Unveiling the Accretion Disks That Fuel Active Galactic

The Pennsylvania State University

The Graduate School

Department of Astronomy and Astrophysics

UNVEILING THE ACCRETION DISKS THAT FUEL ACTIVE

GALACTIC NUCLEI

A Thesis in

Astronomy and Astrophysics

Karen Theresa Lewis

c 2005 Karen Theresa Lewis

Submitted in Partial Fulﬁllment of the Requirements for the Degree of

Doctor of Philosophy

December 2005 The thesis of Karen Theresa Lewis was read and approved1 by the following:

Michael Eracleous Associate Professor of Astronomy and Astrophysics Thesis Adviser Chair of Committee

Steinn Sigurdsson Associate Professor of Astronomy and Astrophysics

W. Niel Brandt Professor of Astronomy and Astrophysics

Donald Schneider Professor of Astronomy and Astrophysics

L. Samuel Finn Professor of Physics

Lawrence Ramsey Professor of Astronomy and Astrophysics Head of the Department of Astronomy and Astrophysics

1Signatures on ﬁle in the Graduate School. iii Abstract

An increasing number of Active Galactic Nuclei (AGN) exhibit broad, double- peaked Balmer emission lines, reminiscent of those observed in Cataclysmic Variables; these double-peaked Balmer lines represent some of the best evidence for the existence of accretion disks in AGNs. There is considerable evidence to support the hypothesis that double-peaked emitters are “clean” systems in which the accretion disk is not veiled by a disk wind. This unobscured view affords the opportunity to study the underlying accretion disk which is believed to exist in all AGNs. In this thesis, I study two aspects of double-peaked emitters, namely the mechanism responsible for diminishing the accretion disk wind and the long-term profile variability of the double-peaked emission lines. It has been argued that double-peaked emitters have accretion flows that transition to a vertically extended, radiatively inefficient accretion flow at small radii. This scenario naturally explains the diminished wind in double-peaked emitters, but also offers a way to illuminate the outer accretion disk, which is necessary to produce the double-peaked emission lines. I critically analyze this hypothesis through robust estimates of the accretion rate in a few objects and also through an investigation of the X-ray spectra, which are sensitive to the structure of the inner accretion disk. I find that this hypothesis may be valid in some, but not all double-peaked emitters. Thus, alternative mechanisms for diminishing the disk wind should be sought; ideally these mechanisms should also offer a way to illuminate the outer accretion disk. Furthermore, robust estimates of the accretion rate should be determined for a much larger sample of double-peaked emitters in order to determine whether the distribution of accretion rates is continuous. A set of 20 double-peaked emitters has been monitored for nearly a decade in order to observe long-term profile variations in the double-peaked emission lines. Variations generally occur on timescales of years, and are attributed to physical changes in the accretion disk. The profile variability requires the use of non-axisymmetric accretion disk models; a few of the best observed objects have been modeled, with varying degrees of success, by invoking circular accretion disks with bright spots or spiral arms, or elliptical disks. I have characterized the variability of a group of seven double-peaked emitters in a model independent way and found that variability is caused primarily by the presence of one or more lumps of excess emission that change in amplitude, projected velocity, and shape over periods of several years. An elliptical accretion disk does not produce the correct variability patterns, and for those objects with a known black hole mass, the timescale for variability in this model is an order of magnitude longer than is observed. The spiral arm model produces variability on the correct timescale, but it is also unable to reproduce the observations. However, I suggest that with the simple modification of allowing the spiral arm to be clumpy, many of the observed variability patterns could be reproduced. To make further progress, it is important to continue monitoring these objects at least twice per year. Additionally, a few objects which showed significant variability should occasionally be monitored intensively (every few weeks) for several months at a time in order to probe variability taking place on the dynamical timescale. iv Table of Contents

List of Tables ...... vi

List of Figures ...... vii

Acknowledgments ...... viii

Chapter 1. Introduction ...... 1 1.1 The Accretion Disk Paradigm for Active Galactic Nuclei ...... 1 1.2 A Brief History of Double-Peaked Emitters ...... 3 1.2.1 External Illumination by a Radiatively Ineﬃcient Accretion Flow...... 3 1.2.2 Connection Between Double-peaked Emitters and the General AGNPopulation ...... 5 1.2.3 Challenges to the RIAF hypothesis ...... 6 1.2.4 Variability of the Double-Peaked Balmer Emission Lines . . . 7 1.3 Double-Peaked Emitters — Who Needs Them? ...... 8 1.4 TheGoalsofthisThesis ...... 9

Chapter 2. Black Hole Masses in Double-Peaked Emitters ...... 11 2.1 Introduction...... 11 2.2 Sample Selection, Observations, and Data Reduction ...... 12 2.3 AnalysisandResults ...... 17 2.3.1 FittingMethod ...... 17 2.3.2 SourcesofSystematicError ...... 17 2.3.3 Notes on Individual Objects ...... 18 2.4 DiscussionandConclusions ...... 19

Chapter 3. XMM and RXTE Observation of 3C 111 ...... 23 3.1 Introduction...... 23 3.2 Propertiesof3C111 ...... 25 3.3 Observations and Data Reductions ...... 27 3.3.1 XMM-Newton ...... 27 3.3.2 Rossi X-ray Timing Explorer ...... 29 3.4 TimingAnalysis ...... 30 3.5 SpectralAnalysis ...... 30 3.5.1 ContinuumModels ...... 32 3.5.2 Models for the Fe Kα Line...... 37 3.5.3 Combined Continuum and Fe Kα Emission Models ...... 42 3.5.3.1 Truncated Accretion Disk - Models #7a,b ...... 42 3.5.3.2 Highly Ionized Accretion Disk - Models #8a,b,c . . 43 3.5.3.3 Partial Covering - Model #9 ...... 46 v

3.6 Discussion...... 49 3.6.1 Origin of the Low Energy Component ...... 49 3.6.2 Interpretation of the Spectral Models ...... 50 3.7 Conclusions ...... 52

Chapter 4. Long-Term Profile Variability in Double-Peaked Emitters ...... 54 4.1 Introduction...... 54 4.2 Observations and Data Reductions ...... 57 4.3 ProfileAnlysis...... 61 4.3.1 Relative Narrow and Broad Line Fluxes ...... 64 4.3.2 DifferenceSpectra ...... 67 4.3.3 Variations in Profile Parameters ...... 75 4.3.4 Variations with Integrated Broad Hα Flux...... 79 4.4 Model Profile Characterization ...... 81 4.4.1 Physical Motivation ...... 81 4.4.2 Calculation of the Model Profiles ...... 83 4.4.3 Model Characterization ...... 84 4.5 Discussion and Interpretations ...... 89 4.6 Conclusions ...... 91

Chapter 5. Conclusions and Suggestions for Future Work ...... 93 5.1 ViabilityoftheRIAFscenario...... 93 5.2 Characterization of the Long-term Proﬁle variability ...... 94 5.3 Accretion Disk Winds in Double-Peaked Emitters ...... 95

Appendix A. Telluric Correction Method ...... 96

Appendix B. Inclination Angle of the Disk in 3C 111 ...... 99

Appendix C. Total Galactic Hydrogen Column Density Towards 3C 111 ..... 100

Bibliography ...... 101 vi List of Tables

2.1 GalaxyProperties ...... 13 2.2 TemplateStars ...... 13 2.3 Velocity Dispersions and Derived Properties ...... 21

3.1 ObservationLog ...... 28 3.2 Best Fit Continuum Parameters ...... 33 3.3 Best Fit Parameters for Combined Continuum and Fe Kα Line Models . 47

4.1 GalaxyProperties ...... 56 4.2 InstrumentalConﬁgurations ...... 58 4.3 LogofObservations ...... 59 vii List of Figures

2.1 Observed Ca ii linesandBest-ﬁtModels ...... 16

3.1 3C111X-rayLightcurve...... 31 3.2 X-raySpectraof3C111 ...... 34 3.3 Confidence Contours in the Column Density and Photon Index..... 35 3.4 Confidence Contour in the Folding Energy and Reflection Fraction . . . 36 3.5 Confidence Contours for Gaussian Fits to the Fe Kα Line ...... 39 3.6 Confidence Contours for the Disk-line Fits to the Fe Kα Line (Powerlaw +softGaussianContinuum) ...... 40 3.7 Confidence Contours for the Disk-line Fits to the Fe Kα Line (Powerlaw +Comptonreflectioncontinuum)...... 41 3.8 Ratio of the REFSCH and Power Law Models ...... 44 3.9 Confidence Contour in the Ionization Parameter and Reflection Fraction 45 3.10 Confidence Contour in the Column Density and Covering Fraction for thePartialCoveringModel ...... 49 3.11 Confidence Contours for the Gaussian Fit to the Fe Kα Line (Partial CoveringModel) ...... 53

4.1 Double-Peaked Balmer Emission Line Profile ...... 61 4.2 Broad Hα LightCurves ...... 66 4.3 DifferenceSpectra ...... 68 4.4 Variations in Profile Properties as a Function of Time ...... 77 4.5 Variations in Peak Separation with Broad Hα Flux...... 80 4.6 ModelProfiles...... 85

A.1 Un-corrected Spectrum of IRAS 0236.6–3101 ...... 96 A.2 Example Fit of a Rapidly Rotating B-star and the Residuals ...... 98 viii Acknowledgments

I am first and foremost grateful to my parents, my first teachers, and all of my family for their moral support during my twenty-three years in school! Secondly, I am indebted to my advisor Mike Eracleous for his willing advice and assistance, in matters both large and small, and also his (seemingly) endless reserves of patience. Finally I would like to thank the Penn State Astronomy Department for providing such a wonder- ful group of people to work (and relax) with. Numerous members of the department have helped me during the last six years and I cannot possibly give credit to everybody, but I would especially thank Tamara Bogdnović, John Debes, Jie Ding, Julian van Eyken, Suvrath Mahadevan, Mike Sipior, and Michele Stark for helping me to keep the loss of sanity to a minimum, especially during the “early years”. I would like to thank my thesis committee for their useful advice and suggestions. Chapter 3 of this thesis was previously published (ApJ, 622, 2, 618); I would like to acknowledge the work of Mario Gliozzi who prepared the Rossi X-ray Timing Explorer data for analysis and also performed the timing analysis. I also gratefully acknowledge the work of Rita Sambruna and Richard Mushotzky who provided valuable scientific input. The code I used to perform the relativistic blurring in Chapter 3 was generously provided by Andy Fabian. There are numerous individuals with whom I’ve had engaging scientific conversations during the course of working on this thesis, including but certainly not limited to Mateo Guinazzi, Aaron Barth, Jonathon Gelbord, David Ballantyne, Suvi Gezari, and John Everett. The results on the long-term variability of the double-peaked Balmer lines in AGNs would not have been possible without the dedication of many people, in particular Mike Eracleous and Jules Halpern who logged over 100 nights of observing. Alex Filippenko, a dedicated fan of this project, has carried out many observations and provided several spectra. Thaisa Storchi-Bergmann, another dedicated collaborator, has carried out many observations at CTIO. We also thank Sue Simkin and Bob Becker for providing specific spectra of Pictor A. Sarah Gallagher, Ann Hornschemeier, and Mike Sipior also assisted with some of the observations. Finally I would like to thank the staff at the Kitt Peak National Observatory and the Cerro-Tololo Interamerican Observatory for their expertise and helpfulness. Many of the spectra for the variability study were obtained at the National Optical Astronomy Observatories and some with the Low Resolution Spectrograph on the Hobby Eberly Telescope. The National Optical Astronomy Observatories, which operates the Kitt Peak National Observatory and the Cerro-Tololo Interamerican Observatory, is operated by AURA, under a cooperative agreement with the National Science Foundation. The Hobby-Eberly Telescope (HET) is a joint project of the University of Texas at Austin, the Pennsylvania State University, Stanford University, Ludwig-Maximilians- Universität München, and Georg-August-Universität Göttingen. The HET is named in honor of its principal benefactors, William P. Hobby and Robert E. Eberly. The Marcario Low Resolution Spectrograph is named for Mike Marcario of High Lonesome Optics who fabricated several optics for the instrument but died before its completion. ix

The LRS is a joint project of the Hobby-Eberly Telescope partnership and the Instituto de Astronom´ıa de la Universidad Nacional Autonoma de M´exico. Finally, I am extremely grateful for the ﬁnancial support of NASA, through the Graduate Student Researchers Program (NGT5-50387) and the Pennsylvania Space Grant Consortium Fellowship program. The research presented in Chapter 3 was also partially supported by an XMM-Newton grant from NASA (NAG-9982). Additional support for travel to observatories was provided by the Zaccheus Daniel Foundation and the Association of University for Research in Astronomy, Inc. (AURA). 1

Chapter 1

Introduction

1.1 The Accretion Disk Paradigm for Active Galactic Nuclei

Active Galactic Nuclei (AGNs) are among the most energetic objects in the Uni- verse, with bolometric luminosities often exceeding 1046 erg s−1. It is now generally accepted that at the heart of these powerhouses lies a supermassive black hole with a mass 6 in excess of 10 M that is accreting matter from its host galaxy (see, e.g., Salpeter 1964; Rees 1984; Peterson 1997). As matter spirals inwards, it forms an equatorial accretion disk around the black hole (such as that described by Shakura & Sunyaev 1973), whose inner portions are heated to UV-emitting temperatures by “viscous” stresses which tap into the potential energy of the accreting gas. During the past decade, simulations have made clear the importance of magneto-hydrodynamical processes in accretion disks, and a more modern theory attributes the “viscosity” to magnetic tension and reconnection. In particular a magneto-rotational instability may play a critical role in allowing the gas in the disk to shed its angular momentum (for a review, see Balbus & Hawley 2003). The inner accretion flow acts as a central engine, driving many of the observed phenomena. In particular, two of the hallmarks of AGNs—broad emission lines and the emission of hard X-rays—are directly related to the accretion flow that fuels the AGN, as described below. The broad emission lines observed in AGNs typically have full widths at half maximum (FWHM) of & 5000 km s−1, although some AGNs exhibit lines with FWHM & 104 km s−1. The widths of these lines are far in excess of the thermal velocity of the gas, which is estimated to be 10 km s−1 (T 104 K), and a large bulk velocity of the ∼ ∼ emitting gas is required to explain them. Furthermore, the broad emission line fluxes respond to changes in the ionizing continuum on timescales of days to a few light months (e.g. Peterson 1993; Korista 1995; Peterson et al. 2004), indicating that gas in the Broad Emission Line Region (BELR) is located within the inner 0.1 pc of the AGN. During the past several decades, many observational studies have been undertaken to unravel the nature of the BELR, including large scale surveys, detailed observations of individual objects, and intensive reverberation mapping campaigns, in which the response of the broad emission lines to changes in the ionizing continuum flux is systematically monitored. Although a definitive theory of the nature of the BELR remains elusive, during the past decade the idea that the BELR is related to the accretion flow that fuels the AGN has been steadily gaining in support in the AGN community, both observationally and theoretically. An historical description of the emergence of this disk plus wind scenario for the BELR is beyond the scope of this introduction, but brief reviews are given by Gaskell & Snedden (1999) and Eracleous (2004). 2

The dense accretion disk itself, with densities of 1013−15cm−3, is thought to be a source of the broad, low-ionization Hα, Hβ, and Mg iiλ2798 lines, as first argued by Collin-Souffrin (1987) and Collin-Souffrin & Dumont (1989). This idea is bolstered by the fact that some AGNs emit double-peaked Hα and Hβ profiles (Eracleous & Halpern 1994; Strateva et al. 2003), reminiscent of emission lines from the disks of Cataclysmic Variables, where they are an regarded as the kinematic signature of the accretion disk (e.g. Horne & Marsh 1986). These double-peaked emitters are the subject of this thesis and will be described in more detail below and in subsequent chapters. The accretion disk proper cannot be the sole source of broad emission-lines, however, because it is unable to produce the strong C iii]λ1909, C ivλ1549 and Lyα emission lines which are commonly observed. The disk is not sufficiently ionized to produce the C iii]λ1909 and C ivλ1549 lines, and the Lyα line photons are effectively trapped in the dense disk. Furthermore, the majority of AGNs do not emit double-peaked low- ionization emission lines. However, there are several ways in which the disk-signature of these lines can be masked; I will return to this issue in 1.2.2. Instead, an accretion § disk wind is thought to be an important source of both high- and low-ionization broad emission lines as discussed by many authors (e.g., Shields 1977; Emmering, Blandford, & Shlosman 1992; de Kool & Begelman 1995; Chiang & Murray 1996; Elvis 2000). This wind is thought to be propelled by a combination of resonant line driving by the UV emission from the inner disk and/or magneto-hydrodynamical forces (e.g. Murray, Chi- ang, Grossman, & Voit 1995; Königl & Kartje 1994; Proga, Stone, & Kallman 2000; Everett 2004). The bulk of the line emission is expected to arise in the base of the wind, which is qualitatively similar to an accretion disk atmosphere. In this scenario, the low-ionization lines form in the densest portions of the wind which sits just above the disk (or in the disk itself) while the high-ionization lines form in the less dense portions of the wind which have been slightly accelerated. As clearly demonstrated by Chiang & Murray (1996), although the disk wind is rotating, the lines arising from it are single-peaked because the optical depth is not isotropic; photons will escape much more easily along lines of sight with a low projected velocity, thereby enhancing the core of the profile and suppressing the wings. The other ubiquitous feature of AGNs is X-ray emission (e.g. Mushotzky, Done, & Pounds 1993, and also references in Chapter 3), both in the form of continuum emission which extends to hundreds of keV and emission lines, the most prominent being the Fe Kα fluorescence line at 6.4 keV. The accretion disk around a supermassive black hole never reaches X-ray emitting temperatures, so it is postulated that a corona of energetic electrons forms in the inner regions of the accretion flow and UV photons from the inner accretion disk are inverse Compton scattered into the X-ray regime. The exact form of this corona (i.e. thermal or non-thermal origin, geometry, etc.) are as yet uncertain, but in all current models, the formation of this corona is intimately related to processes in the underlying accretion disk (see, e.g., Haardt & Maraschi 1991, 1993; Poutanen 1998, and references therein). Some of these hard photons are scattered and irradiate the underlying cold, dense disk, leading to several reprocessing features, namely fluorescent emission lines (most notably from Fe Kα ) and continuum emission in the form of a 3

Compton reflection bump, an excess of emission that peaks at 30 keV. These reflection ∼ features are described in more detail in Chapter 3. Thus the accretion disk is an integral part of an AGN. Not only does it provide a means to feed the supermassive black hole and harness the potential energy of the accreting matter, but many of the most energetic and ubiquitous features of AGNs are formed in the accretion disk and the associated wind and corona. Although the presence of an accretion disk in an AGN is now commonly accepted, and in some ways taken for granted, the direct evidence of this accretion disk is limited. A broad excess of UV emission at wavelengths shorter than 4000A,˚ known as the “big blue bump”, is thought ∼ to be the blackbody emission from inner UV emitting portion of the accretion disk (see, e.g., Malkan & Sargent 1982). Additionally, the Fe Kα lines in some AGNs are distorted and their profiles are well-fitted with a model attributing their origin to a disk (see e.g., Fabian et al. 1995; Nandra et al. 1997, and references in Chapter 3). As mentioned above, some AGNs emit double-peaked Balmer emission lines. These disk-like emission lines, represent the most direct, kinematic evidence for the presence of a large-scale accretion disk in AGNs. Numerous studies are devoted to the study of Fe Kα emission lines in AGNs, and some of these results are summarized in Chapter 3. The main focus of this thesis is to study those objects which emit double-peaked Balmer emission lines.

1.2 A Brief History of Double-Peaked Emitters

Double-peaked emission lines were ﬁrst observed in the broad-line radio galaxies (BLRGs) Arp 102B (Stauﬀer, Schild, & Keel 1983; Chen & Halpern 1989; Chen, Halpern, & Filippenko 1989), 3C 390.3 (Oke 1987; Perez et al. 1988), and 3C 332 (Halpern 1990). Given the similarity between these lines and those observed in CVs, these authors suggested that the lines originated in an accretion disk surrounding the AGN at distances 2 of hundreds to thousands of gravitational radii (rg = GMBH /c , where MBH is the mass of the black hole). Later, a survey of BLRGs was undertaken and it was found that 20% of the observed objects exhibited double-peaked Balmer emission lines (Er- ∼ acleous & Halpern 1994, 2003). In addition to possessing double-peaked lines, these objects have several properties which set them apart from other BLRGs. Their Balmer lines are twice as broad as those found in other BLRGs (the most extreme examples have FWHM 40 000 km s−1; Wang et al. 2005), the low-ionization forbidden lines ∼ ([O i]λ6300 and [S ii]λ6717, 6731) are unusually strong compared to other BLRGs, and the non-stellar continuum is weaker, with 50% of the continuum near Hα being at- ∼ tributed to starlight from the host galaxy (compared to . 10% for other BLRGs spanning a similar luminosity range.)

1.2.1 External Illumination by a Radiatively Inefficient Accretion Flow One of the main objections to an accretion-disk origin for the double-peaked (and other) broad emission lines, was that the local dissipative and gravitational energy released by the disk at these large radii (at least in a standard Shakura & Sunyaev thin disk; Shakura & Sunyaev 1973) was insufficient to power the observed lines (Shields 1977; Chen et al. 1989). Thus Chen & Halpern (1989) postulated that the outer accretion disk 4 was illuminated by a vertically extended structure in the inner accretion disk, such as those found in the two-temperature disk solutions described by Shapiro, Lightman, & Eardley (1976) and Rees et al. (1982). These and other radiatively inefficient accretion flows (RIAFs), such as Advection Dominated Accretion Flows (ADAF; Narayan & Yi 1994, 1995) or Convection Dominated Accretion Flow (CDAF; Quataert & Gruzinov 2000; Narayan, Igumenshchev, & Abramowicz 2000), are thought to form when the accretion rate is very low, relative to the Eddington accretion rate,1 because the cooling timescale due to collisions between the electrons and ions is greatly increased; in particular the ions become extremely hot (T 109–1011 K). In the outer disk, the gas can ∼ cool effectively and the disk remains geometrically thin and optically thick. At some radius, the accretion disk becomes dominated by gas pressure (due to the inability of the ions to cool), causing the disk to become geometrically thick and optically thin. I note that when the accretion rate is very high, relative to the Eddington accretion rate, a vertically extended structure (a radiation supported torus, Rees et al. 1982) can also form. This flow is still optically thick and only slightly hotter than the standard thin disk, but is radiatively inefficient because the accretion timescale is shorter than the cooling timescale. The illumination of a dense accretion disk by a weak source of high-energy photons, such as that produced by a RIAF, was investigated by Collin-Souffrin (1987) and a subsequent series of papers (Collin-Souffrin & Dumont 1989, 1990; Dumont & Collin- Souffrin 1990a,b,c). These authors found that the heating of the disk by the external illumination accounted for the “missing” energy and that the physical conditions in the heated gas were in agreement with those required by photoionization models to reproduce the observed low-ionization lines. There is considerable support for the hypothesis, at least among the well-studied objects in the Eracleous & Halpern (1994) sample, that the inner accretion flow in double-peaked emitters has the form of a RIAF instead of a standard geometrically thin accretion disk. The Spectral Energy Distributions (SEDs) of Arp 102B and 3C 390.3 lack the “big blue bump”. This deficit of UV emission is also observed in Low Ionization Nuclear Emission Line Regions (LINERs; Heckman 1980) as shown, for example, by Ho (1999) and those LINERs that are powered by accretion onto black hole are generally considered to harbor a RIAF. Furthermore, the unique properties of the double-peaked emitters in the Eracleous & Halpern (1994) sample, namely strong low-ionization lines and relatively weak non- stellar continuum (which would arise in the inner disk), suggest indirectly a deficit of UV emission in double-peaked emitters in general. Nagao et al. (2002) performed photoionization calculations and showed that the observed [O i]/[O iii] and [O i]/[O iii] line ratios are well-reproduced by an SED similar to those observed in LINERs.

1 38 −1 The Eddington luminosity, given by LEdd = 1.38 10 M/M erg s , is the luminosity at which the radiation pressure of the gas will exceed the× gravitational force of the accreting object. 2 The Eddington accretion rate is given bym ˙ Edd = LEdd/η c , where η is the efficiency with which gravitational energy is converted into radiation. The transition to a two-temperature flow occurs 2 whenm/ ˙ m˙ Edd . α , where α is the Shakura-Sunyaev viscosity parameter (Rees et al. 1982). The ratio of the bolometric luminosity to LEdd is commonly used as a proxy form/ ˙ m˙ Edd; when L /L < ηα2 10−3, the inner accretion flow might take the form of a RIAF. Bol Edd ∼ 5

These similarities to LINERs are underscored by the discovery of double-peaked emission lines in several LINERs, including NGC 1097 (Storchi-Bergmann, Baldwin, & Wilson 1993), M81 (Bower et al. 1996), NGC 4203 (Shields et al. 2000), NGC 4450 (Ho et al. 2000), and NGC 4579 (Barth et al. 2001). Given the difficulty of detecting broad emission lines in LINERs (Ho et al. 1997b) it is likely that a larger number of LINERs exhibit double-peaked emission lines. A final piece of evidence to support the hypothesis that the inner accretion flow has the form of a RIAF is provided by the UV spectra of double-peaked emitters, which are discussed in the next section.

1.2.2 Connection Between Double-peaked Emitters and the General AGN Population If all AGNs have accretion disks and double-peaked emission lines arise in the disk, why don’t all AGNs emit double-peaked emission lines? The UV spectra of double- peaked emitters suggest a possible answer to this question. An important consequence of a deficit in UV emission is that the accretion-disk wind, which is thought to be partially accelerated by UV resonance line driving, should not be as strong as those observed in AGNs with a big blue bump (Proga et al. 2000). The C iii]λ1909, C ivλ1549 and Lyα emission lines of Arp 102B are in fact very weak and furthermore the Mg iiλ2798 line is double-peaked and has a profile that is similar to those of Hα and Hβ (Halpern et al. 1996). Six other double-peaked emitters were observed with the Hubble Space Telescope and a similar phenomenon is observed; preliminary results have been discussed by Eracleous (2004). One of the most striking trends is that as the luminosity of the object increases, the UV resonance lines become more similar to those observed in other AGNs (i.e. stronger and more blueshifted with respect to the narrow component of the line) and the Mg iiλ2798 line becomes less distinctly double- peaked. A similar luminosity effect is seen in CVs where more luminous systems (i.e. those in outburst) tend to have single-peaked Balmer emission lines as opposed to the more commonly observed double-peaked lines (e.g. Marsh & Horne 1990; Still, Dhillon, & Jones 1995). Murray & Chiang (1996) suggested that this phenomenon is simply due to the more luminous CVs having a much “stronger” disk wind component in the emission line. Thus, a reasonable hypothesis is that the accretion disk-wind does not make as significant a contribution to the broad emission lines in double-peaked emitters as in other AGNs, but that as the luminosity (which should be linked to the accretion rate) increases, the wind begins to make a larger contribution. Certainly the underlying accretion disk does not “disappear” as the wind becomes stronger. As demonstrated by Chiang & Murray (1996) and Murray & Chiang (1998), the wind effectively masks the direct emission from the underlying accretion disk, because the photons emitted by the disk must pass through the wind and take on a single-peaked velocity profile. Thus, double-peaked emitters may be very similar to other AGNs with the exception that their accretion-disk winds are weaker and therefore the double-peaked disk emission lines can actually be observed. I also note that even in the absence of an accretion-disk wind, there are other ways to mask the signature of the accretion disk. As stressed by Eracleous & Halpern 6

(2003), an emission line from a disk will only appear double-peaked when the ratio of the outer to inner radius is small (i.e. an accretion ring, as opposed to a disk). As this ratio increases the two peaks merge together and the resulting proﬁle is similar to those seen in many AGNs, particularly when the narrow emission lines are included. Thus it is also possible that the outer boundary of the Balmer line-emitting region of the disk in double-peaked emitters is unusually small in comparison to other AGNs. However, I note that this eﬀect alone would not naturally explain the other properties of double-peaked emitters.

1.2.3 Challenges to the RIAF hypothesis The hypothesis that the inner accretion flows in double-peaked emitters has the form of a RIAF is nicely consistent with the observed properties of the double-peaked emitters in the Eracleous & Halpern (1994) sample and offers a way to link the double- peaked emitters to the general AGN population. However, there is an important chal- lenge to this scenario. Strateva et al. (2003) found over 100 double-peaked emitters in the Sloan Digital Sky Survey (SDSS; York 2000), representing 3% of the z < 0.332 AGN population. The properties of the double-peaked profiles in this sample are very similar to those of the objects in the Eracleous & Halpern (1994) sample and they appear to originate in the same region of the accretion disk. However, Strateva et al. (2003) found that the general properties of the double-peaked emitters are indistinguishable from those of the general AGN population, in sharp contrast to the findings of Eracleous & Halpern (1994). Although the equivalent widths of [O i]λ6300 and the [O i]λ6300/[O iii]λ5007 ratio were larger than in the parent population, there was little difference in the [S ii]λλ6716, 6731 equivalent width or the [O ii]λ3727/[O iii]λ5007 ratio. In particular, only 12% of the SDSS sample were classified as LINERs by their emission line ratios. Taken at face value, these results seem to call into question the hypothesis that all or even most double-peaked emitters have a RIAF. However, there are few caveats. Firstly, among the Eracleous & Halpern (1994) objects only a handful of objects were strictly classified as LINERs; most were merely “reminiscent” of LINERs. LINERs are rather extreme objects with L /L 10−6–10−3 (Ho 1999). It is possible that an Bol Edd ∼ object with a slightly larger Eddington ratio could possess a RIAF that transitions to a thin disk at a smaller radius than in LINERs. As a case in point, Gammie, Narayan, & Blandford (1999) performed detailed modeling of the SED in NGC 4258 which has 7 a well-known black hole mass of (4.0 0.25) 10 M from water maser measurements ± × (Herrnstein et al. 1999). They concluded that the SED was best fitted by a ADAF −1 that had a mass accretion rate of 0.01M yr , which transitions to a thin disk at ∼ r 10–100r . The mass accretion rate corresponds tom/ ˙ m˙ 0.01, and this object ∼ g Edd ∼ is classified as a Seyfert 1.9 by Ho, Filippenko, & Sargent (1997a), not as a LINER. This demonstrates that it is possible for non-LINERs and objects with rather high mass accretion rates (compared to LINERs) to possess a small-scale ADAF or other RIAF nevertheless. Furthermore, as pointed out by Chen et al. (1989), if the accretion rate is very low, the RIAF may extend so far that there is no thin disk to illuminate. 7

I stress that the primary requirements that RIAF scenario satisfies are (1) external illumination of the outer accretion disk and (2) a decreased contribution from an accretion-disk wind due to a deficit of UV photons. While the presence of a RIAF explains both of these nicely, there may be other ways to achieve these conditions. The fact that most of the double-peaked emitters in the SDSS are not LINERs and are not obviously different from the general z < 0.332 population is puzzling, and the RIAF hypothesis should be carefully scrutinized (indeed this is a primary goal of this thesis) but it does not necessarily invalidate the general hypothesis that we see double-peaked emission lines in those objects with weak contributions from accretion disk winds.

1.2.4 Variability of the Double-Peaked Balmer Emission Lines As described in more detail in 4.1, several of the original double-peaked emitters § (Arp 102B, 3C 390.3, and 3C 332) exhibited variations in their line profiles that took place on timescales of years. This long-term variability was intriguing because it was unrelated to more rapid changes in luminosity but might be related to dynamical changes in the disk. Thus a campaign was undertaken to observe the double-peaked emitters found by Eracleous & Halpern (1994). Many objects were observed 2–3 times per year, although some objects could only be observed once per year. The goals of the monitoring campaign were to determine (1) whether all double-peaked emitters exhibited such long- term profile variability (2) to test alternative scenarios for the origin of the double-peaked emission lines and (3) to ultimately learn about physical processes that take place in the outer accretion disks in AGNs and more generally in the BELR. The first goal has clearly been met. Long-term profile variability was found to be a ubiquitous feature of double-peaked emitters, as described in detail in Chapter 4. The second goal has also been successfully met and many alternative origins for the double-peaked lines have been ruled out because they were inconsistent with the observed variability as well as the observed properties of double-peaked emitters. The full details of why the alternative models were ruled out are given in Eracleous & Halpern (2003), but here I briefly summarize the results. It was proposed that the double-peaked lines originated in the BELRs around a binary black hole pair and that the lines were displaced by the orbital motion of the binary pair (e.g., Gaskell 1996). This was ruled out for several reasons, but one of the most critical was that this model predicted that the peaks would drift in opposite directions, which is not observed (Eracleous et al. 1997). A second suggestion was emission from a bipolar outflow, such as a jet (e.g. Zheng, Sulentic, & Binette 1990); the primary problem with this model is that the blue side of the profile should respond to continuum variations before the red, and this was not observed in reverberation mapping of 3C 390.3 (Dietrich 1998). A similar theory, proposed by Wanders et al. (1995) and Goad & Wanders (1996), is that the BELR is a spherically symmetric distribution of clouds that is illuminated anisotropically. This theory is not ruled out by the observed variability; however, it is in disagreement with the other properties of double-peaked emitters. Furthermore, this scenario suffers from the same criticisms that apply to BELR “cloud” models in general; there are numerous ways to destroy the clouds, but very few to create them (see e.g., Mathews & Capriotti 1985). 8

The third, and most lofty, goal remains a work in progress. As will be described in more detail in Chapter 4, proﬁle variability is very common and the most common and easily discernible variability trend is a modulation in the strengths of the red and blue peaks; in some instances the red peak is even stronger than the blue peak, in contrast to the expectation that the blue peak is Doppler boosted. Thus it was clear early on that a simple, circular accretion disk could not explain the observations and more general, non-axisymmetric models were required. These models include disks with bright spots or spiral arms or elliptical accretion disks. Although these models have been applied to a few objects (see Chapter 4), preliminary results for the double-peaked emitters as a whole are only now available, a decade into the monitoring campaign. These results are described in this thesis (Chapter 4) as well as in Gezari (2005).

1.3 Double-Peaked Emitters — Who Needs Them?

It is clear that double-peaked emitters are very interesting objects in their own right, but one can question what will be learned about the broad line regions and accretion disks in the general AGN population by studying them. After all, double-peaked emitters only represent 5–10% of z < 0.4 radio-loud AGNs and 3% of the z < 0.3 AGN population. Why should so much time and energy be devoted to such a small fraction of the AGN population? Here, I provide several compelling reasons to study these objects; this list is by no means exhaustive.

1. Double-peaked emitters represent an “extreme” segment of the AGN population and, like Narrow-line Seyfert 1s and Broad Absorption Line QSOs, oﬀer a unique view of the AGN phenomenon. Any ultimate theory for the physics of AGN BELRs and AGN physics in general must meet the challenges set by these extreme objects.

2. Just because most AGNs do not emit double-peaked emission lines does not mean that the disk does not make a contribution to the broad emission lines in those AGNs. The broad emission lines in some well-studied AGNs also exhibit profile variability on long timescales which is not connected to variations in the illuminating continuum (Kassebaum et al. 1997; Wanders & Peterson 1996; Corbin & Smith 2000). In addition, the profile of the Balmer lines in the Seyfert galaxy Akn 120 are occasionally double-peaked (Peterson et al. 1985) and difference spectra of NGC 5548 (Stirpe, de Bruyn, & van Groningen 1988) also revealed a highly variable, double-peaked component to the broad Balmer lines in that object. Fur- thermore, the line profiles in many AGNs are asymmetric and several authors have suggested that the low-ionization emission-lines of most, if not all, AGNs have some component from an accretion disk (Gaskell & Snedden 1999; Popovićet al. 2004). Double-peaked emitters thus represent “clean” systems in which this disk component can be more easily studied and modeled. In principle there should be little difference in the structure of the outer regions of the accretion disks in double-peaked emitters and other AGNs. Thus, the physical insight into the origin of long-term profile variability of broad emission lines gained from the study of these objects should apply to all AGNs. 9

3. The immediate goal of studying double-peaked emitters is to understand variability in AGN BELRs and its physical origin, but a more lofty goal is to use that variability to learn about the physics of accretion disks. Detailed studies of the physics of accretion flows have been possible in CVs and other stellar binary systems, which are numerous, fairly bright, and vary on short timescales so that the effects of many physical processes can be investigated (see, e.g., Frank, King, & Raine 2002). Do the same general principles of the accretion process apply to all accretion systems, or are the accretion disks of AGNs somehow different? This is admittedly a long-term goal which will take decades of observations of many objects, however this goal should be kept in mind.

1.4 The Goals of this Thesis

In this thesis, I have set out with two primary goals. The ﬁrst is to test the hypothesis that double-peaked emitters possess a RIAF (see 1.2.1). This hypothesis is § a fundamental component to the current theories for explaining the unique properties of double-peaked emitters and their connection to the AGN population as a whole. There is some direct evidence for this hypothesis for some objects, but it must be tested, especially in light of the results of Strateva et al. (2003) which were described in 1.2.3. § Secondly, it must be determined whether the variability trends seen in the handful of well-studied objects applies to double-peaked emitters in general. To meet these goals, I have conducted three inter-related projects.

1. In Chapter 2, I present near-IR observations of the Ca iiλλ8495, 8542, 8662 triplet in ﬁve double-peaked emitters. These lines are used to measure the stellar velocity dispersion in these objects and to estimate the black hole masses in these AGNs via the well-known correlation between the velocity dispersion and the black hole mass (e.g. Tremaine et al. 2002). Black hole masses were also estimated for other objects whose velocity dispersions were obtained by similar means and published in the literature. These masses serve two purposes. Firstly, the RIAF scenario can be tested directly by determining the Eddington ratio in these systems. Secondly, the physical timescales in the accretion disk are set by the black hole mass; to get a better sense of the physical origin of the proﬁle variability in double-peaked emitters, it is necessary to have a robust estimate of the black hole mass.

2. In Chapter 3, I present the results of a simultaneous observation of the BLRG 3C 111 with two X-ray observatories: XMM-Newton and the Rossi X-ray Timing Explorer. The presence of a RIAF in the inner accretion disk greatly affects the X-ray spectrum of an AGN, in particular the reprocessing features (the Fe Kα fluorescent emission line and the Compton bump in the high energy continuum). As described in Chapter 2, BLRGs have been observed to have systematically different spectra than Seyferts, their radio-quiet counterparts. Currently there are two popular theories to explain this difference: 1) the inner thin disk is truncated due to the presence of a RIAF 2) the inner disk could be highly ionized due to a large accretion rate. I test these two hypotheses in the case of 3C 111 and also draw upon recent results from the literature to asses these scenarios in general. This 10

object is not a double-peaked emitter, but previous observations have shown that the X-ray properties of double-peaked BLRGs are indistinguishable from those of BLRGs in general; what is true for the parent population is likely to be true for double-peaked emitters.

3. In Chapter 4, I present a detailed, model-independent characterization of the variability trends in seven double-peaked emitters which were part of the long-term monitoring campaign. This characterization will clearly define the challenges that must be met by current and future models that are developed to describe the physical processes that drive the variability in these objects. I note that the models which can be tested against such a model-independent characterization are not necessarily limited to those based on an accretion disk origin for the double-peaked lines. These trends are compared to those predicted by two of the currently used models, that of an accretion disk with a spiral arm and an elliptical accretion disk and I offer suggestions for their refinement. This analysis represents a first, but very important step, towards the long-term goal of using the double-peaked emitters to better understand the physical origins of the AGN BELR and its long-term variability.

Each of these chapters is self-contained and the necessary background information, as well as principle conclusions are presented in each chapter. In Chapter 5, I summarize the results of these three projects in relationship to each other and the goals of this thesis and also oﬀer several ideas for further study. Throughout this thesis, I assume a Wilkin- −1 −1 son Microwave Anisotropy Probe cosmology (H0 = 70kms Mpc , ΩM = 0.27, ΩΛ = 0.73; Spergel et al. 2003). 11

Chapter 2

Black Hole Masses in Active Galaxies with Double-Peaked Balmer Emission Lines

2.1 Introduction

As described in in Chapter 1, an increasing number of AGNs exhibit double- peaked Balmer emission lines which are thought to arise in the outer regions (r 1000 r ) ∼ g of the accretion disk that feeds the supermassive black hole. These double-peaked lines vary on timescales of years to decades and the variations are unrelated to fluctuations in the illuminating continuum; rather these variations occur due to local, physical changes within the accretion disk itself. The variability of the double-peaked profiles demands the use of non-axisymmetric accretion disk models such as accretion disks with emissivity enhancements (such as bright spots or spiral arms) or elliptical disks; however, these represent only the most simple extensions to a circular accretion disk. Beginning in the mid-1990s, a campaign was undertaken to monitor 20 double-peaked emitters for ∼ approximately a decade, with the hope of being able to test these, and other, models for dynamical phenomena that occur in AGN accretion disks. To successfully interpret and model the long-term profile variability, it is not sufficient to reproduce the observed sequence of line profiles; the variability must occur on a physical time scale that is consistent with the chosen model. The timescales of interest are the dynamical, thermal, and sound crossing times, which are set by the black hole mass (MBH ) and are given by:

3/2 τ 6 M8 ξ months (2.1) dyn ∼ 3 τ τ /α (2.2) th ∼ dyn −1/2 τ 70 M8 ξ3 T years (2.3) s ∼ 5 8 3 5 where M8 = M /10 M , ξ3 = r/10 r , T5 = T/10 K, and α ( 0.1) is the Shakura- BH g ∼ Sunyaev viscosity parameter (Shakura & Sunyaev 1973). The above models all predict variability on different timescales. For example, matter embedded in the disk orbits over τdyn, thermal instabilities will dissipate over τth, density perturbations, such as a spiral wave, precess on timescales that are an order of magnitude longer than τdyn up to τs, and an elliptical disk will precess over even longer timescales. Many of the models described above yield strikingly similar sequences of profiles. In order to determine which physical mechanism is responsible for the profile variability, it is necessary to connect the observed variability timescale (in years) with one of the above physical timescales. Using an estimate of the black hole mass in NGC 1097, Storchi-Bergmann et al. (2003) estimated the physical timescales in the outer accretion 12 disk which favored a spiral arm model over an elliptical disk model for this object. However, in order to discriminate between the above timescales, the black hole mass must be known to better than an order of magnitude. As discussed in Chapter 1, the assumption that the inner accretion flow is radiatively inefficient is fundamental to the current ideas for the formation of the double- peaked emission lines as well as the relationship between double-peaked emitters and the AGN population in general. It is extremely important to test this hypothesis more directly by obtaining estimates of the black hole masses, and thusm/ ˙ m˙ Edd, for double- peaked emitters. Most of the double-peaked emitters are too distant to obtain black hole masses directly via spatially resolved stellar and gas kinematics. However, it is possible to determine the black hole masses indirectly through the well-known correlation between MBH and the stellar velocity dispersion (σ) measured on the scale of the effective radius of the bulge (Ferrarese & Merritt 2000; Gebhardt et al. 2000; Tremaine et al. 2002). Therefore I have begun a program to measure the stellar velocity dispersions in double-peaked emitters using the Ca ii λ8594, 8542, 8662 triplet and present here the results for five objects. After obtaining estimates of the black hole masses via the MBH –σ relationship, the physical timescales in the outer accretion disk andm/ ˙ m˙ Edd can be inferred for each object. In 2.2, I describe the target selection, observations and data reductions. § The analysis of the data, including an examination of the various sources of error, are described in 2.3 and the results and their implications are presented and discussed in § 2.4. In Appendix A, I present a detailed description of the procedure used to correct § the telluric water vapor absorptions lines in these spectra.

2.2 Sample Selection, Observations, and Data Reduction

My primary motivation for obtaining robust black hole masses is to assist with modeling and interpreting the long-term profile variability of the double-peaked emitters. Therefore, I selected targets that were part of the long-term monitoring campaign and have shown interesting variability. Absorption due to telluric water vapor becomes severe beyond 9200 A,˚ so only objects with z < 0.062 were selected. Finally, I only selected objects with declinations less than 5◦. Five objects, NGC 1097, Pictor A, PKS 0921– − 213, 1E 0450.3–1718, and IRAS 0236.6–3103, met these criteria; the properties of these objects are given in Table 2.1. All of the targets are hosted by elliptical galaxies or early- type spirals, whose stellar spectra in the Ca ii region are dominated by G and K giants (Worthey 1994). Therefore I observed numerous G and K giant stars, with spectral types ranging from K5 to G6, to serve as stellar templates. The stars used in the final fits are given in Table 2.2. The spectra were obtained on 2003 Dec 4–8 using the RC spectrograph on the 4m Blanco telescope at the Cerro-Tololo Ineteramerican Observatory. The G 380 grating was used in conjunction with the RG 610 order-separating filter and the spectra covered 7690 – 9350 A.˚ The slit had a width of 100. 33 and was oriented east to west. The galaxies and the template stars were observed at an airmass of less than 1.13 and differential atmospheric refraction was not significant over the small wavelength interval of interest. Thus it was not necessary to orient the slit at the parallactic angle. The resulting spectral 13

Table 2.1. Galaxy Properties

Galaxy fν(8500 A)˚ Extraction Exposure Name za (mJy)b size (00, kpc) Time (s) S/Nc

NGC 1097 0.0043 26.0 5.0 (0.45) 5400 115 Pictor A 0.035 3.0 4.5 (3.3) 18000 45 PKS 0921–213 0.0531 3.0 7.0 (7.7) 14400 40 1E 0450.3-1817 0.0616 0.5 5.0 (6.1) 18000 25 IRAS 0236.6-3101 0.0623 3.2 5.5 (7.1) 12600 80

aRedshifts taken from Eracleous & Halpern (2004) bThese ﬂuxes were determined from spectra taken with a narrow slit and can thus be uncertain by up to a factor of two. cS/N per pixel in the continuum regions in the vicinity of the Ca ii triplet absorption lines.

Table 2.2. Template Stars

Star Spectral Star Spectral Name Type Name Type

HD 3013 K5 III HD21 K1III HD 79413 K1 III HD 3809 K0 III HD 3909 K4 III HD 2224 G8 III HD 2066 K3 III HD 5722 G7 III HD 225283 K2 III HD 14834 G6 III 14 resolution was 1.35 A˚ FWHM, as measured from the arc lamp spectra, corresponding to a velocity resolution of 50 km s−1. The atmospheric seeing during the observations of ∼ 00 00 the galaxies was less than 1. 5 and frequently less than 1. 0, however the template stars were typically observed during worse conditions, when the seeing was as large as 200. 0. A set of bias frames and HeNeAr comparison lamp spectra were taken at the beginning and end of each night and quartz flats were taken immediately preceding or following each galaxy. Rapidly rotating B stars (with rotational velocities in excess of 200 km s−1) were observed periodically throughout the night at similar airmass as the galaxies and template stars. The spectra of these B stars were used to correct for the deep telluric water vapor lines at wavelengths longer than 8900 A˚ (in the galaxies) and shorter than 8400 A˚ (in the template stars). The exposure times for the galaxies are listed in Table 2.1. The primary data reductions—the bias level correction, flat fielding, sky subtraction, removal of bad columns and cosmic rays, extraction of the spectra, and wavelength calibration—were performed with the Image Reduction and Analysis Facility (IRAF)1. The difference in dispersion from one night to the next was less than 0.1%, so the same dispersion was applied to the spectra from all five nights. The error spectrum was calculated by adding in quadrature the Poisson noise in the spectra of the night sky and the object spectrum (prior to the removal of cosmic rays and bad columns). None of the host galaxies in my sample has reported values of the effective radius, and in all cases except NGC 1097 and Pictor A, suitable images were not available from which to determine the effective radius. To obtain a rough estimate of the effective radius, the spectra were collapsed along the spectral direction to obtain a spatial profile, which was then fitted with the sum of a de Vaucouleurs (R1/4) profile and an AGN point source, convolved with a Gaussian, and background. The effective radius determined for Pictor A from a ground-based image was in agreement with that obtained from the spatial profile. There was considerable uncertainty in the effective radii, however, and I chose to extract the spectra using the smallest effective radius that still yielded an acceptable fit to the spatial profile (see also the discussion in 2.3.2). In the case of NGC § 1097, which is a spiral galaxy with a nuclear star forming ring, the extraction radius was chosen to lie just within the star forming ring (see 2.3.3 for more details). The § extraction radii used for the galaxies are listed in Table 2.1. The removal of the telluric water vapor lines at wavelengths longer than 8900 A˚ was an essential step in the data reductions, because many of the Ca ii lines in the observed galaxies were redshifted into this region of the spectrum. As described in more detail in Appendix A, a telluric template of the atmospheric transmission was derived from the spectrum of a rapidly rotating B-star. The spectra obtained from individual exposures of the galaxies were divided by this telluric template allowing for the possibility of a slight wavelength shift between the template and galaxy spectra. Because the humidity was quite low and stable throughout the run (25-35%) and all of

1IRAF is distributed by the National Optical Astronomy Observatories, which are operated by the Association of Universities for Research in Astronomy, Inc., under cooperative agreement with the National Science Foundation. 15 the objects were observed at low airmass, I found that it was possible to correct all of the galaxies with the same telluric template. Before individual exposures of a galaxy were combined, the spectra were normalized with a low order polynomial and small shifts (determined from measurements of the night sky lines and typically less than 1A)˚ were applied to the spectra to ensure that the absorption lines were not artificially broadened. The G and K giant star spectra were also normalized with a low order polynomial. In most instances, the average galaxy spectrum could be successfully fitted without any further normalization. However, the spectrum of NGC 1097 was re-normalized with a low-order polynomial in the region of the Ca ii triplet. The resulting galaxy spectra and a representative template star are show in Fig. 2.1. The average S/N of the galaxy spectra are listed in Table 2.1 and the S/N of the stellar templates ranged from 150–300. 16 . ror bars . est-fit model for each galaxy is overplotted and in ˚ A) re indicated with black dots along the bottom of the spectrum d with a thick solid line. Data points with unusually large er Rest Wavelength ( 8618 emission lines, when present, are indicated by an arrow λ ] ii spectral region in a template star and the five galaxies. The b ii 8446 and [Fe λ i (due to a strongThe night O sky line, a cosmic ray, or a bad column) a Fig. 2.1 The Ca the intervals used to perform the fit, the model is overplotte 17

2.3 Analysis and Results

2.3.1 Fitting Method The velocity dispersion in the host galaxies of these ﬁve AGNs were determined by directly ﬁtting the galaxy spectra with a model given by:

M(λ)= f T (λ) G(λ)+ P (λ) (2.4) · ⊗ where T (λ) is a stellar template, f (< 1) is a dilution factor, G(λ) is a Gaussian with a dispersion σ, P (λ) is a low-order (< 3nd order) polynomial, and denotes a convolution. ⊗ The stellar template was either an individual G or K giant star or a linear combination of G and K giants (with weights of 25% and 75% respectively; following, Worthey 1994). The fitting intervals were selected to exclude emission lines (O i λ8446 and [Fe ii] λ8618) however it was not necessary to explicitly exclude pixels contaminated by strong night sky emission lines, bad columns, or cosmic rays; the large error bars on such data points resulted in these data being effectively ignored in determining the best fitting model. To find the best fitting velocity dispersion and its uncertainty I scanned the 2- dimensional parameter space defined by σ and f in small steps. The coefficients of the low-order polynomial describing the non-stellar continuum were evaluated at each grid point by minimizing the χ2 statistic. For all of the galaxies, including NGC 1097, 2 2 2 the value of χ per degree of freedom (χν ) for the best fit was typically χν 3. This 2 ∼ large value of χν suggests that the error bars on the flux density of the pixels in the spectra were underestimated and in fact the root mean square (RMS) dispersion of the data in featureless regions of the spectra was typically 1.5–1.9 times larger than the formal error bars assigned to the individual pixels. I attribute this increased scatter to imperfect subtraction of the strong night-sky emission lines appearing throughout the spectral range (this was caused by curvature of the sky lines along the direction of the slit). I note that the discrepancy between the RMS deviation and the formal error bars is only slightly greater at wavelengths larger than 8900 A,˚ where the telluric absorption correction was performed. If I increased the error bars on the spectral pixels by the above factor, the best fit models would have had χ2 1 in most cases. Thus the values ν ∼ of χ2 for each fit were rescaled such that χ2 1 and the 68%-confidence error contour ν ≡ in σ and f was defined by ∆χ2 = 2.3. This is equivalent to the practice of rescaling the error bars on each pixel, adopted by Barth, Ho, & Sargent (2002). The galaxy spectra and their best fitting models are shown in Fig. 2.1. This fitting method is not identical to that presented in Barth et al. (2002), however I have fitted the spectrum of Arp 102B obtained by those authors and the velocity dispersion we obtain is consistent with theirs, within the 68%-confidence error bars.

2.3.2 Sources of Systematic Error There are some potential sources of systematic error which were not accounted for by the error analysis described above. Extraction Size. – The eﬀective radii of the galaxies determined here are quite uncertain, however I found that best-ﬁt velocity dispersion was not sensitive to the extraction 18

radius. I also note that the S/N of the spectra was not significantly affected by the size of the extraction radius, with the exception of Pictor A which is discussed further in 2.3.3. § Telluric absorption correction. – Although considerable care was taken to perform the telluric absorption correction, the telluric template is most likely overestimated in the interval from 9195 A–9215˚ A,˚ as described in Appendix A. Consequently, the telluric correction in this interval is not complete and the flux in the corrected galaxy spectra could be underestimated by as much as 4%. This artifact cannot be accounted for in any statistical way; however, when it is obvious in a spectrum, this interval is not included in the fit.

Template Mismatch. – A primary advantage of using the Ca ii triplet to measure the stellar velocity dispersion is the fact that the strength of the absorption lines are relatively insensitive to the stellar population (Pritchet 1978; Dressler 1984). I found that the RMS scatter in the values of σ obtained using the different templates (both individual stars and linear combinations of stars) was only a few kms−1. However, I note that the dilution factor (f) varied with stellar type, due to the variation in line depth with stellar type. In the case of NGC 1097, this RMS scatter was added in quadrature with the 68% error on the best-fit. However, for the other galaxies the uncertainty due to template mismatch was negligible compared to the uncertainty on the best-fit and was not included.

2.3.3 Notes on Individual Objects NGC 1097. – The nuclear structure of the barred spiral galaxy NGC 1097 is extremely complicated, as evidenced by 12CO observations which suggest the presence of a cold nuclear disk being fed by matter streaming along the bar (Emsellem et al. 2001). These authors find that within the inner 500, the velocity dispersion ranges from 145 km s−1 in the center to 220 km s−1 at the inner edge of the bar, as measured from the broadening of the CO band head. There is a star forming ring extending from 500–1000, as mapped by Hα and radio emission (Hummel, van der Hulst, & Keel 1987), which is mostly excluded in these data. However Storchi- Bergmann et al. (2005) recently demonstrated that there is a starburst within 000. 1 (9 pc) of the central black hole. Despite these difficulties, I find an excellent fit to the data is obtained using a single K giant star, although the featureless continuum is more complex and must be described with a 3rd order polynomial. The interval from 8600-8640 A˚ (rest wavelength) was excluded from the fit because contamination from [Fe II] emission might be responsible for the poor fit to the data in this interval. If this interval is included in the fit, the best-fit velocity dispersion increases to 208 5 kms−1. ± Pictor A. – The Ca ii triplet is located in the region of the spectrum with the strongest night sky emission lines. Although the effective radius was estimated to be 900 (6.5 kpc), the use of a smaller extraction radius (400. 5) allowed for a much improved subtraction of the night sky emission, and consequently an increased S/N and 19

decrease in the error bars. When a 900 extraction radius was used, the best-ﬁt velocity dispersion was the same, although the error bars were larger. Both O i and [Fe ii] emission lines are present in the spectrum, and the regions around these lines were ignored in the ﬁt.

PKS 0921–213. – The Ca ii λ 8662 line is strongly contaminated by several bad columns and as a result the blue side of this line does not contribute to the determination of the velocity dispersion.

1E 0450.3–1817. – This object is extremely faint and the S/N is much lower than for the other objects (S/N 15 per pixel), therefore the spectrum was smoothed by ∼ a 3-pixel wide boxcar function in order to increase the S/N to 25 per pixel. The ∼ best-fit velocity dispersion is the same as when the un-smoothed spectrum is fitted, but the error bars are decreased by 30%. I have verified that for NGC 1097 and ∼ IRAS 0236.6–3101, neither the best-fit velocity dispersion nor the uncertainty are changed when the spectra are similarly smoothed.

IRAS 0236.6–3101. – The Ca ii λ8662 line has an observed wavelength of 9202 A,˚ which places this line in the portion of the telluric template that was underestimated, as described in 2.3.2 and Appendix A. The λ8662 A˚ line is deeper than would § be expected from the two other Ca ii lines, and I chose to ignore this line when performing the fit. If this line were included, the best-fit value of σ would increase to 200 km s−1, however a model with this large velocity dispersion is an extremely ∼ poor fit to the λ8542 A˚ line. An O i emission line is present, and the interval around this line was excluded from the fit. Like other IRAS galaxies of similar luminosity, IRAS 0236.6–3101 is probably undergoing active star formation or is in a post-starburst phase (see, e.g., Lipari & Macchetto 1992; Goto 2005). Thus it is not surprising that, as for NGC 1097, it was necessary to use a 3rd order polynomial to describe the featureless continuum.

2.4 Discussion and Conclusions

Using the measurements of the velocity dispersion found above, I determine the black hole masses, the Eddington ratios, and the physical timescales in the accretion disk. All of these quantities are given in Table 2.3. The black hole masses are estimated via the MBH –σ relationship (Tremaine et al. 2002), namely, M σ log = α + β log , (2.5) M σ 0 −1 where α = 8.13 0.06, β = 4.02 0.32, and σ0 = 200 kms . The scatter in the ± ± MBH –σ relationship is not included in the error, but the error bars on the coeﬃcients (α and β) are. For completeness, I include in this table the double-peaked emitters Arp 102B, NGC 4203, and NGC 4579, for which the stellar velocity dispersion has been measured by Barth et al. (2002). The inferred black hole masses for these eight objects 7 8 range from 4 10 M – 1.2 10 M . I also include M81, for which a mass estimate of × × 20

7 M 6–7 10 M is based on resolved stellar and gas kinematics (Bower et al. 2000; BH ∼ × Devereux et al. 2003). I note that the estimate of the black hole mass for NGC 4203 7 used here (MBH 6 10 M ) is in disagreement with the results of Shields et al. ∼ × 6 (2000); Schoenmakers et al. (2001), which placed an upper limit of 6 10 M on the × black hole mass from spatially resolved gas kinematics, however I have chosen to use the mass obtained from the stellar velocity dispersion for two reasons. First of all, as these authors noted, gas dynamical measurements are subject to non-gravitational forces. In this object in particular, it is suspected that there is a warp in the disk (Schoenmakers et al. 2001). Interestingly, if the orientation of the accretion disk is fixed to be the 7 same as that of the dust lanes, the obtained black hole mass was 5.2 10 M , however × this constraint yielded a fit which was significantly worse than when the orientation was allowed to be a free parameter. The physical timescales in the outer accretion disk are computed using Eqn. 2.1– 2.3, assuming that ξ3 = 1, α = 0.1 and T5 = 1. For these black hole masses, the dynamical timescale ranges from a few months to one year, the thermal timescale ranges from one year to a decade, and the sound crossing timescale ranges from a decade to 100 years. The observed profile variations occur on timescales of several years (see Chapter 4) for most objects, which is an order of magnitude longer than the dynamical timescale found here. This strongly suggests that in general the profile variability might be due to either a thermal phenomenon or the precession of a large scale emissivity pattern (e.g. a spiral arm). One exception is Arp 102B, in which the relative fluxes of the red and blue peaks of the profile varied sinusoidally over an interval of four years. Newman et al. (1997) modeled this variability as a bright spot in the accretion disk that was orbiting with a period of 2.16 years. These authors found that the profile variability was satisfactorily fitted when the bright spot was located at a radius between 355–485 rg from the black hole. Using the black hole mass estimated here, the corresponding Keplerian period ranges from 2.1–3.5 years which is consistent with the observed periodicity. The Eddington ratios are computed by comparing the total bolometric luminos- 38 −1 ity to the Eddington luminosity (L = 1.38 10 M/M erg s ). In the case of Edd × NGC 1097 and Arp 102B, the Spectral Energy Distributions (SEDs) are well sampled (Nemmen, Storchi-Bergmann, Eracleous, Terashima, & Wilson 2004; Nemmen, Storchi- Bergmann, Yuan, Eracleous, Terashima, & Wilson 2005; Eracleous, Halpern, & Charlton 2003) and the bolometric luminosity was obtained by integrating the SED directly. For most of the other objects, the bolometric luminosity was obtained by scaling the SED of Arp 102B by the 2–10 keV luminosity. Both PKS 0921–213 and Pictor A are much more luminous than the other objects and the mean radio-loud quasar SED of Elvis et al. (1994) was used instead. For the SED of Arp 102B, I find that L2−10 keV/Lbol = 0.016 whereas for the radio-loud quasar SED, L2−10 keV/Lbol = 0.07. I note that the 2–10 keV flux of an AGN is variable by a factor of a few, so these bolometric luminosities, and thus the deduced Eddington ratios, should be regarded as approximate. The Eddington ratios vary widely, but there is a general positive trend between bolometric luminosity and Eddington ratio. The three low-luminosity LINERs 42 −1 −4 (NGC 1097, M81, NGC 4203, Lbol . 10 erg s ) have Lbol/LEdd . 10 . The moder- ate luminosity objects (Arp 102B, NGC 4579 IRAS 0236.6-3110, and 1E 0450.3–1817, 43 −1 −3 Lbol 10 erg s ) have Lbol/L 10 . The most luminous objects (Pictor A, ∼ Edd ∼ 21 s s τ 2004, 2005; th τ 2.4. § dyn 3–4 2.5–4 35–50 2–4 2–3 35–50 5–8 4–7 60–90 6–8 5–7 70–100 τ 3–4.5 2.5–4 35–50 3–4.5 2.5–4 35–50 0.5–5 0.5–4 10–55 (months) (years) (years) 2 2 d 4 4 3 3 5 − − − − − − − 10 10 10 10 10 10 10 Edd × × × × × × × /L bol L al. (2003); derived from spatially-resolved stellar . For more details, see = 0.07 (radio-loud quasar SED), whereas for all 9); (4) Lewis et al. (2005); (5) Stocke et al. (1983); Bol c 1994) L 9 (3–4) 6 (0.5–1) 8 (2–3) 7 (1–2) 5 (0.4–2) 4 0.08–0.2 1–4 1–3 15–40 1 (3–4) 10 (2–3) 2,3 0.07–0.2 1–4 1–3 15–40 10 keV cleous et al. (2002); (9) Iyomoto et al. (1998); (10) X-ray − X-ray Range in ain 2 /L ) Ref. 40 40 41 41 41 41 43 43 40 1 Bol 10 10 10 10 10 10 10 10 10 − L 10 keV – (1) Nemmen et al. (2004); Nemmen et al. (2005); (2) Eracleou × × × × × × × × × − 1 0 9 1 5 1 2 0 4 2 ...... L 3 8 ) (erg s b

2 4 2 4 3 1 1 4 1 2 2 8 2 4 2 4 M ± ± +4 − BH ± ± ± ± ± ± 7 4 6 6 4 11 = 0.016 (Arp 102B SED). For NGC 1097 and Arp 102B (Nemmen et al. Table 2.3. Velocity Dispersions and Derived Properties a a a ) (10 1 5 12 25 16 3 4 8 15 5 20 4 10 keV − +30 − +18 − ± ± ± ± − ± ± σ M 2 /L Bol L Name (km s Galaxy For Pictor A and PKS 0921–213, I assumed Black hole mass taken from Bower et al. (2000) and Devereux et Velocity dispersions taken from Barth et al. (2002). References for X-ray luminosity and spectra. d a b c M81 ...... 6 NGC 4203 167 IRAS 0236.6–3101 154 NGC 4579 165 1E 0450.3–1817 150 PKS 0921–213 144 Pictor A 145 NGC 1097 196 Arp 102B 188 others I assumed Eracleous et al. 2003, respectively) were integrated to obt & Halpern (1998); (3)(6) Sambruna, Boller Eracleous, et & al. Mushotzkyﬂux (1992); (199 from (7) Turner Eracleous et et al. al. (2002), (2003); distance (8) from Era Freedman et al. ( and gas kinematics, respectively 22

44 −1 PKS 0921–213, Lbol 10 erg s ) have Lbol/L 0.1. Thus it is clear that while ∼ Edd ∼ many double-peaked emitters may harbor radiatively inefficient accretion flows, the accretion flow in some, such as Pictor A and PKS 0921–213, may be more similar to those found in Seyfert 1s. A vertically extended structure may still be required in these objects to illuminate the outer accretion disk and may take the form of a spherical corona such as that used by Dumont & Collin-Souffrin (1990b), a beamed corona (Beloborodov 1999; Malzac, Beloborodov, & Poutanen 2001), and/or the base of the jet (Markoff, Falcke, & Fender 2001). That the double-peaked emitters in even this small sample have a wide range in Eddington ratios is not completely surprising. The double-peaked emitters found in the SDSS by Strateva et al. (2003) are very heterogeneous; although 12% were classified as LINERs, many others had properties which were indistinguishable from the general z < 0.332 AGN population. I note that my results confirm the general findings of Wu & Liu (2004), who estimated black hole masses for both the Eracleous & Halpern (1994) and Strateva et al. (2003) samples through a series of correlations obtained from reverberation mapping of AGNs (Kaspi et al. 2000). These authors found that (1) double-peaked emitters are a heterogeneous group with a wide range in Eddington ratios and (2) those objects with higher bolometric luminosities tend to have larger Eddington ratios. How- ever, the black hole masses of individual double-peaked emitters deduced through that method can be very inaccurate, as acknowledged by these authors. Although the black hole masses for 1E 0450.3–1817 and Arp 102B inferred by these authors are consistent with those obtained through the MBH –σ relationship, the black hole masses for Pictor A and IRAS 0236.6–3101 were overestimated by an order of magnitude. Thus we caution that for the purposes of modeling and interpreting individual objects, it is necessary to obtain black hole masses through a method which is more well tested and calibrated in the range of BH masses relevant to these objects, such as the MBH –σ relationship. 23

Chapter 3

A Simultaneous RXTE and XMM-Newton Observation of the Broad-Line Radio Galaxy 3C 111

3.1 Introduction

In the early 1990s, observations with the Ginga satellite revealed that the spectra of many Seyfert 1 galaxies contain an Fe Kα emission line, with an equivalent width (EW ) of 100–300 eV, as well as a hard excess above 10 keV, relative to the simple power law spectrum fitted over the interval from 2–8 keV (Pounds, Nandra, Stewart, George, & Fabian 1990; Piro, Yamauchi, & Matsuoka 1990; Matsuoka, Piro, Yamauchi, & Murakami 1990; Nandra & Pounds 1994). These features were readily interpreted as signatures of the reprocessing of the primary X-ray continuum emission by nearby Compton-thick material, such as the accretion disk or the obscuring torus (see, e.g. Lightman & White 1988; George & Fabian 1991; Matt, Perola, & Piro 1991; Matt, Perola, Piro, & Stella 1992). These features are important diagnostics of the geometry, dynamics, and physical conditions of the reprocessing medium. In some Seyfert 1s, most notably MCG –6-30-15 (Tanaka et al. 1995; Iwasawa et al. 1996), ASCA observations showed that the Fe Kα line profile had a narrow core at 6.4 keV and a very broad red wing; Fabian et al. (1995) found that the line profile was best modeled as emission from the inner regions of the accretion disk. In a sample of 18 Seyfert 1s observed with ASCA, Nandra et al. (1997) found that many had a broad Fe Kα emission line with an average Gaussian energy dispersion of σ = 0.43 0.12 h i ± keV. The averaged profile of the Fe Kα emission lines, as well as some individual profiles, were not symmetric however, indicating that multiple line components (i.e. broad and narrow) were present. In some objects the line profile had a broad red wing and, like MCG –6-30-15, were well modeled as emission from an accretion disk. However, recent observations with XMM-Newton and have shown that the picture is much more complex. Broad lines, typically with equivalent widths of 100 eV, were ∼ detected in several Seyfert 1s, for example: MCG –6-30-15 (Wilms et al. 2001); MCG – 5-23-16 (Dewangan, Griffiths, & Schurch 2003); NGC 3516 (Turner et al. 2002); Mrk 766 (Pounds et al. 2003); and Ark 120 (Vaughan et al. 2004). On the other hand, XMM has shown narrow Fe Kα lines (typically unresolved by EPIC) to be a ubiquitous feature of Seyfert 1 spectra and in fact some Seyfert 1s showed only a narrow, neutral Fe Kα line, with EW 75 eV (e.g., Reeves 2003). It must be noted, however, that the upper limits ∼ for the equivalent width of a broad line were sometimes quite generous (EW 100 eV), ∼ so a broad component, although not required, was not always ruled out by the data. The hard excess from 10–18 keV observed by Ginga is only the low-energy tail of the Compton reflection bump, which is a high energy continuum component that peaks at 30 keV and continues up to energies of 100 keV. The strength of the Compton ∼ 24 reflection bump is parameterized by Ω/2π, where Ω is interpreted as the solid angle subtended by the reprocessing material to the primary X-ray source. (In the case of a standard accretion disk, Ω/2π = 1.) Nandra & Pounds (1994) measured Ω/2π = 0.5–0.7 using Ginga data, but constraining the properties of the Compton bump well requires very broad spectral coverage. Gondek et al. (1996) combined data from Exosat, Ginga, HEAO-1 and GRO/OSSE to obtain an average 1–500 keV spectrum of 7 Seyfert 1s and found that Ω/2π = 0.76 0.15. More recent observations with BeppoSAX (Perola et al. ± 2002; Bianchi et al. 2004) and the Rossi X-ray Timing Explorer (Lee, Fabian, Brandt, Reynolds, & Iwasawa 1999; Weaver, Krolik, & Pier 1998) also indicated that Ω/2π 0.7, ∼ on average. Observations with numerous X-ray satellites have shown that these reprocessing features are stronger in Seyfert 1s than in their radio-loud counterparts, the Broad-Line Radio Galaxies (BLRGs). The Compton reflection bump is typically much weaker, with Ω/2π . 0.5 (see, e.g., Zdziarski, Johnson, Done, Smith, & McNaron-Brown 1995; Woz- niak, Zdziarski, Smith, Madejski, & Johnson 1998; Eracleous, Sambruna, & Mushotzky 2000; Zdziarski & Grandi 2001; Grandi, Maraschi, Urry, & Matt 2001; Grandi 2001). The Fe Kα line is also weaker, with EW . 100 eV (see, e.g., Eracleous, Halpern, & Livio 1996; Wozniak, Zdziarski, Smith, Madejski, & Johnson 1998; Eracleous & Halpern 1998; Eracleous, Sambruna, & Mushotzky 2000; Grandi, Maraschi, Urry, & Matt 2001; Grandi 2001; Zdziarski & Grandi 2001). Ballantyne, Fabian, & Iwasawa (2004) analyzed simultaneous XMM and RXTE observations of the BLRG 3C 120 and found that Ω/2π 0.5 and the Fe Kα line had ∼ an EW 50 eV. The line, when fitted with a Gaussian, had a width σ 0.1 keV, ∼ ∼ which is much narrower than those found in Seyfert 1s (Nandra et al. 1997). However, a contribution from a broad, distorted emission line from the inner accretion disk could not be ruled out completely. Similar results were obtained by Ogle et al. (2005), using the XMM data only. However, more recently an XMM Newton spectrum of the BLRG − 4C 74.26 revealed a broad emission line with a EW 150 eV and Ω/2π 1 (Ballantyne ∼ ∼ & Fabian 2005). There are several viable scenarios which could explain the weakness of the reprocessing features in BLRGs:

1. The inner accretion disk in BLRGs might have the form of an ion torus (Rees et al. 1982), or other similar radiatively inefficient accretion flow. As a result, the primary X-ray continuum can only be reprocessed by either the outer accretion disk or the obscuring torus, leading to Ω/2π < 0.5 (see, e.g. Eracleous et al. 2000). In this case, the Fe Kα emission line should be narrow (FWHM . 15, 000 km s−1) and produced by Fe atoms which are not highly ionized (E 6.4 keV). ∼ 2. Ballantyne, Ross, & Fabian (2002) suggest that the reprocessing features are weaker in BLRGs because the accretion disk is highly ionized, rather than because the geometry of the accretion disk is changed. Reprocessing of the X-ray emission by ionized media has been studied extensively (see, e.g., Ross & Fabian 1993; Zycki et al. 1994; Nayakshin & Kallman 2001). These authors find that reprocessing by a moderately ionized accretion disk results in numerous low-energy emission and absorption features, due to ionized species of O, C, and N. However, as the 25

ionization increases further, the disk becomes a nearly perfect reﬂector, making the reprocessing features very weak. In this case, the Fe Kα line should be emitted in the inner accretion disk and it should be broad, but with E & 6.7 keV.

3. The weak reprocessing features could be the result of a mildly relativistic outflow, as suggested by Wozniak et al. (1998). This hypothesis is supported by detailed modeling of the effects of bulk motion on the Fe Kα emission line (Reynolds & Fabian 1997) and the Compton reflection bump (Beloborodov 1999).

In this chapter, I present the results of a simultaneous observation of the BLRG 3C 111 with XMM and RXTE with the aim of testing the first two of these hypotheses, which make very clear predictions for the properties of the Fe Kα emission line. The general properties of 3C 111, including the results of previous X-ray studies, and the specific goals of this analysis, with respect to 3C 111, are given in 3.2. In 3.3, I describe § § the data reductions. The results of the timing and spectral analysis are presented in 3.4 § and 3.5 respectively. I discuss the implications of these results in 3.6 and summarize § § my findings in 3.7. § 3.2 Properties of 3C 111

3C 111 is a nearby (z = 0.0485, d = 210 Mpc) BLRG. The host galaxy is marginally resolved in the R-band with the Hubble Space Telescope and although the morphology of the host is somewhat uncertain, it is likely to be a small elliptical-type galaxy (Martel et al. 1999). The radio source has an FR II radio morphology (Fanaroff & Riley 1974) with a single-sided jet (Linfield & Perley 1984). The jet exhibits superluminal motion (Vermeulen & Cohen 1994), which along with the apparent size of the radio lobes (Nilsson et al. 1993), allows one to place constraints upon the jet inclination. As described fully in Appendix B, the jet is inclined at an angle of 21◦–26◦. If 3C 111 happens to be a rare Giant Radio Galaxy, however, the inclination angle could be as small as 10◦. A giant molecular cloud lies along the line of sight to 3C 111 and some care must be taken to estimate the total Galactic hydrogen column density, as described fully in Appendix C. Not only is there a significant contribution from molecular hydrogen, but the molecular hydrogen column density is expected to vary with time due to the presence of AU-scale structures in the molecular cloud. Using the H i map of Elvis, Lockman, & Wilkes (1989) and detailed H2CO studies of the foreground molecular cloud by Marscher, Moore, & Bania (1993) and Moore & Marscher (1995), the total Galactic column density towards 3C 111 is estimated to be 1.2 1022 cm−2. This value, is expected to vary by 21 −2 × several 10 cm , however. Numerous X-ray satellites have been used to observe 3C 111, most importantly Ginga (Nandra & Pounds 1994; Wozniak et al. 1998), ASCA (Wozniak, Zdziarski, Smith, Madejski, & Johnson 1998; Reynolds, Iwasawa, Crawford, & Fabian 1998; Sambruna, Eracleous, & Mushotzky 1999), and RXTE (Eracleous et al. 2000), which have the high energy coverage necessary to study the reprocessing features. There are many remaining 26 unanswered questions regarding the X-ray spectral properties of 3C 111, that under- score the difficulties encountered when analyzing and interpreting the weak reprocessing features in many BLRGs. First, it is uncertain whether a reflection component is even necessary to fit the spectrum of 3C 111. Using RXTE and Ginga data respectively, Eracleous et al. (2000) and Wozniak et al. (1998) found that Ω/2π is consistent with 0 at the 90% confidence level. It is important to place tight constraints upon the continuum emission since it has important implications for the geometry of the reprocessing medium. Equally important is the need to robustly fit the continuum to ensure that the residual Fe Kα emission line can be properly fitted. For example, when X-ray spectra of 3C 120 were fitted with absorbed power law models (Reynolds 1997; Grandi et al. 1997; Sambruna et al. 1999), the Fe Kα line was found to be very strong and broad (EW 0.5 – 1 keV and ∼ σ 1 keV). However, when Compton reflection or a broken power law models were used ∼ (Wozniak et al. 1998; Eracleous et al. 2000; Zdziarski & Grandi 2001), the equivalent width and line width were significantly reduced (EW 100 eV and σ 0.3 keV.) ∼ ∼ Secondly, while the equivalent width of the Fe Kα emission line in 3C 111 was found to be weak (EW 60 +20 eV) by both Eracleous et al. (2000) and Wozniak et al. ∼ −10 (1998), the energy (E) and width of the line were uncertain. Eracleous et al. (2000) fitted the line with a Gaussian with a fixed energy of 6.4 keV and constrained the FWHM of the line to be less than 44,000 km s−1, whereas Wozniak et al. (1998) fitted the line with +0.4 a narrow Gaussian (σ = 0.1 keV) and found that E = 6.7 −0.3 keV. Furthermore, if the Fe Kα line originates in the accretion disk, a Gaussian model is a poor approximation to the true disk line profile and yields misleading estimates of the line energy and width. The line was marginally detected in an ASCA observation (Reynolds et al. 1998) and the line properties were not well constrained. Thus, neither the origin of the line (i.e. inner disk vs. outer disk, or an even more distant reprocessor, such as the torus) nor the ionization state of the reprocessor are known. In order to evaluate the competing scenarios to explain the weakness of the reprocessing features of BLRGs presented in 3.1, these parameters must be better constrained. § A simultaneous XMM-Newton and RXTE observation of 3C 111 can help address these and other issues. It is important to obtain simultaneous observations, since the X-ray flux and spectral parameters of AGNs in general, and BLRGs in particular, are known to vary on timescales of several days (e.g. Gliozzi, Sambruna, & Eracleous 2003). The high energy sensitivity of RXTE is critical for accurately fitting the continuum and detecting the Compton reflection bump, which peaks at 30 keV. On the other hand, the good spectral resolution and large collecting area of XMM-Newton in the 0.4-10 keV range make it ideal for use in a detailed study of the Fe Kα emission line properties. Additionally, the spectrum at low energies will be useful for constraining the ionization of the disk because reprocessing by a moderately ionized disk leads to numerous emission and absorption features at low energies (Ross & Fabian 1993; Zycki et al. 1994; Nayakshin & Kallman 2001) which should be detectable in the XMM-Newton spectrum, despite the large absorbing column. 27

3.3 Observations and Data Reductions

3.3.1 XMM-Newton 3C 111 was observed on 14 March 2001 with the European Photon Imaging Cam- era (EPIC) and the Reflection Grating Spectrometer (RGS) on-board the XMM-Netwon satellite for a duration 40 ks. The p-n data were obtained in Large Window mode and ∼ the MOS data were taken in the Partial Window mode, using the thin filter. The exact exposure times and count rates for each instrument are given in Table 3.1. The data were processed using the XMM-Netwon Science Analysis Software (SAS v5.4.1) using the calibration files released on 29 January 2003. The EPIC data sets (p-n, MOS 1 and MOS 2) were filtered to remove all flagged events (e.g. events suspected to be cosmic rays, bad pixels, etc.). There were intense particle flares during most of the second half of the observation in which the count rate increased by a factor of 20–100. The flares were successfully removed using the Good Time Interval tables provided, which also removed several smaller flares that were present. As a result, the exposure times were reduced by 40% (see Table 3.1). 3C 111 is quite bright and normally the effect of the flares might ∼ have been adequately corrected for with background subtraction. However, I noticed that the source count rate actually decreased dramatically during the flaring intervals, leading me to believe that the flares must have been intense enough to saturate the telemetry, despite the fact that the count rates were well below the expected threshold for telemetry saturation (the saturation thresholds are 1150 and 300 counts s−1 for the p-n and MOS detectors, respectively.) Thus, unfortunately, the data collected during the flaring intervals could not be used. Finally, the EPIC p-n data were filtered to include only single and double pixel events (i.e. PATTERN 4) whereas the MOS data were ≤ filtered to also include triple pixel events (i.e. PATTERN 12). ≤ 00 The source counts were extracted from a circle with a radius of 44 for the p-n data, for which the fractional encircled energy is greater than 90%. The background was extracted from an annulus centered on the source with an outer radius of 11000 and inner radius of 4500; we experimented with several background regions and found little difference in the results. Since the MOS data were obtained in Small Window mode, the largest circular region which could be used to extract the source spectrum had a radius of 3900, which encompasses 88% of the encircled energy. The background could not be extracted from the same chip, because it was not read out. Instead a circle of radius 12500 was extracted from a neighboring chip. As with the p-n data, several measurements of the background were found to be similar. The response matrices (RMFs) and ancillary response functions (ARFs) were generated with the calibration files released on 29 Jan 2003. When constructing the ARFs, the source was treated as a point source, but the encircled energy was modeled as a function of photon energy. The RGS data were reduced using the SAS routine rgsproc, which automatically filtered the data, traced the 1st and 2nd order dispersed image of the source, and selected regions to exclude in the determination of the background spectrum. The RGS data suffered from the same flares as the EPIC data, but I found that it was unnecessary to remove the flares; the background subtracted spectra with and without the flares removed were similar. When the spectra were extracted, a separate background file was created for each order. 28

Table 3.1. Observation Log

Instrument Exposure Time (ks) Exposure Time (ks) Count Rate (counts s−1)a post-ﬁltering post-ﬁltering

XMM - Observation Date: 2001 Mar 14 13:03 – 2001 Mar 15 01:16 pn 42.3 23.4 10.89 MOS 1 44.0 27.8 3.22 MOS 2 44.0 28.4 3.22 RGS 1 44.6 43.5 0.08, 0.05 b RGS 2 44.6 42.5 0.10, 0.04 b RXTE - Observation Date: 2001 Mar 14 08:49 – 2001 Mar 17 03:36 PCA 56.6 6.57 HEXTE 0 18.2 1.14 HEXTE 1 17.8 0.86

aThe count rate was measured for the same interval used for ﬁtting. (For the pn data, 0.5–10 keV was used.) bThe 1st and the 2nd order count rates, respectively. 29

In the preliminary spectral analysis, it became clear that the p-n data suffered from X-ray loading, which occurred because the frames used to calculate the offset map (i.e. the zero-energy level) had a count rate which was too large. The offset map, and thus the gain, were therefore incorrect on a pixel-by-pixel basis. Additionally, because the zero-energy level was too high, double-pixel events were interpreted as single-pixel events of a higher energy. In many cases X-ray loading is an extreme effect of photon pile-up. However, I note that based on the count rate of the filtered p-n data, the chance of photon pile-up is < 1% both from the estimates from the XMM-Newton handbook and my own estimates based on Poisson statistics. The X-ray loading in this instance likely was the result of a flare in one or more of the frames used to calculate the offset map. At this time, there is no method to reliably correct for X-ray loading and I was forced to exclude the p-n data from the spectral analysis. With the total loss of the p-n data and the 40% loss of MOS data, due to flaring, the total count rate was reduced by 70% from that anticipated. The X-ray loading manifested itself in several ways, but was difficult to diagnose. There were two revealing symptoms of X-ray loading, which distinguished the phenomenon from pile-up. First, when the p-n and MOS spectra were fitted with simple absorbed power-law models, there was an absorption feature between 1.8 – 2.1 keV in the p-n data residuals which was absent in the MOS data. I initially suspected a calibration error in the effective area near the Si edge, however this feature was absent in other p-n data with similarly high signal-to-noise ratio (S/N). Secondly, the results of the SAS epatplot routine indicated that the fraction of single pixel events was slightly higher than expected from theory whereas the fraction of double pixel events was slightly lower, the opposite of what was expected for photon pile-up. However, this effect was not dramatic and might have been easily overlooked had there not been suspicious residuals in the p-n data near 2 keV.

3.3.2 Rossi X-ray Timing Explorer 3C 111 was observed with the Proportional Counter Array (PCA) and the High Energy X-ray Timing Experiment (HEXTE) onboard the Rossi X-ray Timing Explorer (RXTE) satellite from 14–17 March 2001. The exposure time for each instrument is given in Table 3.1. The reduction procedure is described in detail in Gliozzi et al. (2003). Briefly, the PCA and HEXTE data were screened to exclude events taken when the Earth elevation angle was 10◦ and the pointing offset from the optical position ◦ ≤ was 0.02 . The PCA data were also filtered to include only events obtained when the ≥ satellite was out of the South Atlantic Anomaly for more than 30 minutes and also those events whose ELECTON-0 parameter was 0.1. The PCA background and light curve ≤ were determined with the L7-240 background developed at the RXTE Guest Observer Facility (GOF), using the FTOOLS task pcabackset, v2.1b. The appropriate response matrices and effective area curves for the observation epoch were produced using the FTOOLS v.5.1 software package and with the help of the REX script provided by the RXTE GOF. Only PCUs 0 and 2 were combined, since PCUs 1,3, and 4 were not always turned on. The background applicable to the HEXTE clusters was obtained during the observation by dithering the instrument slowly on and off the source. 30

3.4 Timing Analysis

To perform the timing analysis of the XMM data, the MOS 1 and MOS 2 data are combined and the p-n data are used, as a slight offset in the gain should not affect the timing results. The data from 0.2–10 keV are binned in 2000 s intervals. The mean count rate of the p-n data is 8.89 s−1 and that of MOS 1+2 is 6.98 s−1. The count rate is moderately variable, with an amplitude of 2.7%, where the amplitude is defined as the difference between the maximum and minimum count rate, divided by the mean count rate. However, the variability is significant and the probability that the count rate is constant is only 2.1%. The data are consistent with a monotonic increase in flux during the course of the observation, but the hardness ratio (defined as the ratio of the flux in the 2–10 keV band to that in the 0.2–2 keV band) is not significantly variable, and the probability that it is constant is 43%. The 2–20 keV data from the PCA are also binned in 2000 s intervals and the mean count rate is 17.29 s−1 for 2 PCUs. The variability is consistent with the XMM data in the overlapping interval with an amplitude of 2.5%. As can be seen in Fig. 3.1, over the entire observation period, the lightcurve is more variable and the probability that the count rate is constant is < 10−38. The largest excursion has an amplitude of 11% and takes place over a 29 ks time interval. As with the XMM data, the hardness ratio (defined as the ratio of the flux in the 10–20 keV band to that in the 2–10 keV band) is not highly variable and the probability that the hardness ratio is constant is 82%.

3.5 Spectral Analysis

In total I have nine separate data sets, obtained simultaneously, covering the spectral range from 0.4–100 keV: RGS 1 and 2 (0.4–1.65 keV in first order and 0.65– 1.65 keV in 2nd order); MOS 1 and 2 (0.4–10.0 keV); PCA (4.0–30.0 keV); and HEXTE clusters 0 and 1 (20.0–100.0 keV). The RGS data are binned such that each bin has a minimum of 25 counts. The MOS data have an excellent S/N, therefore they are binned such that there are 2–3 bins per resolution element, and at least 25 counts per bin. In particular, the region around the expected Fe Kα line is well sampled. At low energies (E < 0.7 keV), the MOS resolution element was slightly undersampled, because the count rate is not large. However, this region overlaps with the high resolution RGS data. The RXTE data were binned such that there were at least 20 counts per spectral bin to ensure that the χ2 test was valid. The spectral analysis is carried out with the XSPEC vs 11.3 software package (Arnaud 1996). The data were obtained simultaneously, so the fit parameters for the nine data sets are forced to be the same in all models, with the exception of the overall normalization constants which are allowed to vary freely to account for cross calibration uncertainties. In general, there is only a 1% difference between the normalization constants for the two MOS data sets, but the PCA normalization constant is 30% higher than the MOS value. Since the RXTE observations span a longer time period than the XMM observations, it is possible that spectral variability will lead to a systematic difference in model parameters between the XMM and RXTE data, a possibility which I explore below. 31

Fig. 3.1 The observed RXTE PCA lightcurve in the 2-20 keV energy band. The dashed vertical lines indicate the interval during which the XMM data were taken while the dashed horizontal line indicates the average count rate over the entire observation. 32

All errors and upper limits listed in the Tables and the text correspond to the 90% confidence interval for one degree of freedom (d.o.f., i.e. ∆χ2 = 2.7), unless otherwise stated. In comparing different models, I refer to the chance probability Pc, by which I mean the probability that the improvement in the fit statistic would occur by chance, as determined by the F-test. In 3.5.1, I describe fits to the continuum using various § models, excluding the energy interval where the Fe Kα line is expected. The residual Fe Kα line from several continuum fits is modeled in 3.5.2. Then in 3.5.3, I attempt § § to fit the combined continuum and Fe Kα emission simultaneously and self-consistently.

3.5.1 Continuum Models I fit the continuum using several different models, excluding the interval from 4.5–7.5 keV, where an Fe Kα line is expected to be located. The fit parameters for all models are listed in Table 3.2 and discussed below. In all models, I include Galactic photoelectic absorption using the cross-sections of Morrison & McCammon (1983), allowing the column density to be a free parameter because the total Galactic hydrogen column density along the line of sight is uncertain. Throughout, I use the solar abundance pattern of Anders & Ebihara (1982) and in all instances the fitted column density is 8 1021 cm−2, which is consistent with the Galactic hydrogen column density towards ∼ × 1 3C 111. Therefore an additional absorber at the redshift of 3C 111 is not warranted, although absorption within the host galaxy or the AGN itself cannot be ruled out.

Power law (Model #1): – The data are first fitted with an absorbed power law model, whose free parameters are the column density (NH ), the photon index (Γ), and the overall normalization constant. The spectral model and residuals are shown in Fig. 3.2 and the 90% confidence contour in NH and Γ is shown in Fig. 3.3.

I use this simple model to test the assumption that there are no systematic differences between the XMM and RXTE data sets. Because Γ and NH are only loosely constrained by the RGS and HEXTE data, I do not include those data sets in this test. I allow NH and Γ for MOS 1 and MOS 2 and Γ for the PCA data to vary independently, but I set NH for the PCA data equal to that of MOS 1 because NH is unconstrained by the PCA data alone. The MOS 1 and 2 data sets yield consistent values of NH , +0.03 but there are discrepancies in Γ, with ΓMOS1 = 1.74 0.02, ΓMOS2 = 1.78 −0.02 and +0.02 ± ΓPCA = 1.69 −0.01. I find that this discrepancy is not the result of the spectrum, as whole, becoming harder during the longer RXTE observation; fits to several temporal subsets of the PCA data are best-fit with the same value of Γ. When the MOS 1, MOS 2, and PCA data are fitted above 3 keV (excluding the interval between 4.5–7.5 keV), I find +0.09 +0.02 that ΓMOS1 = 1.57 , ΓMOS2 = 1.68 0.08 and ΓPCA = 1.70 . Thus it appears −0.07 ± −0.03 that the discrepancy in the power law index is an indication that the spectrum hardens at high energies. However, it is clear that the photon indices inferred from the MOS 1 and PCA data are inconsistent, although the MOS 2 and PCA data are in excellent

1A similarly good ﬁt is obtained using the abundance pattern of Anders & Grevesse (1989), but only if the oxygen abundance is reduced by 50%. In this case, N 1021 cm−2. ∼ H ∼ 33

Table 3.2. Best Fit Continuum Parameters

Model Model # Model Parameters a,b χ2/d.o.f. 21 −2 Powerlaw ...... 1 NH = (8.1 0.1) 10 cm 1005/895 Γ=1± .72 ×0.01 ± +0.1 21 −2 Brokenpowerlaw ...... 2 NH = (8.7 −0.2) 10 cm 928/893 ×+0.04 Γ1 = 1.86 −0.03 Ebreak = 2.9 0.2 keV Γ = 1.69± 0.01 2 ± 21 −2 Power law + Compton 3 NH = (8.1 0.1) 10 cm 999/892 ± +0×.01 reﬂection (neutral) Γ = 1.74 −0.02 (Magdziarz & Zdziarski 1995) Efold > 440 keV +0.06 Ω/2π = 0.13 −0.07

21 −2 Power law + Compton 4 NH = (8.1 0.01) 10 cm 997/890 ± +0×.02 reﬂection (ionized) Γ = 1.74 −0.01 (Magdziarz & Zdziarski 1995 Efold > 440 keV Doneetal.1992)...... Ω/2π = 0.13 0.06 ξ < 20 erg cm± s−1

c +0.01 21 −2 Power law + bremsstrahlung 5 NH = (8.5 −0.02) 10 cm 938/892 ×+0.02 Γ=1.68 −0.01 kT = 0.9 0.2 keV ± d +0.1 21 −2 Power law + Gaussian ...... 6 NH = (8.0 −0.2) 10 cm 916/892 Γ=1.69 ×0.01 +0±.1 E = 1.5 −0.2 keV σ = 0.5 0.1 keV ±

aThe error bars are the 90% confidence interval for 1 degree of freedom (i.e. ∆χ2 =2.706). bThe observed 2–10 keV flux, as measured with the MOS 1 and 2 cameras is (5.8 −11 −1 −2 ± 0.1) 10 erg s cm . The unabsorbed luminosity, at the distance of 3C 111, is 3 44× −1 × 10 erg s . cThe bremsstrahlung component has a 0.5–2 keV unabsorbed flux of (5 2) 10−12 erg s−1 cm−2 and contributes 10% of the flux in the 0.5–2 keV band and 1% of± the flux× in the 2–10 keV band. dThe unabsorbed flux of the emission line 1.4 +1 10−12 erg cm−2 s−1. −0.4 × 34

Fig. 3.2 Top panel - Observed 0.4–100 keV spectrum with the different continuum models described in 3.5.1 overlaid (the fits excluded the data from 4.5–7.5 keV). The RGS data § have been excluded and the MOS have been binned for clarity. Lower panels - Ratio of the data to the model for, top to bottom, MOS 1, MOS 2, PCA, and HEXTE (cluster 0 - crosses, cluster 1 - filled circles). The position of 6.4 keV Fe Kα emission line at the redshift of 3C 111 is indicated by the vertical dashed line. The line is clearly seen in the PCA data, but is less obvious in the MOS 1 and MOS 2 data. 35

Fig. 3.3 Confidence contours (90%) in the NH – Γ plane for four different continuum models. The best-fit powerlaw index for the Powerlaw + Bremsstrahlung and Powerlaw + Soft Gaussian models is significantly steeper than that found for the Powerlaw and Powerlaw + Compton Reflection models. agreement. This can be seen in Fig. 3.2; the MOS 1 data show clear positive residuals above the Fe Kα line region that are not present in the MOS 2 data. Throughout the rest of this chapter, I will assume that the XMM and RXTE data can be fitted with the same set of parameter values and allow only the normalization constants to vary independently, however the effects of allowing the MOS 1 data to have an independent photon index will also be considered.

Broken Power law (Model #2): – The next simplest model is an absorbed broken power law, whose free parameters are the column density, two power law indices (Γ1 and Γ2), the breaking energy (Eb), and the normalization constant. The fit is greatly improved, with a chance probability of 10−16. The small breaking energy, ∼ Eb = 2.9 0.2, suggests that the addition of a low energy component might improve ± the fit greatly. However, this extra component also increases the complexity of the model; before including it, I first investigate the effects of including Compton reflection from neutral and ionized media. Power law + Compton reflection from a neutral medium (Model #3): – I next fit the data with a model which includes power law emission from the primary X- ray source as well as the continuum from X-rays reprocessed by a cold, dense slab of material. This model, implemented with the pexrav routine in XSPEC (Magdziarz & Zdziarski 1995), does not include Fe Kα line emission nor does it include the gravitational effects which would arise if the reprocessing material is near the black hole. The free parameters are: NH , Γ, the folding energy of the primary X-ray power law continuum (Efold), the cosine of the inclination angle 36

(cos i < 0.95 or i > 18◦), and the solid angle subtended by the disk to the primary X-ray source (Ω/2π). Like Eracleous et al. (2000), I find that the elemental abundances do not greatly affect the fit and I keep them fixed to the solar values. As described in Appendix B, the inclination angle can be further constrained to ◦ be less than 26 . As Efold increases above 850 keV, the improvement in the fit is negligible, so I restrict Efold < 1000 keV. With these additional constraints in place, I find that there is a marginal improvement in the fit as compared to the simple power law (Pc = 17%). The spectral model and residuals are shown in Fig. 3.2 and the 90% contour in NH –Γ space is included in Fig. 3.3. As shown by the Efold–Ω/2π confidence contours, (Fig. 3.4), Ω/2π is small, but non-zero. I note that Ω/2π is insensitive to the inclination angle, within the narrow range of allowed values.

Fig. 3.4 Confidence contours in the Efold –Ω/2π plane for the Powerlaw + Compton Reflection model. The inclination was fixed to 26◦. The reflection fraction is small, but non-zero.

Power law + Compton reflection from an ionized medium (Model #4): – As discussed in 3.1, the ionization state of the disk is important in determining the strength § of the reflection features, so I now consider reflection from an ionized disk, as implemented with the pexriv routine in XSPEC (Magdziarz & Zdziarski 1995; Done et al. 1992). The additional free parameters in this model are the disk 6 temperature, T (< 10 K) and the ionization parameter, ξ = 4π Fion/n, (ξ < −1 5000 erg cm s ) where Fion is the incident 5–20 keV flux and n is the density of the reflecting medium. The improvement in the fit, as compared to a neutral −1 reflection model, is negligible (Pc = 37.6%) and we found that ξ < 20 erg cm s and Ω/2π = 0.13 0.06. ± 37

When the above models are fitted to the data, there are unacceptable residuals at low energies (See Fig. 3.2). Furthermore, the results of the broken power law fit indicate that the addition of a soft component is likely to improve the continuum fit significantly. Thus, I now add a variety of low energy components to a power law continuum in an attempt to improve the fit. Power law + Thermal bremsstrahlung (Model #5): – I first add a thermal bremsstrahlung component to the absorbed power law model, leading to a significant improvement in the fit compared to the power law model (∆χ2 = 67 for three −13 extra d.o.f.; Pc 10 ). The thermal component has a temperature of 1 keV ∼ ∼ and is found to contribute 10% and 1% of the flux in the 0.5–2 keV and 2–10 keV bands, respectively. The spectral model and residuals are shown in Fig. 3.2 and the 90% contour in N –Γ space are included in Fig. 3.3. I note that a 0.3 H ∼ keV blackbody component also provides a good fit. Even with the addition of the bremsstrahlung or blackbody component however, some low-energy residuals remain. Power law + Soft Gaussian (Model #6): – It is also possible that the low-energy residuals are the result of remaining problems in the calibration of the MOS effective area curves. Therefore I also attempt to model the residuals with combinations of emission lines and absorption lines and edges. I find that a single Gaussian emission line with E 1.5 keV, σ 0.5 keV, and a flux of 10−12 erg cm−2 s−1 ∼ ∼ ∼ 2 is quite effective in eliminating the majority of the low energy residuals (∆χ = 89 −18 for three extra d.o.f; Pc 10 ). The spectral model and residuals and the 90% ∼ contour in NH –Γ are shown in Figs. 3.2 and 3.3, respectively. Whether a 1 keV bremsstrahlung, a 0.3 keV blackbody, or a Gaussian emis- ∼ ∼ sion line is added, the best-fit power law photon index decreases to the value I find from fitting only the PCA data (Γ = 1.69). Therefore, it is not surprising that when these components are added to the models which include reprocessed emission, no Compton reflection is necessary and Ω/2π < 0.05. I find that if we exclude the data from 1–3 keV, where the low-energy residuals are the most extreme, I find again that Γ = 1.69; it appears that the addition of a low energy component is truly necessary to obtain a reliable description of the 3–100 keV spectrum. The origin of this component is discussed further in 3.6.1. § In summary, the data are best fitted with a simple Power law (with NH 8 21 −2 ≈ × 10 cm and Γ 1.7) with little or no contribution from Compton reflection of the ≈ primary power law emission. This is in marked contrast with Seyfert galaxies, in which Ω/2π = 0.7 (see 3.1). Throughout the remainder of this chapter, a soft Gaussian com- h i § ponent is included in all models because it provides the best statistical fit, however the results I obtain using a thermal bremsstrahlung (or blackbody) component are similar.

3.5.2 Models for the Fe Kα Line In this section, I study the profile of the residual Fe Kα emission from the Power law + Soft Gaussian and Power law + Compton reflection models (#6 and #3 respectively). Before adding the line component, all data outside the 3.5 – 8.5 keV energy 38 interval are excluded and the continuum parameters, including the normalization constants, are frozen. The emission line is fitted by either a Gaussian line or a relativistically broadened line from an accretion disk, implemented with the diskline routine (see Fabian et al. 1989). The Gaussian parameters are the line energy (E), the energy dispersion (σ), and the line flux. The diskline model has the following parameters: the inner and outer radius of the line-emitting region (Rin and Rout, where R is expressed in units r ), the disk inclination (i), the slope β of the emissivity pattern, ( rβ), the g ∝ energy of the emission line (E), and the line flux. Note that the diskline model does not include the effects of the Comptonization of the Fe Kα line as it emerges from the disk, which asymmetrically broadens it further. The line parameters for all three data sets are forced to be the same, with the exception of the normalization constants, which are linked together such that the equivalent width of the line was the same for all three. Since the various line model parameters are highly interdependent, I show the fit results in the form of contours (Figs. 3.5, 3.6, and 3.7), rather than list them in a table. For the Gaussian model, 90% confidence contours in the line energy and the FWHM of the line profile and contours in the EW and the FWHM are shown. For the disk line model, 90% confidence contours in the inner radius of the disk (Rin) and the line energy ◦ are constructed for models with β = –2, –2.5 or –3 and i = 10, 18 or 26 , with Rout fixed to be 500 rg; the contours are very similar for Rout = 1000 rg. The improvement in the fit with the addition of an emission line, whether modeled by a Gaussian or a disk line, is significant. For the Power law + soft Gaussian continuum model χ2 = 314 for 172 d.o.f. when no line is included, whereas χ2 = 163 for 169 d.o.f. for the Gaussian model. For the disk line model, χ2 = 159, 165, and 173 for 169 d.o.f. for β = –3, –2.5, and –2 respectively and ∆χ2 is not very sensitive to the inclination angle. For the Power law + Reflection continuum, χ2 = 279 for 172 d.o.f. when no line is included, whereas χ2 = 184 for 169 d.o.f. for the Gaussian model. For the disk line model, χ2 = 186, 187, and 190 for 169 d.o.f. for β = –3, –2.5, and –2 respectively. The residuals are equally well modeled by the Gaussian and the disk line, but for the Power law + soft Gaussian continuum model, a disk line model with β = –2 is disfavored by the data because of the large value of χ2. As discussed in 3.5.1 and seen in Fig. 3.2 the MOS 1 data show clear residuals § above the Fe Kα line region because those data prefer a smaller Γ than the MOS 2 and PCA data. Because the properties of a weak line are very dependent upon the continuum fit (see 2), I also present the results for a fit in which a Gaussian line and a simple power § law continuum were fitted simultaneously only over the region from 3–10 keV. The photon indices were allowed to vary independently but N was fixed to be 8 1021 cm−2, because H × it is unconstrained by these high energy data. The parameters of the Gaussian line were initially allowed to vary independently. I found that the line parameters inferred from the MOS 1 (EMOS1 = 6.3 0.2 keV, σMOS1 < 0.5 keV, and EWMOS1 = 50 40 eV) and +0.2 ± +40 ± PCA (EPCA = 6.4 −0.1 keV, σPCA < 0.5 keV, and EWPCA = 50 −10 eV) data were in agreement. The line is not well constrained by the MOS 2 data, but when E and σ are the same as for MOS 1, the upper limit to EW is 70 eV, which is consistent with the EW measured for the MOS 1 and PCA data. Thus for the purposes of investigating the line properties, the line energy, σ and EW were forced to be the same for all three data 39

Fig. 3.5 Confidence contours (90%) for the Gaussian fit to the Fe Kα emission line to the residuals from the Powerlaw + soft Gaussian continuum fit (model #6, solid). For comparison, we show the contours for the Gaussian fit to the residuals from the Powerlaw + Compton Reflection fit (model #3, dashed) as well as the simultaneous fit to Fe Kα and the continuum over the 3–10 keV interval (dotted, see 3.5.2). The equivalent width is § defined to be the line flux divided by the specific continuum flux at the rest energy of the emission line, E. 40

Fig. 3.6 Confidence contours (90%) for the diskline fit to the FeKα emission line in Rin– E space, using the Powerlaw + soft Gaussian continuum model (Model #6). The shape of the line profile is sensitive to the inclination i and the powerlaw emissivity index, ◦ ◦ ◦ β, thus we present a set of Rin–E contours for i = 26 (solid), 18 (dashed), and 10 (dotted) using two different values of β. 41

Fig. 3.7 Confidence contours (90%) for the diskline fit to the Fe Kα emission line in Rin–E space, using the Powerlaw + Compton Reflection continuum model (Model #3) for i = 26◦ (solid), 18◦ (dashed), and 10◦ (dotted) and β = –2.5. 42 sets.2 As with the Power law + soft Gaussian and Power law + Reflection continuum models, the addition of a narrow line leads to a significant improvement of the fit (χ2 = 155 for 220 d.o.f. compared to 210 for 221 d.o.f when the line is not included.) An Fe Kα emission line is clearly required to fit the data regardless of the continuum model. The line is resolved in all cases with a large bulk velocity (FWHM −1 > 20, 000 km s and Rin < 100 rg for the Gaussian and disk line models respectively) which implies an origin in an accretion disk as opposed to a distant reprocessor such as the obscuring torus. However the line need not form in the innermost regions of the accretion disk. The equivalent width also has considerable uncertainty (EW = 40–110 eV), but is much smaller than the total EW of the narrow + broad Fe Kα emission line (EW 200 eV) observed in those Seyferts with broad Fe Kα emission lines. ∼ 3.5.3 Combined Continuum and Fe Kα Emission Models In the previous two sections, I verified that reprocessing features, especially the Compton reflection bump, are much weaker in 3C 111 than in Seyferts. However, the models used are not physically reasonable; the Fe Kα emission line is very broad, implying an origin in the inner accretion disk, but the Compton reflection, which would necessarily accompany this disk line, is not allowed by the high energy continuum. In the following sections, I simultaneously fit the continuum and Fe Kα emission line, with the goals of finding a self-consistent model for the data and assessing whether either of the two scenarios presented in 3.1 are physically reasonable. The soft Gaussian component is § included in all models and the best-fit parameters are listed in Table 3.3 and discussed below.

3.5.3.1 Truncated Accretion Disk - Models #7a,b The total spectrum from 0.4–100 keV is fitted with a model which includes Comp- ton reflection and an Fe Kα emission line, both of which arise from the reprocessing of the primary X-ray emission by a neutral or moderately ionized accretion disk (Model #7a). I use the refsch routine, which implements the ionized disk model described in 3.5.1 (Model #4), but convolved with a disk line model to account for relativistic effects § (see, for example Fabian et al. 1989). The Fe Kα line is modeled as a disk emission line with β 3, i 26◦, R 500 r and the normalization of the emission line is tied ≡ − ≡ out ≡ g to Ω/2π such that EWFe K 160 Ω/2π eV (George & Fabian 1991), which is only α ≡ × appropriate for a solar abundance of Fe. Both the continuum and line parameters are allowed to vary. +350 The inner radius of the accretion disk is Rin = 110 −70 rg and the energy of the emission line is consistent with being emitted by nearly-neutral Fe, thus this model indicates that both the continuum and Fe Kα emission line could arise from a truncated accretion disk. However, there are residuals in the spectrum near 6.4 keV. Using the

2Although the line is not resolved by any individual data set, when all three are ﬁtted together the line must be broad, as seen in Fig. 5. This is because the high S/N PCA data require that the line has EW > 40 eV, but for an unresolved line the upper limit to the MOS 2 EW is 30 eV; to ﬁt all three data sets, the line must be broad. 43

Chandra High Energy Transmission Grating, Yaqoob & Padmanabhan (2004) found that the narrow cores of Fe Kα emission lines in Seyferts have an average energy of 6.4 keV ∼ and σ 0.02 keV. The inclusion of such a narrow Gaussian, in which E and σ are fixed to ∼ 2 these values, improves the fit slightly (∆χ = 8 for one d.o.f., Model #7b). The equivalent width of this narrow line is 25 eV. Ghisellini, Haardt, & Matt (1994) found that for ∼ 23 −2 disk inclinations similar to that of 3C 111, a Compton-thin torus (with N 10 cm ) H ∼ will contribute an Fe Kα emission line with an equivalent width of 30 eV while making a negligible contribution to the Compton reflection bump. The line can also originate by transmission through dense clouds along the line of sight. However, with the addition of the narrow line, the reflection fraction, and consequently the equivalent width of +0.04 the emission line arising from the disk, are very small (Ω/2π = 0.09 −0.05). Given the weakness of the disk line it is impossible, with these data, to place any meaningful constraints on either the energy of the disk emission line or the inner radius of the accretion disk. The results of these fits suggest that only a very weak (EW 10–20 eV) broad ∼ emission line is required and the data are primarily fitted by a narrow emission line. This is in sharp contrast to the results of 3.5.2, which suggest that the Fe Kα line is stronger § −1 (EW = 40–110 eV) and very broad (FWHM > 20, 000km s ). This is due in part to the fact that there is significant broad curvature in the refsch continuum. Using the fakeit command in XSPEC a data set was simulated based upon the refsch model parameters for Model #7a, but without the disk emission line component; when these data are fitted with the Power law + soft Gaussian model (#6), there are broad residuals in the Fe Kα emission line region, as seen in Fig. 3.8. This curvature is likely due to the relativistically blurred Fe Kα edge, as I noticed that the excess emission was greater diskline for larger values of Ω/2π but less noticeable for smaller values of Rin. If the component of Model #7a is replaced with a Gaussian, the line only has EW = 40 10 +0.2 ± eV and σ = 0.3 −0.1 keV, which is not inconsistent with the total EW (narrow + broad) of 40 eV inferred from model #7b. Thus the broad residuals studied in 3.5.2 can ∼ § be fitted with a combination of continuum curvature as well as broad and narrow line emission. The fit results for this model (#7b) are consistent with the emission arising from the reprocessing of the primary X-ray continuum by an accretion disk, although a distant reprocessor also contributes to the flux of the Fe Kα emission line. Because the component of the line arising from the accretion disk is very weak, Rin is not constrained and the data do not require the disk to be truncated. However a larger value of Rin (such as found for Model #7a) is more physically attractive because it is consistent with the small value of Ω/2π found.

3.5.3.2 Highly Ionized Accretion Disk - Models #8a,b,c As discussed in 3.1, the reflection features from a highly ionized disk are very § weak; when fitted with the pexriv and refsch routines in XSPEC, as I have done thus far, the value of Ω/2π is misleadingly low (Ross, Fabian, & Young 1999). Therefore, I now fit the data with the constant density ionized disk model described by Ross & Fabian (1993) and Ballantyne, Iwasawa, & Fabian (2001), which is available as a table model for 44

Fig. 3.8 Ratio of a simulated PCA data set based upon model #7a (without the disk emission line component) to the Powerlaw + soft Gaussian model (Model #6). The shape of the residuals is indicative of curvature in the former continuum model, which can mimic a broad Fe Kα line. 45 use in XSPEC (Model #8a). I refer to this model as rf-pexriv to avoid confusion with the previously discussed pexriv model. Ballantyne et al. (2002) compared this model to the ASCA spectrum of the BLRG 3C 120 and found that the spectrum is well fitted with ξ = 4000 erg cm s−1 and Ω/2π fixed at unity. More recently, Ballantyne & Fabian (2005) modeled the XMM-Newton spectrum of the BLRG 4C 74.26 with an ionized disk model and the data were best fit with Ω/2π 1 and there was clear evidence of a broad ∼ Fe Kα emission line arising from an accretion disk with an inner radius that extends to Rin < 6 rg. The disk is modeled as a slab of gas with a constant density of 1015 cm−3, which is illuminated by a power law with index Γ and a sharp cut-off at 100 keV, and the ionization parameter extends beyond 104. (Here the ionizing flux is defined from 0.01– 100 keV.) This model includes the emission from the Fe Kα line, so the emission line and continuum are fit self-consistently. The Comptonization of the reprocessed photons as they emerge from the dense disk is also included, but gravitational and inclination- dependent radiative transfer effects are not. Even though the best-fit ionization is quite large (ξ 4000) the reflection fraction ∼ is Ω/2π = 0.3 0.1; Ω/2π 1, as observed in Seyferts, is not allowed (see Fig. 3.9). As ± ∼ before, the addition of a narrow emission line with E 6.4 keV improves the fit slightly 2 ≡ (∆χ = 7 for 1 extra d.o.f; Model #8b); the EW of the narrow line is 15 eV but the ∼ reflection fraction is further reduced to Ω/2π = 0.2. The results of these fits suggest that even when the disk is allowed to be highly ionized, Ω/2π is still rather small. However, if gravitational effects are included (as in 5.3.2), larger values of Ω/2π might be allowed, § because the reflection features would be blurred.

Fig. 3.9 Conﬁdence contours in the log ξ–Ω/2π plane for the constant density ionized accretion disk model described in 3.5.3.2. § 46

As a test, I have convolved the rf-pexriv model with a disk emission line3 with β 3, i 26◦, R 6 r ; R 400 r and also included a narrow (unconvolved) 6.4 ≡− ≡ in ≡ g out ≡ g keV Gaussian (Model #8c). The response functions were extended using the extend command in XSPEC and the HEXTE data above 80 keV were ignored because the model is only defined up to 100 keV. I find that the fit is improved when the relativistic effects are included but that the best-fit parameters (see Table 3) are very similar; in particular Ω/2π 0.2 and EW 15 eV. This model is computationally expensive and ∼ n ∼ it is impossible with the available computer facilities to extensively explore parameter space for this model or to determine error bars for the parameters. Instead, I determined best-fit models for specific values of Ω/2π (0.6 and 1.0) and I find that these larger values are not absolutely ruled out, but they are not favored by the data (∆χ2 = +7 and +13 for 1 d.o.f. respectively, compared to the best fit model #8c). The models presented in this section provide very good fits to the data and it is certainly possible that the accretion disk is ionized. However, the second scenario proposed in 3.1, (that Ω/2π 1 and the reprocessing features are weak only because § ∼ the disk is highly ionized), is not preferred by the data, although it is not completely ruled out.

3.5.3.3 Partial Covering - Model #9 Recent observations with XMM have shown that, contrary to the early results from ASCA, many Seyfert 1s possess only a narrow Fe Kα line with no evidence for a broad, distorted line (Reeves 2003). In some of these objects, broad residuals are clearly present in the Fe Kα region of the spectrum, but they are equally well fitted by a broad diskline and a model in which the continuum is partially covered by an absorber with a column density of 1022−23 cm−2 (see, e.g. 1H0419-577, Pounds et al. 2004a; NGC 4051, Pounds et al. 2004b). This heavily absorbed power law spectrum turns over at an energy near the Fe Kα emission line, thereby introducing curvature into the continuum spectrum than can mimic a broad Fe Kα line. Here I test whether the broad residuals in the Fe Kα region of the spectrum of 3C 111 are simply an artifact of partial covering. The data are fitted with an absorbed power law model that includes an additional partial covering absorber with column density NH,2 and covering factor fc, implemented with the pcfabs routine. There are clear residuals in the spectrum so a Gaussian emission line is also included (Model #9). Confidence contours for NH,2 and fc are shown in Fig. 3.10. The emission line has an energy of 6.4 keV and an EW 30 eV ∼ ∼ and, as can be seen in Fig. 3.11, the line is not resolved at the 90% confidence level. I find that even when the data from 3–8 keV are excluded from the fit, the same values of NH,2 and Fc are found, which suggests that continuum itself is better fitted by this partial covering model than a simple power law. I note that this model, in which none of the primary emission is reprocessed by a dense, geometrically thin accretion disk, could be considered to be an extreme case of scenario #1 presented in 3.1 (i.e. the accretion § disk is truncated).

3The convolution was carried out using a code kindly provided by A.C. Fabian. 47 /d.o.f. 2 1000/994 1043/996 1050/997 1031/993 χ 1 1 − − eV 10 1 20 eV +6 − − 02 1039/994 1 05 07 . 02 02 01 01 . . 8 ± − . . . . 04 02 02 03 0 erg cm s 0 . . . erg cm s . 0 0 +6 − 0 0 +0 − 2 1 3 +0 − +0 − ± +0 − +0 − = 32 − 10 10 15 = 30 d 66 64 . 66 65 1032/988 . . 76 63 . . = 14 erg cm s . . × × Line Models 3 erg cm s d 4 5 = 0 EW 3 α EW 10 b c K f erg cm s 10 EW ξ < × ξ < ;Γ=1 ;Γ=1 ; ;Γ=1 eV eV 3 ; ;Γ=1 ;Γ=1 ;Γ=1 ; 2 2 ; 2 2 2 × 2 8 2 2 2 . . keV; − − +4 − 06 − 04 − 10 2 0 04 − 8 05 11 − − . . 1 keV; . . . 1 2 0 +4 − +0 − +12 − 0 +8 − . . 0 × cm cm 0 +0 − cm cm +0 − cm cm cm 8 5 . = 4 +0 − . ± = 15 eV 21 21 23 21 20 = 4 21 21 2 21 ξ 09 4 . . . . n = 15 = 25 10 10 ξ 10 = 5 10 10 =3 10 10 n n × × × × d ξ = 0 × × 1; = 0 = 0 ) × = 6 06;a . ) ) EW E 1) 1) . 4 7 4 1 1 2 . 0 . . . 2 π σ d . 2; . . π . 0 ; . 01) 1 . EW EW 0 0 2 0 0 . 2 g E ± +0 − / 0 r +0 − / +0 − ± ± ± ; 3 5 = 8 2 8 . g . =0 ± . 2 3 . e ; Ω r 17 . . . 9 H 500 π . 2 keV; 2 . = 0 N 70 / < = (1 0 = (8 = (7 +350 − = 0 = (8 = (8 π Ω 2 in , 2 ± = (7 H H π H H / H 2 R 4 N N H 150 keV; Ω . N N 120 keV / N N < = 110 < = 6 in fold E fold R E E 3.5.3.2) 3.5.3.1) 3.5.3.2) 3.5.3.1) 3.5.3.2) 3.5.3.3) § § § § § § Model # Model Parameters f ) ) Ω ) Ω d d d ) ) c c a refsch refsch Table 3.3. Best Fit Parameters for Combined Continuum and Fe rf-pexriv rf-pexriv rf-pexriv Model Powerlaw 9 (see + Diskline + Diskline + Gaussian + Partial Covering Reflection ( Reflection ( + Narrow Gaussian + Narrow Gaussian + Narrow Gaussian Powerlaw + Compton 8b (see Powerlaw + Compton 8a (see Powerlaw + Compton 7a (see Powerlaw + Compton 7b (see Reflection ( Reflection ( Reflection ( Blurred Powerlaw + Compton 8c (see 48

Footnotes to Table 3.3 a The soft Gaussian component (see 3.5.1, Model #6) is included in all of these mod- § els. The fit parameters were allowed to vary, but the best-fit parameters are consistent with those found for Model #6: E 1.5 keV; σ = 0.5 keV; and an unabsorbed flux of −12 −1 2 ∼ 10 erg s cm . ∼ bThe error bars correspond to the 90% confidence interval for 1 degree of freedom (i.e. ∆χ2 = 2.706). c The diskline model has β -3, i 26◦, and R 500r . The normalization of the ≡ ≡ out ≡ g emission line has been tied to Ω/2π, such that EW 160 Ω/2π eV. The energy and ≡ × equivalent width of the diskline are denoted by Ed and EWd. dThe narrow Gaussian has a fixed energy of 6.4 keV and width of 0.02 keV. The equivalent width of the narrow line is denoted by EWn. eThe folding energy was restricted to be greater than 100 keV. f The model was convolved with a diskline model with β -3, i 26◦, R 6r and ≡ ≡ in ≡ g R 400r . out ≡ g 49

Fig. 3.10 Conﬁdence contours in the column density and covering fraction for the partial covering powerlaw model described in 3.5.3.3. As described in 3.5.3.3, the absorber § § is expected to produce an Fe Kα emission line. Overlaid on the contours are lines of constant Fe Kα equivalent width (10, 15, and 20 eV) as a function of NH,2 and fc as prescribed by equation (1).

3.6 Discussion

3.6.1 Origin of the Low Energy Component Many of the continuum models presented in 3.5.1 do not provide a satisfactory § fit to the low energy data. I find that the inclusion of a low energy component, described either by a 1 keV bremsstrahlung (or 0.3 keV blackbody) or a broad Gaussian ∼ ∼ line with E 1.5 keV and σ 0.5 keV, greatly improves the fit. Although it appears ∼ ∼ that this component must be included to obtain a good description of the 3–100 keV spectrum, it would be reassuring to determine whether there are any plausible sources of this low-energy component. Previous ASCA and ROSAT observations have shown no evidence for a soft component in the spectrum of 3C 111, however the large and uncertain column density would have made its detection difficult. Soft components have been observed in other BLRGs. The BLRG 3C 382 shows a soft excess which cannot be entirely explained by the known extended halo of hot gas which surrounds the host galaxy (Grandi et al. 2001). The BLRG 3C 120 exhibits a clear soft excess, that contributes 20% of the 0.6–2 keV flux that Ballantyne et al. (2004) model with a 0.3–0.4 keV bremsstrahlung. However, the ∼ origin of these soft components are not well understood. An alternative possibility is that the soft energy component simply compensates for an error in the calibration of the MOS effective area curves. The count rate near 1 50 keV is very large, with some data bins containing more than 4000 counts. The statistical uncertainties ( 2%) are on the order of the uncertainties in the calibration of the MOS ∼ effective area curves, which can be as large as 10% (XMM-Newton helpdesk, private ∼ communication, 2004). For this reason, I also modeled the low energy component as a Gaussian with E 1.5 keV and σ 0.5 keV. The peak amplitude of the emission ∼ ∼ line is only 8% of the continuum flux, so it is not impossible that the soft Gaussian ∼ component is compensating for an error in the MOS effective area curves. The soft Gaussian component is admittedly ad-hoc. However, this component has been chosen, instead of a thermal component, primarily because it affects a relatively isolated region of the spectrum (compared to the bremsstrahlung model) while at the same time it produces a better fit to the low energy residuals. There is independent evidence that there is a known excess in the MOS data, similar to what are present in these data (see Fig. 11 in Kirsch et al. 2004). I have verified that including the Soft Gaussian component only in the two MOS data sets does not affect the results presented here.

3.6.2 Interpretation of the Spectral Models As seen in 3.5.1 and 3.5.3, the high energy data do not allow for a significant § § Compton reflection bump, even if the disk is highly ionized (see 3.5.3.2, Models #8a,b,c). § Therefore these data disfavor the presence of a standard, optically-thick, geometrically- thin accretion disk (neutral or ionized), which extends to small radii, in which case Ω/2π 1. For comparison, if the outer, thin accretion disk transitions at some radius ∼ to a vertically extended structure, (such as an ion-torus), then Ω/2π < 0.3 (Chen & Halpern 1989; Zdziarski, Lubinski, & Smith 1999), consistent with the values of Ω/2π I have obtained. Although the results of 3.5.2 suggest that there is a broad Fe Kα line (FWHM > −1 § 20, 000 km s in 3C 111, these residuals can be partly attributed to curvature in the continuum, either due to relativistic blurring of the Fe Kα edge ( 3.5.3.1) or partial § covering by a dense absorber ( 3.5.3.3). Weak Fe Kα line emission with EW 40 eV § ∼ is still required, but the majority of the flux in this line most likely arises in a distant reprocessor. Thus, from the point of view of fitting the spectrum of 3C 111, it is not necessary to invoke reprocessing by an accretion disk to model the data and there is no obvious reason to prefer the truncated disk models (#7a,b) over the more simple partial covering model (#9). However, it is not immediately obvious which structure could be obscuring the primary X-ray source. The inclination of 3C 111 is less than 26◦ so unless the obscuring torus has a very large opening angle or is very extended vertically, it cannot be responsible for the partial covering. An alternative source of partial covering is Compton-thin (NH < 1023 cm−2) clouds along the line of sight. These same clouds could also contribute to the unresolved Fe Kα emission line. The expected EW of the Fe Kα line transmitted by such clouds is (Halpern 1982)

Γ−1 350f 6.4 N 2 A EW = c H, Fe eV (3.1) Γ + 2 7.1 1023cm−2 4 10−5 × 51 where fc, NH,2, and Γ are the same covering factor, column density, and power law index as found from the partial covering model fit, and AFe is the Fe mass fraction. Using the 23 −2 measured values, the above equation reduces to EW = 71fc(NH /10 cm ) eV. For a given EW of the Fe Kα emission line, there is one-to-one relationship between NH,2 and fc; tracks for EW = 10, 15, and 20 eV are shown in Fig. 3.10. As can be seen, the partial covering absorber can contribute an Fe Kα line with an equivalent width of at most 15 eV. However, if the clouds have a super-solar abundance, the Fe Kα emission would be enhanced. It is likely that the bulk of the narrow Fe Kα emission line is formed in the obscuring torus, which can contribute 30 eV to the total equivalent 23∼ −2 width. Transmission through clouds with NH,2 > 10 cm could also make a small contribution to the observed Fe Kα line, but it is very difficult, without detailed modeling, to estimate the EW of an Fe Kα emission line in this case. Although BLRGs, on average, do not have large intrinsic column densities (Sam- bruna et al. 1999), the BLRG 3C 445, which also has an FR II radio morphology, is similar to 3C 111 in several respects. The intrinsic absorber has a large column density (N 1023 cm−2, Wozniak et al. 1998; Sambruna et al. 1999) but unlike 3C 111, H,int ∼ the absorber in 3C 445 covers the source almost completely (f 0.8; Sambruna et al. c ∼ 1999). The reflection fraction in 3C 445 is also very small Ω/2π < 0.2, but there is a very strong Fe Kα emission line with EW 150 eV (Wozniak et al. 1998); these authors ∼ postulate that the Fe Kα emission line arises in a shell of cold material with NH (2– 23 −2 ∼ 5) 10 cm which is isotropically irradiated from the central source. Thus, 3C 445 × and 3C 111 might in fact be quite similar, with the only difference being the covering factor of the absorber. The partial covering absorber model for 3C 111 would be best tested by UV and soft X-ray observations, because these absorbing clouds should also exhibit emission lines and absorption edges in these regimes. With combined UV and X-ray information, it would be possible to place constraints on the physical conditions and possibly the location of the absorbing material. Unfortunately, the Galactic column density is so large (N 1022 cm−2) that these observations are not possible. However, high H,Gal ∼ S/N UV and X-ray observations of other BLRGs, which do not suffer from such a high Galactic extinction, might reveal that intrinsic absorption is not uncommon in BLRGs. If the covering factor of the absorber is small, as inferred for 3C 111, then the presence of a partial covering absorber could easily be overlooked because the resulting continuum curvature is in the same region as the expected Fe Kα emission line. Alternatively, the partial covering model presented here might simply be a parameterization of the truncated accretion disk (refsch) model presented in 3.5.3.1, in § which significant continuum curvature near the Fe Kα line is present (Fig. 3.8). When a Gaussian line was added to that model in place of a disk emission line, the EW and σ are very similar to those obtained when using the partial covering continuum (see Fig. 3.11). Furthermore, when Seyfert 1s that possess broad Fe Kα line features are fitted 23 −2 with a partial covering model, one obtains NH 2 3–4 10 cm and f < 0.35 (Gel- , ∼ × c bord 2003). As with 3C 111, even when the data from 3–8 keV are excluded from the fit, the values of NH,2 and fc are similar (Gelbord, private communication, 2004). These values of NH 2 are somewhat larger than found for 3C 111, but as discussed in 3.5.3.1, , § the curvature in the refsch model appears to be more extreme for larger values of Ω/2π; 52

refsch when a simulated data set based on a model with Ω/2π = 0.75 and Rin = 6rg (using the fakeit command in XSPEC) is ﬁtted with a partial covering model, I ﬁnd 23 −2 NH 2 3 10 cm . Thus it is possible that, as with the Seyfert 1s studied by Gelbord , ∼ × (2003), the partial covering model for 3C 111 (Model #9) is simply a parameterization of continuum curvature and is not due to a real absorber.

3.7 Conclusions

In this chapter, I present the results of an analysis of simultaneous observations of the BLRG 3C 111 with XMM-Newton and RXTE. The flux is moderately variable on timescales of a day, but there is no evidence for significant spectral variability. The continuum emission has been fitted with a wide variety of models and in all instances there are unacceptably large residuals at low energies. These are well fitted by a Gaus- sian component, which is included in all of the continuum models and is attributed to uncertainties in the calibration of the MOS detectors. Clear, broad, residuals are also present near the Fe Kα emission line region for all models. These can be fitted with a broad Fe Kα emission line with an equivalent width of 40–100 eV, which is weaker ∼ than those observed in Seyferts 1s. The high energy RXTE data strongly disfavor the presence of a strong Compton reflection bump, even if the disk is highly ionized. This result gives strong support to the hypothesis that the geometry of the accretion flow in at least some BLRGs is different from that in Seyfert 1 galaxies. I note that Bal- lantyne & Fabian (2005) have found conclusive evidence of a broad Fe Kα emission line in 4C 74.26, and I suggest that perhaps BLRGs are heterogenous, with some objects having low accretion rates and truncated accretion disks, whereas others have higher accretion rates and ionized disks. I also note that the reprocessing features in 4C 74.26, even when fitted with non-ionized disk models, are not particularly weak (EW 150 ∼ eV and Ω/2π 1.2) so 4C 74.26 is not necessarily reprsentative of all BLRGs. ∼ I find that continuum curvature is a primary source of the broad residuals seen in the Fe Kα line region, although line emission is also required. The data are consistent with a model in which a weak Compton reflection bump and Fe Kα disk emission line are formed in a truncated accretion disk that transitions to a vertically extended structure (such as an ion torus) at small radii. A less complex model, in which the primary X-ray source is partially covered by a dense (1023 cm−2) absorber, also provides a satisfactory fit to the data. In both models, an Fe Kα emission line that most likely arises in a distant reprocessor, such as a Compton-thin obscuring torus or dense clouds along the line of sight, is also present. Given these data, it is not possible to distinguish between these two models. However, it is very likely that the partial coverage model is simply a parameterization of the more complicated model in which the primary emission is reprocessed by an truncated accretion disk. Moreover, the partial coverage model is unappealing in view of the small inclination angle of the jet in 3C 111. 53

Fig. 3.11 Confidence contours for the Gaussian fit to the residual Fe Kα emission line from the partial covering model (with the soft Gaussian included) described in 3.5.3.3. § The equivalent width is defined to be the line flux divided by the specific continuum flux at energy. 54

Chapter 4

Characterization of the Long Term Proﬁle Variability of Double-Peaked Emission Lines in AGNs

4.1 Introduction

In this Chapter, I present results from the campaign to monitor the long-term profile variability in AGNs with double-peaked emission lines. A sample of 20 double- ∼ peaked emitters from the Eracleous & Halpern (1994) sample, in addition to a few more recently discovered double-peaked emitters, were monitored over the last decade. The goals of this monitoring campaign were described in Chapter 1, however I briefly re- iterate them here. The first goal was to determine whether the variability observed in the well-known double-peaked emitters was common. A second goal was to use the variability of the double-peaked emission lines to assess the viability of non-disk scenarios for the double-peaked emission lines. Both of these goals have been met; profile variability is commonly observed in double-peaked emitters and the alternative scenarios have failed to meet all the observational tests (Eracleous & Halpern 2003). The third and more long-term goal, which is the focus of this Chapter, was to exploit the variability of the double-peaked emission lines to test models for phenomena in the outer accretion disks in AGNs. Ultimately studies of the variability in these “extreme” objects may shed light on the origin of long-term profile variability of AGN broad lines in general. The observed variability of the double-peaked line profiles and the fact that in some objects the red peak is stronger than the blue peak (in contrast to the expectations of relativistic Doppler boosting) mandates that the accretion disk is non-axisymmetric. In general, the profiles of the double-peaked emission lines are observed to vary on timescales of years to decades and show occasional reversals in the relative strength of the red and blue peaks (for examples, see Veilleux & Zheng 1991; Zheng, Veilleux, & Grandi 1991; Gilbert, Eracleous, Filippenko, & Halpern 1999; Shapovalova, Burenkov, Carrasco, Chavushyan, Doroshenko, Dumont, Lyuty, Valdés, Vlasuyk, Bochkarev, Collin, Legrand, Mikhailov, Spiridonova, Kurtanidze, & Nikolashvili 2001; Sergeev, Pronik, & Sergeeva 2000; Sergeev, Pronik, Peterson, Sergeeva, & Zheng 2002; Storchi-Bergmann, Nemmen da Silva, Eracleous, Halpern, Wilson, Filippenko, Ruiz, Smith, & Nagar 2003; Gezari, Halpern, Eracleous, & Filippenko 2004; Lewis, Eracleous, Halpern, & Storchi-Bergmann 2004a,b). A few of the best-monitored objects have been studied in detail using models which are simple extensions to a circular accretion disk. These include circular disks with orbiting bright spots (Arp 102B and 3C 390.3: Newman, Eracleous, Filippenko, & Halpern 1997; Sergeev, Pronik, & Sergeeva 2000; Zheng, Veilleux, & Grandi 1991) or spiral emissivity perturbations (3C 390.3, 3C 332, and NGC 1097: Gilbert et al. 1999; Storchi-Bergmann et al. 2003) and precessing elliptical disks (NGC 1097, Eracleous et al. 1995; Storchi-Bergmann et al. 1995, 1997, 2003). All of these non-axisymmetric 55 models lead to modulations in the ratio of the blue to red peak flux; in the case of the bright-spot model, modulations occur on the dynamical timescale whereas for the elliptical and spiral arm models, the variations occur on longer timescales as the pattern precesses. Although simple, these models are also physically motivated, as described in 4.4. § These modeling efforts have met with various degrees of success. In all of these objects, the peak flux ratio varies slowly and systematically on timescales comparable to or longer than the dynamical timescale, and this phenomenon has been fairly successfully reproduced by these models. In Arp 102B, the peak flux ratio suddenly began to vary quasi-sinusoidally over a period of four years; Newman et al. (1997) successfully modeled this phenomenon as a single bright spot orbiting in the accretion disk with a period of 2.16 years. Alternatively, Sergeev et al. (2000) suggested that the disk is composed of several thousand randomly distributed clouds that rotate in the same plane. In this model, the hemisphere of the cloud that is facing the central source is brighter; an important prediction is that features should only propagate from the red side of the profile to the blue because the illuminated side of a cloud that is traveling from negative to positive projected velocity is invisible to the viewer. This tendency was tentatively observed by Sergeev et al. (2000) and is more clear in data for Arp 102B spanning a longer time period (Gezari 2005). I also note that Gezari (2005) found another epoch during which the blue to red peak ratio varies semi-sinusoidally with with a similar period as found by Newman et al. (1997). Zheng et al. (1991) also suggested a bright “patch” model, among others, for 3C 390.3. The spiral arm model used by Gilbert et al. (1999) also quite successfully reproduced the profile variations in 3C 332, but was less successful for 3C 390.3. The elliptical and spiral arm models used to describe NGC 1097 (Storchi-Bergmann et al. 2003) also reproduced the basic variability patterns well, but only after the power-law illumination index was allowed to vary. (A non-constant index was justified by the significant decrease in illuminating flux during the course of the observation.) However, in all cases, these simple models fail to reproduce the fine details of the profile variability. It is extremely important to determine whether the variability trends observed in the small number of well-monitored objects are common to all double-peaked emitters. Are there universal variability patterns that might point to a global phenomenon that occurs in most AGN accretion disks? Is each double-peaked emitter unique? To this end, I have characterized the profile variability in seven objects from the long-term monitoring campaign whose long-term variability has not been studied in any detail: Pictor A, PKS 0921–213, 1E 0450.3–1817, CBS 74, 3C 59, PKS 1739+18C, and PKS 1020–103. These objects were also selected because they have all been regularly monitored for a similar period of time ( 6–8 yrs) and with a similar frequency ( 1–2 times per year). The ∼ ∼ properties of these objects are given in Table 4.1. The primary goal of this Chapter is to characterize the observed variability patterns in these objects in a model-independent way. Currently, the models which have been used to fit the line profiles represent the most simple extensions to a circular disk. As mentioned above, these models fail to reproduce the fine details of the profile variability. A detailed characterization of the variability will help guide the refinement of these models and more importantly inspire ideas for completely new families of models. 56

Table 4.1. Galaxy Properties

c −1 Object Starlight LX (erg s ) X-ray a b Name mV z E(B-V) Fraction (0.1–2.4 keV obs.) Ref.

Pictor A 16.2 0.0350 0.043 10% 3.9 1043 1 × 43 PKS 0921–213 16.5 0.0531 0.060 30–40% 3.3 10 2 × 42 1E 0450.3-1817 17.8 0.0616 0.043 40–50% 5.4 10 3 × 44 CBS 74 16.0 0.0919 0.036 10% 2.6 10 4 × 44 3C 59 16.0 0.1096 0.064 20–30% 1.6 10 5 × 44 PKS 1739+18C 17.5 0.1859 0.062 10% 5.7 10 1 × 44 PKS 1020–103 16.1 0.1965 0.046 10% 8.4 10 1 ×

aRedshifts from Eracleous & Halpern (2004) bReddening obtained from Schlegel, Finkbeiner, & Davis (1998) cTypical starlight fraction within the spectroscopic aperture used for the observations over a rest wavelength range of 5500–7000A.˚ It may vary due to seeing and/or the aperture size. dReferences for X-ray luminosity – Observed X-ray ﬂuxes for 0.1–2.4 keV band taken from the Rosat All-Sky Survey in an observed band of 0.1–2.4 keV, except 1E 0450.3- 1817 (Einstein) and PKS 0921-213 (XMM-Newton). The ﬂuxes were not corrected for absorption. (1) Brinkmann, Siebert, & Boller (1994); (2) Lewis et al. (2005) (3) Stocke et al. (1983) (4) Bade et al. (1998) (5) Brinkmann et al. (1995) 57

In essence, these results will serve as benchmark that future models can be easily tested against. This model-independent characterization is used to assess the viability of the elliptical disk and spiral arm models. This is done by characterizing a large number of model profiles, spanning a range of properties, in the same manner as the observed data. The goal is not to perform detailed modeling of any particular object, but to diagnose what aspects of the observed profiles and their variability can and cannot be reproduced by these models. Finally, I considered some simple modifications which can be made to these models in order to better reproduce the observations. This Chapter is organized as follows. In 4.2, the observations and data reductions § are described. 4.3 is devoted to describing the methods for characterizing the profile § variability in a model independent way and describing the common trends in the data. In 4.4 I describe in more detail the two models I compare with the observations, including § their physical motivation, the details of the model profile calculation, and the common variability trends. In 4.5 I assess the viability of these two models through a comparison § of the variability trends in the objects and the models and I also suggests a simple refinement to the spiral arm model which might afford a better description of the observed variability trends. Finally, in 4.6 I summarize the primary findings. § 4.2 Observations and Data Reductions

The majority of the spectra were obtained over the period from 1991–2004 using the 2.1m and 4m telescopes at Kitt Peak National Observatory (KPNO), the 1.5m and 4m telescopes at Cerro-Tololo Interamerican Observatory (CTIO), the 3m Shane Tele- scope at Lick Observatory, the 1.3m and 2.4m telescopes at the MDM Observatory, the and the 9.2m Hobby-Eberly Telescope (HET) at the Mconald Observatory. A complete list of the spectrographs and gratings used is provided in Table 4.2 and a journal of the observations is given in Table 4.3. The spectra were taken through a narrow slit (1.0025 – 2.000) and the seeing varied from 100–300. The spectra were extracted from a 400-800 wide region along the slit and the resulting spectral resolution ranged from 300–800; the resolution for each instrumental configuration is listed in Table 4.2. Most of the data reductions were carried out by M. Eracleous, although some spectra were provided in reduced form by J. P. Halpern and A. V. Filippenko (see also the notes to table 4.3.) The reductions were very similar to those described in Chapter 2, with two important differences. The flux scale was calibrated using spectrophotometric standard stars drawn from Oke & Gunn (1983), Stone & Baldwin (1983) or Baldwin & Stone (1984) that were observed at the beginning and end of each evening. Although the relative flux scale is accurate to better than 5%, the absolute flux scale is uncertain by up a factor of two due to the small slit used and the occasional non-photometric sky conditions. The observed wavelengths of the Hα line profiles presented here do not extend to long wavelengths (λ . 8000A)˚ so telluric correction was not as vital as for the observations presented in Chapter 2. Never- theless, there were several atmospheric features which were important to remove, namely the O2 bands near 6890A˚ and 7600A˚ (the A- and B- bands, respectively) and the water vapor bands near 7200A˚ and 8200A.˚ The observed spectrophotometric standards were sufficiently featureless to be used to correct these bands. 58 ˚ A) ) Spec. Res. ( 00 Table 4.2. Instrumental Configurations Code1a Telescope1b2a Spectrograph KPNO 2.1m3a KPNO 2.1m3b KPNO 4m3c Grating GoldCam CTIO 1.5m3d GoldCam CTIO 1.5m4a Slit Width CTIO ( 1.5m4b CTIO 1.5m RC4c Cass 240 CTIO4d 4m Cass 35 CTIO5a 4m Cass CTIO6a 4m Cass CTIO7a 4m BL 1.8–1.9 1817b RC Lick 3m MDM 35 1.3m 1.8–2.08a RC 26 MDM 2.4m8b RC MDM 2.4m 99a RC 4.5–5.5 32 HET MK 9.2m 1.7 III Kast Modspec HET 3.5–4.4 9.2m G510 du Pont Modspec KPGL2 2.5m 1.8 Boller 1.8 BL & 181 Chivens 600 LRS 600 l/mm BL l/mm G250 1.3 1200 l/mm LRS 1.8 830 l/mm 1.5–2.0 6.0 1.5 2 3.0–4.0 1.25–1.5 1.5–2.0 3.5–4.0 1.5 2.0 Grism 3 1.25 1.5 6.0–8.0 Grism 1 6.0–7.0 8.0 7.0–8.0 5.0–6.0 4–5 2.0 1.0–2.0 1.5 5.0 1.0 2.9 8.0 6.3 4.8 9.3 59

Table 4.3: Log of Observations

Exposure Instr. Exposure Instr. UT date Time (s) Code UT date Time (s) Code

Pictor A 1983 Aug 11a 4200 9 1998 Jan 02 3600 3a 1987 Jul 21b 140000 4c 1998 Oct 20 5400 3a 1994 Feb 18 3000 4b 1999 Nov 01 3000 3a 1994 Dec 08c 300 4d 1999 Nov 03 1800 3c 1995 Sep 29d 1800 4a 2001 Jan 21 3600 3a 1997 Jan 02 3600 3a 2003 Jan 04 3600 3a PKS 0921–213 1995 Mar 24 3000 1a 1998 Apr 07 1800 7a 1995 Mar 25 3600 1a 1998 Apr 08 1800 7a 1996 Feb 15 7200 1a 1998 Dec 20 3000 7a 1996 Feb 16 3600 1b 1999 Dec 02 3000 1a 1997 Jan 02 3600 3a 1999 Dec 03 3000 1a 1997 Jan 04 3000 3a 2000 Mar 16 6000 6a 1997 Mar 24 1700 6a 2001 Jan 21 6000 3a 1998 Jan 02 4800 3a 2001 Oct 24 2000 1a 1998 Jan 28 3600 1a 2003 Jan 02 7200 3a 1E 0450.3–1817 1989 Nov 06 3600 6a 1998 Jan 02 7200 3a 1989 Nov 07 5400 6a 1998 Oct 15 7200 1a 1991 Feb 05 1800 2a 1998 Dec 19 6000 7a 1992 Jan 16 7200 4a 1999 Feb 09 3600 7a 1994 Feb 16 6000 4a 1999 Nov 01 6600 3a 1995 Jan 23 5447 1b 1999 Dec 02 3000 1a 1996 Feb 15 3600 1a 1999 Dec 03 3000 1a 1996 Feb 16 5400 1b 1999 Dec 04 3600 1b 1996 Oct 10 5400 5a 2000 Sep 23 3600 1a 1997 Jan 02 3600 3a 2000 Sep 24 3000 1a 1997 Jan 04 3600 3a 2001 Jan 24 8100 3c 1997 Sep 27 5700 1a 2003 Jan 02 7200 3a CBS 74 1998 Jan 30 3600 1b 2001 Jan 24 1800 7a 1998 Apr 09 1200 6a 2002 Oct 10 3000 1a 1998 Oct 14 3600 1a 2002 Oct 11 3600 1b 1999 Feb 11 3600 6a 2003 Mar 25 300 8a 1999 Dec 02 3000 1a 2003 Apr 05 300 8a Continued on Next Page. . . 60

Table 4.3 – Concluded

Exposure Instr. Exposure Instr. UT date Time (s) Code UT date Time (s) Code 1999 Dec 04 3000 1b 2003 Oct 24 1800 7a 2000 Mar 14 6000 6a 2003 Dec 24 1200 8a 2000 Sep 24 3000 1b 2004 Jan 13 300 8a 3C 59 1991 Feb 04 1800 2a 2000 Sep 24 3600 1b 1991 Feb 05 1500 3a 2001 Jul 21 2700 1b 1997 Feb 07 3000 1a 2002 Oct 10 3600 1a 1997 Sep 28 3600 3a 2002 Oct 11 3600 1b 1998 Oct 13 3600 3a 2003 Oct 21 600 8a 1998 Oct 15 3600 1a 2004 Jan 19 600 8a 1998 Dec 20 2400 7a 2004 Sep 09 600 8a 1999 Dec 04 3600 1b ··· ··· ··· PKS 1739+18C 1992 Jul 09 1800 2a 2000 Jun 04 2400 1b 1996 Jun 15 3600 1b 2000 Sep 24 2400 1b 1997 Jun 09 3600 1b 2001 Jul 21 3000 1b 1997 Sep 28 3600 1b 2002 Jun 14 2700 1b 1998 Apr 09 3000 7b 2002 Oct 11 3600 1b 1998 Jun 27 3000 7a 2003 Apr 02 300 8a 1998 Oct 13 3600 1b 2004 Jun 22 300 8a 1999 Jun 15 3000 1b 2004 Aug 05 300 8a PKS 1020–103 1991 Feb 04 600 2a 1998 Jan 05 3600 3a 1991 Feb 05 2400 2a 1998 Dec 20 2400 7a 1992 Jan 16 1800 4a 1999 Dec 04 3600 1a 1996 Feb 16 3600 1a 2001 Jan 22 7200 3d 1997 Jan 04 3600 3a 2003 Jan 03 6000 3d

a Kindly provided in reduced form by A. V. Filippenko, see Filippenko (1985) b Sum of seven exposures over three nights, 1987 Jul 20-22. Kindly provided by S. Simkin (see also, Sulentic et al. 1995). c Kindly provided in reduced form by T. Storchi-Bergmann. d Kindly provided in reduced form by R. H. Becker. 61

4.3 Proﬁle Anlysis

Before analyzing the spectra, they were shifted into the rest frame and corrected for Galactic reddening, using the redshifts and color excesses given in Table 4.1. Then the underlying continuum (from both stellar and non-stellar sources) was modeled, using the spectrum of an inactive elliptical galaxy plus a powerlaw continuum, and subtracted. There was often residual curvature around the broad Hα profile that was removed using a low (1st or 2nd) order polynomial. These processes of continuum subtraction, particularly that of the local continuum, introduce some uncertainty which will be discussed and quantified below as necessary. The profiles after continuum subtraction are displayed in Fig. 4.1. With the exception of PKS 1739+18C, the spectra have been scaled such that the narrow emission line fluxes are identical, as described in 4.3.1. The fluxes of § the narrow emission lines in PKS 1739+18C could not be reliably measured, thus the line profiles were scaled by the total flux in the line profile. These profiles, and their variations with time, will be characterized in the following sections.

Fig. 4.1 Observed double-peaked Balmer line profiles, after continuum subtraction. For each object, each spectrum has been scaled such that the fluxes of the narrow emission lines are identical and an arbitrary constant offset has been applied. In PKS 1739+18C, the narrow emission lines are too weak to be measured reliably, and the observed flux was scaled such that the total flux of the line profile is constant. This plot is continued on the next two pages . . . 62

Figure 4.1 continued . . . 63

Figure 4.1 concluded. 64

4.3.1 Relative Narrow and Broad Line Fluxes The absolute flux scale is uncertain by up to a factor of two, because the spectra were observed though a narrow slit and in variable sky conditions. A technique commonly used in studies of AGN broad line variability is to use the narrow emission lines for internal flux calibration. These lines form in low-density gas located several light decades from the central engine and they are not expected to respond to rapid variations in continuum flux (see, e.g., Peterson 1993). The narrow emission lines covered by these spectra include [O i]λλ6300,6363, [N ii]λλ6548,6583, [S ii]λλ6716,6730, and the narrow component of Hα. Scaling factors are determined for each spectrum to achieve a uniform flux scale for all of the spectra, relative to a chosen reference spectrum. Unlike the narrow emission lines, the total flux in the broad Hα line does vary in response to changes in the external illumination. It is important to determine how the total broad Hα flux changes with time, again relative to some reference spectrum, for two reasons. Firstly, if one wants to study variations in the shape of the broad Hα line profiles, it is necessary to scale the profiles such that they have the same total broad Hα flux. Short-term fluctuations in the illumination will cause the total broad line flux to change (i.e. through light reverberation), as observed in 3C 390.3 (Dietrich 1998; Shapovalova et al. 2001). These changes in flux are not of interest when studying long-term profile variability that is due to physical changes in the accretion disk, and must be corrected for. Secondly, as discussed in more detail in 4.3.4, the properties § of the double-peaked line profile may vary in response to long-term variations in the illuminating flux, for example as seen in NGC 1097 (Storchi-Bergmann et al. 2003); it would be interesting to see if such effects are common in other objects. To measure the fluxes of the narrow lines, the underlying broad Hα profile must be subtracted. The [O i]λ6300, [O i]λ6363 and [S ii]λλ6716,6730 lines lie on the wings of the broad Hα profile, which are well approximated by a 3rd order polynomial. There is typically less than a 10% difference between the fluxes of the strong [O i]λ6300 and [S ii]λλ6716,6730 lines obtained for different fits to the wings of the broad Hα profile, however the weaker [O i]λ6363 line flux has a larger ( 20%) uncertainty. ∼ Fitting the [N ii]λλ6548,6583 and narrow Hα lines is more challenging for two reasons. The [N ii]λλ6548,6583 lines can be blended with Hα, but this difficulty is easily overcome. The [N ii] lines are fitted first, with the assumption that the [N ii]λ6583 line flux is three times larger than that of [N ii]λ6548, as dictated by atomic physics, and that the FWHM of the line profiles are identical. After the [N ii] lines are subtracted, the residual Hα flux is easily determined. In some objects, there are clearly two components of narrow Hα, one with a FWHM similar to that of the forbidden lines ( 1000 km s−1) ∼−1 and another (sometimes blue-shifted) with a FWHM of 2000–3000 km s ; the fluxes ∼ of both components are included in the narrow line flux. The line decomposition is quite robust and for a given fit to the underlying broad Hα profile and the uncertainty in the narrow Hα and [N ii] fluxes is similar to that for [O i] and [S ii] ( 10–15%). ∼ Of greater concern is the fact that any fit to the underlying broad Hα profile is necessarily based upon an assumption about the general shape of the broad profile. Thus the Hα and [N ii] flux measurements are subject to systematic uncertainty. Using a few test spectra, the effective continuum underlying the Hα+[N ii]narrow line was 65 subtracted using a number of different models and the line fluxes of the residual narrow lines were measured as described above. Not all fits to the underlying broad Hα profile lead to reasonable Hα+[N ii] residuals; the [N ii] lines must be fitted with a flux ratio of 3:1 and the remaining Hα profile must be approximately fitted by one or two Gaussian profiles. Within this constraint, the systematic uncertainty due to fitting the broad profile was estimated to be approximately 30%. It is possible to guard against this systematic uncertainty by considering the ratios of the narrow line fluxes (i.e. [O i]/[S ii], [S ii]/Hα, [O i]/Hα, and [N ii]/Hα), which like the narrow line fluxes, should not change significantly with time. These line ratios are very stable from one observation of an object to the next, with only a handful of outliers. If the fit to the underlying broad Hα profile were wildly different from one observation to the next, this would be manifested in the line ratios, with the exception of [O i]/[S ii]. It is still possible that the Hα+[N ii] flux can be systematically uncertain by up to 30%, however as long as the flux is systematically under- or over-estimated by roughly the same amount for all of the observations, this systematic uncertainty will have little effect on the relative broad line fluxes. The narrow line scaling factor was determined for each spectrum by dividing the measured narrow line fluxes by those from a reference spectrum. The scaling factors from the individual lines were averaged together, and the standard deviation of the values was assigned as an approximate error bar. The individual scaling factors generally lay within two standard deviations (and in most cases within 1 standard deviation) of the mean. In a few instances an individual scaling factor, typically from Hα or [N ii], was clearly skewing the mean; this point was eliminated and the mean and standard deviation were recalculated. The fractional uncertainty in the narrow line scaling factors was typically 10–15%. To determine the broad line scaling factor, the total narrow line flux determined above was subtracted from flux of the entire profile (narrow + broad). The local continuum subtraction introduces a 5–10% uncertainty in the flux of the total profile, but does not introduce any additional uncertainty into the total narrow line flux. The broad line scaling factor is simply the ratio of the total observed broad line flux for each observation divided by that for the reference spectrum. When all the errors are taken into account, the uncertainty on the broad line scaling factors are less than 10% for all objects. Given that there are 15% and 10% fractional errors on the broad and narrow line scale factors, respectively, the “true” broad line fluxes, relative to a reference spectrum, are generally uncertain by 20%. In Fig. 4.2, the broad Hα lightcurves are shown. ∼ Despite the large uncertainty in the broad Hα fluxes, there are several objects (1E 0450.3, Pictor A, and PKS 0921-213) for which there are obvious variations in the broad Hα flux; the impact of this varying broad Hα flux on the profile shape is explored in 4.3.4. § In the next two sub-sections, I present two different methods for characterizing the variations in double-peaked line profiles in a model-independent way. After the results for each method are presented, I first describe general trends and then highlight results from specific objects. 66 on. The relative fluxes and 4.3.1. § flux as a function of time, relative to a particular observati α H broad Fig. 4.2 Variations in the total their errors were calculated as described in 67

4.3.2 Difference Spectra One simple way to visualize how a line profile is changing with time is to create difference spectra, in which one plots the residuals from some reference spectrum. Here I present difference spectra that result from subtracting either the average or the “minimum” profile. The former is simply the average of all the profiles whereas the latter is constructed by selecting, at each wavelength, the minimum flux from among all the spectra. This minimum spectrum represents a base profile which is common to all the profiles. As an idealized example, if the emission arises from a circular accretion disk with an emissivity enhancement (such as a hot spot or spiral arm), then the minimum profile would be that of the underlying circular accretion disk, and the difference spectra would clearly reveal the emissivity enhancement as it travels through the disk and affects different portions of the emission line profile. Before constructing the difference spectra, the original spectra were renormalized to the same total broad line flux (see 4.3.1). This procedure eliminates the effects of § large changes in the total illumination (such as those observed in 3C 390.3 due to light reverberation Dietrich 1998; Shapovalova et al. 2001) that would obscure changes in the profile shape. The spectra were also resampled to the wavelength scale of the spectrum with the highest spectral resolution (typically 3A),˚ so they could be subtracted pixel- by-pixel. Then the two “comparison” spectra were subtracted from each individual spectrum. The resulting difference spectra are shown in Fig. 4.3. Because the spectra were rescaled to have similar broad-line fluxes, the narrow emission lines fluxes vary from one observation to the next. Rather than remove the narrow lines, which would require assumptions about the shape of the underlying profile, these regions were simply not plotted in the difference spectra and the omitted regions are indicated by vertical gray stripes. As can be seen in Fig. 4.3, the prominent features in the difference spectra are generally broader than the omitted spectral regions, thus these gaps do not greatly impair the interpretation of the difference spectra. There are several trends commonly seen in the difference spectra of most objects. In general, trends are best illustrated by the differences from the minimum spectrum.

1. The difference spectra reveal the presence of distinct lumps of emission, and typically there are multiple lumps, sometimes at both negative and positive projected velocities, at any given time. The amplitude of these lumps, relative to the flux in the minimum spectrum at the position of the lump, is typically 20–30% although when the lump is located at the wings of the profile, the relative flux is much larger ( 50%). An exception is Pictor A, in which the spectrum has undergone ∼ significant changes in profile shape; the flux in the “lumps” of emission is several times larger than in either the minimum or the average spectrum. The changes in these distinct lumps of excess emission, either in shape, amplitude, or projected velocity, are primarily responsible for the variability in the line profiles.

2. Neither the average nor, more importantly, the minimum spectra are consistent with the expected proﬁle from a circular accretion disk, because the red peak is often as strong as, or stronger, than the blue peak. If the proﬁle variability can indeed be explained as an underlying disk with extra, transient lumps of emission, 68

Fig. 4.3 Difference between each individual spectrum and either the average or minimum spectrum, as described in 4.3.2. The average or minimum profile is shown on the top, § and the difference spectra are stacked below. Please note that the time interval between each spectrum is not uniform. The vertical gray stripes indicate regions of the spectrum occupied by narrow emission lines which are not being plotted for clarity. The dotted horizontal line indicates the null flux level for each spectrum. A dotted vertical line in the spectra of PKS 1739+18C indicates the position of the B-band. This figure is continued over the next five pages . . . 69

Figure 4.3 continued . . . 70

Figure 4.3 continued . . . 71

Figure 4.3 continued . . . 72

Figure 4.3 continued . . . 73

Figure 4.3 concluded.

then the underlying disk itself cannot be modeled as an axisymmetric disk and thus the “base” (i.e. minimum) proﬁle can be expected to vary on long timescales. Thus one should be cautious in associating the minimum spectrum with the proﬁle of an underlying “static” disk to which all other features are added.

3. The less luminous systems – Pictor A, PKS 0921–213, and 1E 0450.3–1817, vary on shorter timescales than the more luminous systems that presumably have more massive black holes. The three lowest redshift objects have known black hole 7 masses (M 4 10 M ), determined in Chapter 2, and will be discussed in BH ∼ × more detail in 4.5. § 4. If one assumes that lumps from one observation epoch can be tracked in following observations, several general conclusions can be drawn.

(a) The lumps can persist for several years; in a few objects, some lumps appear to remain throughout the duration of the monitoring observations such as the red lumps in 3C 59 and PKS 1739+18C. These lumps still do vary slightly in amplitude and shape, but they do not dissipate completely. (b) When the lumps are variable, they typically arise on rather short timescales (less than one year) and dissipate more slowly (over several years), generally becoming broader in the process of decaying. In an extreme case, a lump disappeared on timescales of less than a year, and a new one appeared in 74

approximately the same place within another year (e.g. 1E 0450.3–1817, 1995– 1996). (c) The lumps are often observed to drift across the proﬁle from positive to negative projected velocity and only very rarely does the lump drift in the opposite sense, reminiscent of Arp 102B (Sergeev et al. 2000; Gezari 2005).

Pictor A, PKS 0921–213, 1E 0450.3–1817 and CBS 74 all show striking variability which I describe in more detail here.

Pictor A. – This object shows the most dramatic variability, in the sense that the line was not observed to be double-peaked prior to 1994. The blue peak originated at a projected velocity of 7500 km s−1, implying that ∼ the emitting gas is located at characteristic radius (rc) of less than 1500 rg [(r /r )−1/2 v/c]. This peak steadily drifted to less negative projected ve- c g ∼ locity (∆v 3000 km s−1) over a period of approximately 4 years and has p ∼ been dissipating slowly over at least 7 years (the blue peak is still present in the last observation epoch). It is interesting to note that the red peak emerged earlier than the blue peak, and has all but disappeared at the last observation epoch. PKS 0921–213. – The profile of this object in the early observations is very unusual and appears to have two double-peaked components, one with more widely separated peaks underlying another with more closely spaced peaks. This latter component appears to have mostly disappeared within a year. After this, the most prominent feature is a lump in the red peak, at a projected velocity of 4500 km s−1 (implying a characteristic radius of less than ∼ 4000 rg) that strengthened over two years and also began to dissipate over a similar timescale. Additionally during this time, the lump drifted towards lower projected velocity by 2000 km s−1. ∼ 1E 0450.3–1817. – This object is unique in the large number of distinct lumps which have appeared, drifted across the profile, and dissipated; during some observation epochs there are up to three different lumps present at once. The lumps emerged at both small and large (as well as negative and positive) projected velocities. Some of the lumps are long-lived and do not vary significantly in strength, shape, or position over a period of 8 years, whereas ∼ other lumps emerge, drift, and dissipate over 4 years. CBS 74. – A lump located in the red side of the profile drifted by 2000 km s−1 (projected velocity) and dissipated over approximately two years. A second broad feature on the blue side of the profile steadily increased in strength and then began to decrease again over 5 years. The rapid variability in this object may be related to the fact that the inner radius of the disk extends down to 200 r . Although the black hole mass is probably larger in this object than ∼ g in Pictor A, PKS 0921–213, and 1E 0450.3–1817, the physical timescales for variability may be similar. 75

In the above descriptions, I have assumed that one can track a lump from one epoch in subsequent observations. This does not imply that the same parcel of gas is responsible for the emissivity enhancement and in fact this cannot be the case, at least for PKS 0921–213 and 1E 0450.3–1817. The dynamical timescales for these objects are known to be only a few months (see Chapter 2) and each of these objects were closely monitored at times (with gaps of only 1–2 months between observations); there was clearly no orbital motion. Thus, these lumps of emission should not be likened to the bright spot postulated by Newman et al. (1997). The fact that the properties of the lumps (projected velocity, width, and amplitude) are not randomly distributed from one observation to the next strongly suggests some kind of physical connection between these lumps, but the lumps are associated with a location in the disk, rather than any particular parcel of matter within the disk. I offer a possible interpretation of this phenomenon in 4.5 § 4.3.3 Variations in Profile Parameters As shown in 4.3.2, difference spectra are powerful tools for diagnosing the root § cause of the observed variability, such as an emissivity enhancement that is traveling across the profile or the sudden dissipation of a feature. However, it is also desirable to characterize the data in a more quantitative way. In particular, even if a given family of models is able to qualitatively reproduce the observed profiles, there is a large region of parameter space to be explored in order to find the best-fitting model for a sequence of profiles; a quantitative description of the profile variations will be of great utility when detailed modeling is performed. Although the profiles are extremely complex, to first order they can be reduced to a small set of easily measurable quantities: the velocity shifts of the red and blue peaks, the ratio of the blue to red peak flux, the full-width of the profile at half-maximum and quarter maximum (FWHM and FWQM), the velocity separation of the two peaks, the velocity shifts of the profile centroid at FWHM and FWQM, and the average velocity shift of the peaks. The maximum is defined to be the flux of the higher peak and all velocities are measured with respect to the narrow Hα emission line. The resulting plots of profile parameters as a function of time are shown in Fig. 4.4. In determining the profile parameters for a given spectrum, there are three major sources of uncertainty. First and foremost is the error in measuring the various parameters. The velocity shifts of the red and blue peaks and the shifts of the profile at FWHM and FWQM can be quite uncertain. For example, when a peak is flat-topped, such as the blue peak of PKS 1739+18C, there is considerable ambiguity in the location of the peak. In objects where the peak fluxes are relatively low, or the [S ii]λλ6716, 6731 and [O i]λ6363 lines are strong, the profile at the half- and/or quarter-maximum is contaminated by these narrow lines, making it difficult to determine the width and especially the shifts of the profile. To take into account this uncertainty, each profile property was measured three times, the values were averaged and the standard deviation in the measurements was assigned as an error bar. In some objects (1E 0450.3–1817, PKS 0921– 213), and CBS 74), there appear to be multiple peaks in a small number of spectra and in all cases the stronger peak (which was also had the less extreme projected velocity) 76 was adopted rather than measuring the positions of both peaks and assigning extremely large errors to the peak position and flux ratio. In addition, continuum subtraction introduces additional uncertainty in the measured profile properties, particularly in the FWHM and FWQM. It was not practical to explore the effect of continuum subtraction for each spectrum individually. Instead, the effects of continuum subtraction were explored in great detail for two representative spectra, one from PKS 0921–213, which has a large starlight fraction, and one from PKS 1739+18C, which has a negligible contribution from starlight. The errors incurred from continuum subtraction were very similar for both of these objects, and this extra error was adopted for the other spectra, for which only a set of three measurements was made. As a consistency check, for a few spectra from each object the local continuum subtraction was performed several times to ensure that the error from the continuum subtraction was similar to what was estimated from the test cases. When the error for a particular object was larger than in the test cases, the error bars were increased accordingly. Additionally, a few objects were observed multiple times within a few days, thus it was also possible to estimate the errors that are induced by different observing conditions. This error is relatively small compared to those from continuum subtraction and the measurement process, but not negligible. When multiple spectra were obtained within one month of each other, the results were averaged together, and in these instances it was not necessary to factor in this latter error. The variability plots reveal a few additional trends, which were not readily apparent from the difference spectra presented in 4.3.2. § 1. Blueshifts in the profile of the FWHM, FWQM and the average peak velocity are quite common, and the blueshifts can be as large as 2000 km s−1, although they are more typically 500 km s−1. There is also considerable variability in these ∼ shifts.

2. With the exception of the proﬁle widths at FWHM and FWQM, the various parameters rarely vary in concert. In some objects the peak separation varies in a similar way as the proﬁle widths, but not in all instances.

3. It is not uncommon for the red peak to be stronger than the blue peak. With the exception of CBS 74, for which the Doppler boosting of the blue peak is quite strong, all of the objects in this sample have a red peak that is stronger than the blue in at least 50% of the observations! I note that due to the non-uniform sampling, this does not imply that the red peak is stronger than the blue 50% of the time. More puzzling is that in some cases the peak reversal is extreme, with the red peak being 20-30% stronger than the blue. The most striking instances of a peak reversal are PKS 0921-213 and Pictor A, in which the blue peak has at best achieved a ﬂux equal to that of the red peak. In the context of a disk model, a rather dramatic asymmetry is required to produce such a strong peak reversal. If this asymmetry is persistent, then objects which exhibit these strong red peaks should, at some epoch exhibit, extremely large ratios of the blue to red peak ﬂux, especially considering that any asymmetry in the emissivity will be Doppler boosted on the blue side. These objects may simply have not been 77

Fig. 4.4 Variations in the profile properties: (1) the velocity shift of the red and blue peak, (2) the ratio of the blue to red peak flux ratio, (3) the full-widths at half and quarter maximum and the peak separation (denoted by open red triangles, filled blue circles, and filled black stars, respectively) (4) and the velocity shift of the profile at the FWHM, FWQM and the average peak velocity offset. All velocities are referenced with respect to the narrow Hα emission line. The measurement procedure is described in 4.3.3. This figure is continued on the next two pages . . . § 78

Figure 4.4 continued . . . 79

Figure 4.4 concluded.

observed over a long enough time period to observe large blue peak ﬂuxes yet, however the lack of strong blue peaks is very curious.

4.3.4 Variations with Integrated Broad Hα Flux As mentioned in 4.3.1, variations in the illuminating flux might lead to variations § in the line profiles. The accretion disk itself is not truncated at the inner and outer radii, rather these regions do not emit Balmer emission lines efficiently because the gas is over-ionized or not adequately illuminated. Thus the effective inner and outer radii are not rigid and may be influenced by the changes in the illuminating flux. Thus, as the illuminating flux increases, the inner and outer radii might become larger; this will result in a decrease in the peak separation. The profile widths at FWHM and FWQM are also affected, but changes in the peak ratio and the line profile shifts are less pronounced. The errors on the FWHM and FWQM are generally quite large, thus I focused on searching for negative correlations between variations in the peak separation and the broad Hα flux. Such correlations have in fact been observed in NGC 1097 (Storchi-Bergmann et al. 2003), 3C 332 (Gezari 2005), and also possibly 3C 390.3 (Shapovalova et al. 2001). Four of the seven objects (Pictor A, PKS 0921–213, 1E 0450.3–1817, and PKS 1020–103) show variability in the broad Hα flux, and in Fig. 4.5 the peak separation as a function of the broad Hα flux is shown for these objects. 80

Fig. 4.5 Variations in the velocity separation of the red and blue peaks as a function of the relative broad Hα ﬂux, as determined in 4.3.1, for Pictor A, PKS 0921–213, § 1E 0450.3–1817, and PKS 1020–103. Each data point is labeled with the observation number. In some instances two observations overlap, and only one data point is visible. 81

As can be seen in Fig. 4.5, there is no clear correlation between variations in the peak separation and the broad Hα flux. In many instances there are dramatic variations in either the peak separation or flux, without a corresponding variation in the other. For example, the most dramatic variations in the peak separation in Pictor A occur from 1994 to 1997, when the broad Hα flux is nearly constant. In PKS 0921–213, the flux decreases by a factor of two from 1995 March to 1997 March, and it is suggestive that the profile changes dramatically over this time period. However, the peak separation actually increases over this interval. As described in 4.3.2 § above, the initial spectrum of PKS 0921-213 (1995 March) may actually have two double- peaked components, one with a lower flux level and broader peaks and a second with narrower peaks, the latter of which seems to disappear over the course of the next several observations. If this is indeed the case, the peaks which should be measured are those of the persistent, lower-flux component, not those of the transient component, as has been done. Therefore the peak separations determined for the initial several spectra may be incorrect. If one considers only the last five observations, there is a hint of a negative trend between the broad Hα flux and the peak separation. However, this is a very tentative conclusion, and the correlation could be spurious, particularly in light of the fact that this phenomenon was not noticed in any of the other objects. The variations in the flux of the illuminating continuum most likely have impacted the shape of the profile in PKS 0921–213, but the simple model in which the inner radius of the disk increases with flux probably does not apply.

4.4 Model Proﬁle Characterization

In this section I characterize the profile variability of two families of models—an elliptical disk and a circular disk with a spiral emissivity perturbation—which have been used in the past to model the profile variations in some double-peaked emitters (see 4.1.) Although these models are simple, there is strong theoretical motivation for each; § they are not simply convenient ways to “parameterize” the observed profile variability. Thus before describing the details of the profile calculation ( 4.4.2) and the variability § patterns for these models ( 4.4.3), I set the stage by describing the mechanisms which § could lead to the formation of elliptical disks and disks with spiral arms.

4.4.1 Physical Motivation Elliptical Disk – An elliptical accretion disk could form due to the presence of a second supermassive black hole, analogous to the formation of an elliptical disk in some CVs due to perturbations from the companion star (see the discussion of Eracleous et al. 1995). Because galaxies (and in particular the massive early-type galaxies that typically host radio galaxies) are thought to form through mergers, it is likely that some galaxies contain binary black holes pairs (Begelman, Blandford, & Rees 1980). These authors estimated that for conditions similar to those found in the cores of massive ellipticals, the black holes in the two merging galaxies would quickly form a “hard” state binary. The binary will remain hard for a long time (108–1010 years) because eventually the stars in the central regions of the galaxy 82

that are capable of interacting with the binary, leading to further hardening, will be depleted. The timescale for hardening is then controlled by the rate at which stars refill the loss cone, i.e. the relaxation time. This idea has been recently confirmed by large N-body simulations by Makino & Funato (2004). Due to the continuous perturbation by the second black hole, the accretion disk around the primary will quickly attain a uniform eccentricity and maintain that eccentricity for some time. Alternatively, if the accretion disk is only perturbed temporarily by a massive body, the inner regions of the disk will begin to circularize due to differential precession. A second formation mechanism for an elliptical disk suggested by Eracleous et al. (1995), inspired by the sudden appearance of double-peaked emission lines in the LINER NGC 1097, is the tidal disruption of a star by a black hole. For a black 8 hole with a MBH . 10 M , a solar-mass star will be tidally disrupted before it is accreted. This tidal debris initially has very eccentric orbits, and Syer & Clarke (1992) find that the debris will form an eccentric accretion disk within a viscous timescale.

Spiral Arms. – The second model I consider is that of a circular disk with a spiral emissivity perturbation which is an attractive model for numerous reasons. Firstly, spiral arms are directly observed in galactic disks and indirectly CVs (see, e.g., Steeghs, Harlaftis, & Horne 1997; Baptista, Harlaftis, Marsch, & Steeghs 2005; Papadimitriou, Harlaftis, Steeghs, & Niarchos 2005). Spiral waves are thought to form in CVs due to tidal interactions with the binary companion whereas in self-gravitating galaxy disks they form due to gravitational instabilities. In AGN accretion disks, only the outer portions (ξ 1000 r ) are self-gravitating. However ≥ g the passage of a massive star cluster, a second supermassive black hole, or even the effects of a triaxial stellar distribution could induce the formation of a spiral arm. Chakrabarti & Wiita (1993, 1994) first suggested that spiral shocks might be a major contributor to the variability in the double-peaked emission lines in CVs and AGNs (Arp 102B and 3C 390.3), respectively. Secondly, spiral arms can play a role in angular momentum transport (see, e.g, Livio & Spruit 1991; Matsuda et al. 1989). In particular, other proposed mechanisms for angular momentum transport (i.e. “viscosity” and hydromagnetic winds) may be ineffective in the outer disk and an alternative mechanism such as a spiral arm may be necessary. Finally, Chakrabarti & Wiita (1993) suggest that spiral shocks in an accretion disk may contribute to the photometric variability commonly observed in AGNs because the spiral shocks fragment and reform, naturally leading to the formation of transitory “hot spots” on the disk.

In all of the scenarios presented above, either the eccentric disk or the spiral arm will precess on long timescales, leading to variability in the observed line proﬁle. The timescales over which the variability is expected is discussed further in 4.5. § 83

4.4.2 Calculation of the Model Profiles The symmetric double-peaked emission lines seen in CVs result from the red- and blueshifting of the emission line arising from different portions of the rotating accretion disk. When the emission from the entire disk is considered, there is comparatively little gas traveling with very small or very large projected velocity, and most of the emission arises from gas that is moderately red- or blueshifted. The line profile from an AGN accretion disk is affected by several relativistic effects. These include (1) Doppler boosting of the blue peak relative to the red peak, (2) gravitational redshifting, which shifts the entire line profile to longer wavelengths and can cause the red wing of the line to become extended and (3) light bending, that leads to a solid angle (and thus flux) that changes with the radius in the disk and the inclination of the disk with respect to the the observer. Although these relativistic effects make it more difficult to compute the line profile, they effectively break the degeneracies between the inner and outer radius and the inclination angle that are problematic in the study of CVs. A formalism for calculating the double-peaked emission line profiles in the weak- field limit was laid out by Chen et al. (1989) and Chen & Halpern (1989). The line profile is computed numerically by integrating:

3 f ξdϕ dξ I (ξ,ϕ,ν) D (ξ, ϕ) Ψ(ξ, ϕ) (4.1) ν ∝ ν Z Z over the surface of the disk. Here, ϕ is the azimuthal angle in the frame of the disk and ξ is the distance from the central black hole in units of rg. The emissivity of the disk and information about the local line profile are given in Iν(ξ,ϕ,ν). The velocity structure of the disk is contained in D(ξ, ϕ), the Doppler factor, and gravitational redshifting and light bending are incorporated through Ψ(ξ, ϕ). It is important to note that the weak- field approximations made in this calculation are invalid for ξ . 100 (see, e.g., Fabian et al. 1989). The codes used to implement both the elliptical and spiral-arm models for the line profiles are the same as those used by Storchi-Bergmann et al. (2003). For an elliptical disk, the Doppler factor was recomputed by Eracleous et al. (1995) to account for the non-circular orbits. The specific intensity is given by:

−q 2 1 ε0ξ (νe ν0) Iν = exp − 2 (4.2) 4π √2πσ "− 2σ # where the exponential term accounts for local broadening of the emission line due to turbulence and the factor of ξ−q accounts for the decrease in incident ﬂux from the illuminating source. The value of q lies between 1–3 (see, e.g. Chen & Halpern 1989; Dumont & Collin-Souﬀrin 1990b; Strateva et al. 2003; Eracleous & Halpern 2003). The elliptical disk model used here has seven parameters: ξ1 and ξ2, the inner and outer pericenter of the line-emitting region, the inclination angle i (i 0 for a face-on disk), the ≡ broadening parameter σ, and the eccentricity which is described with three parameters— e1, e2, and ξe. The disk has an eccentricity e1 from ξ1 to ξe and from ξe to ξ2, the eccentricity increases linearly from e1 to e2. Thus a wide range of disks can be described, from disks with constant eccentricity to those in which the inner regions have circularized. 84

Time evolution of the proﬁles via precession of the elliptical disk is achieved by varying the angle that the major axis makes to the line of sight (φ). The spiral arm model was implemented by introducing an emissivity perturbation with the shape of a spiral arm, given by:

0 A 4 ln2 2 A 4 ln2 2 I (ξ, φ)= I (ξ) 1+ exp (φ ψ0) + exp (2π φ + ψ0) (4.3) ν ν 2 δ2 − 2 δ2 − 0 where Iν (ξ) is given by Eqn. 4.2 and

log ξ/ξsp ψ0 = φ0 + . (4.4) tan p Here ξ1 and ξ2 represent the inner and outer radius and q, i and σ are the same as above. The spiral perturbation introduces several additional parameters: A, the amplitude of the arm (i.e. the contrast relative to the underlying disk); the pitch angle, p; the angular width of the arm, δ; and the termination radius of the arm, ξsp (the arm is launched from the outer edge of the disk). Again time variability is achieved by varying the viewing angle φ. In CVs, pairs of spiral arms are often observed, however Gilbert et al. (1999) found that the presence of multiple spiral arms led to a disk which was too symmetric to reproduce the peak reversals observed in 3C 332 and 3C 390.3. Moreover, a single armed spiral is preferred in a disk that has a Keplerian rotation curve in the absence of an external tidal perturbation (Adams, Ruden, & Shu 1989). In principle, one should alter the Doppler factor to account for the perturbation to the velocity ﬁeld as well, as was done by Chakrabarti & Wiita (1994). Using time-resolved spectroscopy and eclipse mapping of the CV IP Pegasi, Baptista et al. (2005) found that the velocities along the two spiral arms (which are located at diﬀerent distances from the center) are lower than the Keplerian velocity by up 10-15% in the distant arm and 40% lower in the closer arm.

4.4.3 Model Characterization Rather than fitting these models directly to the observed profiles, I have characterized extensive libraries of these model profiles in exactly the same way as for the data, namely the construction of difference spectra and plots of the profile parameters as a function of time. A few representative examples from the library of models are shown in Fig. 4.6. In each, the primary disk parameters (ξ1,ξ2,i,σ) were chosen to specifi- cally match some of the objects being studied here to facilitate a comparison. However, the other properties were chosen simply for illustration. Only the differences from the minimum spectra are shown. The elliptical disk model leads to surprisingly similar difference spectra and patterns in the profile parameter variations over a wide range of model parameters, thus only one example is shown. The amplitude of the variations are certainly dependent on specific model parameters, but the general trends are not. On the other hand, the spiral arm models exhibit more diverse variability patterns, and a larger number of examples are shown. 85 , s, ines, Right: . . . were chosen simply for illustration. e ξ law illumination index, and broadening parameter were rofile shifts) are represented by solid, dashed, and dotted l e eccentricity and Differences from the minimum spectrum at various precession el; for a description of the models and their input parameter identical to those in Fig. 4.4. Here the profile widths at FWHM spiral arm model on the next two pages Center: Grayscale image of the disk emissivity. 4.4. The inner and outer radii, inclination angle, and power § Left: FWQM, and the peakrespectively. separation This (as figure well is as continued the with corresponding examples p for the Plot of profile parameters as a function of precession phase, please see chosen to approximate the profile of PKS 0921–213, however th Fig. 4.6 phases. The model parameters are listed at the top of this pan 86 213 gure perties of the spiral arm affect the variability behavior. Fi isk parameters chosen to approximate the profile of PKS 0921– . . . continued on the next page and spiral arms parameters chosen to demonstrate how the pro Figure 4.6 continued. Both models shown here have circular d 87 een chosen simply as examples of the variability behavior. sk parameters chosen to approximate PKS 1739+18C (top) and Figure 4.6 concluded. The models shown here have circular di CBS 74 (bottom) and as before the spiral arm parameters have b 88

Both the difference spectra and the variability plots reveal that there are some important differences between these two models. However, before highlighting these, I note that both models very naturally produce reversals in the peak flux ratio for a wide range of input parameters. Blue-shifting of the profiles at certain precession phases are also not uncommon, but not as ubiquitous as the peak reversals.

Difference spectra – see Fig. 4.6 (center panels). The difference spectra for the spiral arm models, as expected, reveal a lump of excess emission that travels through the profile. When the arm has a small pitch angle, at certain phase angles the line of sight passes through the arm at two different locations, leading to two lumps of emission at different velocities. However, in general there is a single lump of emission and the difference spectra are quite simple. Additionally, this lump of emission broadens and sharpens depending on whether the line of sight passes across or along the spiral arm. The difference spectra for the elliptical disk models are comparatively complex. This is not surprising as the velocity field itself has been altered. There are always two lumps of emission, one blue-shifted and another red-shifted, which vary in strength with time. Each of these lumps sometimes fragments into two sub-lumps to form a double-peaked structure.

Variability plots – see Fig. 4.6 (right panels). When considering the variability plots, it is the elliptical disk model which yields extremely simple variability patterns. The following common behaviors were noticed. All of the profile parameters, with the exception of the peak separation, vary nearly in concert (i.e. the minima and max- ima co-incide). Every parameter, except the profile width at FWHM and FWQM, has exactly one minimum and maximum per precession phase, and the profile variations occur smoothly and symmetrically. The profile widths and sometimes the shifts at FWHM and FWQM undergo very little change in comparison to the peak separation and the shift in the average velocity of the peaks. In the example shown in Fig. 4.6, the eccentricity increased linearly from 0 to 0.15. If e=0.15 throughout the disk, the variation in the peak flux ratio is much more significant, but the variability patterns are qualitatively similar to those shown in Fig. 4.6 On the other hand, the profile properties of the spiral arm model vary in a non- uniform manner. A parameter can remain nearly constant and then undergo a significant change over a small fraction of the precession period; this is best exem- plified by the velocities of the red and blue peaks. This is not unexpected because the spiral arm is localized in projected velocity space. The profile widths, FWHM and FWQM, are the only parameters that vary at similar times and by the same magnitude. Although the shifts in the profile centroids at FWHM and FWQM tend to vary at roughly the same time, they can vary by different magnitudes. All of the models shown are for spiral arms with an amplitude of two to facilitate a comparison; when the amplitude is increased, the variability patterns are not qualitatively different, but the amplitude of the variations increases. It must be noted that there are certainly instances, particularly when the pitch angle and/or arm width is large, that the spiral arm model predicts smoothly varying profile parameters that are more similar to those seen from the elliptical 89

disk models; smooth variability does not preclude a spiral arm but non-uniform variability does eliminate the possibility that the accretion disk is elliptical unless additional perturbations are added.

This characterization has shown that the spiral arm and elliptical disk models are (in most instances) clearly distinguishable. Furthermore, in the case of the spiral arm model, the variability plots are very sensitive to the input model parameters; therefore plots of this type will indeed be extremely useful in quickly pinpointing which spiral arm parameters are most likely to reproduce a sequence of observed proﬁles.

4.5 Discussion and Interpretations

Now that the data and the two simple models have been fully characterized, it is possible to assess the viability of these two models. Both of these models are successful, at least qualitatively, in the sense that the peak reversals are commonly seen in the data are naturally reproduced without fine-tuning. The elliptical disk model is somewhat more successful in producing extreme peak reversals than the spiral arm model. The elliptical disk more easily produces profiles that can be blue-shifted (since the variations are always symmetric); however, neither model naturally reproduces the very large blueshifts of more than 1000 km s−1 that are observed in nearly every object on occasion. Each model enjoys individual successes and failures as well. The presence of multiple lumps of excess emission in the difference spectra, both at positive and negative projected velocity, is consistent with the elliptical disk model. The spiral arm model can only produce multiple lumps that are both red- or blueshifted at specific lines of sight through the spiral arm. On the other hand, the lack of coordinated variability between the profile parameters in the observed data certainly disfavors the elliptical disk model but could be explained by disks with spiral arms. Despite the various differences in profile variability, the most significant difference between these two models is the pattern precession timescale. A spiral arm will precess on timescales that are several times to an order of magnitude longer than the dynamical timescale, with the sound crossing timescale as an extreme upper limit (Laughlin & Korchagin 1996). As described in Eracleous et al. (1995), an elliptical disk can precess due to two effects, the precession of the pericenter due to relativistic effects, which occurs on a timescale of

3 5/2 PG 10 M8ξ˜ yr (4.5) R ∼ 3 or in the case of a binary, also due to the tidal forces of the secondary

3 q4 3/2 −1/2 Ptidal 500 a17 M8 yr (4.6) ∼ 1+ q4 ! 8 where M8 is the mass of the black hole(s) in units of 10 M , ξ˜3 is the pericenter distance 3 in units of 10 rg, q4 is the mass ratio divded by four, and a17 is the binary separation in units of 1017 cm. Equation 4.5 is only applicable to an accretion ring (i.e. when the ratio of the outer to inner radius is small) because in a large disk (and in the absence of 90 a constant perturbing force) the inner disk will circularize before the outer disk has had an opportunity to precess. For three objects in this study, Pictor A, PKS 0921-213, and 1E 0450.3-1817, the 7 black hole masses are known (M 4 10 M ), giving dynamical and sound crossing BH ∼ × timescales of 1–4 months and 15–40 years respectively (where the spread is the result ∼ of a 50% uncertainty in the black hole mass). Thus the precession timescale for a spiral arm is on the order of a decade, whereas for an elliptical disk it is 400 years! The fact that clear variability is observed within one decade in these objects strongly suggests that the elliptical disk model is untenable. Although black hole masses have not been obtained for the other objects in this study, these objects are unlikely to have black hole masses that are significantly smaller than these and it would appear that the elliptical disk model may not be tenable for any objects in this sample. At face value, it would appear that neither of these models will be able to reproduce the observed variability; however, both of these models represent the simplest extension to the circular disk. There is much room for improvement and it is possible that simple modifications to the above models might yield better agreement with the observed profile variability. Here, I focus on the spiral arm model, which at least based on the variability timescale, is still a viable scenario. One simple modification which has already been used successfully in NGC 1097 Storchi-Bergmann et al. (2003) is to allow the emissivity power-law index to vary with both radius and time. As described in 4.3.4, no trends between the peak separation and broad Hα flux were found and such a § modification is not justified for these objects. In the detailed descriptions of the observed difference spectra, I noted that while it was implausible that the lumps of emission were due to bright spots which were orbiting in the disk, the non-random distribution of lumps in projected velocity strongly suggested that the various lumps were somehow physically associated with each other. Consider a disk with a spiral arm; from the point of view of a parcel of gas that is orbiting in the disk, this spiral arm is a nearly stationary shock that must be traversed. The gas parcel is temporarily heated and compressed, and if it is slightly denser than the material around it, it will appear as a temporary bright spot. However, after traversing the arm (if it has not been destroyed) the gas parcel will continue to orbit and perhaps intersect the spiral arm on the next orbit. Now, if instead of a single clump of gas, an azimuthal stream of gas encounters the shock front, a series of bright spots will appear, which might give the appearance of a persistent bright spot at a fixed projected velocity. Over time, the spiral arm will precess, and the location of the bright spots will shift. This scenario can naturally explain the variety of behaviors of the lumps in the difference spectra, with some seeming to persist for years and others disappearing and reappearing. Certainly all of the gas will be slightly heated as it passes through the arm, but the emissivity will be dominated by the dense parcels of gas, leading to a clumpy spiral arm whose radial profile changes with time. Furthermore, the situation of having more than one lump of emission, on both sides of the profile, can be accommodated. Although Gilbert et al. (1999) found that multiple spiral arms led to a disk which was too axisymmetric to reproduce the observed peak reversals, this was for a uniform spiral perturbation, not one in which only isolated bright spots appeared. In this scenario, there is no observational obstacle to the presence 91 of multiple spiral arms, as generally seen in the CVs. However, as pointed out in 4.4.2 § the one-armed spiral is preferred in a disk with Keplerian orbits in the absence of a perturbing force (Adams et al. 1989), thus this scenario requires at least a temporary external perturbation. There are some significant problems that this clumpy spiral arm model does not remedy, however, namely the fact that the lumps rarely drift from the blue side of the profile to the red, which should occur half of the time in a spiral arm model, the large blueshifts that are observed at times, and the fact that most objects have red peaks stronger than the blue at least 50% of the time. The modification described above is in some ways a simple “fix” to the spiral arm model. The greater utility of the detailed characterization of the data presented here is the ability to quickly compare almost any model with the observations in a flexible, semi-quantitative way. There are numerous disk-based models which can be envisioned, such as radiation-induced warps in an accretion disk (see, e.g., Pringle 1996). Popović, Mediavilla, & Muñoz (2001) have suggested that gravitational microlensing may play a role in the profile variability in AGN broad lines. This mechanism would not lead to persistent bright spots such as those observed here however, so this model is probably not generally applicable. All of these models, including even the basic models characterized here, have several free parameters. In the case of the clumpy spiral arm model, there is even more flexibility in the model parameters and it will be more difficult to constrain and test the model. Thus it would be essential to use Monte Carlo simulations, as done by Sergeev et al. (2000) (who were using a similar complex model to describe the profile variations in Arp 102B), to determine the likelihood of a random configuration reproducing the observed variability trends. With long, and well-sampled time-series of data, a clumpy spiral arm model can be well-tested, because the bright spots must trace out the path of the precessing spiral arm(s). Such generalizations of the disk model open up a Pandora’s box on one hand, but on the other hand, they appear necessary in order to describe the observed data. The application of such detailed models will be truly useful when applied to large numbers of objects. Only then can one progress beyond describing in minute detail the variability of one particular object and begin to learn about the physics taking place in accretion disks that leads to variability in the AGN BELR in general. Thus it is necessary to monitor a large number of objects, however it is also necessary to continue monitoring a smaller number of objects very frequently and over long time periods to test the viability of the various models.

4.6 Conclusions

In this Chapter I have characterized, in a model-independent way, the variability of the broad, double-peaked Hα emission lines in seven objects, namely Pictor A, PKS 0921–213, 1E 0450.3–1817, CBS 74, 3C 59, PKS 1739+18, and PKS 1020–103 which have been monitored over the past decade. The variability in these objects is caused primarily be excess lumps of emission, which change in morphology, amplitude, and projected velocity with time, generally on timescales of less than five years, but some lumps of emission remain unchanged for nearly duration of the observations. Those objects that are less luminous or have accretion disks that extend to small radii (r 200 r ) exhibit ∼ g variability on shorter timescales. There are often multiple lumps of emission observed 92 at a single epoch and they are generally located at both positive and negative projected velocities. These lumps appear to drift very slowly from one observation epoch to the next. For some objects the dynamical timescale is known to be only a few months, thus these lumps cannot be orbiting brights spots such as those used to model Arp 102B (Newman et al. 1997; Sergeev et al. 2000). The most striking profile variations observed are changes in the ratio of the blue to red peak, thus this trend which has long been observed in some of the more well- studied double-peaked emitters seems to be very common. In fact, with the exception of CBS 74, all of the objects in this study have a red peak that is stronger than the blue in at least 50% of the observations. Some objects in this sample are very extreme, most notably PKS 0921–213 and Pictor A, in that the blue peaks, which are supposed to be boosted due to relativistic effects, are rarely observed to be significantly stronger than the red peak. I then compared these variability trends with those expected from two simple models, an elliptical accretion disk and a circular disk with a single-armed spiral emissivity perturbation. In general, neither of these models reproduces the observed variability trends in detail; spiral arm models do not predict the presence of multiple lumps of emission at a single epoch and the elliptical disk model predict profile variability that is extremely smooth, uniform, and symmetric, which is not observed. From a considera- tion of physical timescales, at least for the three objects with a known black hole mass (Pictor A, PKS 0921–213, and 1E 0450.3–1817) the spiral arm model is able to produce variability on a reasonable timescale. Thus I proposed a simple alteration to the spiral arm model in which one or more clumpy spiral arms are present in the accretion disk. This model retains many of the general theoretical characteristics of the simple, uniform spiral arm studied in this chapter but is likely to produce the observed variability more successfully. Finally, I note that significant variability appears to be occurring on timescales of less than a year. In particular lumps of emission were seen to disappear and reappear within a year, or at least change significantly in shape or amplitude, and it is very likely that much of the variability is being missed by the current observing strategy. It would be very useful to monitor all objects at least twice per year, and to intersperse periods of intense monitoring (perhaps as often as every few weeks) for a few of the more variable objects such as 1E 0450.3–1817 or PKS 0921–213. In particular, to test the clumpy spiral arm model, and determine how long an individual bright spot remains “turned-on” such short-timescale observations are absolutely necessary. 93

Chapter 5

Conclusions and Suggestions for Future Work

In this thesis, I set out with two goals. The first goal was to test the hypothesis that the inner accretion flows in double-peaked emitters have the form of a radiatively inefficient accretion flow (RIAF). The presence of such a structure would serve the dual purposes of providing a source of external illumination to the outer disk and diminishing the accretion disk wind. The second goal was to characterize the variability of seven double-peaked emitters that have been systematically monitored over the past decade and assess several currently used models in light of these observations.

5.1 Viability of the RIAF scenario

In Chapter 2, I demonstrated that the accretion rates in double-peaked emitters, relative to the Eddington accretion rate, vary widely. Of the nine objects presented, −2 seven had LBol/LEdd . 10 and the inner accretion disks in these objects may be some variety of RIAF. I note that the radial extent of the RIAF in the two objects with L /L 10−3–10−2 are likely to be smaller than in the other objects. However, Bol Edd ∼ the two remaining objects have L /L 0.1, implying accretion rates that are Bol Edd ∼ a significant fraction of the Eddington accretion rate. In Chapter 3, I modeled the spectrum of 3C 111 and found that a model in which the inner disk had the form of a RIAF best explained the weakness of the reprocessing features. On the other hand, the BLRG 4C 74.26 was convincingly shown by Ballantyne & Fabian (2005) to have an inner disk that extended to small radii and in that object, the weakness of the reprocessing features was the result of the high ionization of the gas in the disk. However, I note that 4C 74.26 has unusually strong reprocessing features and may not be representative of BLRGs as a class. These results suggest that BLRGs (the parent population of the Eracleous & Halpern 1994 sample of double-peaked emitters) and double-peaked emitters themselves are a “mixed bag”; some objects have low accretion rates compatible with a RIAF and others have much larger accretion rates. Thus, it is clearly no longer valid to assume that the inner accretion flows in all double-peaked emitters have the form of a RIAF. In the objects studied in Chapter 2, three out of the five BLRGs (1E 0450.3–1817, IRAS 0236.6-3101, and Arp 102B) and naturally all of the LINERs have accretion rates compatible with a RIAF. However, this sample is small and it is not yet possible to determine what fraction of double-peaked emitters contain a RIAF, particularly since black hole masses have not been determined for objects in the more heterogeneous SDSS sample. As I described in 1.2.3, the absence of a RIAF in a double-peaked emitter does § not necessarily invalidate the basic hypothesis that an AGN exhibits clear double-peaked emission lines only when the contribution of the disk-wind is decreased. 94

I suggest that future eﬀorts should be directed towards investigating alternative mechanisms that diminish the contribution of the disk-wind to the broad emission lines in double-peaked emitters. Ideally, this alternative mechanism should also provide a means for illuminating the outer disk. One possibility is that the accretion rates in PKS 0921– 213 and Pictor A are large enough for a radiation-supported torus to form. This structure is vertically extended, however it is unknown whether the accretion disk wind in such a system would be diminished in any way. Additionally, robust black hole masses have only been determined for nine double-peaked emitters; obtaining masses for 30, and ∼ preferably more, double-peaked emitters would allow one to determine whether there are any trends between the accretion rates and other properties of the double-peaked emitters. In particular, it would be interesting to determine whether there is a continuous distribution of accretion rates or whether double-peaked emitters tend to have either low or high accretion rates. Additional X-ray observations of BLRGs would be useful as well and XMM Newton and RXTE data have been obtained for the double-peaked − emitter PKS 0921–213, although the data were not obtained simultaneously. However in my opinion, more progress will be made by obtaining more black hole masses (although study of the X-ray spectra of BLRGs are interesting in their own right.) There are six objects from the Eracleous & Halpern (1994) sample with z . 0.06 which can be observed from the northern hemisphere and more importantly black hole masses should be obtained for objects from the Strateva et al. (2003) sample. Additionally, one could expand the redshift range greatly by using the Mg ib λλ5167, 5173, 5184 triplet instead of the near-IR Ca ii triplet. However, the faintness of objects with z > 0.1 dictates the use of an 8m class telescope if a useful number of objects are to be observed.

5.2 Characterization of the Long-term Proﬁle variability

The seven objects studied here all showed variability over the decade-long observation campaign and those objects that are less luminous or have accretion disks which extend to small radii (r 200r ) vary on shorter timescales than other objects. As ∼ g observed in other well-studied double-peaked emitters, the most prominent variability occurs in the ratio of the blue to red peak flux. The line profile variability is primarily driven by the presence of lumps of excess emission which change in morphology, amplitude, and projected velocity on timescales of less than five years. Neither of the two commonly used models (elliptical accretion disks and disks with spiral arms) can satisfactorily reproduce the variability trends observed in these objects. In addition, based on the black hole masses for the double-peaked emitters that have been determined thus far, the elliptical disk model predicts variability on timescales of 100s of years, which is completely inconsistent with the observations. I suggested in 4.6 that if the spiral arm § were clumpy, many of the observed variability patterns could be explained. Significant progress could be made by increasing the frequency with which objects are monitored, at least for objects such as 1E 0450.3–1817 and PKS 0921–213. These objects were each monitored intensely (several times per year, but only for a few years) and significant variability was observed. Furthermore, the dynamical timescales in these objects are now known to be only a few months. With the current observing strategy, variations occurring on the dynamical timescale have not even been probed in most 95 objects. However, one cannot loose sight of the long-term profile variability and I suggest that these objects should be observed at a minimum twice a year for several more years. This long-term monitoring should be coupled with periods of a few months in which the objects are observed every few weeks. In particular, the “clumpy” spiral-arm model predicts that bright spots will turn on and off, but remain in nearly the same position. It is important to verify that we are not seeing a bright spot in the disk that is orbiting, as in the models of Sergeev et al. (2000) or Newman et al. (1997). This can only be done if the objects are monitored several times within a dynamical timescale. Furthermore, it would be worthwhile to place limits on how long the bright spots remain active and the frequency with which they appear and disappear. It is also important to start monitoring a subset of the objects from the heterogeneous Strateva et al. (2003) sample to verify that similar variability trends occur in these objects. Finally, much work remains to be done in the modeling of these systems. Many models have not even been tried yet, such as a warped disk (Pringle 1996) or spiral arm models in which the velocity field is perturbed. The results from this study suggest that gravitational microlensing, which was suggested as a possible source of broad line variability in AGNs (Popovićet al. 2001) is not generally applicable because the bright spots persist at roughly the same projected velocity for several years.

5.3 Accretion Disk Winds in Double-Peaked Emitters

Finally, I would brieﬂy like to mention that the UV spectra of double-peaked emitters were fundamental to the development of the idea that the disk-winds in double- peaked emitters are weaker than in other systems. Only six objects have been observed, and to verify this general idea and to make further progress, it is important to extend these studies to more objects. This can be done by future UV observations of low- redshift double-peaked emitters and especially those from the Strateva et al. (2003) sample, perhaps with the Cosmic Origins Spectrograph that will (hopefully) be placed on the Hubble Space Telescope. Alternatively, one could ﬁnd high redshift double-peaked Mg ii emitters in the SDSS and obtain follow-up J, H, and K band IR spectroscopy to observe the Balmer emission lines. 96

Appendix A

Telluric Correction Method

Removal of the telluric water vapor absorption lines was an essential component of the data reductions because, with the exception of NGC 1097, at least one of the Ca ii absorption lines in each target galaxy was embedded in the telluric water vapor lines at wavelengths longer than 8900 A.˚ To illustrate this I show in Fig. A.1 the uncorrected spectrum of IRAS 0236.6 –3101 in which all three Ca ii lines are embedded in the telluric water vapor bands. To create a template of the atmospheric transmission, it is common to observe an object with a nearly featureless continuum in which any sharp features can be attributed to atmospheric absorption. The atmospheric transmission changes not only with position in the sky and airmass, but also throughout the night. Because I wished to observe telluric standards several times each night, I opted to observe rapidly rotating B-stars, with rotational velocities in excess of 200 km s−1. Although these stars are not as featureless as white dwarfs, they have the advantage of being more common, allowing us to ﬁnd a telluric standard in a similar region of the sky as my targets. More importantly, they are very bright, requiring only 30s of exposure. ∼

Fig. A.1 Average of the IRAS 0236.6-3101 spectra in the absence of telluric correction. (The spectral resolution is 1.35A˚ FWHM.) The positions of the Ca ii triplet lines are marked. For comparison, the ﬁnal spectrum obtained when the telluric correction is performed can be seen in Fig. 2.1. 97

Despite the rotational broadening, the Hydrogen Paschen lines were still narrow enough that they could not be fitted as part of the continuum as shown in Fig. A.2. I modeled the intrinsic spectrum of the B-star as a combination of a low-order (n< 4) polynomial and a set of Paschen absorption lines represented with Voigt profiles. The best-fit Voigt profile parameters were determined by fitting the interval from 8650– 8875 A,˚ a portion of the B-star spectrum which had only three relatively un-blended Paschen lines and was free from telluric absorption lines. Restricting the profile parameters to narrow ranges around the best-fit values, the B-star spectrum was fitted over 8725–9300 A,˚ which included four Paschen lines, two of which were fitted in the previous step. A sequence of rejection iterations was performed, in which the upper and lower rejection levels became progressively smaller to force the fit to model the upper envelope of the spectrum. An example of this fit is shown in Fig.A.2. The B-star spectrum was then divided by this model fit to generate a template of the telluric absorption lines. Despite the rejection iterations, the residual template still exceeded unity in near the red end of the spectrum, particularly in the interval from 9195–9215A.˚ Therefore a first- order polynomial was fitted post facto to the envelope of the template to ensure that all values were less than unity. Nevertheless this renormalization, I suspect that the telluric template is overestimed in this region and I exercise caution when using the template over this range of wavelengths. The telluric template was set to unity at wavelengths less than 8900 A.˚ Of the six rapidly rotating B-stars observed (HD 23227, HD 34863, HD 43445, HD 58127, HD 65622, HD 75869), the Paschen lines were well fitted by the simple Voigt profile only in HD 34863, for which the residuals at wavelengths less than 8900 A˚ were less than 1%. This object was also the most rapidly rotating (v sin i = 370 kms−1, Glebocki & Stawikowski 2000). The Paschen lines in the other objects deviated by less than 4% from the simple model, with the exception of HD 58127, which turned out to be a Be star, but the residuals were systematic and would have induced unacceptably large features into the spectra they were used to correct. Unfortunately, HD 34863 was only observed on the first night. However, because the humidity was so low (25–35%) and (more importantly) stable throughout the run, I found that I could successfully correct the spectra from all five nights using a single template, derived from HD 34863. I tested the success of this template by applying it to the other B-stars that were observed throughout the observing run; the root mean square (RMS) deviation in the corrected B-stars was less than 3% and sometimes less than 1%, which is negligible compared to RMS deviation in individual exposures of the galaxies. To correct the galaxy spectra, the spectrum was divided by the telluric template derived from HD 34863, allowing for a small relative shift between the galaxy and template. The telluric absorption lines are not resolved in these data and it is possible that the lines are so severely blended that they form a pseudo-continuum that would not be corrected for by my method. However, I have verified that this is not the case. A spectrum of the telluric water vapor lines derived from a high resolution (∆λ/λ 150 000) ∼ near-IR spectrum of Arcturus (Hinkle et al. 2000) was convolved by a Gaussian with dispersion of 1.35 A.˚ There was no pseudo-continuum associated with the convolved spectrum and there was excellent agreement between the overall shape of the convolved telluric template and the telluric template used in this paper. 98

Fig. A.2 Top: Spectrum of the rapidly rotating star HD 23683 (histogram) and the best- ﬁtting model (solid line) for the hydrogen Paschen lines. Bottom: The residuals to the ﬁt, which are used as a template of the telluric water vapor lines. Before this template is applied to the galaxy spectra, the values of data points with wavelengths less than 8900 A˚ are set to unity and the template is renormalized with a 1st order polynomial.

The spectrum of HD 34863 was also used to derive a telluric template to correct the spectra of the G and K giant stars from 8160–8400 A.˚ In this region, the B-star spectrum is featureless, and deriving the template was straight forward. This template was applied to the stars using the same method as for the galaxies. 99

Appendix B

Inclination Angle of the Disk in 3C 111

Following the methods of Eracleous et al. (1996), we use the observed superluminal motion in the radio jet of 3C 111 and the projected linear size of the radio lobes to place constraints upon the inclination angle of the jet, and thus the accretion disk. The measured proper motion, µ, in the jet is 1.54 0.2 mas yr−1 (Vermeulen ± & Cohen 1994). The apparent velocity, relative to the speed of light, is given by −1 βapp 47.4 µ z h , where z is the redshift, and h is Hubble’s constant, in units of ' −1 −1 100 km s Mpc . Thus, for 3C 111 βapp = 5.1 0.7, assuming h = 0.7, which implies ◦ ◦ ◦ ± either i< 13 or 10 25.0), the radio lobes have an average intrinsic size of 180 h kpc, with a tail of sources which extends to 500 h−1 kpc. Assuming the radio lobes of 3C 111 have an intrinsic size of 500 h−1 kpc , the inclination angle must be greater than 21.7◦. However, it is possible that 3C 111 is a giant radio galaxy (GRG) and the intrinsic size of its radio lobes could be much larger. There are some enormous giant radio galaxies, such as 3C 236 (3 h−1 Mpc; Nilsson et al. 1993) and NVSS 2146+82 (2 h−1 Mpc; Palma et al. 2000), but these sources are rare and the size distribution of GRGs drops oﬀ rapidly for sizes greater than 1 h−1 Mpc (Schoenmakers et al. 2001). Assuming 3C 111 is not larger than 1 Mpc h−1, then i > 10.6◦. Thus I adopt a conservative range in the inclination angle, 10◦

Appendix C

Total Galactic Hydrogen Column Density Towards 3C 111

The X-ray emission of AGNs are modified by photoelectic absorption by material from within our own Galaxy, that of the host galaxy and from obscuring structures in the AGN itself. To study the absorption intrinsic to the AGN, the absorption from within the Galaxy must be properly accounted for. Unfortunately, 3C 111 is located behind a molecular cloud, which makes estimating the total Galactic column density difficult in two ways: there are significant contributions to the column from molecular hydrogen, which is not probed by 21 cm studies; and the gas in the molecular cloud could be quite clumpy, causing the column density to vary with time. Fortunately, 3C 111 provides a bright radio continuum source which can be used to study the molecular hydrogen in the cloud in great detail through molecular absorption line measurements. The molecular hydrogen column density is 9 1021atoms cm−2 (Bania, Marscher, & Barvainis ≈ × 21 −2 1991). Combined with the atomic hydrogen column, N 3.3 10 atoms cm HI ≈ × (Elvis, Lockman, & Wilkes 1989), this yields a total neutral hydrogen column density of 1.2 1022atoms cm−2. This total column density is in rough agreement with ≈ × 22 −2 N 10 atoms cm , inferred from diffuse IR dust emission measurements made H,tot ≈× by Schlegel, Finkbeiner, & Davis (1998), assuming the Galactic dust-to-gas ratio. Using 3C 111 as a the background continuum source, Marscher et al. (1993) and Moore & Marscher (1995) observed variations in the strength and profile of the 4.83 GHz H2CO absorption line. The line profile variability suggests that the molecular cloud is structured on sub-parsec to AU scales. Marscher et al. (1993) modeled the line profiles as a superposition of many dense clumps of gas. Assuming that the variations in the column density and the line profile are due to fluctuations in the number of clumps that lie along the line of sight to 3C 111. They estimated that there are 30–100 clumps along the line of sight, with column densities of 2 1020atoms cm−2. Therefore, the total Galactic ≈ × 21 −2 column density towards 3C 111 can be expected to vary by several 10 atoms cm 22 −2 × from the measured value of 1.2 10 atoms cm . × 101 Bibliography

Adams, F. C., Ruden, S. P., & Shu, F. H. 1989, ApJ, 347, 959

Anders, E. & Ebihara, M. 1982, Geochim. Cosmochim. Acta, 46, 2363

Anders, E. & Grevesse, N. 1989, Geochim. Cosmochim. Acta, 53, 197

Arnaud, K. A. 1996, in ASP Conf. Ser.101, Astronomical Data Analysis Software and Systems V (San Francisco:ASP), 17

Bade, N., Engels, D., Voges, W., Beckmann, V., Boller, T., Cordis, L., Dahlem, M., Englhauser, J., Molthagen, K., Nass, P., Studt, J., & Reimers, D. 1998, A&AS, 127, 145

Balbus, S. A. & Hawley, J. F. 2003, Lecture Notes in Physics Vol. 614: Turbulence and Magnetic Fields in Astrophysics, 614, 329

Baldwin, J. A. & Stone, R. P. S. 1984, MNRAS, 206, 241

Ballantyne, D. R. & Fabian, A. C. 2005, ApJL, 622, L97

Ballantyne, D. R., Fabian, A. C., & Iwasawa, K. 2004, MNRAS, 354, 839

Ballantyne, D. R., Iwasawa, K., & Fabian, A. C. 2001, MNRAS, 323, 506

Ballantyne, D. R., Ross, R. R., & Fabian, A. C. 2002, MNRAS, 332, L45

Bania, T. M., Marscher, A. P., & Barvainis, R. 1991, AJ, 101, 2147

Baptista, R. Morales-Rueda, L., Harlaftis, E. T., Marsch, T. R., & Steeghs, D. 2005, A&A, in press, astro-ph/0507188

Barth, A. J., Ho, L. C., Filippenko, A. V., Rix, H., & Sargent, W. L. W. 2001, ApJ, 546, 205

Barth, A. J., Ho, L. C., & Sargent, W. L. W. 2002, AJ, 124, 2607

Begelman, M. C., Blandford, R. D., & Rees, M. J. 1980, Nature, 287, 307

Beloborodov, A. M. 1999, ApJL, 510, L123

Bianchi, S., Matt, G., Balestra, I., Guainazzi, M., & Perola, G. C. 2004, A&A, 422, 65

Boller, T., Meurs, E. J. A., Brinkmann, W., Fink, H., Zimmermann, U., & Adorf, H.-M. 1992, A&A, 261, 57

Bower, G. A., Wilson, A. S., Heckman, T. M., Magorrian, J., Gebhardt, K., Richstone, D. O., Peterson, B. M., & Green, R. F. 2000, Bulletin of the American Astronomical Society, 32, 1566 102

Bower, G. A., Wilson, A. S., Heckman, T. M., & Richstone, D. O. 1996, AJ, 111, 1901

Brinkmann, W., Siebert, J., & Boller, T. 1994, A&A, 281, 355

Brinkmann, W., Siebert, J., Reich, W., Fuerst, E., Reich, P., Voges, W., Truemper, J., & Wielebinski, R. 1995, A&AS, 109, 147

Chakrabarti, S. K. & Wiita, P. J. 1993, A&A, 271, 216

—. 1994, ApJ, 434, 518

Chen, K. & Halpern, J. P. 1989, ApJ, 344, 115

Chen, K., Halpern, J. P., & Filippenko, A. V. 1989, ApJ, 339, 742

Chiang, J. & Murray, N. 1996, ApJ, 466, 704

Collin-Souﬀrin, S. 1987, A&A, 179, 60

Collin-Souﬀrin, S. & Dumont, A. M. 1989, A&A, 213, 29

—. 1990, A&A, 229, 292

Corbin, M. R. & Smith, P. S. 2000, ApJ, 532, 136 de Kool, M. & Begelman, M. C. 1995, ApJ, 455, 448

Devereux, N., Ford, H., Tsvetanov, Z., & Jacoby, G. 2003, AJ, 125, 1226

Dewangan, G. C., Griﬃths, R. E., & Schurch, N. J. 2003, ApJ, 592, 52

Dietrich, M. e. a. 1998, ApJS, 115, 185

Done, C., Mulchaey, J. S., Mushotzky, R. F., & Arnaud, K. A. 1992, ApJ, 395, 275

Dressler, A. 1984, ApJ, 286, 97

Dumont, A. M. & Collin-Souﬀrin, S. 1990a, A&A, 229, 313

—. 1990b, A&A, 229, 302

—. 1990c, A&AS, 83, 71

Elvis, M. 2000, ApJ, 545, 63

Elvis, M., Lockman, F. J., & Wilkes, B. J. 1989, AJ, 97, 777

Elvis, M., Wilkes, B. J., McDowell, J. C., Green, R. F., Bechtold, J., Willner, S. P., Oey, M. S., Polomski, E., & Cutri, R. 1994, ApJS, 95, 1

Emmering, R. T., Blandford, R. D., & Shlosman, I. 1992, ApJ, 385, 460

Emsellem, E., Greusard, D., Combes, F., Friedli, D., Leon, S., P´econtal, E., & Wozniak, H. 2001, A&A, 368, 52 103

Eracleous, M. 2004, in ASP Conf. Ser. 001, AGN Variability from X-rays to Radio Waves (San Franciso:ASP), 001

Eracleous, M. & Halpern, J. P. 1994, ApJS, 90, 1

Eracleous, M. & Halpern, J. P. 1998, in AIP Conf. Proc. 431, Accretion Power in Astro- physical Systems: Some Like it Hot! (New York:AIP), 261

—. 2003, ApJ, 599, 886

—. 2004, ApJS, 150, 181

Eracleous, M., Halpern, J. P., & Charlton, J. C. 2003, ApJ, 582, 633

Eracleous, M., Halpern, J. P., Gilbert, A. M., Newman, J. A., & Filippenko, A. V. 1997, ApJ, 490, 216

Eracleous, M., Halpern, J. P., & Livio, M. 1996, ApJ, 459, 89

Eracleous, M., Livio, M., Halpern, J. P., & Storchi-Bergmann, T. 1995, ApJ, 438, 610

Eracleous, M., Sambruna, R., & Mushotzky, R. F. 2000, ApJ, 537, 654

Eracleous, M., Shields, J. C., Chartas, G., & Moran, E. C. 2002, ApJ, 565, 108

Everett, J. 2004, in ASP Conf. Ser. 311, AGN Physics with the Sloan Digital Sky Survey (San Francisco: ASP), 147

Fabian, A. C., Nandra, K., Reynolds, C. S., Brandt, W. N., Otani, C., Tanaka, Y., Inoue, H., & Iwasawa, K. 1995, MNRAS, 277, L11

Fabian, A. C., Rees, M. J., Stella, L., & White, N. E. 1989, MNRAS, 238, 729

Fanaroﬀ, B. L. & Riley, J. M. 1974, MNRAS, 167, 31P

Ferrarese, L. & Merritt, D. 2000, ApJL, 539, L9

Filippenko, A. V. 1985, ApJ, 289, 475

Frank, J., King, A., & Raine, D. J. 2002, Accretion Power in Astrophysics: Third Edition (Cambridge University Press, 398 pp.)

Freedman, W. L., Hughes, S. M., Madore, B. F., Mould, J. R., Lee, M. G., Stetson, P., Kennicutt, R. C., Turner, A., Ferrarese, L., Ford, H., Graham, J. A., Hill, R., Hoessel, J. G., Huchra, J., & Illingworth, G. D. 1994, ApJ, 427, 628

Gammie, C. F., Narayan, R., & Blandford, R. 1999, ApJ, 516, 177

Gaskell, C. M. 1996, ApJL, 464, L107+

Gaskell, C. M. & Snedden, S. A. 1999, in ASP Conf. Ser. 175, Structure and Kinematics of Quasar Broad Line Regions (San Francisco: ASP), 157 104

Gavazzi, G. & Perola, G. C. 1978, A&A, 66, 407

Gebhardt, K., Bender, R., Bower, G., Dressler, A., Faber, S. M., Filippenko, A. V., Green, R., Grillmair, C., Ho, L. C., Kormendy, J., Lauer, T. R., Magorrian, J., Pinkney, J., Richstone, D., & Tremaine, S. 2000, ApJL, 539, L13

Gelbord, J. M. 2003, PhD thesis, Johns Hopkins University

George, I. M. & Fabian, A. C. 1991, MNRAS, 249, 352

Gezari, S. 2005, PhD thesis, Columbia University

Gezari, S., Halpern, J. P., Eracleous, M., & Filippenko, A. V. 2004, in IAU Conf. Ser. Volume 222, The Interplay of Black Holes, Stars, and ISM in Galactic Nuclei, ed. T. Storchi-Bergmann, L. C. Ho, & H. R. Schmitt (Cambridge, UK: Cambridge Uni- versity Press), 95

Ghisellini, G., Haardt, F., & Matt, G. 1994, MNRAS, 267, 743

Gilbert, A. M., Eracleous, M., Filippenko, A. V., & Halpern, J. P. 1999, in ASP Conf. Ser, 175: Structure and Kinematics of Quasar Broad Line Regions, ed. C. M. Gaskell, W. N. Brandt, M. Dietrich, D. Dultzin-Hacyan, & M. Eracleous (San Francisco: ASP), 189

Glebocki, R. & Stawikowski, A. 2000, Acta Astronomica, 50, 509

Gliozzi, M., Sambruna, R. M., & Eracleous, M. 2003, ApJ, 584, 176

Goad, M. & Wanders, I. 1996, ApJ, 469, 113

Gondek, D., Zdziarski, A. A., Johnson, W. N., George, I. M., McNaron-Brown, K., Magdziarz, P., Smith, D., & Gruber, D. E. 1996, MNRAS, 282, 646

Goto, T. 2005, MNRAS, 413

Grandi, P. 2001, Mem. Soc. Astron. Italiana, 72, 91

Grandi, P., Maraschi, L., Urry, C. M., & Matt, G. 2001, ApJ, 556, 35

Grandi, P., Sambruna, R. M., Maraschi, L., Matt, G., Urry, C. M., & Mushotzky, R. F. 1997, ApJ, 487, 636

Haardt, F. & Maraschi, L. 1991, ApJL, 380, L51

—. 1993, ApJ, 413, 507

Halpern, J. P. 1982, PhD thesis, Harvard University

—. 1990, ApJL, 365, L51

Halpern, J. P., Eracleous, M., Filippenko, A. V., & Chen, K. 1996, ApJ, 464, 704

Heckman, T. M. 1980, A&A, 87, 152 105

Herrnstein, J. R., Moran, J. M., Greenhill, L. J., Diamond, P. J., Inoue, M., Nakai, N., Miyoshi, M., Henkel, C., & Riess, A. 1999, Nature, 400, 539

Hinkle, K., Wallace, L., Valenti, J., & Harmer, D. 2000, Visible and Near Infrared Atlas of the Arcturus Spectrum 3727-9300 A˚ (San Francisco: ASP)

Ho, L. C. 1999, ApJ, 516, 672

Ho, L. C., Filippenko, A. V., & Sargent, W. L. W. 1997a, ApJS, 112, 315

Ho, L. C., Filippenko, A. V., Sargent, W. L. W., & Peng, C. Y. 1997b, ApJS, 112, 391

Ho, L. C., Rudnick, G., Rix, H., Shields, J. C., McIntosh, D. H., Filippenko, A. V., Sargent, W. L. W., & Eracleous, M. 2000, ApJ, 541, 120

Horne, K. & Marsh, T. R. 1986, MNRAS, 218, 761

Hummel, E., van der Hulst, J. M., & Keel, W. C. 1987, A&A, 172, 32

Iwasawa, K., Fabian, A. C., Reynolds, C. S., Nandra, K., Otani, C., Inoue, H., Hayashida, K., Brandt, W. N., Dotani, T., Kunieda, H., Matsuoka, M., & Tanaka, Y. 1996, MNRAS, 282, 1038

Iyomoto, N., Makishima, K., Matsushita, K., Fukazawa, Y., Tashiro, M., & Ohashi, T. 1998, ApJ, 503, 168

K¨onigl, A. & Kartje, J. F. 1994, ApJ, 434, 446

Kaspi, S., Smith, P. S., Netzer, H., Maoz, D., Jannuzi, B. T., & Giveon, U. 2000, ApJ, 533, 631

Kassebaum, T. M., Peterson, B. M., Wanders, I., Pogge, R. W., Bertram, R., & Wagner, R. M. 1997, ApJ, 475, 106

Kirsch, M. G. F., Altieri, B., Chen, B., Haberl, F., Metcalfe, L., Pollock, A. M. T., Read, A. M., Saxton, R. D., Sembay, S., & Smith, M. J. S. 2004, Proc. SPIE, 5488, in press

Korista, K. T. e. a. 1995, ApJS, 97, 285

Laughlin, G. & Korchagin, V. 1996, ApJ, 460, 855

Lee, J. C., Fabian, A. C., Brandt, W. N., Reynolds, C. S., & Iwasawa, K. 1999, MNRAS, 310, 973

Lewis, K., Eracleous, M., Halpern, J., & Storchi-Bergmann, T. 2004a, in ASP Conf. Ser. 311, AGN Physics with the Sloan Digital Sky Survey, ed. G. T. Richards & P. B. Hall (San Francisco: ASP), 193

Lewis, K., Eracleous, M., Halpern, J., & Storchi-Bergmann, T. 2004b, in IAU Conf. Ser. Volume 222, The Interplay of Black Holes, Stars, and ISM in Galactic Nuclei (Cambridge;New York: Cambridge University Press), 107 106

Lewis, K., Eracleous, M. Gliozzi, M., Sambruna, R., & Mushotzky, R. F. 2005, ApJ, in prep.

Lightman, A. P. & White, T. R. 1988, ApJ, 335, 57

Linﬁeld, R. & Perley, R. 1984, ApJ, 279, 60

Lipari, S. & Macchetto, F. 1992, ApJ, 387, 522

Livio, M. & Spruit, H. C. 1991, A&A, 252, 189

Magdziarz, P. & Zdziarski, A. A. 1995, MNRAS, 273, 837

Makino, J. & Funato, Y. 2004, ApJ, 602, 93

Malkan, M. A. & Sargent, W. L. W. 1982, ApJ, 254, 22

Malzac, J., Beloborodov, A. M., & Poutanen, J. 2001, MNRAS, 326, 417

Markoﬀ, S., Falcke, H., & Fender, R. 2001, A&A, 372, L25

Marscher, A. P., Moore, E. M., & Bania, T. M. 1993, ApJL, 419, L101

Marsh, T. R. & Horne, K. 1990, ApJ, 349, 593

Martel, A. R., Baum, S. A., Sparks, W. B., Wyckoﬀ, E., Biretta, J. A., Golombek, D., Macchetto, F. D., de Koﬀ, S., McCarthy, P. J., & Miley, G. K. 1999, ApJS, 122, 81

Mathews, W. G. & Capriotti, E. R. 1985, in Astrophysics of Active Galaxies and Quasi- Stellar Objects, 185–233

Matsuda, T., Sekino, N., Shima, T., Sawada, K., & Spruit, H. C. 1989, in NATO ASIC Proc. 290: Theory of Accretion Disks (Dordrecht: Kluwer), 355–371

Matsuoka, M., Piro, L., Yamauchi, M., & Murakami, T. 1990, ApJ, 361, 440

Matt, G., Perola, G. C., & Piro, L. 1991, A&A, 247, 25

Matt, G., Perola, G. C., Piro, L., & Stella, L. 1992, A&A, 257, 63

Moore, E. M. & Marscher, A. P. 1995, ApJ, 452, 671

Morrison, R. & McCammon, D. 1983, ApJ, 270, 119

Murray, N. & Chiang, J. 1996, Nature, 382, 789

—. 1998, ApJ, 494, 125

Murray, N., Chiang, J., Grossman, S. A., & Voit, G. M. 1995, ApJ, 451, 498

Mushotzky, R. F., Done, C., & Pounds, K. A. 1993, ARAA, 31, 717

Nagao, T., Murayama, T., Shioya, Y., & Taniguchi, Y. 2002, ApJ, 567, 73 107

Nandra, K., George, I. M., Mushotzky, R. F., Turner, T. J., & Yaqoob, T. 1997, ApJ, 477, 602

Nandra, K. & Pounds, K. A. 1994, MNRAS, 268, 405

Narayan, R., Igumenshchev, I. V., & Abramowicz, M. A. 2000, ApJ, 539, 798

Narayan, R. & Yi, I. 1994, ApJL, 428, L13

—. 1995, ApJ, 444, 231

Nayakshin, S. & Kallman, T. R. 2001, ApJ, 546, 406

Nemmen, R. S., Storchi-Bergmann, T., Eracleous, M., Terashima, Y., & Wilson, A. 2004, in IAU Conf. Ser. Volume 222, The Interplay of Black Holes, Stars, and ISM in Galactic Nuclei (Cambridge;New York: Cambridge University Press), 53

Nemmen, R. S., Storchi-Bergmann, T., Yuan, F., Eracleous, M., Terashima, Y., & Wil- son, A. S. 2005, ApJ, in prep

Newman, J. A., Eracleous, M., Filippenko, A. V., & Halpern, J. P. 1997, ApJ, 485, 570

Nilsson, K., Valtonen, M. J., Kotilainen, J., & Jaakkola, T. 1993, ApJ, 413, 453

Ogle, P. M., Davis, S. W., Antonucci, R. R. J., Colbert, J. W., Malkan, M. A., Page, M. J., Sasseen, T. P., & Tornikoski, M. 2005, ApJ, 618, 139

Oke, J. B. 1987, in Superluminal Radio Sources, ed. A. J. Zensus & T. J. Pearson (Cambridge, UK: Cambridge University Press), 267–272

Oke, J. B. & Gunn, J. E. 1983, ApJ, 266, 713

Palma, C., Bauer, F. E., Cotton, W. D., Bridle, A. H., Majewski, S. R., & Sarazin, C. L. 2000, AJ, 119, 2068

Papadimitriou, C. J., Harlaftis, E. T., Steeghs, D., & Niarchos, P. G. 2005, in Astro- nomical Society of the Paciﬁc Conference Series (San Francisco: ASP), 395

Perez, E., Mediavilla, E., Penston, M. V., Tadhunter, C., & Moles, M. 1988, MNRAS, 230, 353

Perola, G. C., Matt, G., Cappi, M., Fiore, F., Guainazzi, M., Maraschi, L., Petrucci, P. O., & Piro, L. 2002, A&A, 389, 802

Peterson, B. M. 1993, PASP, 105, 247

—. 1997, An introduction to active galactic nuclei (Cambridge, New York Cambridge University Press, 238 p.)

Peterson, B. M., Ferrarese, L., Gilbert, K. M., Kaspi, K., Malkan, M. A., Maoz, D., Merritt, D., Netzer, H., Onken, C. A., Pogge, R. W., Vestergaard, M., & Wandel, A. 2004, ApJ, 613, 682 108

Peterson, B. M., Meyers, K. A., Carpriotti, E. R., Foltz, C. B., Wilkes, B. J., & Miller, H. R. 1985, ApJ, 292, 164

Piro, L., Yamauchi, M., & Matsuoka, M. 1990, ApJL, 360, L35

Popovi´c, L. C.,ˇ Mediavilla, E., Bon, E., & Ili´c, D. 2004, A&A, 423, 909

Popovi´c, L. C.,ˇ Mediavilla, E. G., & Mu˜noz, J. A. 2001, A&A, 378, 295

Pounds, K. A., Nandra, K., Stewart, G. C., George, I. M., & Fabian, A. C. 1990, Nature, 344, 132

Pounds, K. A., Reeves, J. N., Page, K. L., Wynn, G. A., & O’Brien, P. T. 2003, MNRAS, 342, 1147

Poutanen, J. 1998, in Theory of Black Hole Accretion Disks (Cambridge, New York: Cambridge University Press), 100–+

Pringle, J. E. 1996, MNRAS, 281, 357

Pritchet, C. 1978, ApJ, 221, 507

Proga, D., Stone, J. M., & Kallman, T. R. 2000, ApJ, 543, 686

Quataert, E. & Gruzinov, A. 2000, ApJ, 539, 809

Rees, M. J. 1984, ARAA, 22, 471

Rees, M. J., Phinney, E. S., Begelman, M. C., & Blandford, R. D. 1982, Nature, 295, 17

Reeves, J. 2003, in ASP Conf. Ser. 290, Active Galactic Nuclei: from Central Engine to Host Galaxy (San Francisco: ASP), 35

Reynolds, C. S. 1997, MNRAS, 286, 513

Reynolds, C. S. & Fabian, A. C. 1997, MNRAS, 290, L1

Reynolds, C. S., Iwasawa, K., Crawford, C. S., & Fabian, A. C. 1998, MNRAS, 299, 410

Ross, R. R. & Fabian, A. C. 1993, MNRAS, 261, 74

Ross, R. R., Fabian, A. C., & Young, A. J. 1999, MNRAS, 306, 461

Salpeter, E. E. 1964, ApJ, 140, 796

Sambruna, R. M., Eracleous, M., & Mushotzky, R. F. 1999, ApJ, 526, 60

Schlegel, D. J., Finkbeiner, D. P., & Davis, M. 1998, ApJ, 500, 525

Schoenmakers, A. P., de Bruyn, A. G., R¨ottgering, H. J. A., & van der Laan, H. 2001, A&A, 374, 861

Sergeev, S. G., Pronik, V. I., Peterson, B. M., Sergeeva, E. A., & Zheng, W. 2002, ApJ, 576, 660 109

Sergeev, S. G., Pronik, V. I., & Sergeeva, E. A. 2000, A&A, 356, 41

Shakura, N. I. & Sunyaev, R. A. 1973, A&A, 24, 337

Shapiro, S. L., Lightman, A. P., & Eardley, D. M. 1976, ApJ, 204, 187

Shapovalova, A. I., Burenkov, A. N., Carrasco, L., Chavushyan, V. H., Doroshenko, V. T., Dumont, A. M., Lyuty, V. M., Vald´es, J. R., Vlasuyk, V. V., Bochkarev, N. G., Collin, S., Legrand, F., Mikhailov, V. P., Spiridonova, O. I., Kurtanidze, O., & Nikolashvili, M. G. 2001, A&A, 376, 775

Shields, G. A. 1977, ApJL, 18, 119

Shields, J. C., Rix, H., McIntosh, D. H., Ho, L. C., Rudnick, G., Filippenko, A. V., Sargent, W. L. W., & Sarzi, M. 2000, ApJL, 534, L27

Spergel, D. N., Verde, L., Peiris, H. V., Komatsu, E., Nolta, M. R., Bennett, C. L., Halpern, M., Hinshaw, G., Jarosik, N., Kogut, A., Limon, M., Meyer, S. S., Page, L., Tucker, G. S., Weiland, J. L., Wollack, E., & Wright, E. L. 2003, ApJS, 148, 175

Stauﬀer, J., Schild, R., & Keel, W. 1983, ApJ, 270, 465

Steeghs, D., Harlaftis, E. T., & Horne, K. 1997, MNRAS, 290, L28

Still, M. D., Dhillon, V. S., & Jones, D. H. P. 1995, MNRAS, 273, 863

Stirpe, G. M., de Bruyn, A. G., & van Groningen, E. 1988, A&A, 200, 9

Stocke, J. T., Liebert, J., Gioia, I. M., Maccacaro, T., Griﬃths, R. E., Danziger, I. J., Kunth, D., & Lub, J. 1983, ApJ, 273, 458

Stone, R. P. S. & Baldwin, J. A. 1983, MNRAS, 204, 347

Storchi-Bergmann, T., Baldwin, J. A., & Wilson, A. S. 1993, ApJL, 410, L11

Storchi-Bergmann, T., Eracleous, M., Livio, M., Wilson, A. S., Filippenko, A. V., & Halpern, J. P. 1995, ApJ, 443, 617

Storchi-Bergmann, T., Eracleous, M., Ruiz, M. T., Livio, M., Wilson, A. S., & Filip- penko, A. V. 1997, ApJ, 489, 87

Storchi-Bergmann, T., Nemmen, R. S., Spinelli, P. F., Eracleous, M., Wilson, A. S., Filippenko, A. V., & Livio, M. 2005, ApJL, 624, L13

Storchi-Bergmann, T., Nemmen da Silva, R., Eracleous, M., Halpern, J. P., Wilson, A. S., Filippenko, A. V., Ruiz, M. T., Smith, R. C., & Nagar, N. M. 2003, ApJ, 598, 956

Strateva, I. V., Strauss, M. A., Hao, L., Schlegel, D. J., Hall, P. B., Gunn, J. E., Li, L., Ivezi´c, Z.,ˇ Richards, G. T., Zakamska, N. L., Voges, W., Anderson, S. F., Lupton, R. H., Schneider, D. P., Brinkmann, J., & Nichol, R. C. 2003, AJ, 126, 1720 110

Sulentic, J. W., Marziani, P., Zwitter, T., & Calvani, M. 1995, ApJL, 438, L1

Syer, D. & Clarke, C. J. 1992, MNRAS, 255, 92

Tanaka, Y., Nandra, K., Fabian, A. C., Inoue, H., Otani, C., Dotani, T., Hayashida, K., Iwasawa, K., Kii, T., Kunieda, H., Makino, F., & Matsuoka, M. 1995, Nature, 375, 659

Tremaine, S., Gebhardt, K., Bender, R., Bower, G., Dressler, A., Faber, S. M., Fil- ippenko, A. V., Green, R., Grillmair, C., Ho, L. C., Kormendy, J., Lauer, T. R., Magorrian, J., Pinkney, J., & Richstone, D. 2002, ApJ, 574, 740

Turner, T. J., Mushotzky, R. F., Yaqoob, T., George, I. M., Snowden, S. L., Netzer, H., Kraemer, S. B., Nandra, K., & Chelouche, D. 2002, ApJL, 574, L123

Vaughan, S., Fabian, A. C., Ballantyne, D. R., De Rosa, A., Piro, L., & Matt, G. 2004, MNRAS, 351, 193

Veilleux, S. & Zheng, W. 1991, ApJ, 377, 89

Vermeulen, R. C. & Cohen, M. H. 1994, ApJ, 430, 467

Wanders, I., Goad, M. R., Korista, K. T., Peterson, B. M., Horne, K., Ferland, G. J., Koratkar, A. P., Pogge, R. W., & Shields, J. C. 1995, ApJL, 453, L87+

Wanders, I. & Peterson, B. M. 1996, ApJ, 466, 174

Wang, T.-G., Dong, X.-B., Zhang, X.-G., Zhou, H.-Y., Wang, J.-X., & Lu, Y.-J. 2005, ApJL, 625, L35

Weaver, K. A., Krolik, J. H., & Pier, E. A. 1998, ApJ, 498, 213

Wilms, J., Reynolds, C. S., Begelman, M. C., Reeves, J., Molendi, S., Staubert, R., & Kendziorra, E. 2001, MNRAS, 328, L27

Worthey, G. 1994, ApJS, 95, 107

Wozniak, P. R., Zdziarski, A. A., Smith, D., Madejski, G. M., & Johnson, W. N. 1998, MNRAS, 299, 449

Wu, X. & Liu, F. K. 2004, ApJ, 614, 91

Yaqoob, T. & Padmanabhan, U. 2004, ApJ, 604, 63

York, D. G. e. a. 2000, AJ, 120, 1579

Zdziarski, A. A. & Grandi, P. 2001, ApJ, 551, 186

Zdziarski, A. A., Johnson, W. N., Done, C., Smith, D., & McNaron-Brown, K. 1995, ApJL, 438, L63

Zdziarski, A. A., Lubinski, P., & Smith, D. A. 1999, MNRAS, 303, L11 111

Zheng, W., Sulentic, J. W., & Binette, L. 1990, ApJ, 365, 115

Zheng, W., Veilleux, S., & Grandi, S. A. 1991, ApJ, 381, 418

Zycki, P. T., Krolik, J. H., Zdziarski, A. A., & Kallman, T. R. 1994, ApJ, 437, 597 Vita Karen Theresa Lewis Education The Pennsylvania State University University Park, Pennsylvania 1999–2005 Ph.D. in Astronomy & Astrophysics, December 2005 Astronomy & Astrophysics University of Wisconsin – Madison Madison, WI 1995–1999 B.S. in Physics (with Honors) and Mathematics Awards and Honors National Research Council Postdoctoral Fellowship 2005 NASA Graduate Student Research Program Fellowship 2002–Present TA of the Year, Department of Astronomy & Astrophysics 2002 Pennsylvania Space Grant Consortium Fellowship 2000–2003 Zaccheus Daniel Foundation for Astronomical Science Grant 2000–2002 Braddock Fellowship, Eberly College of Science 1999–2001

Research Experience Doctoral Research The Pennsylvania State University 2002–2005 Thesis Advisor: Prof. Michael Eracleous The main focus of this project is to test models for dynamical phenomena in AGN accretion disks using the long-term profile variability of the double-peaked Balmer disk emission lines observed in some AGNs. Graduate Research The Pennsylvania State University 1999–2001 Research Advisor: Prof. Michael Eracleous The optical emission lines properties of Weak-Line Radio Galaxies were used to constrain the properties of the accretion flow in these objects. Graduate Research The Pennsylvania State University 1999–2001 Research Advisor: Dr. David Burrows I analyzed a Chandra ACIS observation of the young supernova remnant N103B. I also assisted the ACIS team at Penn State with efforts to characterize and mitigate the effects of Charge Transfer Inefficiency in the ACIS CCDs. Teaching Experience Certificate in College Teaching The Pennsylvania State University 2005 I attended a ten-week Course in College Teaching offered by the Schreyer Institute for Teaching Excellence at Penn State University which focused on various aspects of teaching pedagogy as well as practical issues such as course planning. Teaching Assistant The Pennsylvania State University 1999,2001–2002 I taught an Astronomy Laboratory for non-majors. As part of this I developed several new labs to explore Kepler’s Laws, the Inverse Square Law, Navigation, Stellar Evolution, and the search for Extra-Solar Planets.