n i=1 k i outbursts galaxy yr, (2.8) where k i = 1 if T i,sens > 0, and k i = 0 if T i,sens = 0. Ṅ th as a function of L X,burst for different spectral shapes and Eddington ratios is also plotted in Figures 2.9 and Comparing these predictions with our survey constraints, we see that the upper limits derived in this survey are consistent with the predicted rates, except that around an X-ray luminosity of ergs s 1, this deep-field survey sets a tighter constraint on the rate of Eddington-limited events than the theory (Figure 2.9a, Figure 2.10a). If we reduce f Edd to 0.1, the discrepancy decreases and almost disappears (Figure 2.9b, Figure 2.10b). There are other sources that may help to resolve this discrepancy: (1) As mentioned in 2.1, a tidal-disruption event may not exhibit the expected outburst characteristics, due to a short duration of the event or X-ray obscuration. It is then likely to be missed by current and previous surveys. (2) Equation 2.4 is derived under the assumption that the galactic nucleus is a singular isothermal sphere. For other kinds of density distributions, the predicted rate will become smaller according to Figure 5b of Wang& Merritt(2004), which gives the computed tidaldisruption rates and SMBH masses for all the galaxies in their sample. A straight-line fit to the data points in this plot gives ( ) 0.52 N WM Rev (M BH ) yr 1 MBH 10 6, (2.9) M with a smaller rate and stronger dependence on SMBH mass. A comparison with the predicted tidal-disruption rate derived from this modified relation is shown in Figure 2.11.

31 21 Fig % confidence upper limits on the event rates of outbursts derived from this survey and expected event rates from the theoretical prediction as functions of the lower limit on the SB X-ray luminosity of outbursts. Solid, dotted, dashed, and dash-dotted lines represent photon indices Γ of 2, 3, 4, and 5, respectively. SMBHs that were not capable of producing outbursts of luminosity L SB,burst were removed from the sample. We assumed T burst = 6 months, f bc = 10 and f Edd = 1.0 (a) or 0.1 (b). The predicted rates were derived under the assumption of isothermal stellar density distributions in galactic nuclei. Fig The same as Figure 2.9, but for the HB X-ray luminosity of outbursts.

32 The revised rates N th Rev are 0.5 dex smaller and now are slightly below our observed upper limits for the full luminosity range. (3) The M BH -σ relation in Equation 2.5 only holds for local SMBHs. The relation could evolve in the sense of velocity dispersion decreasing (Woo et al. 2006) with redshift for a fixed BH mass, making the predicted rate decrease at high redshift. All these effects may cause the upper limits on event rate given by this survey to be above the predicted stellar tidal-disruption rates. On the other hand, the galaxy sample in Wang & Merritt (2004) is a set of 61 elliptical galaxies, while our sample is dominated by late-type galaxies. By considering the contribution from the bulges of spirals, the predicted rates could be significantly increased (Wang & Merritt 2004), as the masses of the bulges are smaller than the total masses of the galaxies that we were using. This could increase the discrepancy between the observations and theoretical predictions. 22 Fig % confidence upper limits on the event rates of outbursts derived from this survey and expected event rates from the theoretical prediction for the (a) SB and (b) HB, under the assumptions of T burst = 6 months, f bc = 10 and f Edd = 1.0. Solid, dotted, dashed, and dash-dotted lines represent photon indices Γ of 2, 3, 4, and 5, respectively. The observational upper limits are the same as those in Figures 2.9 and 2.10, while the predicted rates were derived from a modified relation which does not require isothermal stellar density distributions in galactic nuclei (Equation 2.9). We employed the Eddington limit above based on the assumption that the X-rays come from a transient accretion disk and its corona. Alternatively, X-rays could instead come from tidal-stream collisions (e.g., Kochanek 1994), in which case the luminosity could perhaps exceed the Eddington limit. If we remove the constraint on the SMBH mass set by Equation 2.7, the upper limits on the event rates get tighter for high X-ray luminosities and are the same as those in Figure 2.8, and the predicted rates also change dueto their dependenceon M BH. Theresults are plotted in Figure 2.12, which shows the predicted rates from both the original relation (Equation 2.6) and its revision (Equation 2.9). Our survey constraints for L X,burst ergs s 1 are 0.5 dex tighter than the analytical predictions assuming isothermal stellar density distributions in galactic nuclei,

33 and are consistent with the revised tidal-disruption rates derived from the full galaxy sample in Wang & Merritt (2004). We have assumed A(z) = 1 in the above analyses, which represents no redshift evolution of the outburst rate. However, the rate of X-ray outbursts could increase with redshift owing to changes in galactic nuclei and SMBH masses. Milosavljević et al. (2006) have proposed that stellar tidal disruptions are largely responsible for the AGN X-ray luminosity function at luminosities below erg s 1. If this is indeed the case, then the order-of-magnitude increase in the comoving number density of such AGNs out to z 1 (e.g., Brandt & Hasinger 2005 and references therein) implies that the outburst rate must correspondingly increase. Scaling the Wang & Merritt (2004) rates upward by A(z = 0.8) 10 to account for the median redshift of our sample would lead to significant disagreement with our observational constraints, even if the revised rate (Equation 2.9) is adopted, suggesting that the X-ray luminosity function at luminosities below erg s 1 is not primarily due to stellar tidal disruptions. 23 Fig % confidence upper limits on the event rates of outbursts derived from this surveyandexpectedeventratesfromthetheoretical predictionforthe(a)sband(b)hb, under the assumption of T burst = 6 months. Solid, dotted, dashed, and dash-dotted lines represent photon indices Γ of 2, 3, 4, and 5, respectively. The X-ray luminosities were not constrained by the Eddington limit in these plots. The observational upper limits are the same as those in Figure 2.8. The predicted rates were derived from Equation 2.6 which is the analytic relation for galactic nuclei with isothermal stellar density distributions, and from Equation 2.9 which does not require isothermal stellar density distributions. On the observational side, Donley et al. (2002) performed a systematic survey for X-ray outbursts using the ROSAT database (although this survey had some substantial systematic uncertainties owing to complex selection effects). They detected five outbursts and placed the first constraints on the rate of such outbursts. The rate of large-amplitude X-ray outbursts from inactive galaxies in the local Universe is 10 5 galaxy 1 yr 1, estimated from the survey volume and the galaxy space density. Compared to their results, our survey constraints are based on uniform observational data for optical galaxies which are less biased, and there are no uncertainties introduced in the estimation

34 of survey volume or the galaxy space density. Moreover, we are able to probe X-ray outbursts with higher redshifts and in a harder X-ray band for the first time, and we derived luminosity-dependent rate constraints which offer insight into the low-luminosity regime. HB X-ray surveys are able to detect obscured outbursts which could have been missed in previous surveys; intrinsic column densities as low as cm 2, perhaps associated with gas from the tidally disrupted star, would greatly reduce the detectability of soft-spectrum outbursts in the ROSAT band. Here we adopted T burst = 6 months, the same as in Donley et al. (2002). The SB and HB constraints are shown in Figure 2.8. From L X,burst down to ergs s 1, the upper limit on the event rate increases from 10 4 galaxy 1 yr 1 to 10 2 galaxy 1 yr 1, mainly due to the limited sensitivity in the low-luminosity regime. Further observations are required to assess whether low-amplitude events, which could come from partial tidal disruptions or the accretion of brown dwarfs, planets, or small gas clouds, are truly more common than high-amplitude events, as for other transient phenomena in nature. When L X,burst ergs s 1, the upper limit on the event rate is 10 4 galaxy 1 yr 1, either in the SBor thehb. Comparedto theresults in Donley et al. (2002), these constraints allow an amplification of the outburst rate up to a factor of 10 at most when considering obscured X-ray outbursts and redshift evolution from z 0 to z Conclusions and Future Work In summary, we constructed a sample of optical galaxies in the CDF-N and CDF-S with redshifts obtained from the literature; the median redshift is 0.8. We analyzed exceptionally sensitive Chandra observations of these galaxies, which span 798 days for the CDF-N and 1828 days for the CDF-S. We searched for X-ray outburstswith thecriterion that thecountratevaries byaminimumfactor of 20inoneofthreestandard bands. No outbursts were found, and thus we set upper limits on the rate of such events in the Universe, which depend on the X-ray luminosity of outbursts. If we only consider those galaxies hosting SMBHs, and those with SMBHs not massive enough to swallow a whole star without disruption, we derive an upper limit on the rate of an outburst with L X,burst ergs s 1 and T burst = 6 months to be 10 4 galaxy 1 yr 1 (Figure 2.8), without any other assumptions about the physical model producing X-ray outbursts. Compared to the survey by Donley et al. (2002), our survey probes both higher redshifts and harder X-ray energies. The outburst rate may increase by a maximum factor of 10 when taking into account both obscured X-ray outbursts and redshift evolution from z 0 to z 0.8. We are able to set constraints on low-luminosity events down to ergs s 1, though at this luminosity the event rate is limited by sensitivity. We also compared our constraints to the predicted tidal-disruption rates of Wang & Merritt (2004), exploring several possibilities in physical parameter space. If the outburstluminosity is limited by theeddingtonluminosity with f Edd 0.1, or thepredicted rate is computed from a modified relation which does not require isothermal stellar density distributions in galactic nuclei, our results are roughly consistent with theoretical predictions. Otherwise, our constraints are tighter than the predictions. Moreover, if stellar tidal disruptions are largely responsible for the AGN X-ray luminosity function at luminosities below erg s 1 as proposed by Milosavljević et al. (2006), the 24

35 predicted rate should be scaled upward by A(z = 0.8) 10 to account for the median redshift of our sample. This scaling leads to significant discrepancies with our observational constraints, suggesting that the X-ray luminosity function at low luminosities is not likely to be dominated by stellar tidal disruptions. The constraints set by this study could be significantly improved by further deep-field surveys and new missions. For example, the 1 Ms of additional CDF-S exposure starting in 2007 September will increase the monitoring baseline significantly and provide 13 new observations which could be grouped into several additional epochs for outburst searching and other variability studies. Future missions such as Lobster and erosita in the soft X-ray band and the BHFP in the hard X-ray band will also be capable of detecting and studying outbursts, benefiting from their large sky coverage. Grindlay (2004) predicted that the Energetic X-ray Imaging Survey Telescope (EXIST) implementation of the BHFP would detect X-ray outbursts out to 100 Mpc at a rate of 30 yr 1, adopting the rates of Wang & Merritt (2004). Such missions will put better constraints on the outburst rate, the fraction of obscured outbursts, and redshift evolution, enhancing our knowledge about the nature of X-ray outbursts. 25

36 26 Chapter 3 Discovery of the Most-Distant Double-Peaked Emitter at z = Introduction Double-peaked emitters are active galactic nuclei (AGNs) emitting broad and double-peaked low-ionization lines. A survey of 100 radio-loud AGNs (z < 0.4) suggests that 20% of these sources are double-peaked emitters (Eracleous & Halpern 1994, 2003). The frequency of double-peaked emitters is much lower ( 3%) among the general population of 3000 low-redshift (z < 0.33) AGNs selected by the Sloan Digital Sky Survey (Strateva et al. 2003), and these double-peaked emitters are predominantly (76%) radio-quiet. Since the discovery of the first examples of double-peaked emitters in the 1980s (Oke 1987; Chen et al. 1989), more than 150 such sources have been found. Most have been identified based on their Hα, and sometimes Hβ, lines. Spectroscopy of several of these sources showed that the Mg ii λ2799 line 1 also has a double-peaked profile similar to those of Hα and Hβ (Halpern et al. 1996; Eracleous et al. 2004). Selection based on the Mg ii line profile could in principle find higher redshift candidates; however, due to contamination from the underlying Fe ii and Fe iii emission-line complexes (e.g., Wills et al. 1985) and possibly Mg ii self-absorption, it is difficult to create a complete sample of double-peaked Mg ii λ2799 emitters, and only a few such objects have been reported (Strateva et al. 2003). The highest redshift double-peaked emitters discovered to date have z 0.6. Observational results and basic physical considerations indicate that the most-plausible origin of the double-peaked emission lines is the accretion disk (e.g., Chen & Halpern 1989; Eracleous et al. 1995; Eracleous & Halpern 2003). The line profile can be well fit by emission from the outer regions of a Keplerian disk, typically at distances from the black hole of hundreds to thousands of gravitational radii (R G = GM/c 2 ). In many cases, the viscous energy available locally in the line-emitting region is insufficient to power the observed lines, and it has been suggested that these strong lines are produced by external illumination of the disk, probably from an X-ray-emitting elevated structure in the center (e.g., Chen et al. 1989). In this paper, we report the discovery of the highest redshift double-peaked emitter known to date. This source, CXOECDFS J (hereafter J ), was detected as an X-ray point source in the 250 ks observations of the Extended Chandra Deep Field-South (E-CDF-S; Lehmer et al. 2005), and was identified as a 1 ThisMg iilineisactuallyaλ2796/2803doublet. Duetolinebroadening,the twocomponents are usually blended and cannot be resolved. Thus we treat the doublet as a single line when discussing the observed emission feature throughout this paper.

37 double-peaked Mg ii emitter at z = by Keck/DEIMOS optical spectroscopy. We adopt a cosmology with H 0 = 70 km s 1 Mpc 1, Ω M = 0.3, and Ω Λ = 0.7 throughout this paper. 3.2 Keck Observations and Disk-Model Fit Optical spectroscopic observations of J were carried out using the DEIMOS spectrograph (Faber et al. 2003) on the 10 m Keck II telescope on January 15, 2007 (UT), as part of the E-CDF-S spectroscopic program (PIs: P. Capak, M. Salvato, J. Kartaltepe; Silverman et al. 2009, in preparation). We used the 600 l/mm grism and the GG455 filter. The wavelength coverage was Å with a resolution of 3.5 Å. The seeing was 0. 6, and the airmass was The total exposure time was 9000 s in five individual exposures. The wavelength-dependent response was corrected by an observation of a single flux standard star while the overall normalization was set to match the COMBO-17 R-band magnitude (Wolf et al. 2004, 2008). The redshift of J , 1.369, was determined using the Mg ii λ2799 and Ne v λ3426 narrow lines (FWHM 400 km s 1 ). The rest-frame near-uv (NUV) spectrum around the Mg ii λ2799 line is shown in Figure 3.1, smoothed to a resolution of 3.6 Å. The spectrum shows a clearly double-peaked broad Mg ii line along with a central narrow line, similar to previously discovered double-peaked Hα or Mg ii emitters. The Mg ii absorption doublet at 2600 Å is likely produced by an intervening absorber at z = 1.21, as inferred from the narrow velocity dispersion (FWHM 300 km s 1 ) and large blueshift (e.g., Ganguly et al. 2007). An alternative interpretation of the doublet as arising in an AGN outflow would require outflow velocities of km s 1. Assuming that the J Mg ii line emission originates from the accretion disk, we can use the line shape to determine a set of parameters characterizing the emission region (see Chen & Halpern 1989; Eracleous et al. 1995, for a description of line-profile accretion-disk modeling). We start by subtracting the underlying continuum and Fe emission-line complexes, as shown in Figure 3.1. The continuum is represented by a simple power law, F λ λ 1.6 (Vanden Berk et al. 2001). The Fe pseudo-continuum is modeled by the broadened Fe-emission template of Vestergaard & Wilkes (2001), fit to the Å spectrum excluding the Mg ii line, where we assumed that the Fe-line broadening is similar to that of the Mg ii line complex, FWHM km s 1. Such Fe-line broadening could result if the line is emitted from the base of a wind launched from the accretion disk. The double-peaked Mg ii line profile does not differ much for any reasonable choice of the Fe-line broadening (from 5000 to km s 1 ). The continuum and Fe-line complex subtracted Mg ii profile of J cannot be well fit by a circular relativistic Keplerian disk model. In the absence of line-profile variability, which can help distinguish between the different non-axisymmetric disk models, we choose the elliptical disk model of Eracleous et al. (1995), which has the smallest number of extra free parameters(for a total of 7 disk-fit parameters). The model assumes an external source of illumination represented by a power law with emissivity, ǫ R q, and we fix q = 3, equivalent to an illuminating point source on the disk axis high above the disk. The central narrow-line part of the spectrum was excluded from the fit. From 27

38 thebest-fitmodel, thediskinclination withrespecttoourlineofsightisi = degrees, the inner and outer radii of the emission ringare R 1 = R G and R 2 = R G, the turbulent broadening parameter is σ = km s 1, and the disk ellipticity is e = , with a semi-major axis orientation of φ 0 = 110±20 degrees with respect to our line of sight. The integrated line flux is F Mg II = (9.3±0.2) erg cm 2 s 1. These emission-line region parameters are similar to those obtained for lower-redshift double-peaked emitters, e.g., Eracleous & Halpern (2003), Strateva et al. (2003), and Strateva et al. (2008). 3.3 Multiwavelength Properties The numerous multiwavelength deep surveys of the E-CDF-S allow us to study the properties of J from radio to X-ray wavelengths. Figure 3.2 presents the radio, infrared (IR), optical, and X-ray images of J A broad-band spectral energy distribution (SED) of the source is shown in Figure 3.3, with the majority of the SED data displayed in Table It is one of the best-sampled double-peaked emitter SEDs, comparable to that of the prototype source Arp 102B (Eracleous et al. 2003; Strateva et al. 2008). Details of the broad-band properties are discussed below. Radio J was observed by the Very Large Array (VLA) at 1.4 GHz in (Kellermann et al. 2008) and 2007 (Miller et al. 2008). 2 The reported core radio flux densities are 1.14 ± 0.03 mjy and 1.23 ±0.02 mjy, respectively. It was also detected by the Australia Telescope Compact Array (ATCA) at 1.4 GHz, with a 30% higher flux density (Rovilos et al. 2007). Because of the low resolution (beam size 17 7 ), we do not use the ATCA results. The average VLA flux density is plotted in Figure 3.3. We show the radio image from Miller et al. (2008) in Figure 3.2a. Most of the data were obtained when the VLA was in configuration A, and some were obtained in configuration BnA. The beam size is with a position angle near zero (i.e., N-S). The source is associated with two brighter radio lobes, which extend to a few hundred kpc away from the center, displaying an FR II morphology. The core and the lobes are sources 20, 19, and 21 in the catalog of Miller et al. (2008), and the total radio power at 1.4 GHz is erg s 1 Hz 1. According to the radio contours, there could also be a weak jet leading to the lobe in the western feature. We did not find any clear counterparts for the lobes/jet at other wavelengths, except for a faint optical source at the position of the eastern lobe. Moreover, these structures are more extended than nearby point sources. Thus they are not likely physically associated with galaxies. The radio loudness parameter, defined as R = f 5 GHz /f (the ratio of 4400 Å flux densities in the rest frame; e.g., Kellermann et al. 1989), is computed using the 1.4 GHz and 914 nm COMBO-17 (Wolf et al. 2004, 2008) flux densities with the assumption of a radio power-law slope of α r = 0.8 and an optical power-law slope of α o = nmiller/vlaecdfs main.html.

39 Fig. 3.1 Rest-frame NUV spectrum of J showing the double-peaked Mg ii line. The vertical dash-dotted line indicates the expected position of Mg ii λ2799. Two gaps in the spectrum are caused by the DEIMOS CCDs and telluric O 2 absorption. Dashed, dotted, grey, and thick curves are different emission components used to fit the observed double-peaked Mg ii line profile, as indicated in the plot. The rms noise of the spectrum is erg cm 2 s 1 Å 1. 29

40 Fig. 3.2 (a) Radio 1.4 GHz, (b) IR 5.8 µm, (c) optical R-band, and (d) X-ray kev images of J Each image is 60 (0.51 Mpc at z=1.369) on a side. The last three images are overlaid with radio contours, ranging from % of the maximum pixel value following a logarithmic scale. The downward arrows point to the position of J The ellipse at the lower left corner of (a) shows the beam size of the radio observations. Flux density or flux in each band is indicated; the X-ray flux is in units of erg cm 2 s 1. The apparent extension in the Chandra image is due to the large point spread function at its location. 30

41 Fig. 3.3 Radio through X-ray SED of J The names of observatories/surveys and years of observations are indicated for the data points. The dotted and dashed curves show the Elvis et al. (1994) and Richards et al. (2006) SED templates, normalized to the COMBO-17 R-band data point, respectively. The broad-band SED of J is in general agreement with typical radio-loud quasar SEDs. The Arp 102B SED (Eracleous et al. 2003; Strateva et al. 2008, and references therein) is also shown for comparison. The J SED and both of the SED templates show a big blue bump in the UV (marked with a thick downward-pointing arrow), while the Arp 102B SED does not. 31

42 (F ν ν α ). The source has a radio loudness of R 429, including the contributions from the extended lobe emission. 3 IR The E-CDF-S was covered by the Spitzer Far Infrared Deep Extragalactic Legacy Survey (FIDEL) at 24 and 70 µm, 4 and by the Spitzer IRAC/MUSYC Public Legacy Survey in the E-CDF-S (SIMPLE) at 3.6, 4.5, 5.8, and 8.0 µm. 5 J was detected at 3.6, 4.5, 5.8, 8.0, and 24 µm with flux densities of 0.09, 0.12, 0.14, 0.19, and 0.50 mjy, respectively (e.g., see Figure 3.2b). It was not detected at 70 µm; we estimate the 2 σ flux-density upper limit to be 1.5 mjy. Optical, UV In the optical band, J is present in the COMBO-17 catalog (Wolf et al. 2004, 2008) and the deep MPG/ESO Wide Field Imager (WFI) R-band catalog (Giavalisco et al. 2004). It is outside the field-of-view of the Galaxy Evolution from Morphologies and SEDs (GEMS) survey (Caldwell et al. 2008). The B-band absolute AB magnitude is M B = The WFI R-band image is shown in Figure 3.2c and the 17-band photometry data points from the COMBO-17 observations are shown in Figure 3.3. The galactic extinction in the optical band is small (a correction factor of 1.02 for the V-band). J was observed by GALEX in 2003 and The observed NUV (λ eff = 2267 Å) flux densities are 5.95 ± 0.13 and 7.69 ± 0.23 µjy, brightening by 30% in two years. This variability amplitude is typical for double-peaked emitters (e.g., Gezari et al. 2007; Strateva et al. 2008). The correction factor for the NUV Galactic extinction is 1.07, following the Galactic extinction law of A NUV = 8.0E(B V) (e.g., Gil de Paz et al. 2007). The source was not detected in the far-uv band (λ eff = 1516 Å), due to the Lyman break at rest-frame 912 Å. The average NUV flux density is shown in Figure 3.3. X-ray J was detected during the Chandra E-CDF-S survey in 2004, with an effective exposure time of 210 ks. It is located at the edge of the E-CDF-S field and has an off-axis angle of 8. The Chandra image is shown in Figure 3.2d. J has one of the best X-ray spectra available for double-peaked emitters; the number of kev source counts is We performed spectrum extraction on the reduced and cleaned level 2 event files (Lehmer et al. 2005) using the reduction tool acis extract (AE; Broos et al. 2010). The spectrum was binned at a signal-to-noise ratio of 5 using AE and then fit with XSPEC (Version ; Arnaud 1996). The kev spectrum can be well modeled with a power law modified by Galactic absorption, with χ 2 = 18.0 for 22 degrees of freedom (see Fig. 3.4). We adopted a Galactic neutral hydrogen column density N H,G = cm 2 (Stark et al. 1992). The best-fit photon index is Γ = 1.72 ± 0.10; the uncertainties are quoted at 90% confidence. This photon index is typical for radio-loud quasars, Γ RL (e.g., Reeves et al. 1997). The rest-frameunabsorbed kev luminosity is ergs 1, well withinthequasar regime. Intrinsic absorption (N H,i at z = 1.369) is not required to fit the spectrum, with a 90% confidence upper limit of cm The integrated radio flux density was used to compute the radio loudness in order to be consistent with the typical definition in the literature. The radio loudness is 41 if only the radio flux from the core component is taken into account. 4 See 5 See 6 See

43 Fig. 3.4 X-ray spectrum of J overlaid with the best-fit model. The bottom panel shows the deviation of the data from the model in units of σ, with error bars of size unity. The spectrum can be modeled with a simple power law modified by Galactic absorption; the best-fit photon index for this model is Γ = 1.72±

44 To compare the SED of J to those for typical quasars, we show in Figure 3.3 the mean quasar SED from Richards et al. (2006) and the mean radio-loud quasar SED from Elvis et al. (1994), both normalized to the COMBO-17 R-band flux of J The Richards et al. (2006) SED template is derived from a sample of radio-quiet sources, and does not cover the radio and X-ray bands. The optical to X-ray data agree reasonably well with the mean quasar SEDs. The excess emission in the IR bands suggests a contribution from the host galaxy, similar to some other doublepeaked emitters (e.g., Strateva et al. 2008). The weaker radio emission than the mean radio-loud SED probably arises because we only include the radio emission from the core, while some of the Elvis et al. (1994) objects included extended radio emission when only low-resolution observations were available. As the multiwavelength data were not taken simultaneously, variations in the fluxes at different wavebands are likely responsible for part of the discrepancies between the data and SED templates. Thus the broad-band SED of J does not differ significantly from those of typical radio-loud quasars, despite the double-peaked nature of this source. The bolometric luminosity estimated based on the Elvis et al. (1994) template is L bol erg s 1, corresponding to an accretion rate of 1 M yr 1 under the assumption of accretion efficiency η = 0.1. Note that the template SED will double-count the IR emission if the IR bump consists of reprocessed nuclear continuum emission, which could result in an overestimate of the bolometric luminosity by up to 20 30%. 3.4 Discussion The discovery of J doubles the redshift range of known double-peaked emitters, from 0.6 to The rest-frame 2500 Å monochromatic luminosity of J is erg s 1 Hz 1, interpolated from the COMBO-17 flux densities. It is thus among the few most optically luminous double-peaked emitters known. Figure 3.5 shows the position of J in the redshift versus 2500 Å monochromaticluminosity plane. Data for the other double-peaked emitters are from Strateva et al. (2008) and references therein. The X-ray and bolometric luminosities of J are also comparable to those for the brightest double-peaked emitters. The UV to X-ray index, defined as α OX = log[F ν (2500 Å)/F ν (2 kev)], is Compared to the α OX L relation for typical radio-quiet AGNs in Steffen et al. (2006), J Å has a flatter α OX (predicted α OX = 1.45 at this L ). This factor of 3 X-ray 2500 Å enhancement is expected given the radio-loud nature of the source; jet-linked X-ray emission is likely making a substantial contribution to the X-ray spectrum. Following Eracleous & Halpern (1994), we estimated the viscous power released in the line-emitting region W d. Assuming an accretion efficiency of η 0.1 (L bol = ηmc 2 ), the gravitational power output is given by [ ( ( )] 1 8 W d = 7.7 L bol 1 ) 1ζ2 8 1 erg s 1, (3.1) ζ 1 3ζ 1 3ζ 2 where ζ 1 and ζ 2 are the inner and outer radii of the emission region in units of R G. For J , L bol erg s 1, ζ 1 200, ζ 2 900, and W d

45 Fig. 3.5 Redshift-luminosity distribution of double-peaked emitters (adapted from Strateva et al. 2008). Filled dots represent radio-quiet sources, and open dots represent radio-loud sources. The positions of J , Arp 102B, and a few high-redshift sources are indicated. 35

46 erg s 1. Theratio of themg ii luminosity to theviscous power is then L Mg II /W d Assuming that J has the same Hα to Mg ii line ratio as Arp 102B ( 5.0; Halpern et al. 1996), the Hα luminosity is 35% of the total energy available locally. Based on computations of the emission from the accretion disks of cataclysmic variables in Williams (1980), we expect that no more than 20% of the local viscous energy would be emitted as Hα. Thus the local energy is probably insufficient to power the strong lines, and external illumination of the accretion disk appears necessary to explain the observed Mg ii line luminosity even for this luminous, and likely efficiently accreting, double-peaked emitter. The source of the external illumination is still uncertain. For the prototype double-peaked emitter Arp 102B, Chen & Halpern (1989) proposed that the outer accretion disk was illuminated by a vertical extended structure in the inner disk, such as a geometrically thick and optically thin X-ray-emitting flow produced by radiatively inefficient accretion (RIAF; e.g., Rees et al. 1982; Narayan & Yi 1994). This mechanism has also been proposed to apply to other low Eddington ratio, low-luminosity double-peaked emitters. However, recent studies have revealed that some sources are actually efficient accretors, with Eddington ratio L/L Edd 0.1, a regime where RIAFs cannot exist (e.g., Lewis & Eracleous 2006; see also Strateva et al. 2008). The SED of J shows a big blue bump (BBB) in the UV, as well as a typical X-ray photon index and luminosity for radio-loud quasars, also indicating a relatively high-efficiency accretion flow in the central engine. In Figure 3.3, we show the SED of Arp 102B (Eracleous et al. 2003; Strateva et al. 2008, and references therein). Compared to J , Arp 102B has a luminosity about two orders of magnitude fainter and lacks a BBB. For efficiently accreting double-peaked emitters, external photons may come from a different kind of disk-illuminating structure, for example, disk photons scattered by electrons in jets or slow outflows as suggested by Cao & Wang (2006). Another open question regarding double-peaked emitters is their connection to the general AGN population. Observationally, except for their double-peaked and generally broader emission lines, many double-peaked emitters resemble typical AGNs (e.g., through broad-band optical to X-ray SEDs). Despite the requirement of X-ray illumination of the disk, their X-ray properties (e.g., spectra and power output) do not differ greatly from those of single-peaked AGNs (e.g., Strateva et al. 2003, 2006), though recent study suggested enhanced X-ray emission relative to the UV/optical emission in a sample of the broadest double-peaked emitters (Strateva et al. 2008). Theoretically, all luminous AGNs are expected to have accretion disks, and Keplerian disks are capable of producing double-peaked emission lines. Although the disk parameters, such as the inclination and the ratio of the inner to outer radius, could affect the appearance of the line profile, they are not sufficient to explain why only 3% of AGNs are double-peaked emitters. Murray & Chiang (1997) suggested that a varying optical depth in an accretion-disk wind could determine the presence of single or double-peaked line profiles. The underlying double-peaked lines become single-peaked due to radiative transfer in a strong radiation-driven disk wind, and therefore double-peaked emission lines are mostly existent in low-luminosity AGNs, where disk winds are weak with small optical depths. However, this model cannot explain the existence of the most luminous double-peaked emitters, including J Alternatively, the presence of single or 36

47 double-peaked emission lines could be related to the structure and kinematics of the broad-line region (BLR). It is commonly believed that the BLR is highly stratified. The low-ionization BLR could consist of two kinematically distinct regions: an outer region of the accretion disk and a standard kinematically hot broad-line cloud component. If in the majority of AGNs line emission from the latter component dominates, we will see single-peaked broad emission lines, while in the remaining 3% we see double-peaked lines from the accretion disk. However, the detailed structure of the BLR is poorly understood. It seems likely that the small fraction of double-peaked emitters among AGNs is the result of a combination of these factors, i.e., the influence of external illumination, the position and inclination of the line-emitting region of the accretion disk, the presence of disk winds, and the complex geometric nature of the BLR. The properties of J indicate that at higher redshift, double-peaked emitters still possess typical AGN SEDs and X-ray spectra, and the double-peaked line profile can also be explained by the Keplerian disk model. Long-term profile variability has been found to be an ubiquitous property of double-peaked emitters (e.g., Gezari et al. 2007). As the E-CDF-S field will continue to be surveyed in spectroscopic campaigns, it is likely that additional optical/near-ir spectroscopic data for J can be obtained in the future. These will help to study line-profile variability and constrain better the structure and kinematics of the accretion disk, which could shed light on some of the unresolved problems discussed above. 37

48 38 Table 3.1. SED Data for J Band log νl ν (erg s 1 ) Radio VLA 1.4 GHz Infrared FIDEL 70 µm < FIDEL 24 µm SIMPLE 8.0 µm SIMPLE 5.8 µm SIMPLE 4.5 µm SIMPLE 3.6 µm Optical a COMBO-17 I COMBO-17 R COMBO-17 U UV GALEX 2267 Å X-ray Chandra 2 kev Chandra 5 kev a The full set of COMBO-17 SED data points are available in Wolf et al. (2004, 2008).

49 39 Chapter 4 The Chandra Deep Field-South Survey: 2 Ms Source Catalogs 4.1 Introduction One of the greatest successes of the Chandra X-Ray Observatory (Chandra) has been the characterization of the sources creating the kev cosmic X-ray background (CXRB), and the deepest Chandra surveys form a central part of this effort. The two deepest Chandra surveys, the Chandra Deep Field-North and Chandra Deep Field-South (CDF-N and CDF-S, jointly CDFs; see Brandt & Hasinger 2005 for a review), have each detected hundreds of X-ray sources over 450 arcmin 2 areas with enormous multiwavelength observational investments. They have measured the highest sky density of accreting supermassive black holes (SMBHs) to date and have also enabled novel X-ray studies of starburst and normal galaxies, groups and clusters of galaxies, large-scale structures in the distant universe, and Galactic stars. As part of an effort to create still deeper X-ray surveys, we proposed for substantial additional exposure on the CDF-S during Chandra Cycle 9. The CDF-S has superb and improving coverage at optical, infrared, and radio wavelengths; it will continue to be a premiere multiwavelength deep-survey field for the coming decades as additional large facilities are deployed in the southern hemisphere. Furthermore, owing to the 1 Ms of Chandra exposure already available (Giacconi et al. 2002, hereafter G02), the CDF-S is a natural field to observe more sensitively. Although our proposal was not approved in the peer review, subsequently 1 Ms of Director s Discretionary Time was allocated for deeper CDF-S observations. The allocated observations were successfully executed in 2007 September, October and November, raising the CDF-S exposure to 2 Ms and improving its sensitivity to be comparable to that of the CDF-N (e.g., Alexander et al. 2003, hereafter A03). Additional sky coverage at such flux levels is critically important as it substantially improves the statistical sample sizes of the faintest X-ray sources and also allows a basic assessment of the effects of cosmic variance. Furthermore, approximately doubling the exposure on previously detected sources substantially improves the constraints on their positions, spectral properties, and variability properties. In this paper, we present up-to-date Chandra source catalogs and data products derived from the full 2 Ms CDF-S data set along with details of the observations, data processing, and technical analysis. Detailed subsequent investigations and scientific interpretation of the new CDF-S sources will be presented in future papers, e.g., studies of heavily obscured and Compton-thick active galactic nuclei (AGNs), high-redshift AGNs, AGN spectra and variability, starburst and normal galaxies, and clusters and groups of galaxies. In 4.2 we describe the observations and data reduction, and in 4.3 we present the main and supplementary point source catalogs and describe the methods used to

50 create these catalogs. In 4.4 we estimate the background and sensitivity across the survey region. We also present basic number-count results for point sources in 4.5. We summarize in 4.6. The Galactic column density along the line of sight to the CDF-S is remarkably low: N H = cm 2 (e.g., Stark et al. 1992). The coordinates throughout this paper are J2000. A H 0 = 70 km s 1 Mpc 1, Ω M = 0.3, and Ω Λ = 0.7 cosmology is adopted. 4.2 Observations and Data Reduction Observations and Observing Conditions The CDF-S consists of 23 separate observations described in Table 4.1. The 1 Ms catalogs for the first 11 observations taken between 1999 October 14 and 2000 December 23 were presented in G02 and A03. Note that observation 581 (1999 October 14) was excluded from the data reduction and is not listed in Table 4.1 due to telemetry saturation and other problems. The second 1 Ms exposure consisted of 12 observations taken between 2007 September 20 and 2007 November 4. The Advanced CCD Imaging Spectrometer imaging array (ACIS-I; Garmire et al. 2003) was used for all of the Chandra observations. The ACIS-I is composed of four pixel CCDs (CCDs I0 I3), covering a field of view of ( 285 arcmin 2 ), and the pixel size of the CCDs is The focal-plane temperature was 110 C for observations and , and 120 C for the others. The 12 new observations were taken in Very Faint mode to improve the screening of background events and thus increase the sensitivity of ACIS in detecting faint X-ray sources (Vikhlinin 2001). The background light curves for all 23 observations were inspected using EVENT BROWSER in the Tools for ACIS Real-time Analysis(TARA; Broos et al. 2010) software package. Aside from a mild flare during observation (factor of 3 increase for 5 ks), all data sets are free from significant flaring, and the background is stable within 20% of typical quiescent Chandra values. After filtering on good-time intervals and removing the one mild flare, we are left with Ms of total exposure time for the 23 observations. Because of the differences in pointings and roll angles for the individual exposures, the total region covered by the entire CDF-S is arcmin 2, considerably larger than the ACIS-I field of view. Combining the 23 observations, the average aim point(weighted by exposure time) is α J = 03 h 32 m s, δ J = Data Reduction The basic archive data products were processed with the Chandra X-ray Center (CXC) pipeline software versions listed in Table 1. The reduction and analysis of the data used Chandra Interactive Analysis of Observations (CIAO) tools whenever possible 1 ; 40 1 See for details on CIAO.

51 however, custom software, including the TARA package, was also used. Each observation was reprocessed using the CIAO tool acis process events, to correct for the radiation damage sustained by the CCDs during the first few months of Chandra operations using a Charge Transfer Inefficiency (CTI) correction procedure (Townsley et al. 2000, 2002) 2, to remove the standard pixel randomization which blurs the Chandra point spread function (PSF), and to apply a modified bad-pixel file as detailed below. One important deviation from the standard Chandra reduction procedure outlined by the CXC is implementation of a stripped-down bad-pixel file. We note that the standard bad-pixel file supplied with all Chandra data currently excludes 6 7% of the total effective area on front-illuminated devices (e.g., ACIS-I). A large fraction of the bad-pixel locations identified in this file, however, appear to be flagged solely because they show a few extra events (per Ms) almost exclusively below kev. 3 Good events with energies above 0.7 kev that fall on these bad pixels are likely to be perfectly acceptable for source searching, as well as for photometry and spectral analysis albeit with a few mild caveats regarding misinterpretation. Rather than reject all events falling on such columns, we instead adopted a procedure to only exclude events below a rowdependent energy of kev. 4 To this end, we generated a stripped-down bad-pixel file, only selecting obvious bad columns and pixels above 1 kev; this excluded 1.5% of the total effective area on front-illuminated devices. Once the entire 2 Ms data set was combined, we isolated hot soft columns as those where the total number of events with energies below 0.7 kev was 5σ or more above the mean. We then rejected any events in those columns that fell below a row-dependent kev; this removed 1% of all events. Through inspection of the data in CCD coordinates, we additionally discovered that the CXC-preferred CIAO tool acis run hotpix failed to flag a substantial number of obvious cosmic-ray afterglows ( per observation, depending on exposure length), elevating the overall background and, in egregious cases, leaving afterglows to be mistaken as real sources. This problem appeared to be worse for Faint mode data, presumably because the additional 5 5 screening applied in Very Faint mode rejects the strongest afterglows (Vikhlinin 2001). To remedy this situation, we reverted to using the more stringent acis detect afterglow algorithm on all of our data. Notably, none of our sources has a count rate high enough that acis detect afterglow would reject true source counts, which we verified by inspection of events flagged by this routine. Even acis detect afterglow failed to reject all afterglows, and thus we created custom software to remove many remaining faint afterglows from the data. Working in CCD coordinates, we removed additional faint afterglows with three or more total counts occuring within 20 s (or equivalently 6 consecutive frames). In total, we removed Note that the CXC CTI correction procedure is only available for 120 C data; thus we did not CTI-correct observations and See prods/badpix/index.html 4 Theenergyrangeof keVandfrequencyofoccurrencewereverifiedbyvisualinspection of such columns in our 2 Ms data set. We found that such hot soft columns were not clearly seen in any individual observations. The upper energy bound appears to vary as a function of distance from the readout edge of the front-illuminated CCDs, such that rows closest to the readout edge only have extra events below 0.5 kev, while those furthest away have extra events extending up to 0.7 kev.

52 total events associated with afterglows. In all cases, we inspected the data set and found that such flagged events were isolated and not associated with apparent legitimate X-ray sources. 4.3 Production of the Point-Source Catalogs The production of the point-source catalogs largely followed the procedure described in 3 of A03. The main differences in the catalog-production procedure used here are the following: 1. Our main Chandra catalog includes sources detected by running wavdetect(freeman et al. 2002) at a false-positive probability threshold of 10 6, less conservative than the 10 7 value adopted by A03. Even with this revised threshold, we expect the fraction of false sources to be small; see for details. 2. Additional sensitivity can be obtained by merging the 250 ks Extended Chandra Deep Field-South (E-CDF-S; Lehmer et al. 2005, hereafter L05) with the 2 Ms CDF-S. An additional 86 X-ray sources were detected with this approach. These sources are presented in a supplementary catalog described in Image and Exposure Map Creation We registered the observations in the following manner. wavdetect was run on each individual cleaned image to generate an initial source list. Centroid positions for each detected source were determined using the reduction tool acis extract (AE; Broos et al. 2010). 5 The observations were registered to a common astrometric frame by matching X-ray centroid positions to optical sources detected in deep R-band images taken with the Wide Field Imager (WFI) of the MPG/ ESO telescope at La Silla (see 4.2 of Giavalisco et al. 2004). The matching was performed using the CIAO tools reproject aspect and wcs update adopting a 3 matching radius and a residual rejection limit 6 of 0. 6; sources were typically used in each observation for the final astrometric solution. The tool wcs update applied linear translations ranging from to 0. 34, rotations ranging from to 0.009, and scale stretches ranging from to ; individual registrations are accurate to All of the observations were then reprojected to the frame of observation 2406, since this data set required the smallest translation to align it with the optical astrometric frame. We constructed images using the standard ASCA grade set (ASCA grades 0, 2, 3, 4, 6) for three standard bands: kev (full band; FB), kev (soft band; SB), and 2 8 kev (hard band; HB). Figure 4.1 shows the full-band raw image. Exposure maps in the three standard bands were created following the basic procedure outlined in of Hornschemeier et al. (2001) and were normalized to the effective exposures of a source located at the average aim point. Briefly, this procedure takes into account the 42 5 Theacis extractsoftwarecanbeaccessedfromhttp:// users guide.h 6 This is a parameter used in wcs update to remove source pairs based on pair positional offsets.

53 effects of vignetting, gaps between the CCDs, bad-column filtering, bad-pixel filtering, and the spatially dependent degradation in quantum efficiency due to contamination on the ACIS optical-blocking filters. A photon index of Γ = 1.4 was assumed in creating the exposure maps, which is approximately the slope of the X-ray background in the kev band (e.g., Marshall et al. 1980; Gendreau et al. 1995; Hasinger et al. 1998). We show the full-band exposure map in Figure 4.2. Using the full-band exposure map, we calculated the survey solid angle as a function of the minimum full-band effective exposure; the result is plotted in Figure 4.3. Approximately 56% and 42% of the CDF-S field has a full-band effective exposure greater than 1 Ms and 1.5 Ms, respectively, with a maximum effective exposure of Ms (note this is slightly smaller than the Ms total exposure since the aim points of all the Chandra observations were not exactly thesame). Thesurvey solid angles arecomparable tothose of the 2Ms CDF-N (A03; dashed curve in Fig. 4.3). Adaptively smoothed images were created using the CIAO tool csmooth on the raw images. Exposure-corrected smoothed images were then constructed following of Baganoff et al. (2003). We show in Figure 4.4 a color composite of the exposure-corrected smoothed images in the kev (red), 2 4 kev (green), and 4 8 kev (blue) bands. Source searching was performed using only the raw images, while many of the detected X-ray sources are shown more clearly in the adaptively smoothed images Point-Source Detection Point-source detection was performed in each of the three standard bands with wavdetect using a 2 sequence of wavelet scales (i.e., 1, 2, 2, 2 2, 4, 4 2, 8, 8 2, and 16 pixels). The criterion for source detection is that a source must be found with a given false-positive probability threshold in at least one of the three standard bands. For the main Chandra source catalog discussed in , the false-positive probability threshold in each band was set to If we conservatively consider the three images searched to be independent, 18 false detections are expected in the main Chandra source catalog for the case of a uniform background. However, this false-source estimate is conservative, since a single pixel usually should not be considered a source-detection cell, particularly at large off-axis angles (wavdetect suppresses fluctuations on scales smaller than the PSF). As quantified in of A03, the number of false-sources is likely 2 3 times less than our conservative estimate. We also provide additional source-significance information by running wavdetect using false-positive probability thresholds of and These results are presented in , which can be utilized to perform more conservative source screening if desired Point-Source Catalogs Main Chandra Source Catalog The source lists resulting from the wavdetect runs discussed in with false-positiveprobabilitythresholdof weremergedtocreatethemainpoint-source 43

54 Fig. 4.1 Full-band ( kev) raw image of the 2 Ms CDF-S. The gray scales are linear. The apparent scarcity of sources near the field center is largely due to the small PSF at that location (see Figs. 4.4 and 4.10 for clarification). The black outline surrounding the image indicates the extent of all the CDF-S observations. The large rectangle indicates the GOODS-S (Giavalisco et al. 2004) region, and the central square indicates the Hubble Ultra Deep Field (UDF; Beckwith et al. 2006) region. The cross near the center of the images indicates the average aim point, weighted by exposure time (see Table 4.1). 44

55 Fig. 4.2 Full-band ( kev) exposure map of the 2 Ms CDF-S. The darkest areas represent the highest effective exposure times (the maximum value is Ms). The gray scales are logarithmic. The regions and the cross symbol have the same meaning as those in Fig