Linking the Power Sources of Emission-Line Galaxy Nuclei from the Highest to the Lowest Redshifts

A dissertation presented to the faculty of the College of Arts and Sciences of Ohio University

In partial fulﬁllment of the requirements for the degree Doctor of Philosophy

ANCA CONSTANTIN

has been approved for the Department of Physics and Astronomy and the College of Arts and Sciences by

Joseph C. Shields Associate Professor of Physics and Astronomy

Leslie A. Flemming Dean, College of Arts and Sciences CONSTANTIN, ANCA. Ph.D. August 2004. Physics

Linking the Power Sources of Emission-Line Galaxy Nuclei from the Highest to the Lowest Redshifts (185pp.)

Director of Dissertation: Joseph C. Shields

This dissertation searches for common grounds for the diversity of properties exhibited by the emission-line nuclei of galaxies, from large look-back times to the local universe. I present results of (1) a program of high signal-to-noise spectroscopy for > 44 z ∼ 4 quasars using the MMT and Keck observatories; (2) a detailed analysis of the ultraviolet and optical spectral behavior of 22 Narrow Line Seyfert 1 (NLS1) galaxies based on archival Hubble Space Telescope (HST) spectra; (3) an in-depth > investigation of the proposed link between NLS1s and z ∼ 4 quasars, by means of comparison of composite spectra, and a Principal Component Analysis; (4) a simulation of Seyferts/quasars designed to explore the role of dust in modifying their observed spectral energy distribution; and (5) a sensitive search for accretion signatures in a large sample of nearby emission-line galaxy nuclei, employing a quantitative comparison of the nebular line ﬂux ratios in small (HST) and large (ground-based) apertures. The low and high redshift quasars are found very similar in their emission characteristics, although diﬀerences exist. In particular, the data bolster indications of > supersolar metallicities in the luminous, z ∼ 4 sources, which support scenarios that assume substantial star formation concurrent or preceding the quasar phenomena. Because high-z sources are more metal enriched and more spectroscopically heterogeneous than the NLS1s, a close connection between these objects remains doubtful. The results show that NLS1s have redder UV-blue continua than those measured in other quasars and Seyferts. The sources with UV line absorption are in general less powerful and show redder spectra, indicating that a luminosity-dependent dust absorption may be important in modifying their continua. A receding-torus –like geometry seems to explain these trends and other observed correlations between quasar luminosity and continuum slope. Finally, in most of the nearby emission-line nuclei, the expected increased AGN-like behavior at smaller scales is not seen, although the nuclear emission is resolved. This suggests that these sources are not necessarily powered by accretion onto a compact object, and that the composite model proposed for the LINER/H II transition nuclei (that assumes a central accreting-type nucleus surrounded by star-forming regions) is not generally supported.

Approved

Joseph Shields Associate Professor of Physics and Astronomy To my parents, Ludmila and Timis Pana Acknowledgments

Finishing up this work and writing it down took a lot of teaching, learning, and... a little playing. I am the one who learned, mostly. There are, however, quite a few people that made this process not only possible, but also smooth, engaging and gratifying, and I want to thank them, especially for putting up with me for all these years. My greatest salutation goes to my advisor, Joe Shields, for giving me the opportunity to work on such interesting projects, and for all his support, sage counsel and patience. Most of all, I am hopeful that I learned to be at least half as patient and tactful as he was with me. Joe, you rank 1st in my role model teacher list. (I’ll let you know of any change, albeit very unlikely. In the meantime, I count on your advice.) Special thanks go to my dissertation committee members, especially for their thoughtful reading of my work, and extend to all physics and astronomy faculty at Ohio University, who, through their catching enthusiasm and genuine comity, have been constant sources of encouragement. Would I be so lucky as to plunge in similarly friendly environments in my future jobs? I really hope so, but I’ve heard it is hard to find such places... Speaking of places, and people: Don Roth, you are above words of appreciation for being there when I needed you. “Herschel” was good to me, however, much better after you “talked” to “him” first, especially about IRAF and its accompanying bugs. Of course, many warm thanks are for my fellow astro and physics colleagues, from whom I benefited invaluable intellectual stimulation, technical assistance, and of course comfort that comes from true friendship. Robert, we have spent three years of enchanting “office mate-ism.” Gabi and Andi, you had great influence (mostly good) on me, starting with choosing Ohio University as grad school, and with your help and hospitality that I so much needed in my first year in Athens, that will never be forgotten. Flori, God sent you in my way to show me what good souls are like. Bassem, you believed in me and showed me how “simple” things (particularly the codes) are. Steven, your “on deck” presence for everybody is highly appreciated. I felt honored to be trusted with your drafts. Yurii, you are so unique, I’m looking forward to seeing you again. Mangala, it has been so easy to talk to you. David and Laura, I didn’t hear you much, but your silent presence remained sensed. Keep up the good work and try to let others know about it. Swati, Manasvita, Justin, Zack, and the whole astrogroup, you have been such an enjoyable team; the wonderful greetings (with the flowers) sent to me on that January 5th meant a huge deal. May your life in the LAIR be full of adventure. Suhita and Aparna, there was always too little time spent with you, especially lately. You made India come to me. My WIPHA allies, to which I gladly add Ruth, Heather, Ennice, Karen, and Tracy, your patronage, care, and generosity overwhelmed me. The many others, whose names I will forgo mentioning here, you collectively filled up my years in Athens with zing.

Cristoi, I owe you my greatest debt, for everything we’ve been (and will be) through together. Your passion for life continuously inspires me. Above all, thanks Mia for choosing us, and for appearing at the right moment, with the right smile. I know you would answer with all the candor of your heart: “moecome.”

My first two years of graduate education were supported through a teaching assis- tantship appointment from Ohio University. Support for my fifth year was provided through the John Cady Graduate Fellowship. For the rest of the time spent in grad school, I benefited from financial assistance from Prof. Shields (and Prof. Statler, for one summer, if I recall correctly) through external grants. The Ohio University Graduate Student Senate is acknowledged for partial funding awarded to attend the 203rd Meeting of the American Astronomical Society, held in January, 2004, in At- lanta, Georgia. The financial burden of many other meetings, summer schools and workshops was generously supported by Prof. Shields through his grants. Contents

Abstract 4

Acknowledgements 7

List of Figures 11

List of Tables 17

1 Introduction 18 1.1 Rationale ...... 18 1.2 ActiveGalacticNuclei ...... 19 1.2.1 The Nucleus Types and their Oddities ...... 20 1.2.2 The Central Power Source of Low Luminosity Nuclei . . . . . 22 1.3 LimitationsofPreviousWork ...... 24 1.4 Thiswork ...... 26

2 Emission-Line Properties of z > 4 Quasars 27 2.1 Introduction...... 27 2.2 Observations&dataanalysis ...... 28 2.3 Spectroscopicproperties ...... 39 2.3.1 Generalcharacteristics ...... 39 2.3.2 OIandNV...... 45 2.3.3 The EW vs. Luminosity correlation ...... 47 2.3.4 Selectionbiases ...... 51 2.4 Discussion...... 54 2.5 Conclusions ...... 57

3 Ultraviolet and optical properties of Narrow-Line Seyfert 1 galaxies 59 3.1 Introduction...... 59 3.2 The NLS1 sample & data processing ...... 61 3.3 TheNLS1compositespectra...... 68 3.3.1 Overallcontinuum ...... 69

9 10

3.3.2 ReddeningandAbsorption...... 72 3.3.3 TheUVspectrum...... 75 3.3.4 Theopticalcomposite ...... 75 3.3.5 EmissionLines ...... 77 3.4 Are NLS1s the analogues of high z QSOs? ...... 88 3.4.1 Direct comparison of composite spectra ...... 90 3.4.2 Principal Component Analysis ...... 93 3.5 Conclusions ...... 97

4 Dust Reddening and the AGN Spectral Energy Distribution 99 4.1 Introduction...... 99 4.2 Themodel:recedingtorus ...... 103 4.2.1 Simulationparameters ...... 104 4.3 Results...... 110 4.3.1 Anoteofcaution ...... 118 4.3.2 The fraction of obscured accretion power ...... 119 4.4 Summary&Conclusions ...... 120

5 The Power Sources of the Low Luminosity Emission-Line Galaxy Nuclei 122 5.1 Introduction...... 122 5.2 The nuclei sample & data processing ...... 125 5.2.1 Thedata...... 125 5.2.2 Measurements of emission-lines ...... 131 5.2.3 New evidence for broad emission ...... 139 5.2.4 General nuclear properties of the sample ...... 144 5.3 Is the nuclear emission resolved? ...... 149 5.4 The nebular excitation and the central engines ...... 152 5.5 Discussion ...... 158 5.5.1 TheremarkablenucleusofNGC4736 ...... 161 5.6 Summary&Conclusions ...... 164

6 Summary 167 6.1 BasicResults ...... 167 6.1.1 Nuclear activity at the highest redshifts ...... 167 6.1.2 Getting closer to home: the NLS1 galaxies ...... 168 6.1.3 Is dust shaping the quasar SED? ...... 168 6.1.4 Emission-line nuclei in the nearby universe ...... 169 6.2 FutureDirections ...... 170

Bibliography 173 List of Figures

1.1 Example of line flux-ratio diagnostic diagrams used in classifying galaxy nuclei in the Palomar survey (adapted from Ho et al. 1997a). H ii nuclei are shown as crosses, Seyferts as filled squares, LINERs as filled circles, and transition objects as open circles...... 21

> −17 2.1 Observed spectra of QSOs at z ∼4. The flux, Fλ, is in units of 10 erg s−1 cm−2 A˚−1. The prominent emission features are marked. . . . 35 > 2.2 Observed spectra of QSOs at z ∼4...... 36 > 2.3 Observed spectra of QSOs at z ∼4...... 37 > 2.4 Observed spectra of QSOs at z ∼4...... 38 2.5 Composite spectra (average -upper panel, and median -lower panel) > of z ∼4 QSOs included in this observational program are shown in comparison with the LBQS composite, representative of z ≈ 2 objects. The y-axis has been normalized to unit mean flux over the wavelength range 1430A–˚ 1470A.˚ The redshift measured from the Civ feature was > used for the Doppler correction of the z ∼4 spectra. As a consequence of the different velocity shifts presented by the low- and high-ionization lines, the alignment of the high-z spectral features with those in the low z composite required a slight additional shift of our composites. 41 2.6 Equivalent widths and line peak/continuum ratios for Civ emission as a function of Civ line widths (full width at half maximum, FWHM), measuredinoursample...... 42 2.7 Upper panel: Flux standard deviation in Fλ relative to the average composite spectrum, as a function of wavelength for the individual spectra comprising the composites. Lower panel: The standard devia- tionofthemean,expressedin%...... 44

11 12

2.8 Rest-frame EW of the Civ emission line as a function of absolute luminosity of the quasar (at 1450 A˚ rest frame). The Baldwin Effect reported by Osmer, Porter & Green (1994), based on measurements at lower redshift, is shown by the dotted line. The upper panel shows measurements from our sample only, with different symbols for the color (×) and grism (•) selected objects. The lower panel adds mea- > surements for z ∼4 QSOs from Schneider, Schmidt & Gunn (1991) and the SDSS survey. Although consistent with the low redshift trend, the measurements do not, by themselves, show the expected correlation. 48 > 2.9 Equivalent width of Lyα + Nv versus continuum luminosity for z ∼4 QSOs. SymbolsarethesameasinFig.2.8...... 49 2.10 Comparison between median and average composites computed for two sets of quasars, grouped according to their luminosity at 1450A˚ (Lν expressed in ergs s−1 Hz−1). The same normalization as in Figure 2.5 is used. A weak drop in EW with luminosity can be noted for Lyα but not for Civ orotherlines...... 50 2.11 The distributions of the rest-frame (upper panels) and observed (lower panels) EWs for Lyα and Civ; three different samples are compared using a KS test criterion which indicates that the distributions are consistent with a common parent population...... 53 2.12 Comparison of composite spectra for color-selected and grism-selected QSOs. The composites include spectra published by SSG in addition to the data presented here. Average and median spectra are shown in the upper and lower panels, respectively...... 55

α 3.1 Histogram of continuum spectral indices, α, where Fν ∝ ν , for all but one (Mrk110) NLS1 in the sample. The average and median values are indicated...... 68 3.2 Top panel: NLS1 composite spectrum plotted as log(Fλ) vs. rest- frame wavelength, with the principal emission features identified. The flux has been normalized to unit mean flux over the wavelength range 1430A–˚ 1470A.˚ Middle panel: The median composite spectrum, plotted on a linear scale, and zoomed near the continuum level for a better visualization of the weak features in the optical range. A more detailed UV line identification is presented in Figure 3.4. The power-law continuum fits are overplotted as dashed and dotted lines. Bottom panel: Number of NLS1s contributing to the composite as a function ofrest-framewavelength...... 71 13

3.3 Spectral indices plotted vs. redshift (upper panel) and 1450A˚ luminos- −1 −1 ity (lower panel), with Lν expressed in ergs s Hz . The Spearman rank coefficient and the probability of the correlation happening by chance are indicated. The error bars in both directions are smaller than the symbol size, and therefore not indicated...... 73 3.4 Top panel: NLS1 average composite spectrum plotted for the UV range only, where at least 18 objects are contributing. The same normalization as in Figure 2.5 is used. The prominent emission features are marked. Middle panel: Flux standard deviation in Fλ relative to the average composite spectrum, as a function of rest-frame wavelength for the individual spectra comprising the composite. Bottom panel: The standard deviation of the mean, expressed in percent...... 76 3.5 The optical range of the HST NLS1 median composite, constructed using only the 3 objects that span the whole spectral range. The same flux normalization as in Figure 3.2 is used. The Sulentic et al. (2002) NLS1 median composite, based on 24 ground-based object spectra, is shown for comparison. The strong similarity between the two medians suggests that the HST NLS1 composites are representative of these objects...... 77 3.6 Emission-line velocity offsets, relative to the rest frame (defined by Civ λ1549), as a function of ionization potential for selected emission lines (see text). Error bars show the 1 σ uncertainty in the velocity measurement. Permitted and semi-forbidden lines are shown in the top panel, and forbidden lines are shown in the bottom panel. Only one measurement is shown for each line, and the point type indicates the adopted profile. Overlapping points are slightly offset horizontally from each other for clarity. The points are labeled by ion...... 87 > 3.7 Direct comparison between NLS1 and z ∼ 4 QSO average composite spectra. The same normalization as in Figure 3.2 is used...... 91 3.8 Rest-frame EW of the Civ emission line, in A,˚ as a function of 1450A˚ −1 −1 > luminosity, in ergs s Hz , for individual NLS1 and z ∼ 4 QSO spectra. The Baldwin relation found by Osmer, Porter & Green (1994) > for a sample of 186 luminous quasars [logLν(1450) ∼ 29, 0 the NLS1 sample (dotted line), the z ∼ 4 QSOs (dashed line), and the combined sample (continuous line). The principal components show similar modulations in the spectral variation in both categories of objects;theamplitudediffers...... 95 14

4.1 Upper panel: Spectral indices measured in the HST NLS1 sample plot- −1 −1 ted vs. 1450A˚ luminosity, with Lν expressed in erg s Hz . Lower panel: Optical spectral indices (measured by Neugebauer et al. 1987) vs. continuum luminosity Lν (1450A),˚ for the PG quasars for which HST-FOS spectra are available in the archive. The filled circles and the asterisks denote objects with and without UV resonance line absorption respectively. The α − L correlation is definitely present in both samples (the Spearman rank coefficients and the probabilities of the correlations to happen by chance are indicated in each plot), along with a trend for the low L objects to be more prone to gas absorption. 102 4.2 Schematic representation of the receding torus model. The predicted behavior of the geometry of the absorbing material is presented for the low-luminosity (top) and the high-luminosity (bottom) objects: a larger inner radius (RS) of the torus produces an increased opening angle (θc) for the more luminous sources, provided the half-height of the torus (h)remainsunchanged...... 105 4.3 Covering factor as a function of intrinsic luminosity for different parameter values of the half-height of the torus (h), expressed in parsecs. Note that, with the chosen parametrization, in this simple picture, most often h ≈ 0.5, and objects less luminous than 1045 erg s−1 are > 53 −1 highly obscured, while luminous sources with L ∼10 erg s are barely affectedbydustabsorption...... 108 4.4 The observed (reddened) power-law slopes plotted against the quasar absorbed monochromatic absolute luminosities measured at 1450A˚ [Lν (1450A),˚ top] and at 4800A˚ [Lν (4800A),˚ bottom]. The results are presented for −3/2 two particle density value sets: nH = Rs (black filled symbols), 3/2 −3/2 and nH = 10 × Rs (orange open symbols). The symbol size is proportional to the quasar intrinsic bolometric luminosity...... 112 4.5 Same α − L plot as in Figure 4.4, for luminosities measured at 1450A˚ −3/2 only, calculated with absorption through dust with nH = Rs , for two different sets of intrinsic SED shapes: open, black symbols correspond to a Gaussian distribution of αintrinsic around α0 = −0.5, while for the filled, blue symbols α0 = +0.5...... 114 4.6 Same α − L plot as in Figure 4.4, for luminosities measured at 1450A˚ −3/2 only, calculated with absorption through dust with nH = Rs , where different symbol/colors are assigned to different quasar orientation ranges: blue, double-lined symbols for the more pole-on objects (θ = 0◦ − 30◦), and filled, orange symbols for the more edge-on sources (θ = 70◦ − 90◦)...... 115 15

4.7 Same α − L plot as in Figure 4.4, for luminosities measured at 1450A˚ (filled symbols) and at 4800A˚ (open symbols), for three different extinction laws, as indicated in each panel: Small Magelanic Cloud type (top), starburst galaxy-like (middle), and AGN-type (bottom). The −3/2 particle density values are given by nH = Rs , in all three cases. . 117 5.1 Ratio of EW of total (narrow+broad) Hα line emission in HST and Palomar apertures, on a logarithmic scale, as a function of nuclear spectral class, as defined by Ho et al. (1997a). The size of each data point is scaled by the respective distance to the host galaxy. The SUNNSdataarenotincludedhere...... 134 5.2 [N ii]+Hα to [S ii] region in NGC3245, NGC4429, NGC4594, NGC4736 and NGC5921. The thin continuous lines show the individual Gaussian components (2 per narrow line), the dashed line represents the broad Hα feature, and the dotted line is the final fit to the observed spectrum, which is presented in thick continuous line. The residuals after the subtraction of the model fit are shown at the bottom of each panel. . 142 5.3 The STIS-HST [N ii] λ 6583/Hα line flux ratio plotted against the galaxy distance. The data points are symbol (and color) coded based on the ground-based classification of their nuclear emission into: H ii systems (red diamonds), transition nuclei (T, green triangles), narrow- lined LINERs (L2, blue squares), broad-line LINERs (L1, cyan crosses), and the Seyferts (S, magenta filled dots). The Transition and L2 objects for which HST-STIS observations suggest the presence of broad Hα features are indicated by big circles, while the big squares indicate the T, L2 and H ii nuclei with narrow features that show broad wings. 147 5.4 The narrow Hα luminosity – measured in the STIS-HST aperture – as a function of nuclear spectral class, as defined by Ho et al. (1997a). The size of each data point is scaled by the respective distance to the host galaxy. The objects with possible broad Hα features are indicated by big circles (L1s are all broad-lined but are not indicated as big circles). The big squares indicate the nuclei with narrow features that show broad wings, and whose profiles are fitted by multiple Gaussian components...... 148 5.5 RatioofHα flux measured with STIS-HST and by Palomar, as a function of nuclear spectral class, as defined by Ho et al. (1997a). The size of each data point is scaled by the respective distance to the host galaxy. The T and L2 objects with possible broad Hα features are indicated by big circles. The big squares indicate the ambiguous nuclei withnarrowfeaturesthatshowbroadwings...... 151 16

5.6 [N ii]/Hα from HST spectra ratioed to the Palomar value, as a function of nuclear spectral class (ground based classification). Same symbol codeasinFigure 5.5...... 153 5.7 [S ii] λλ 6716, 6731/Hα from HST spectra ratioed to the Palomar value, as a function of nuclear spectral class (ground based classification). SamesymbolcodeasinFigure 5.5...... 154 5.8 [O i]/Hα from HST spectra ratioed to the Palomar value, as a function of nuclear spectral class (ground based classification). Same symbol codeasinFigure 5.5...... 155 5.9 [S ii] 6716/6731 from HST spectra ratioed to the Palomar value, as a function of nuclear spectral class (ground based classification). Same symbolcodeasinFigure 5.5...... 157 5.10 [S ii] 6716/6731 from HST spectra compared to the same ratio from the Palomar measurements. Same color/symbol code as in Figure 5.3. 159 List of Tables

2.1 QSOs observed at MMT and Keck telescopes...... 31 2.1 QSOs observed at MMT and Keck telescopes...... 32 2.1 QSOs observed at MMT and Keck telescopes...... 34

3.1 NLS1Sample ...... 63 3.1 NLS1Sample ...... 64 3.2 NLS1Properties ...... 65 3.2 NLS1Properties ...... 67 3.3 Emission-line measurements ...... 79 3.3 Emission-line measurements ...... 80 3.3 Emission-line measurements ...... 81 3.3 Emission-line measurements ...... 82 3.3 Emission-line measurements ...... 83 3.3 Emission-line measurements ...... 84 3.4 Proportions of PCs in the C iv emissionregion ...... 96

5.1 Nuclei Sample, HST observations ...... 126 5.1 Nuclei Sample, HST observations ...... 127 5.1 Nuclei Sample, HST observations ...... 128 5.1 Nuclei Sample, HST observations ...... 129 5.1 Nuclei Sample, HST observations ...... 130 5.2 HST Fluxes of Strong Emission Lines ...... 135 5.2 HST Fluxes of Strong Emission Lines ...... 136 5.2 HST Fluxes of Strong Emission Lines ...... 137 5.2 HST Fluxes of Strong Emission Lines ...... 138 5.2 HST Fluxes of Strong Emission Lines ...... 139 5.3 Galaxies with Broad Hα Emission ...... 143 5.3 Galaxies with Broad Hα Emission ...... 145 5.4 HSTdata,KSstatistics ...... 162 5.5 Palomardata,KSstatistics ...... 163

17 18

Chapter 1 Introduction

1.1 Rationale

The past forty years of discoveries advance the fact that galaxies are much more than hundreds of billions of stars and gas held together by gravity: evidence increasingly suggests that nearly every galaxy harbors a massive black hole at its center (e.g., Richstone et al. 1998). Even the inner regions of the Milky Way indicate the presence of a compact dark mass of a few million solar masses (e.g., Ghez et al. 2003; Schödel et al. 2003). The initial indication of the existence of the supermassive black holes came from relatively distant galaxies: the exotic quasars, with their tremendous power and compact nature that could (and still can) only be explained as originating from matter falling onto an enormous black hole, with a mass as large as a billion times that of the Sun (Lynden-Bell 1969). The fact that the quasars were very numerous at high redshift, and hence in the past, but are almost absent in the environs of the Milky Way, suggests that they fade over time, and that the massive black holes that drove these systems should persist in the centers of normal galaxies today. Making a positive identification of a black hole is extraordinarily challenging, albeit not impossible, at least when they are close to us. The strength of their influence on the surroundings is in fact what makes them discernable. The detection of high orbital velocities of stars and gas in the centers of many bright nearby galaxies reveals the gravitational field of a supermassive black hole. Dynamical studies also indicate that there are significant correlations between the nuclear black hole mass 19

and the mass, luminosity, and velocity dispersion of the host galaxy bulge (Magorrian et al. 1998; Ferrarese & Merritt 2000; Gebhardt et al. 2000). This means that the “monster” in the middle “knows” its host, strengthening the idea that formation and evolution of galaxies and their active nuclei are intimately related. In nearby galaxy nuclei, the black hole is relatively dormant; there is modest or no emission associated with accretion of the surrounding matter onto it, unlike what is seen at high redshift, in quasars. Thus, if quasars evolve and cease the accreting process to become our quiescent neighbours, it is important to ﬁnd out how this happened, at what point in their lives, and why. Is there evidence for an evolution of active galaxies consistent with these ideas? Answering these questions is of great importance for understanding whether the black hole formation is a generic consequence of galaxy formation and evolution, and where in this picture the active phase ﬁts in.

1.2 Active Galactic Nuclei

Active Galactic Nuclei (AGN) exist over a wide range of redshifts, and in recent years this has grown to include objects at z > 6. The comoving density of luminous AGN peaks at z ≈ 2 − 3. Relatively little attention has been given to the emission properties of sources at redshifts beyond this “quasar era.” Such objects are nonetheless of interest as tracers of accretion phenomena in young galaxies. Their spectra can be used to investigate accretion physics as well as chemical enrichment of their emitting gas, which in turn can reveal information on early star formation episodes. In the local universe, where quasars seem to vanish, low-luminosity emission-line galaxy nuclei remain abundant. The Palomar spectroscopic survey of nearby galaxies (Filippenko & Sargent 1985; Ho, Filippenko & Sargent 1997a), which is the most sensitive of its kind, reveals that a fraction as high as ∼ 86% of all bright sources reveal such activity. The expectations are that a great majority are powered to some degree by an AGN-type nucleus, as these objects are the most natural evolutionary endpoints of quasars. Testing this hypothesis is complicated in that for a large fraction of these nearby emission-line nuclei (∼ 30%) the nature of their central source is ambiguous. 20

It is therefore important to investigate the ionization mechanisms in these objects, so as to obtain an accurate census of nearby nuclei that are powered by accretion, and hence to place robust constraints on the black hole demography.

1.2.1 The Nucleus Types and their Oddities

The diversity of phenomena exhibited by emission-line galaxies has led to classification schemes that at least in part reflect fundamental physical distinctions among sources. An important characteristic that differentiates emission-line objects comes from the appearance of their optical spectra. The presence or absence of the broad permitted lines (e.g. Hβ, with line widths of thousands of km s−1) gave rise to two main classes of objects. Sources that show the broad features are assigned to a type 1 classification, while the ones showing only the narrow lines, both permitted and forbidden, with widths of hundreds of km s−1, are labeled type 2. In terms of their narrow line properties alone, the emission-line nuclei are considered to fall into four major categories: star-forming or H ii nuclei, Seyfert galaxies, Low-Ionization Nuclear Emission-line Regions (LINERs), and LINER/H ii transition systems. The distinction between these subclasses is based on optical line flux ratios, which trace the warm ionized component of the interstellar medium, thus providing potential information on the physical mechanism responsible for the ionization. The diagnostic diagrams (e.g., Baldwin, Phillips & Terlevich 1981; Veilleux & Osterbrock 1987; Figure 1.1) are shown to be valuable tools in distinguishing H ii regions from the objects presumed to be excited by the accretion onto a black hole (the power-law radiation) or other potential heating processes. Seyferts and LINERs, unlike the other subgroups of nuclei, include significant fractions of objects that exhibit broad lines. Together with their more luminous counterparts, the quasars, these Seyfert 1s and LINER 1s comprise the class of emission-line galaxies that are unambiguously powered by an accreting-type nucleus. The presence of the broad components indicates the existence of rapidly orbiting emission-line 21

Figure 1.1 Example of line flux-ratio diagnostic diagrams used in classifying galaxy nuclei in the Palomar survey (adapted from Ho et al. 1997a). H ii nuclei are shown as crosses, Seyferts as filled squares, LINERs as filled circles, and transition objects as open circles. 22

clouds under the influence of the strong pull of a supermassive black hole. An important factor that is believed to differentiate between these subgroups is the nuclear energetic output, ultimately determined by the reservoir and rate of accretion of matter. Quasars are at the extreme of high accretion rates, hence the most luminous ones, while LINER 1s are the weakest AGNs. An important role in determining the observed spectral type is considered to be played by the orientation of an AGN relative to our line of sight (e.g., Peterson 1997, Fig. 7.1.), due to some degree of axisymmetric structure and anisotropic dust obscuration. There is increasing evidence that a highly opaque dusty torus surrounds the central engine, and that, when this geometry is viewed more edge-on, the central accretion disk and the broad line region are hidden, resulting in a type 2 appearance; conversely, for high inclination angles, a direct view of the optical broad-lined AGN is obtained. It remains however an open question whether this picture applies to all type 2 sources. Seyfert 1s with relatively narrow Balmer lines show other correlated behavior that has led to their identification as a distinct subclass. Narrow Line Seyfert 1s (NLS1) tend to display extreme X-ray variability, soft X-ray continua, strong optical emission-line asymmetries and distinct line ratios. Their characteristics are potentially explained by either a relatively small black hole, that accretes at high rates, or a more pole-on view of a planar broad line emission region. If the first scenario is correct, these galaxies could be in an early evolutionary phase, in relation to other AGNs (Mathur 2000). This picture has fueled speculation about possible connections between NLS1s and the high z QSOs.

1.2.2 The Central Power Source of Low Luminosity Nuclei

The division between the different types of emission-line nuclei turns out to be somewhat arbitrary, as the observed line flux ratios seem to follow a rather continuous distribution (e.g., Ho, Filippenko & Sargent 1993; see also Figure 1.1). The intermediate properties of the transition objects, between those of LINERs and typical H ii regions, suggest a composite nature, where a central LINER nucleus is surrounded 23 by star-forming regions, producing thus a mix of these sources’ emission (Ho et al. 1993). Likewise, if all LINERs are genuine AGNs, the transition objects would simply be accretion sources that are highly contaminated by starlight, and they would have to be counted when estimating the faint end of the local (z ≈ 0) AGN luminosity function and in the black hole mass statistics. One problem is that interpretation of LINERs themselves, especially the narrow- lined ones, has been a source of controversy because their optical line ratios can be reproduced reasonably well by models based on a variety of different physical mechanisms. These processes include shock heating (Dopita & Sutherland 1995), photoionization by a non-stellar continuum (Ferland & Netzer 1983; Halpern & Steiner 1983), or by hot stars (Shields 1992; Barth & Shields 2000). The viability of the shock scenario loses ground in the light of a recent analysis of the emission-line profiles of the Palomar nuclei (Ho et al. 2003). The values of the velocity dispersions of the nuclear gas are found, in general, to be smaller than those required for the shock excitation to be important. Other studies also show that the high excitation spectrum that the shocked gas should produce in the UV is never observed, i.e., the intensities of the high-excitation lines such as C IV λ1450 and He II λ1640 are much weaker than predicted (Barth et al. 1996, 1997; Gabel et al. 2000). New empirical facts also show that invoking ionization by hot stars has problems too. For this model to be successful, the star cluster has to be formed in an instantaneous burst that must be only ∼ 3 − 5 Myr old in order to have sufficient Wolf-Rayet stars available for supplying the extreme-UV photons necessary to boost the low-ionization lines. The main difficulty with this scenario comes from the fact that the nuclear stellar population of the host galaxy of the majority of these nuclei is demonstrably old (Ho et al. 2003; Sarzi et al. 2004). Several lines of arguments suggest that in fact LINERs are accretion powered. The optical spectra of LINERs are readily reproduced in photoionization calculations with an AGN-like continuum by adjusting (to an order of magnitude lower than in the Seyferts) the ionization parameter U, defined as the ratio of the density of ionizing photons and the density of nucleons at the illuminated face of the cloud. Moreover, 24

both type 1 and 2 LINERs resemble AGNs in their compact, point-like hard X-ray and radio emission. Broad Hα has been detected in about 25% of LINERs (Ho et al. 1997b), which constitutes strong evidence for an accreting nature of their central source. Some LINER 2s are found to contain hidden broad features through spec- tropolarimetric observations (Barth, Filippenko & Moran 1999a,b). LINER SEDs approximate those predicted for radiatively inefficient accretion flows onto black holes (Quataert et al. 1999; Ptak et al. 2004), while they bear very little resemblance to those characteristic of normal stellar systems (Schneider, Schmidt & Gunn 1997), especially in their contribution from X-rays. Finally, a significant fraction of the galaxies that present kinematic evidence for black holes are in fact LINERs. Never- theless, these AGN-like characteristics are not found in all LINERs. Thus, for the class of LINERs as a whole, it remains difficult to assess with certainty the AGN content. While a clear view of their central engine is desirable, obtaining it is, however, challenging. The overwhelming dominance of the starlight in the optical spectra of most of these rather weak nuclei makes it difficult to isolate unique signatures that identify the nature of the central source.

1.3 Limitations of Previous Work

The eﬃciency with which emission-line nuclei are detected and analysed is almost always prone to observational constraints and selection biases. As mentioned above, one key practical complication in the simple scheme used in diagnosing the low-luminosity emission-line nuclei is the starlight contamination. Any integrated spectrum of galactic nuclei would include, to some degree, emission from stars in the host galaxy. For the high z systems, which are also the most luminous ones, the contrast between the nuclear and stellar background radiation is so high that the contamination is minimal, and can be safely neglected. In most of the nearby systems however, the central source is weak (a typical LINER bolometric luminosity is < 42 −1 ∼10 erg s , which is orders of magnitude less luminous than powerful Seyferts and QSOs), and the stellar emission may completely swamp the nebular line emission. 25

Thus, whether or not LINER 2s and transition objects exhibit broad lines, the sign- posts of the accretion mechanism, might be a matter of the observational sensitivity to the nuclear emission alone (i.e., quality of spectra, aperture size, e.g., Osterbrock & Shaw, 1988; Osterbrock & Martel, 1993; this work). High quality, small aperture observations are increasingly available for such nuclei, and may be the key to solving this contrast problem. > Previous studies of quasars at z ∼ 4 have included only very limited investigations of emission-line properties, or their implications for the quasar host. Studies that would address these issues, especially in an evolutionary context, by employing comparisons with quasars at lower redshifts are clearly desirable. In particular, the possible connection between the high z quasars and NLS1s needs quantitative tests. Comparisons between the z > 4 spectra and most published NLS1 data have been so far impracticable since the former cover only the UV rest-frame wavelengths while the latter mostly the optical range. Studies of their emission characteristics that employ data spanning the same bandpass are needed to fully investigate their relative behavior. To date, nearly all AGN samples are ﬂux limited at blue optical wavelengths. The extent to which this selection biases AGN samples is not fully understood. Given the steepness of the luminosity function for luminous QSOs, most objects are expected to lie close to the magnitude limit of a survey, so even small amounts of extinction would eliminate them from a blue-selected sample. Determining the population of red QSOs has been the subject of intense work lately, and some have been found through other selection techniques that employ wavelengths other than optical and UV. However, whether their redness is intrinsic or caused by obscuring material or other factors external to the central source of radiation remains unknown, and still diﬃcult to approach observationally. Simulations of various reddening scenarios might be able to constrain the nature of these red objects, and maybe even predict the total fraction of obscured accretion power in the universe. 26 1.4 This work

This dissertation employs primarily optical and UV observations to analyse the emission-line behavior on small spatial scales within galaxy nuclei, from early epochs of galaxy formation to the local universe. Objects at the highest observable redshifts > (z ∼ 4) are systematically studied by means of their emission-line properties, hence revealing information on the timescale of star formation and chemical enrichment in the young universe (Chapter 2). Direct comparisons of the emission properties of > z ∼ 4 sources with those at low and intermediate redshift provide valuable insights regarding accretion-powered activity in relation to star formation and galaxy evolution (Chapters 2 & 3). In particular, a detailed study of the optical and UV properties of the NLS1s is performed, and a comparative analysis of their emission characteristics > relative to those of the z ∼ 4 QSOs is conducted in order to probe the proposed NLS1 – high z QSO connection (Chapter 3). Observational biases that almost invariably accompany any AGN survey are investigated in-depth along with the role of the absorbing dust in modifying the shape of the quasar spectral energy density distribution (Chapters 2, 3 & 4). Nearby galactic nuclei are studied through high quality HST archival spectra that expose spatial scales much smaller than those observed in the past surveys, allowing us to probe in detail the degree to which the underlying continuum sources are related or not to some form of activity substantially diﬀerent from normal star formation (Chapter 5). 27

Chapter 2 Emission-Line Properties of z > 4 Quasars

Originally published as:

Constantin, A., Shields, J. C., Hamann, F., Foltz, G. B., & Chaﬀee, F. H., 2002, ApJ, 565, 50

2.1 Introduction

The properties of QSOs at the highest known redshifts are important for understanding the early history of galaxies and the formation of supermassive black holes. Over the past decade, a growing number of quasars have been reported at redshifts z > 4, corresponding in look-back time to more than ∼ 90% of the age of the Universe. Objects in this regime are thus inherently young, and accordingly of great interest for studying the onset of accretion-powered activity in relation to galaxy formation and evolution. > Optical studies of QSOs at z ∼4 to date have mostly emphasized their discovery, luminosity function, and use as probes of intervening matter seen in absorption. Investigations of the emission-line properties of these objects have been far more limited (Schneider, Schmidt & Gunn 1991; Storrie-Lombardi et al. 1996). Further scrutiny of their emission spectra in relation to those of their low-redshift counterparts is clearly desirable. Speciﬁcally, such comparisons may reveal information on the timescale of star formation and chemical evolution associated with the QSO phenomenon (Hamann & Ferland 1999), and on time-dependent behavior of the accretion process (Mathur 2000). This paper presents a detailed analysis of the optical emission line properties 28

> for a large sample of z ∼4 QSOs. High signal-to-noise, moderate resolution spectra of 44 quasars were obtained over multiple observing runs at the MMT and Keck observatories. The study employs direct measurements of the emisson lines as well as comparisons of composite spectra. Preliminary results from this survey were reported by Shields & Hamann (1997). Here we compare our spectra with those of redshift ∼ 2 quasars and find general agreement, although subtle differences are present in the strengths and profiles of some emission features. Our objects show little evidence for a Baldwin Effect, the empirical anticorrelation between line equivalent width and luminosity L often found in lower redshift samples (Baldwin 1977; see Osmer & Shields 1999 for a review). We also investigate the degree to which our findings may be influenced by selection effects. We compare the properties of our targets > with published measurements for other z ∼4 QSO samples representing a variety of discovery methods. This analysis indicates that our sources are representative of known quasars in this redshift regime, and suggests that the observed differences from lower-z sources are at least partially intrinsic. In subsequent papers (e.g., Dietrich et al. 2002), we will present more detailed comparisons of quasar emission-line properties as a function of redshift and luminosity, based on a sample of > 800 objects that span a much wider range in z and L. Although the values of the adopted cosmological parameters play an important −1 role at these extreme redshifts, we adopt throughout the paper an H0 = 50 km s −1 Mpc , q0 = 0.5, Λ = 0 cosmology, to ease the comparison with earlier published work in this field.

2.2 Observations & data analysis

> We began a program of spectroscopy of z ∼4 QSOs in 1994, with targets selected from the sources known at that time (∼ 50) or reported in subsequent years in the literature. Data were obtained in a series of 7 observing runs at the MMT Observatory during 1994 – 1996, and 2 runs at the W. M. Keck Observatory in 1997 October and 1999 June. The goal of this program was to obtain a statistically signiﬁcant description of quasar spectra in this high redshift regime; 32 objects were observed 29

at the MMT, and 12 at Keck, for a total of 44 QSOs. Specific targets were selected based on their observability, with the brightest available sources given priority in order to maximize the resulting signal-to-noise ratio. We generally avoided sources known to exhibit strong absorption lines near the emission redshift. The majority of the observed objects are taken from the APM Color Survey (Storrie-Lombardi et al. 1996, BR/BRI prefix) and the Second Palomar Sky Survey (Kennefick et al. 1995b, PSS prefix). Table 2.1 lists the redshifts, r magnitudes, and the discovery reference for each source. Data were acquired with the Red Channel of the MMT Spectrograph (Schmidt, Weymann & Foltz 1989) and with LRIS at the Keck ii telescope (Oke et al. 1995). Observations were obtained with a long slit of 100width, yielding full width at half maximum (FWHM) spectral resolution of ∼ 10 A˚ and ∼ 7 A,˚ with pixel sampling of 3.53 A˚ pixel−1 and 1.86 A˚ pixel−1, respectively. These resolutions correspond to 300 – 400 km s−1 across the C IV λ1549 emission line, or typically 0.07 – 0.1 times this line’s FWHM. The spectral coverage was chosen to span the redshifted Lyα λ1216 – He II λ1640 interval; the He II feature is of specific interest since the N V λ1240/He II λ1640 ratio can be used as an abundance diagnostic (Hamann & Ferland 1993; Ferland et al. 1996). The observational setup was essentially identical for the two telescopes. A long-pass filter (LP495 at the MMT, GG495 at Keck) was used to prevent second-order contamination. Multiple exposures totaling ∼ 1 − 2 hrs were obtained for each quasar, yielding a typical signal-to-noise ratio of ∼ 50 per pixel. Observations were obtained at an < airmass ∼1.5, and the slit was oriented approximately perpendicular to the horizon in order to minimize the effects of atmospheric differential refraction (Filippenko 1982). The atmospheric seeing was typically 100 (or better for the Keck ii telescope), but not all the observations were obtained under absolute photometric conditions. Calibration and reduction of the resulting spectra were carried out using standard techniques as implemented in the IRAF software package1. Observations of nearly

1The Image Reduction and Analysis Facility (IRAF) is distributed by the National Optical As- tronomy Observatories, which is operated by the Association of Universities for Research in Astron- omy Inc. (AURA), under cooperative agreement with the National Science Foundation. 30

featureless standard stars were used to provide flux calibrations and to correct for the atmospheric absorption bands. He-Ne-Ar arc-lamps were used for the wavelength calibrations. Multiple spectra acquired for the same object were averaged. Figures 2.1, 2.2, 2.3, and 2.4 present all the reduced spectra included in this observational program, sorted on RA. A small number of known Broad Absorption Line (BAL) QSOs were observed as part of the sample in order to enable additional comparisons with low-redshift quasars. These sources are PSS 1048+4407, BR 1144-0723, and PSS 1438+2538. Our spectra, which have greater S/N and wavelength coverage than many of the published discovery spectra, also reveal for the first time broad absorption lines in PC 0027+0525 and PSS 0137+2837. The primary focus of the present work is on the emission lines. Accurate measurement of these features requires correction for absorption in many cases. For the majority of objects, the absorption appears as narrow lines due to associated or intervening systems, superimposed on either the emission lines or continuum. In these cases we removed the absorption feature through a simple interpolation. For the BAL QSOs and several additional objects (BRI 0151-0025, PSS 1159+1337, BRI 1346- 0322, PC 1415+3408, and PSS 1435+3057), the absorption was severe enough that a reliable reconstruction of the unabsorbed spectrum was not possible. We excluded these sources from our measurements of the emission lines and from our composite spectra. All objects in this sample show strong Lyα forest absorption, which clearly influences the measured profile and flux of the Lyα emission line in many cases; we did not attempt to correct for this effect. Objects in our sample are subject to small amounts of Galactic foreground extinction, spanning AV =0.04−0.49, with a typical value of ∼ 0.1 (Schlegel, Finkbeiner & Davis 1998). We corrected the spectra for the resulting reddening using the empirical selective extinction function of Cardelli, Clayton & Mathis (1989). The spectra were Doppler-corrected to the rest frame using z values that we measured for the majority of sources from the C IV λ1549 line. The observed λ for each case was established by fitting a Gaussian to the top 20% of the profile. 31

Table 2.1. QSOs observed at MMT and Keck telescopes.

a b Object z r C IV Lyα + N V logLν(1450) Ref

c d c EW1 (A)˚ EW2 (A)˚ EW (A)˚

PSS0003+2730 4.23 19.3 33 32 95 31.30f 3

BR0019-1522 4.52 19.0 43 34 97 31.53 6

PC0027+0521 4.21 21.9 57 43 133 30.09 13

PC 0027+0525e 4.0921.5 – – – 13

SDSS003525+00 4.75 21.2 69 64 125 31.11 4

PSS0059+0003 4.15 19.5 26 23 78 31.24f 7

BRI0103+0032 4.43 18.6 44 36 87 31.52 15

PC0131+0130 3.78 19.4 17 15 72 31.37 10

PSS 0137+2837e 4.3019.0 – – – 3

BRI0151-0025 4.19 18.9 – – 72 31.44 15

PSS0152+0735 4.05 19.6 27 27 60 31.06f 3

BRI0241-0146 4.01 18.2 27 24 57 31.62 15

PSS0248+1802 4.42 18.4 32 31 95 31.76f 8

PC0307+0222 4.39 20.4 20 20 75 30.81 11

BR0401-1711 4.22 18.7 68 61 195 31.49f 16

BR0951-0450 4.34 18.9 28 28 66 31.35 9,16

BRI0952-0115 4.40 18.7 28 27 50 31.55 16

BRI1013+0035 4.38 18.8 36 35 56 31.43 14

BR1033-0327 4.49 18.5 26 24 72 31.53 6,16

PSS 1048+4407e 4.4519.3 – – – 8

BRI1050-0000 4.28 18.7 73 73 119 31.26 14 32

Table 2.1 (cont’d)

a b Object z r C IV Lyα + N V logLν(1450) Ref

c d c EW1 (A)˚ EW2 (A)˚ EW (A)˚

PSS1057+4555 4.10 17.7 24 21 74 31.79 8

BRI1108-0747 3.91 18.1 28 25 93 31.44 16

BRI1110+0106 3.91 18.3 20 19 42 31.40 16

BR 1144-0723e 4.1618.6 – – – 16

PSS1159+1337 4.08 18.5 – – 15 31.91f 3

BR1202-0725 4.69 18.7 17 17 22 31.87 6,16

PC1301+4747 3.99 21.3 46 46 27 30.32 12

PSS1317+3531 4.36 19.1 30 29 85 31.39f 8

BRI1328-0433 4.20 19.0 39 26 71 31.37 9,16

BRI1346-0322 4.00 18.8 – – 104 31.21 16

PC1415+3408 4.58 21.4 – – 70 30.66 13

PSS1435+30574.3519.3 – – 20 7

PSS 1438+2538e 4.2419.5 – – – 8

PC1450+3404 4.19 20.8 44 38 102 30.68 13

BRI1500+0824 3.93 18.8 23 18 44 31.32 15

PSS1618+4125 4.21 19.6 31 26 73 31.06f 3

Q1745+6226 3.89 18.8 38 34 79 31.80f 1

RX1759+6638 4.32 20.0 45 40 68 31.30 5

Q2203+2900 4.41 20.8 31 26 53 30.87 2

BR2212-1626 3.99 18.1 51 50 87 31.53 16

BR2237-0607 4.55 18.3 31 26 83 31.74f 6, 16 33

For situations where the measured line was significantly affected by absorption or low signal-to-noise, we verified the result with fits to the top 50% of the profile. For the BAL QSOs, the redshifts listed in Table 2.1 are taken from the discovery reference. For the remaining objects where C IV was not measurable due to absorption (see above), Lyα was used instead. The redshift measurements in these cases relied on Gaussian fits that used as a template only the red side of the profile. The resulting z determinations are in good agreement with previously published values, with an estimated uncertainty of ≤ 0.01. Luminosities for our sample targets were derived using published photometric data. Rest-frame 1450A˚ fluxes (AB magnitudes at 1450A)˚ were obtained directly from the discovery papers for the majority of the objects in our sample. For the objects where these numbers were not available, the flux measured from our spectra was used, after the absolute flux calibration was adjusted to obtain consistency with published broad-band photometry. The luminosity values, expressed in ergs s−1 Hz−1, are also included in Table 2.1. To investigate the emission-line properties of the observed QSOs, we constructed composite spectra, and employed measurements of the lines in individual objects. The equivalent widths (EWs) of the strong lines, specifically for Lyα + N V and C IV λ1549, were measured interactively, using a direct integration of the line flux referenced to the interpolated continuum level. For the Lyα+N V feature we used the extension of the continuum from the red side only, and did not attempt any deblend- ing. The resulting values for C IV are very similar to those obtained by modelling the lines with 2 Gaussians, representing broad and narrow components. Possible in- accuracies in these measurements are dominated by ambiguities in placement of the continuum; experiments with alternative choices of the continuum level suggest that this contributes a systematic uncertainty of ∼ 20%. In our analysis, we found it useful to combine our measurements with published > emission-line data for other z ∼4 QSO samples, and in particular those discovered via the Palomar CCD Grism Survey (Schneider, Schmidt & Gunn 1991, 1997; Schmidt, Schneider & Gunn 1995, hereafter referred to as SSG, ∼ 15 objects), and early 34

Table 2.1 (cont’d)

a b Object z r C IV Lyα + N V logLν(1450) Ref

c d c EW1 (A)˚ EW2 (A)˚ EW (A)˚

BR2248-1242 4.15 18.5 106 98 211 31.24 6,16

PC2331+0216 4.09 20.0 17 15 85 30.93 11

acalculated from our spectra using the C IV emission-line b ∗ photometry in r or analogous bandpass (R for the APM survey, r for PSS, r for SDSS, r4 for SSG) as reported in the respective discovery papers crest-frame EWs calculated by direct integration of the line ﬂux

drest-frame EWs calculated by ﬁtting the line to a single Gaussian

eexhibits broad absorption lines (BAL); z values are taken from the discovery papers f values derived using the ﬂux fλ(1450) measured from our spectra; estimated errors are approximately ± 0.2 dex.

References. — (1) Becker et al. (1992); (2) Dickinson & McCarthy 1987; (3) S. Djorgovski, private comunication; (4) Fan et al. 1999; (5) Henry et al. 1994; (6) Isaak et al. 1994; (7) Kennefick et al. 1995a; (8) Kennefick et al. 1995b; (9) Kennefick et al. 1996; (10) Schmidt et al. 1987; (11) Schneider et al. 1989; (12) Schneider et al. 1991; (13) Schneider et al. 1997; (14) Smith et al. 1994a; (15) Smith et al. 1994b; (16) Storrie-Lombardi et al. 1996. 35

> −17 −1 Figure 2.1 Observed spectra of QSOs at z ∼4. The ﬂux, Fλ, is in units of 10 erg s cm−2 A˚−1. The prominent emission features are marked. 36

> Figure 2.2 Observed spectra of QSOs at z ∼4. 37

> Figure 2.3 Observed spectra of QSOs at z ∼4. 38

> Figure 2.4 Observed spectra of QSOs at z ∼4. 39

discoveries from the Sloan Digital Sky Survey (SDSS; Fan et al. 1999, 2000, 2001, ∼ 35 objects, only those with z > 3.9 were considered in this analysis). In contrast with our measurements, the published EW values for these samples rely on single component Gaussian fitting. Our results and the published EWs can be meaningfully combined only if the line measurement procedures are consistent. We tested different methods of measuring the emission lines with our data. For the C IV line, the single Gaussian fits give, in general, lower EWs than those obtained with our direct integration technique, as they tend to lose flux from the wings if they are prominent, and from the peak if the cores dominate. In order to obtain a reasonable consistency in the comparisons involving the SSG and the SDSS samples, we measured the C IV EWs in both ways. For Lyα + N V, the fit results are sensitive to the method of treating the two lines and the asymmetry resulting from Lyα forest absorption; consequently, we employed only the EWs obtained by direct integration of the profile. The values are recorded in Table 2.1. Finally, we constructed composite spectra using the data for individual sources normalized to unit mean flux over the wavelength range 1430 A–˚ 1470 A.˚ Average and median composites were calculated over the wavelength range common for all objects. The average composite may be influenced by several extreme objects with unusually strong and narrow lines (BR0401-1711, BRI1050-0000, BR2212-1626, BR2248-1242). The median spectrum may thus be more representative of the typical object in our sample.

2.3 Spectroscopic properties

2.3.1 General characteristics

An initial scrutiny of the spectra of z > 4 QSOs reveals substantial agreement with those of their lower redshift counterparts. This similarity is noteworthy, given the widely differing amounts of time available for the evolution of the AGN and > underlying host galaxy. In Figure 2.5 (upper panel) we compare our z ∼4 average 40 composite spectrum with the composite spectrum derived as an average2 from the total Large Bright Quasar Survey (LBQS) sample, which is weighted toward z ≈ 2 objects in this wavelength range (S. Morris 1999, private comunication; see Brotherton et al. 2001 for additional information). This comparison is especially useful given that our sample and the subset of the LBQS contributing to this wavelength interval have nearly identical average luminosities (Lν (1450A)˚ ≈ 31.3), so that any differences should reflect a redshift dependence or differing selection criteria. We estimated the average LBQS Lν(1450A)˚ from the absolute B magnitude given by Francis et al. (1991), their Figure 8, following the methodology of Schneider, Schmidt & Gunn (1989) to relate the flux level in the two different bandpasses, while correcting for differences in cosmological parameters. In general terms, the high and intermediate redshift composites agree remarkably well; they have the same strong emission lines, similar continuum shape redward of the Lyα feature, and comparable strengths and > profiles of the lines. The pronounced Lyα forest in the z ∼4 composites does not reflect an intrinsic difference in the nature of the emission sources, but rather an increase in the opacity of the intergalactic medium at larger redshifts. > Similarities between the QSOs at z ∼4and those at lower redshift can also be seen in measurements of spectral features for individual sources. Surveys of QSOs at z ≈ 2−3 (Francis et al. 1992; Francis, Hooper & Impey 1993) display strong (negative) correlations of line widths (FWHM) with line EWs and with line peak/continuum ratios. Similar tendencies are present in the high z sources. Figure 2.6 illustrates these results for all quasars in our sample, for which the C IV lines were measurable. Sources with narrow lines also tend to show the strongest emission. Furthermore, < spectra of QSOs at z ∼3 exhibit systematic line velocity shifts that correlate with the degree of ionization of the emitting species (Gaskell 1982; Espey et al. 1989; Tytler & Fan 1992; McIntosh et al. 1999). As reported previously by Storrie-Lombardi et al. (1996), this behavior is also present in z > 4 samples. A detailed comparison of the high and intermediate redshift composites in Fig- ure 2.5 shows excellent agreement for some lines but dissimilarities in others. C II

2The algorithm used in constructing the LBQS composite is described by Francis et al. (1991) 41

> Figure 2.5 Composite spectra (average -upper panel, and median -lower panel) of z ∼4 QSOs included in this observational program are shown in comparison with the LBQS composite, representative of z ≈ 2 objects. The y-axis has been normalized to unit mean ﬂux over the wavelength range 1430A–˚ 1470A.˚ The redshift measured from the Civ feature was used for > the Doppler correction of the z ∼4 spectra. As a consequence of the diﬀerent velocity shifts presented by the low- and high-ionization lines, the alignment of the high-z spectral features with those in the low z composite required a slight additional shift of our composites. 42

Figure 2.6 Equivalent widths and line peak/continuum ratios for Civ emission as a function of Civ line widths (full width at half maximum, FWHM), measured in our sample. 43

λ1335 and Si IV+O IV] λ1400 remain basically unchanged, while differences are present in the strengths of Lyα, N V, C IV, and O I λ1304. The Lyα emission in z > ∼4quasars is considerably affected by the forest absorption, which may remove up to ∼ 50% of the line, but the high-z composites nonetheless show a stronger Lyα feature than in the z ≈ 2 average. The N V emission is also enhanced in the high z systems, and the same is true for the O I λ1304 emission. The rest-frame equivalent width of > the latter feature is ∼ 3A˚ in the composite z ∼4 spectra, which is about 50% larger than that seen in spectroscopic surveys at lower redshifts (e. g., Brotherton et al. 1994; Forster et al. 2000, EW(O I) ≈ 2A).˚ > The C IV line is stronger in the z ∼4 average than in the intermediate z composite spectrum. The difference is most pronounced in the line core, while the line wings are more nearly comparable in the two composites. As noted previously, the sample includes several remarkable objects with unusually strong and narrow lines, which potentially distort the comparison when using an average composite for this modest sample. To address this question, the high z median and the redshift ∼ 2 composite are shown for comparison in Figure 2.5, lower panel. The Lyα and C IV features are clearly reduced in the high z median, but still stronger, especially in the core, than those in the LBQS composite. Aside from the differences in Lyα and C IV, the high redshift median and average spectra are remarkably similar. In particular, the O I and N V emissions are equally strong in both high z composites. The degree of variation between the members of our sample can be seen in the standard deviation of the normalized spectra as a function of wavelength, plotted in Figure 2.7. The largest variations are present in the cores of the prominent emission lines (Lyα and C IV), consistent with the previous results for lower redshift samples (Francis et al. 1992; Brotherton et al. 2001). The lower panel of Figure 2.7 shows the standard deviation of the mean as a function of wavelength, which provides a measure of the uncertainty of the average spectrum. 44

Figure 2.7 Upper panel: Flux standard deviation in Fλ relative to the average composite spectrum, as a function of wavelength for the individual spectra comprising the composites. Lower panel: The standard deviation of the mean, expressed in %. 45

2.3.2 O I and N V

The O I λ1304 and N V λ1240 lines show differences between the intermediate and high redshift spectrum composites that merit further attention. These features potentially convey information about the structure of the broad-line region and the physical conditions in the emitting plasma. Additionally, both lines are possible tracers of metallicity in the QSOs, which is of particular interest for understanding these sources in an evolutionary framework. We first examine whether we have correctly identified the O I feature. Previous quasar emission line studies have raised the question of whether the λ1304 feature is indeed O I emission or something else. The O I line is a triplet (λ1302.17, λ1304.86, λ1306.03) with a mean laboratory wavelength (obtained by weighting each transition by its Einstein A coefficient) of 1303.50A˚ (Verner et al. 1996). For the high optical depths that may occur in broad line clouds, the individual features in the triplet will approach equal strengths, modifying the mean to 1304.36A.˚ The feature we attribute to O I is seen at a wavelength of ∼ 1307A˚ in these composites. At first glance, a better identification might seem to be with Si II λλ1304.37, 1309.27 (mean laboratory wavelength 1307.63A).˚ However, this doublet is expected to be accompanied by considerable emission in other Si II lines, such as λλ1196, 1264, 1531, 1817 (Gaskell 1982; Dumont & Mathez 1981), which are weak or absent. Furthermore, we note that the high z composites are derived from spectra that are Doppler-corrected using redshifts given by C IV, which is blueshifted relative to the low ionization lines. An identification of the λ1307 feature with O I would be consistent with this trend, while Si II would not. We conclude that the O I λ1304 identification is secure. The enhanced strength of O I in the high redshift sources may reflect a structural difference in their broad-line regions (BLRs). Several studies of AGN ensembles suggest that low ionization lines (e.g. O I, C II) are emitted preferentially in low- velocity BLR components, while Lyα and high ionization lines (N V, Si IV, C IV) tend to trace higher velocity plasma (e. g., Brotherton et al. 1994; Francis et al. 1992; Wills et al. 1993; Baldwin et al. 1996). This is sometimes described in terms of a BLR with two components: the Intermediate Line Region (ILR) and Very Broad 46

Line Region (VBLR). In our high redshift sample, the narrow component makes a larger fractional contribution to the spectra than in the lower redshift systems. The C IV line has a larger velocity width in the LBQS composite (FWHM ≈ 5600 km s−1 > −1 ) than in the z ∼4 composites, both average and median (FWHM ≈ 4170 km s and 4900 km s−1, respectively). The unusual strength of O I emission may thus be the most obvious tracer of a characteristically enhanced ILR in z > 4 QSOs. The strong N V emission is not expected in systems with an enhanced narrow component, however, because of its high ionization. The systematically higher N V emission may instead reflect high metallicity, consistent with standard chemical evolution scenarios for young, massive elliptical galaxies or galaxy bulges. Hamann & Ferland (1992, 1993) suggested that N V, which is a collisionally excited line, may be elevated relative to other lines as a consequence of the secondary nitrogen enrichment in vigorously star-forming environments. Interestingly, enhancements in the O I emission may provide supporting evidence for this interpretation. O I λ1304 is apparently formed via fluorescence with the H Lyβ line (see, e.g. Netzer 1990 for a discussion). The strength of the O I feature is thus expected to scale with the O/H abundance ratio (Netzer & Penston 1976). However, the O I line strength may be influenced by various other factors, such as the sensitivity of the line coincidence to the velocity field in the emitting region. To summarize, the N V and O I features suggest that the broad-line regions of QSOs at z > 4 may be intrinsically different from those at lower redshift. Low-velocity components of the BLR are more prominent in these high-z sources, as signaled by the strength of O I and the narrow velocity width for C IV. In addition to this structural difference, however, the strength of N V emission implicates characteristically > higher metallicities in the z ∼4 objects. High metallicities may also contribute to the enhancement of the O I line. 47

2.3.3 The EW vs. Luminosity correlation

> To investigate the amplitude of the Baldwin Eﬀect for QSOs at z ∼4, we present plots of the rest equivalent width of C IV λ1549 as a function of the continuum lumi-

nosity Lν (measured at 1450 A˚ in the rest frame), for our observed sample (Figure 2.8, upper panel), and for an enlarged sample that includes also measurements obtained from the SSG and SDSS surveys (Figure 2.8, lower panel). The dotted line is the relation found by Osmer et al. (1994) for a sample of 186 quasars covering a wide range of redshift (z < 3.8). The overall results at high redshift (z > 4) appear compatible with the trend and scatter found for AGNs at lower redshift and luminosity, but do not, by themselves, show a Baldwin Eﬀect correlation in C IV. A Spearman test is consistent with the null hypothesis at > 99% level. The corresponding diagrams for Lyα + N V are shown in Figure 2.9, and likewise are consistent with no correlation. These measurements combine Lyα and N V, and therefore, a possible trend in the Lyα line alone, for which the Baldwin Eﬀect has been previously claimed, may be weakened by the presence of N V, which does not show such EW − Lν anticorrelation (Osmer, Porter & Green 1994; Korista et al. 1998). To explore further the luminosity-dependent behavior, we divided the quasars in

our sample into two luminosity bins, with log Lν (1450 A)˚ = 30.34−31.12 and 31.31− 31.97. Average and median composites for these subsets are shown in Figure 2.10. The noise and the intrinsic peculiarities present in single sources are considerably diminished in the composite spectra, making them a valuable tool in identifying line strength variations originating in diﬀerences in luminosity. A simple inspection of the individual lines reveals a small decrease in the EW with luminosity for Lyα, but not for C IV or other emission lines. > The weak or absent Baldwin Eﬀect in our z ∼4 sample may be an intrinsic characteristic of these objects, which are extreme in both redshift and luminosity. Alter- natively, the absence of a clear trend may stem from the limited luminosity range of

our sample and the substantial scatter that always appears in the Lν − EW relation. Selection effects influencing our sample are also a significant concern; if strong-lined objects are over-represented among our high luminosity QSOs, the result would be a 48

Figure 2.8 Rest-frame EW of the Civ emission line as a function of absolute luminosity of the quasar (at 1450 A˚ rest frame). The Baldwin Eﬀect reported by Osmer, Porter & Green (1994), based on measurements at lower redshift, is shown by the dotted line. The upper panel shows measurements from our sample only, with diﬀerent symbols for the color (×) > and grism (•) selected objects. The lower panel adds measurements for z ∼4 QSOs from Schneider, Schmidt & Gunn (1991) and the SDSS survey. Although consistent with the low redshift trend, the measurements do not, by themselves, show the expected correlation. 49

> Figure 2.9 Equivalent width of Lyα + Nv versus continuum luminosity for z ∼4 QSOs. Symbols are the same as in Fig. 2.8. 50

Figure 2.10 Comparison between median and average composites computed for two sets of −1 −1 quasars, grouped according to their luminosity at 1450A˚ (Lν expressed in ergs s Hz ). The same normalization as in Figure 2.5 is used. A weak drop in EW with luminosity can be noted for Lyα but not for Civ or other lines. 51 weakened Baldwin correlation. We explore the role of selection eﬀects further in the next section.

2.3.4 Selection biases

An important consideration in interpreting the properties of our high redshift QSOs is the extent to which selection effects can influence the properties of a sample. > Here we examine the methods used for identifying z ∼4 QSOs and ways that selection criteria may be reflected in emission-line characteristics. The majority of QSOs known at z > 4 have been discovered by color-selection techniques (Irwin et al. 1991; Kennefick et al. 1995a,b; Storrie-Lombardi et al. 1996, 2001) that rely in particular on the contrast between the observed spectral region centered on the Lyα feature, and the continuum at shorter wavelengths, which is strongly absorbed by the Lyα forest. Quasars are thus distinguished from stars primarily on the basis of their extremely red B – R or similar colors, which separate them from the stellar locus in the color-magnitude/color-color diagrams. The emission lines can contribute significantly to the total flux observed in typical photometric bandpasses. Their role is amplified at these particularly high redshifts since the scaling factor (1 + z) boosts the values of the observed EWs. The prominence of Lyα could therefore introduce biases affecting the detectability > of z ∼4 quasars. Most of the objects in our sample come from the APM Color Survey (BR/BRI quasars). These quasars are drawn from a magnitude-limited sample, and as a consequence, the strong-lined sources could be over-represented because of the contribution of the Lyα line to the measured flux in the R band (Malmquist bias). The amplitude of this effect can be roughly evaluated by estimating the change in the apparent (R) magnitude produced by the presence of a strong Lyα line: for a rest frame EW1 = 30A˚ (weak line) and EW2 = 160A˚ (strong line) with respect to the obs unabsorbed continuum, the observed (z ≈ 4.5) equivalent widths would be EW1 ≈ ˚ obs ˚ ˚ 165 A and EW2 ≈ 880 A respectively. With an R bandpass of width ∼ 1000 A, the line increases the measured flux by ∼ 16% and 88% compared with a pure continuum source. The ratio of the two broadband fluxes with strong compared to weak Lyα 52

would be f2/f1 ' 1.62, which corresponds to a difference in magnitude ' 0.5 mag. This value is actually an underestimate of the effect because the continuum blueward of Lyα is highly suppressed by the absorption forest. In addition, for some of the observations in the BRI survey, a narrower R bandpass (∼ 600A)˚ was used, resulting in an increase of the fractional contribution of Lyα to the measured flux. This effect leads to a bias in favor of objects with large EW, that could potentially account for the unusual examples of strong-lined QSOs in our sample. The same trend was suggested previously by Kennefick et al. (1995b), who calculated the selection probabilities of QSOs (from POSS II observations) for three cases of Lyα + N V line strength (see their Fig. 6). These considerations apply to the BR/BRI, PSS, and SDSS sources, which rely on similar bandpasses for QSO identification. Although this effect seems to be important, it could be counteracted by the treatment of the other bandpasses in the selection process. If only sources that are detected in B are included, as is the case for the APM survey, objects with very strong lines near the limiting R magnitude may be excluded, because their R-band flux is dominated by Lyα and the continuum (which dominates in B) is correspondingly weak. However, the selection process is further affected by the fact that, for very weak lines, the contrast between the Lyα emission and the continuum blueward of it may not be high enough to satisfy the necessary color criterion. In the case of the APM survey, the understanding of this bias is complicated by the use of a magnitude-dependent color threshold, due to signal-to-noise considerations (Storrie-Lombardi et al. 2001, see their Figure 1), again introducing a preference for strong-lined sources. The fact that the majority of the color selected objects in our sample fall above the dotted line in the Baldwin diagram, may be due, in part, to this bias. A thorough understanding of the Malmquist and color biases would require extensive modeling based on the choices of filters, limiting magnitudes, and/or colors. An alternative is to investigate the EW properties of samples discovered by different techniques. We compared the EWs and composite profiles of our color-selected sources to a grism-selected sample, which includes objects observed by us and by Schneider, Schmidt & Gunn (1991). The SSG sample consists of quasars selected 53

Figure 2.11 The distributions of the rest-frame (upper panels) and observed (lower panels) EWs for Lyα and Civ; three different samples are compared using a KS test criterion which indicates that the distributions are consistent with a common parent population. based on the detection of Lyα in slitless spectra. As discussed by Schmidt, Schneider & Gunn (1995), this detection technique may also lead to preferential selection of QSOs with strong lines. Figure 2.11 shows the rest-frame and the observed EW distributions, for C IV and Lα, for the color-selected and grism-selected samples. The K-S (Kolmogorov-Smirnov) statistics are consistent with the two samples having the same parent population. The average and median composite spectra should indicate similar trends. Figure 2.12 shows this comparison, which reveals a remarkable similarity between the composites. Likewise, there is no apparent difference in N V and O I; both lines are evidently stronger in the high redshift sources, regardless of the selection technique. We can conclude that the color selection methods (primarily BR/BRI) find QSOs with analogous properties to those of sources found by the SSG grism survey. Therefore, the corresponding biases, if apparent, affect the samples to a similar degree. 54

More useful comparisons may be those with the SDSS sample, which is claimed to be less biased. According to the authors, the selection probability in the SDSS detection of high redshift objects does not strongly depend on the emission line strength, but on the redshift and the SED shape. Comparisons between observed EW(Lyα) distributions for our sample, the grism (SSG) sample and the SDSS survey show that they are consistent with a single distribution function (probability of being drawn from the same distribution is respectively KSSDSS/oursample =0.633, KSSDSS/grism =0.910). The rest and observed frame EW histograms, for both Lyα and C IV, are displayed in Figure 2.11, separately for color-selected QSOs from our sample, for the grism- selected objects from SSG, and for the SDSS quasars. The KS statistic in each case is consistent with a common parent population. One possible complication in this comparison may be that the characteristic luminosities for these different samples are not the same. Inspection of Figures 2.8 and 2.9 indicate that quasars in our sample are somewhat more luminous than those in the SDSS. The Baldwin Effect would then predict smaller EWs for our sources. The fact that the EW distributions are actually similar could thus be a coincidence, due to a bias favoring strong-lined objects in our sample. A closer inspection of the luminosity distributions suggests that such a conspiracy is unlikely to be significant. There is substantial overlap in luminosities for the different samples, and differences in their median values are small; median Lν for our QSOs exceeds that of the SDSS sources by only a factor of 2. We conclude that the selection effects, if present, are not more evident in our sample than in the other high redshift surveys.

2.4 Discussion

Interpretations of the strong cosmological evolution of AGNs are generally based on the functional form of a (simple) observed luminosity function of bright QSOs at high z and prescriptions for the growth and luminosity of the underlying black holes. In the cold dark matter scenario, as a paradigm for a hierarchical cosmogony that 55

Figure 2.12 Comparison of composite spectra for color-selected and grism-selected QSOs. The composites include spectra published by SSG in addition to the data presented here. Average and median spectra are shown in the upper and lower panels, respectively. 56

models the evolution of the luminosity function, quasars are believed to be a short- lived, ﬁrst phase of the formation of a galaxy in the potential well of a dark matter halo (Haehnelt & Rees 1993; Haiman & Loeb 2001). Observations at high redshifts intercept epochs when quasars are very young objects, which thus might be expected to have characteristically low metallicity. How- ever, the trends reported here in the N V and O I broad emission lines (section 2.3.2) suggest that the metallicities are typically higher in higher redshift, and implicitly more luminous QSOs. The prominent emission lines indicate that, in spite of their youth, the high-redshift QSOs have undergone substantial chemical enrichment on a timescale that is short compared with their life span, implying a rapid star formation process in the host galaxy. Therefore, the very existence of objects at z > 4 presents severe timing problems (time to form and turn on), that can be solved only under the assumption that they reside in the most massive objects which have collapsed at these early epochs, associated with the rare high peaks in the primordial density ﬁeld (Turner 1991; Katz et al. 1994). This result may naturally arise from the analogy with the mass-metallicity relationship present in the low z spirals and ellip- ticals: massive galaxies reach higher metallicities because their deeper potential is better able to retain their gas against the galactic winds built by the thermal pres- sure from supernovae, while low-mass systems eject their gas before high enrichments are attained. Thus, quasar metallicities should be similarly tied to the gravitational binding energy of the local star-forming regions, and perhaps also to the total mass of their host galaxies (Hamann & Ferland 1999). Consistent with this picture, recent AGN studies have revealed correlations between the mass of the central black hole and QSO luminosities, QSO host masses, and the stellar velocity dispersion (Laor 1998; Magorrian et al. 1998; Ferrarese & Merritt 2000; Gebhardt et al. 2000). Quasar broad lines trace matter within only the central few parsecs, but the emerging black hole/bulge connections strongly suggest that the emission-line plasma is closely linked to the larger star-forming environment. Recent cosmic-structure simulations (Gnedin & Ostriker 1997) show that protogalactic condensations can form stars and reach higher than solar metallicities 57

> at z ∼6, suggesting that at epochs preceding, or concurrent with the QSO formation period, the bulge star formation already occurred. Therefore, interpretation of the abundance data in these extreme redshift sources yields unique constraints on the evolution of those environments, indirectly probing the epoch and extent of early star formation associated with QSOs. The observed enhancements in the N V and O I emission reinforce the previous evidence of solar or greater metallicities (Z>Z ) that occur before QSOs become observable (Dietrich & Wilhelm-Erkens 2000; Hamann & Ferland 1999 and references therein). The evidence of structural differences reported in Section 2.3.2 for the broad line region in these young sources provides further motivation for examining their evolutionary context. If the enhanced ILR emission is real and not a consequence of selection effects, this behavior may be an important tracer of accretion physics and its relationship to redshift and luminosity. Based on our initial reports of narrow UV > lines in z ∼4 QSOs, Mathur (2000) suggested an analogy between these sources and narrow-line Seyfert 1 (NLS1) galaxies. In this picture, both classes of objects are in an early evolutionary phase, in which accretion proceeds at or near the Eddington limit. While this scenario has several appealing aspects for explaining NLS1 phenomena, assigning it in general to high-z QSOs may be premature. While the latter sources < are young in cosmological terms (ages ∼2 Gyr), this does not ensure that they are in an early phase of accretion, since the timescale for QSO activity in an individual source is estimated to be only a few times 107 yr (e.g., Kauffmann & Haehnelt 2000). Furthermore, narrow profiles in the UV lines are probably not an adequate basis for linking our sources to a NLS1 classification, which is based on optical lines. As discussed by Wills et al. (2000), the UV line widths do not correlate in a simple way with Hβ width or with other NLS1 properties. Further inquiry is needed to understand the implications of line profile behavior in high-redshift QSOs.

2.5 Conclusions

Although an increasing amount of information is becoming available on the properties of QSOs discovered at very high redshift, only limited efforts have been made 58 to survey systematically the emission-line properties of these objects and/or the selection effects related with the techniques by which they were discovered. We have conducted a thorough analysis of 44 high signal-to-noise spectra from the MMT and Keck observatories which yield measurements of the most prominent emission lines in the rest-frame interval 1100A˚ – 1700A.˚ Composite spectra for the whole data set and for subsets were constructed and > analysed in order to investigate: a) the emission properties of z ∼4 QSOs in comparison with those of their lower redshift counterparts, b) the luminosity dependence of the emission features, and c) the role of selection effects in existing samples. There are several conclusions that may be inferred from the work we have outlined here: > 1. In terms of their ultraviolet rest-frame spectra, z ∼4 QSOs strongly resemble quasars at low redshift. Subtle differences are present, however, and in particular, Lyα, N V, C IV and O I are stronger in the high z sources. The C IV line is also > systematically narrower in the z ∼4 objects. 2. Among the high redshift QSOs, a weak Baldwin Effect is possibly present in Lyα but not in C IV or other lines. The lack of a strong trend may reflect the limited > span of luminosity in the existing z ∼4 samples. 3. Selection effects favoring strong-lined objects are a significant concern for surveys of high redshift QSOs. Our sample includes several sources with extremely strong, narrow and peaked lines that may arise from such a bias. However, quantita- > tive comparisons with other existing surveys of z ∼4 QSOs, including those discovered by SDSS, suggest that our sample overall is not strongly biased in this way. 4. All z > 4 composite spectra show strong N V and O I. The unusual strengths of these features and the narrow C IV profile suggest the presence of characteristic structural differences in the BLR, and also high abundances of heavy elements (solar or up to several times solar metallicities) in quasar environments at these early times. These findings are consistent with standard chemical evolution scenarios for young, massive bulge-dominated galaxies. 59

Chapter 3 Ultraviolet and optical properties of Narrow-Line Seyfert 1 galaxies

Originally published as:

Constantin, A., & Shields, J. C., PASP, 115, 592

3.1 Introduction

Narrow-Line Seyfert 1 galaxies (Osterbrock & Pogge 1985; Goodrich 1989) are a subclass of AGNs that manifest a distinctive ensemble of properties. They are rather rare objects that exhibit relatively narrow broad lines, steep X-ray spectra, strong Fe II emission, and weak emission from the narrow line region; they are more variable in X-rays than the Broad-Line Seyfert 1 objects (“normal” Seyfert 1s, hereafter Sy1), and exhibit pronounced soft X-ray excesses. They seem to cluster at one extreme end of the Boroson & Green (1992) “Eigenvector 1” (EV1) relation, as a result of their tendency toward weak [O III] λλ4959, 5007 emission, and narrow and blue-asymmetric Hβ proﬁles. Understanding this EV1 is important for NLS1s in particular and for AGN in general since it may be closely linked and possibly driven by the central engine parameters, in particular L/LEdd, the Eddington ratio (Boroson & Green 1992; Boroson 2002). However, whether this is the main and the only physical parameter that controls the NLS1 classiﬁcation, with its distinctive features, is still a matter of debate. To date, NLS1 studies are built on either individual objects or on samples for which the spectra span only narrow wavelength ranges; also, with the exception of the X-ray observations, these samples are small in general. In particular, detailed investigations 60

of the NLS1 blue/UV emission properties have been limited (e.g., Rodrigues-Pascual et al. 1997; Kuraszkiewicz & Wilkes 2000). Studies of their spectral energy distributions (SEDs) that cover wide bandpasses at a single epoch are almost completely missing. This kind of data is particularly useful for testing and constraining models proposed for these sources. Further examination of larger samples of NLS1 emission spectra is clearly desirable. Understanding the nature of NLS1s requires a detailed description of their average behavior. Therefore, a first goal of the present study is to obtain a comprehensive spectral characterization of the typical NLS1 galaxy. The mounting number of high quality HST spectra of NLS1 sources allows for a better definition of their spectral properties in general, and their UV emission in particular. In this study, we make use of HST archival observations of 22 NLS1s, a sample which is nearly twice as large as any of those used in previous studies of the NLS1 ultraviolet line and continuum emission. This database includes several objects whose observations cover a wide wavelength range (Lyα to Hα region) permitting thus, for the first time, a simultaneous survey of the NLS1 UV and optical spectral features. We construct average and median NLS1 composite spectra and provide measurements of the resulting underlying continuum, and the line strengths, ratios, and widths. The second important question we attempt to address here is the degree to which > NLS1s and high redshift (z ∼4) quasars share common properties. The possible connection between these two classes of objects has been suggested by Mathur (2000), who proposed a scenario in which both NLS1s and high z QSOs are in an early evolutionary phase, such that accretion proceeds at or near the Eddington limit. This analogy is mostly based on the indications for high metallicities in both of these categories of sources, and on the presence of an enhanced low-velocity component in the UV spectra of high redshift quasars. However, no direct comparison of their emission properties has been attempted yet. This is primarily due to their wide separation in redshift, which makes it difficult to observe them in the same spectral ranges. The current availability of high-quality UV spectra for the low redshift objects allows us to test the validity of this picture. We examine and compare the emission properties 61

> of the NLS1s and z ∼4 quasars, in order to determine the extent of the similarities exhibited by these two classes of objects. This comparison provides additional insights into AGN behavior as a function of redshift, luminosity, metallicity, and other physical parameters.

3.2 The NLS1 sample & data processing

Non-proprietary, archival HST Faint Object Spectrograph (FOS), Goddard High Resolution Spectrograph (GHRS), and Space Telescope Imaging Spectrograph (STIS) spectra of all objects known to us that were previously identified in the literature as < −1 NLS1 galaxies (i.e., FWHM(Hβ) ∼2000 km s , [O III]/Hβ < 3), were retrieved from the Space Telescope Science Institute in the form of calibrated data. A number of these objects are borderline sources in terms of this specific classification; however, they resemble the NLS1 characteristics by every other definition, and they have been extensively used in other NLS1 studies. Table 3.1 summarizes the instrumental setup and the resulting total wavelength coverage corresponding to each observation. References relevant to the classification of these objects are also listed. Each line in the table refers to observations obtained under a single observing program; in all but one case (I Zw 1), these observation sets were taken at a single epoch. The data obtained under separate observing programs are listed as different lines. In general, observations of each object were acquired with multiple gratings. The full wavelength coverage was obtained by co-adding the individual spectra, after re- sampling to a common dispersion (the lowest number of A/pixel,˚ to avoid loss of information), and flux-averaging in the overlapping regions. As indicated in Table 3.1, there are several galaxies for which multiple observations were obtained at different epochs and in some cases with different instruments. Because these sources are characterized by significant variability, the continuum level often needed minor < rescalings (∼10%) before averaging. The use of multiple instrumental configurations translated also into different spectral resolutions R. To obtain the final observed spectrum, it was necessary to convert the individual observations first to a common 62

R, by gaussian smoothing of the spectra with high resolution, and second, to a common dispersion. For all objects, the averaging in the common wavelength ranges was performed by weighting the spectra by the reciprocal of their noise variance. Many of these spectra display significant resonance line absorption. The main focus of this analysis is on the emission lines, and therefore, accurate measurements of these features required correction for the intervening absorption. When this appears as narrow lines superimposed on either the emission lines or the continuum, we removed the absorption feature through a simple interpolation. When the absorption was severe, especially near or within some of the strong emission features (Lyα, N V, and C IV), a conservative reconstruction of the line was attempted by low-order polynomial fitting. For these particular cases, as indicated in Table 3.2, the line flux and equivalent width (EW) measurements are considered to be only lower limits. The objects in this sample are subject to small amounts of Galactic foreground extinction, spanning AV = 0.03 − 0.31, with a typical value of ∼ 0.05 (Schlegel, Finkbeiner & Davis 1998). We corrected the spectra for the resulting reddening using the empirical selective extinction function of Cardelli, Clayton & Mathis (1989). No correction for intrinsic reddening was attempted. In order to combine the individual source observations into a single composite spectrum, a proper alignment in wavelength of the emission lines is necessary. This requires, in turn, consistent measurements of the redshift values z used in the Doppler- correction process. Therefore, we remeasured the redshifts using the C IV λ1549 emission-line, which is present in 21 out of 22 objects (the C IV feature is a doublet λλ1548.2, 1550.7, with the simple average of λ1549.5, corresponding to the optically thick case, consistent with photoionization models). The observed wavelength for each case was established by fitting a Gaussian to the top 20% of the profile. For situations where the measured line was significantly affected by absorption or low S/N, we verified the result with fits to the top 50% of the profile. The resulting z determinations (also listed in Table 3.2) are in good agreement with previously published values, with an estimated uncertainty of ≤ 0.001, and lead to a good superposition of the principal emission lines in the rest-frame. 63

Table 3.1. NLS1 Sample

Object Instr. Gratings Coveragea Refb

Ark564 FOS G130H,G190H,G270H,G400H,G570H 1087–6817 6,9,10,11

1H0707-495 STIS G140L,G230L 1122-3155 6

IRAS13224-3809 STIS G140L,G230L 1117-3148 1,6

IRAS13349+2438 FOS G190H,G270H 1572-3294 4

KUG1031+398 FOS G130H,G190H,G270H 1087-3301 8,11

Mrk110 STIS G140M 1194-1250 2,11

Mrk335 FOS G130H,G190H,G270H 1087-3301 2,11

GHRS G160M 1221-1257

Mrk478 FOS G130H,G190H,G270H 1087-3277 2,5,11

STIS G140M 1194-1300

Mrk486 FOS G190H,G270H 1567-3293 2,11

Mrk493 FOS G130H,G190H,G270H,G400H,G570H 1087–6817 7,11

Mrk766 STIS G140L,G230L 1087-3151 3,5,7,11

NGC4051 STIS E140M 1140-1729 6,11

PG1211+143 FOS G130H,G190H,G270H 1087-3275 2,11

GHRS G140L 1190-1477

STIS G140M 1194-1300

PG1351+640 FOS G130H,G190H,G270H 1087-3301 2

STIS G140M 1194-1300

STIS G230L 1568-3151

PG1404+226 FOS G130H,G190H,G270H,G400H 1087-4780 2,11

PG1411+442 FOS G130H,G190H,G270H 1087-3276 2

STIS G230L 1572-3157 64

Table 3.1 (cont’d)

Object Instr. Gratings Coveragea Refb

PG1444+407 FOS G130H,G190H 1087-2330 3

RX J0134-42 FOS G130H,G190H,G270H,G400H,G570H 1087–6817 5

TonS180 STIS G140M 1194-1299 5,11,12

STIS G140L,G230L 1120-3160

WPVS007 FOS G130H,G190H,G270H,G400H,G570H 1087–6817 5,13

IZw1 FOS G130H,G190H,G270H 1087-3276 2,11

FOS G190H,G270H 1568-3295

FOS G270H 2222-3277

GHRS G160M 1221-1257

IIZw136 STIS G140M 1194-1300 2

GHRS G140L 1153-1739

a given in A˚

bReferences to optical emission-line measurements that led to their NLS1 classiﬁcation

References. — (1) Boller et al. (1993); (2) Boroson & Green (1992); (3) Goodrich (1989); (4) Grupe et al. (1998); (5) Grupe et al. (1999); (6) Leighly (1999); (7) Osterbrock & Pogge (1985); (8) Puchnarewicz et al. (1995); (9) Stirpe et al. (1990); (10) van Groningen (1993); (11) V´eron-Cetty, V´eron & Gon¸calves (2001); (12) Winkler (1992); (13) Winkler et al. (1992) 65

Table 3.2. NLS1 Properties

a b α c e Object z logLν(1450) α(fν ∝ ν ) EW(Civ) ΓROSAT Ref

Ark 564d 0.0239 28.25 -1.34±0.01 31.6 3.4±0.1 1

1H0707-495d 0.0355 29.01 -0.46±0.01 20.5 2.3±0.3 3

IRAS13224-3809 0.0580 29.03 -0.44±0.02 33.0 4.5±0.1 1,3

IRAS13349+2438 0.1038 28.55 -3.23±0.06 55.3 2.8±0.1 3

KUG1031+398 0.0432 28.10 -0.99±0.02 92.4 4.3±0.1 1

Mrk110 0.0340 ··· ··· ··· 2.4±0.1 4

Mrk335 0.0260 29.23 -0.64±0.01 77.7 2.9±0.1 3

Mrk478 0.0755 29.86 -0.68±0.01 39.7 3.1±0.1 1,3

Mrk486d 0.0387 27.84 -2.84±0.09 50.0 ··· ···

Mrk493 0.0311 28.35 -0.73±0.01 107.5 2.7±0.2 1

Mrk766d 0.0127 26.13 -4.41±0.04 98.1 2.7±0.1 1,3

NGC4051d 0.00245 26.38 -1.86±0.04 75.5 2.8±0.0 3

PG1211+143 0.0804 29.75 -1.14±0.01 66.5 3.1±0.2 3,4

PG 1351+640d 0.0875 29.65 -1.35±0.01 115.0 2.5±0.6 4

PG 1404+226d 0.0935 29.22 -0.91±0.01 44.7 4.1±0.2 3,4

PG 1411+442d 0.0875 29.68 -0.98±0.03 148.0 3.0±0.5 4

PG1444+407 0.2640 30.55 -0.81±0.04 42.1 ··· ···

RX J0134-42d 0.2360 30.31 -0.12±0.01 27.5 7.7±2.6 2

TonS180 0.0613 29.63 -0.76±0.01 33.0 3.0±0.1 3 66

Measurements of the emission features in the individual spectra were carried out with line profile fitting. This was in general unsuccessful when single Gaussians were used, as they tend to lose flux from the wings if they are prominent, and from the peak if the cores dominate. The overall shapes of the lines measured in these NLS1 spectra were generally well represented by single Lorentzian fits, which is consistent with previous findings (Moran, Halpern & Helfand 1996; Leighly 1999; Véron-Cetty, Véron & Gon¸calves 2001). The C IV equivalent widths are also listed in Table 3.2, along with the rest-frame continuum luminosity measured at 1450A˚ [Lν (1450), expressed in ergs s−1 Hz−1]. The EW measurements may have errors resulting from the choice of continuum placement; experiments with alternative continuum fits suggest a systematic uncertainty of ∼ 15%. Another parameter recorded in Table 3.2 is the spectral index α of the power- α law fit (defined by Fν ∝ ν ) that best approximates the continuum shape of each individual object spectrum in the rest-frame. In determining the continuum solution, we tried to make use of the wavelength ranges which contain pure continuum emission. A reliable fit is obtained when a wide separation of the continuum windows is available (e.g., ∼ 1100 – 4000A),˚ such that the Fe emission, which is strong and prevalent in the UV and the blue part of the spectra of NLS1s, is avoided. In the present sample, there is only a small number of objects for which the available spectra cover completely and/or extend redward of the broad Fe II and Fe III emission-line complexes. For the sources spanning only the UV range, the measured spectral index should be considered a lower limit, as the apparent continuum can be strongly contaminated and reddened by the Fe emission. The full range of slope values and their distribution among the sample are displayed in Figure 3.1, along with the calculated median and average values. In identifying the Fe features, and consequently the emissionless continuum windows, we used information based on high resolution spectra of I Zw1, as provided by Vestergaard & Wilkes (2001) in the UV bandpass, and by Oke & Lauer (1979) in the regions redward of Mg II λ2797.9. The windows chosen in determining the power law continuum shape are placed around 1140, 1285, 1320, 1350, 1450, 3810, 3910, 4040, 4150, 5470, 5770-5800, and 6210A,˚ with the interval lengths ranging from 10 67

Table 3.2 (cont’d)

a b α c e Object z logLν(1450) α(fν ∝ ν ) EW(Civ) ΓROSAT Ref

WPVS007 d 0.0280 27.78 -1.41±0.01 112.2 9.0±2.0 2

IZw1 0.0585 29.40 -1.75±0.01 50.0 3.1±0.1 1,3

IIZw136 0.0619 29.59 -0.73±0.04 61.2 3.2±... 4

a calculated from the observed spectra using the C IV emission-line; the only exception is Mrk110 for which the line is not available, and the redshift is taken from the literature b −1 −1 to ease comparison with earlier published work, an H◦ = 50 km s Mpc ,

q◦ = 0.5, Λ = 0 cosmology was adopted. The values reflect flux measurements which are corrected for Galactic extinction; typical uncertainties are ∼0.05 dex. crest-frame EWs (in A)˚ calculated by fitting a single Lorentzian profile to the line dEmission lines are contaminated by absorption; indicated measurements (EWs), obtained from the polynomial interpolated profiles, should be regarded as lower limits e References for the ΓROSAT values

References. — (1) Boller, Brandt & Fink (1996); (2) Grupe et al. (1998); (3) Leighly (1999); (4) Wang et al. (1996) 68

α Figure 3.1 Histogram of continuum spectral indices, α, where Fν ∝ ν , for all but one (Mrk110) NLS1 in the sample. The average and median values are indicated. to 20A.˚ These wavelength ranges are also used in ﬁtting the composite spectra (see Section 3.3.1), and they are illustrated in Figure 3.2.

3.3 The NLS1 composite spectra

In this section we present NLS1 composite spectra, and resulting measurements of continuum and emission-line characteristics. The use of composite spectra is com- plementary to analysis of individual spectra. Composites oﬀer higher S/N ratios, allowing measurement of weak features, while also directly providing a description of typical NLS1 spectral properties. We examine the results in relation to similar observations of broader AGN samples, in order to explore the extent to which NLS1s represent a distinct subclass. These comparisons potentially provide useful tests for discriminating between model scenarios for the NLS1 phenomenon. 69

3.3.1 Overall continuum

When building spectral composites, one of the most challenging tasks is the gen- eration of an underlying continuum that reflects the typical appearance of the sample as a whole. Francis et al. (1991) noted that “there is no ‘correct’ way of co-adding spectra that exhibit differences on many different scales.” Our sample consists only of NLS1 sources, which, as a subcategory of the AGN classification, share a set of common attributes. The spectra nonetheless show a substantial diversity in their properties. As Figure 3.1 and Table 3.2 reveal, the continuum shape, as described by the power-law index α, exhibits a considerable variety among individual NLS1 spectra. Because only very few objects in the sample span the whole wavelength range, a simple median or average composite displays artificial discontinuities at locations where the number of contributing spectra changes significantly. Some care is thus required in order to use these composites to describe typical continuum properties for NLS1 galaxies. Figure 3.2 presents the NLS1 average and median composites, constructed using data from individual objects normalized to the mean flux in the wavelength range 1430 A–˚ 1470 A.˚ Each spectrum was given equal weight, thus avoiding biasing the resulting composites toward the brightest objects (i.e., those with the highest S/N). The median displays a smoother continuum than does the average, and is well described by a single power law (α = −0.798 ± 0.007) from just redward of Lyα to Hγ. Owing to the relatively high S/N of the spectrum and the wide separation of the fitted regions, the statistical uncertainty in the spectral index, as given by a Chi-square minimization method, is quite small. However, the value itself is rather sensitive to the precise wavelength sections employed in the fitting. Redward of Hγ, the continuum flux density rises above the level predicted by the UV power-law, and is best approximated by a separate power-law, with α = −2.38 ± 0.01. Since these objects are low-luminosity AGNs, contamination by the host galaxy starlight may contribute to this change in the spectral shape. Both fits are shown in the middle panel of Figure 2.5; the continuum windows used in the power-law fits are indicated as horizontal lines below the composite. Prominent emission features in the optical 70

range are labeled. The number of objects contributing to the composites at each wavelength is presented in the bottom panel of Figure 3.2. The UV-blue continuum spectral index for this HST NLS1 composite falls among the steepest values found in other AGN composite measurements (−1 <α< −0.4; see for example, Table 5 in Vanden Berk et al. 2001)1. Such red continua were measured only in the Zheng et al. (1997) and Telfer et al. (2002) HST composite spectra (α = −0.99, and α = −0.71 respectively), and in the Schneider et al. (2001) sample of Sloan Digital Sky Survey very high redshift quasars (average α = −0.93). There are several possible reasons why these particular samples show such steep continua, and we explore them in the following paragraphs, in an effort to understand the origins of the extreme continuum properties of the NLS1 composites. One effect that may contribute to the soft continuum slopes of the high z QSOs and the HST sources derives from the restricted wavelength baseline typically used in fitting the continua. The spectra of objects comprising these steep continuum samples make available only short regions redward of Lyα, in which the iron contamination may produce an artificial rise in the continuum profile (Telfer et al. 2002) . However, in the NLS1 composites, the wide baseline used in fitting the continuum permits a successful accounting for the Fe II and Fe III complexes, therefore making the Fe contamination an unlikely cause for their overall red continua2. The steepness of the HST composites can also potentially be attributed to an evolution of QSOs to softer spectra at lower redshifts and/or an observational bias toward detecting sources with harder continua at higher redshift (Francis 1993). The space-based observations include a significant number of low redshift sources, and this is the case for the NLS1 galaxies as well. The same evolutionary effect and the potential detection biases can also be responsible, at least partially, for the redness

1Rodrigues-Pascual et al. (2000) have also reported evidence that NLS1s are redder than other Seyfert 1s at optical wavelengths. 2The Fe contamination is the primary factor that accounts for the difference between the spectral index that best fits the NLS1 median composite and the median (and average) value of the power- law index distribution of the individual continuum fits (see Figure 3.1). The majority of the objects in the NLS1 sample cover only the UV wavelength region, where the power-law solution is unable to properly estimate the true underlying level below the Fe II and Fe III emissions, and as a consequence, steeper indices are measured. 71

Figure 3.2 Top panel: NLS1 composite spectrum plotted as log(Fλ) vs. rest-frame wavelength, with the principal emission features identified. The flux has been normalized to unit mean flux over the wavelength range 1430A–˚ 1470A.˚ Middle panel: The median composite spectrum, plotted on a linear scale, and zoomed near the continuum level for a better visualization of the weak features in the optical range. A more detailed UV line identification is presented in Figure 3.4. The power-law continuum fits are overplotted as dashed and dotted lines. Bottom panel: Number of NLS1s contributing to the composite as a function of rest-frame wavelength. 72

of the NLS1 composites, especially when compared with the ground-based composite spectra. The present sample shows little evidence of a correlation between continuum slope and z, but a considerably stronger trend relating α and luminosity (Fig. 3.3). Re- gardless of its origin, the luminosity dependence of spectral slope is probably largely responsible for the relatively steep α for the NLS1 composite, since the typical luminosity of our NLS1s is lower than that of sources employed in most other quasar composites.

3.3.2 Reddening and Absorption

The luminosity dependence of spectral slope could be intrinsic to the accretion source, but independent evidence suggests that it is mostly attributable to luminosity- dependent reddening. Internal dust, if present, is expected to be accompanied by gas producing observable absorption lines. In our sample, the signature of strong absorption near the systemic velocity of the host galaxy is present in almost half the objects (see Table 3.2). Median values of the spectral indices corresponding to subsamples of sources with and without strong resonance line absorption are −1.34 and −0.73 respectively, consistent with steeper (redder) continua in the absorbed NLS1s. Me- dian logarithmic luminosity values for the absorbed and unabsorbed subsamples are 29.01 and 29.60 respectively (mean values are 28.4 ± 0.4 and 29.3 ± 0.2 respectively), thus directly linking the presence of absorption with luminosity. The location and physical state of the absorbers in NLS1s may be important for understanding these objects. Intrinsic absorption related to the central accretion source is now known to be present in a signiﬁcant fraction of Seyfert 1 nuclei (Cren- shaw et al. 1999). NLS1s exhibit similarities in emission properties to low-ionization broad absorption-line QSOs (BALQSOs), which Boroson (2002) has interpreted as evidence for high luminosities relative to the Eddington value for both classes of object; in the BALQSOs, the absorption is clearly also closely related to the accretion process. In the NLS1s, the intervening gas may be a warm (highly ionized) absorber 73

Figure 3.3 Spectral indices plotted vs. redshift (upper panel) and 1450A˚ luminosity (lower −1 −1 panel), with Lν expressed in ergs s Hz . The Spearman rank coeﬃcient and the probability of the correlation happening by chance are indicated. The error bars in both directions are smaller than the symbol size, and therefore not indicated. 74

that is potentially dusty (e.g., Komossa & Greiner 1999); alternatively, the absorbing matter may be related to the “lukewarm absorber” identified in case studies by Kraemer et al. (2000) and Crenshaw et al. (2002), which may reside on kpc scales. The luminosity dependence of reddening and absorption for the NLS1s is consistent with a scenario in which the absorbing matter covers a larger solid angle as seen by the central source in lower luminosity objects, a natural expectation for interstellar matter in a disk-like geometry of luminosity-independent scale-height, but with a sublimation radius for dust that scales with luminosity, similar to the ‘receding torus’ model (Lawrence 1991). If the absorbing medium has a flattened distribution aligned with the host galaxy disk, we might then expect a correlation between the host inclination and the spectral (UV) color, as reported for several Seyfert 1s by Crenshaw & Kraemer (2001). Inclination values are available in the literature for 16 of our sample members; the data do not show a statistically significant trend with α, however. In summary, the extent to which the UV/optical absorbers in NLS1 nuclei are associated with the central accretion structure versus the normal host galaxy interstellar medium remains ambiguous. The nature of the dusty absorber in these galaxies can be further tested by ana- lyzing their soft X-ray (0.1 – 2.4 keV) characteristics. A “lukewarm” (low ionization) constitution of the gas would suggest strong absorption, that translates into flatter observed soft X-ray spectra. If a “warm” (high ionization) absorber is present, steeper X-ray continua should be measured (Grupe et al. 1998). The ROSAT spectral indices (Γ ≡ 1 − α) are available for the majority of the NLS1s employed in this study, and are recorded in Table 3.2. Splitting the sample into absorbed and unabsorbed objects, as before, and calculating the median value of the X-ray spectral slope for each subgroup, should provide us a rough criterion for distinguishing between the two types of absorbers. The resulting values are Γabsorbed spectra = 3.4 and Γunabsorbed spectra = 3.1, nominally implicating the presence of highly ionized absorbers (mean values are 4.2 ± 0.8 and 3.3 ± 0.2 respectively); however, the difference in the median values is small, and it appears likely that the absorbing medium in these objects is described by a range of properties. 75

3.3.3 The UV spectrum

Figure 3.4 shows the average spectrum in the UV range, where most of the objects (≥ 18) in the sample are contributing. The standard deviation (RMS) and the standard deviation of the mean spectra show the degree of variation between the members of the sample and the uncertainty of the average composite as a function of wavelength. Near the wavelength corresponding to the common continuum normalization (λ ∼ 1450A),˚ the spectral variation within the NLS1 sample is dominated by the modulations in the cores of the strongest emission lines (Lyα, C IV, He II, and C III]), which are in general under 15%. In this respect, NLS1s behave like the other more general samples of AGNs (Francis et al. 1992; Brotherton et al. 2001). Away from the normalization wavelength interval, the bulk of the spectral variance is accounted for by diﬀerences in the individual continuum shapes and noise.

3.3.4 The optical composite

> For wavelengths λ ∼ 3000A,˚ the HST NLS1 spectrum is based on only a small number of objects. Because fewer than 5 NLS1s make up the composites at these wavelengths, the individual source contributions are more pronounced, making it difficult to identify and measure the spectral features (continuum and emission lines) located at the transition points in the number of contributing objects (e.g., at ∼ 4360A,˚ ∼ 5500A).˚ A smoother spectrum is constructed using only the three spectra that span the whole wavelength range (Ark564, Mrk493, WPVS007), and is presented in Figure 3.5. We compare our optical median composite with the Sulentic et al. (2002) median NLS1 spectrum, constructed from a much bigger sample (24 sources, ground-based spectra). The two spectra are very similar in both the continuum shape and the profile and strength of emission features. Small differences, like the stronger peaks in the emission lines in the HST composite, are mostly due to the different spectral resolution that characterizes the two samples ( 4 – 7A˚ FWHM for the Sulentic spectra, ∼ 2A˚ for the FOS-HST observations). The good agreement between the two medians suggests that the HST composites are representative of typical NLS1 optical spectra, despite the fact that they are constructed from such a 76

Figure 3.4 Top panel: NLS1 average composite spectrum plotted for the UV range only, where at least 18 objects are contributing. The same normalization as in Figure 2.5 is used.

The prominent emission features are marked. Middle panel: Flux standard deviation in Fλ relative to the average composite spectrum, as a function of rest-frame wavelength for the individual spectra comprising the composite. Bottom panel: The standard deviation of the mean, expressed in percent. 77

Figure 3.5 The optical range of the HST NLS1 median composite, constructed using only the 3 objects that span the whole spectral range. The same ﬂux normalization as in Figure 3.2 is used. The Sulentic et al. (2002) NLS1 median composite, based on 24 ground-based object spectra, is shown for comparison. The strong similarity between the two medians suggests that the HST NLS1 composites are representative of these objects. small object sample.

3.3.5 Emission Lines

The richness of the line emission in the AGN spectra is easily distinguishable in the NLS1s due to their narrow widths. Hence, careful identiﬁcation and intensive analysis of the NLS1 emission lines have been performed, though to date, only on individual sources, e.g. I Zw1 (Laor et al. 1997; Vestergaard & Wilkes 2001, and references therein). When composites are built using relatively large samples, higher S/N is achieved, and therefore, higher accuracy is expected in identifying and measuring the emission lines. We have been able to detect and parametrize in the HST NLS1 composites a large number of emission features. Many are barely present 78

or are heavily blended, and therefore difficult to quantify in other AGN composites (Francis et al. 1991; Vanden Berk et al. 2001; Zheng et al. 1997; Telfer et al. 2002). The emission lines are measured from the average composite, as it presents a higher S/N ratio than the median spectrum. An exception is made for the wavelength interval ∼ 2900 − 3170 A,˚ where the emission features in the average spectrum are distorted by the jumps present in the continuum shape (see Section 3.3.1), and where the median is used. Line fluxes, strengths (EWs), widths (FWHM) and line shifts relative to the laboratory central wavelengths (∆v) are listed for all detected features, along with their 1 σ error bars, in Table 3.3. The lines have been identified by matching wavelength positions and respective relative strengths of features found in the Francis et al. (1991), Zheng et al. (1997), Vanden Berk et al. (2001) and Telfer et al. (2002) composites, and in the detailed analyses of I Zw1 (Laor et al. 1997; Vestergaard & Wilkes 2001). Tentative identification of newly discovered features is based mainly on data available from the Atomic Line List3 and Verner et al. (1996).

All measurements, including the peak positions of each emission line, λmean rest, and the respective errors are generated using the task specfit (Kriss 1994) as implemented in the IRAF4 software package. The method employs line and continuum spectral fitting via an interactive χ2 minimization. The fit is performed in 14 separate spectral intervals; their lengths varied, in order to keep a reasonable number of parameters, from ∼ 150A˚ in the UV-blue region, rich in prominent emission features, to ∼ 500 – 900A˚ in the optical range. The equivalent widths are measured relative to the resulting local continua, which differ from the shape estimated in Section 3.3.1, mainly due to the fact that the average and not the median composite is used in the fitting process. Comparisons indicate that line fluxes measured in the average and < median composites are consistent to within ∼10% in most cases.

3The Atomic Line List is hosted by the Department of Physics and Astronomy at the University of Kentucky; http://www.pa.uky.edu/∼peter/atomic. 4The Image Reduction and Analysis Facility (IRAF) is distributed by the National Optical As- tronomy Observatories, which is operated by the Association of Universities for Research in Astron- omy Inc. (AURA), under cooperative agreement with the National Science Foundation. 79

Table 3.3. Emission-line measurements

c d f g h λlab λmean rest ∆v Rel. Flux EW FWHM

Linea (A)˚ (A)˚ (km s−1) [100 ×F/F(Lyα)] (A)˚ (km s−1)

C III? 1175.5 1175.5±0.5 0±130 0.9±0.2 1.1 1190± 690

Si II 1197.4 1199.3±0.7 480±170 0.5±0.2 0.6 790± 500

Si III 1206.5 1206.7±1.9 50±470 0.4±0.6 0.5 780±1350

Lyα 1215.7 ··· ··· 100.0±1.6 131.8 · · ·

..narrow 1215.7 1216.3±0.1 150±25 69.5±0.9 91.6 1480± 10

..broad 1215.7 1212.6±1.4 –760±340 30.6±1.0 40.2 7000± 280

N V 1240.8 1240.0±0.1 –190±20 13.0±1.5 17.6 2110± 130

Si II? 1248.4 1245.0±0.4 –810±90 3.7±1.0 5.0 2200± 400

Si II 1264.8 1262.9±0.9 –450±210 3.5±0.4 4.8 3510± 350

O I 1303.5 1305.4±0.3 440±70 3.7±0.6 5.5 2060± 130

Si II 1309.3 1309.1±1.5 –40±340 1.0±0.7 1.5 2030± 280

C II 1335.3 1336.5±0.3 270±70 2.6±0.3 4.0 1930± 260

N IIb 1344.6 1349.0±0.9 980±200 1.1±0.3 1.7 2020± 430

Si IV+O IV] 1400.0 ··· ··· 10.5±3.2 17.2 · · ·

..Si IV 1393.7 1393.7±0.7 0±150 3.1±2.2 5.1 1520± 600

..O IV] 1401.4 1404.6±1.5 680±320 5.4±2.3 8.9 2030± 440

..Si IV 1402.7 1399.9±2.7 –590±570 2.0±0.7 3.3 1520± 690

N IV] 1486.5 1486.8±0.9 60±180 0.7±0.3 1.2 2020±1910

Si II 1530.1 1536.9±0.2 1330±40 4.6±0.8 7.9 2030± 280

C IV 1549.5 ··· ··· 44.8±1.7 77.1 · · ·

..narrow 1549.5 1549.5±0.1 0±20 33.3±1.3 57.3 1900± 60

..broad 1549.5 1541.6±8.2 –1520±1580 11.5±1.1 19.8 12000±2500 80

Table 3.3 (cont’d)

c d f g h λlab λmean rest ∆v Rel. Flux EW FWHM

Linea (A)˚ (A)˚ (km s−1) [100 ×F/F(Lyα)] (A)˚ (km s−1)

He II 1640.4 ··· ··· 23.3±2.6 41.5 · · ·

..narrow 1640.4 1639.7±0.2 –120±40 6.4±0.8 11.4 1780± 240

..broad 1640.4 1630.4±3.0 –1820±540 16.9±2.5 30.1 11990±4090

O III] 1663.5 1665.4±0.2 340±40 4.1±1.1 6.2 2500± 480

Al II 1670.8 1680.0±1.5 1650±260 1.3±1.0 1.9 2250±1330

N II 1725.2 1725.6±0.9 70±150 2.2±0.9 3.3 4710±2100

N III] 1750.5 1750.5±0.3 0±50 2.0±0.2 3.0 1480± 170

Fe II UV191 1785.4 1785.3±0.7 –20±110 1.2±0.2 1.8 1430± 250

Si II 1814.1 1816.9±0.5 460±80 0.6±0.2 0.9 1250± 730

Al III 1858.8 1853.5±1.1 –850±170 1.3±0.5 1.9 2240± 540

Fe III UV52 1867.9 1864.1±0.8 –610±120 1.7±0.5 2.6 2760± 730

Si III] 1892.0 1892.7±0.1 110±20 3.6±0.5 5.5 1250± 130

C III 1907.9 ··· ··· 21.1±5.7 32.2 · · ·

..narrow 1907.9 1908.2±0.1 50±20 10.0±0.6 15.3 1380± 10

..broad 1907.9 1907.7±4.6 –30±720 11.0±5.7 16.9 12230±1110

N II] 2141.4 2144.0±0.9 360±130 0.3±0.2 0.4 1040±1140

C II] 2327.5 2326.6±0.8 –120±100 3.7±1.3 5.7 2540±430

Fe III UV 47 2419.3 2422.8±0.2 430±30 3.2±0.5 5.1 1560±200

O II 2438.8 2439.3±0.6 60±70 0.7±0.2 1.2 1060±250

[O II] 2470.9 2467.6±0.8 –400±100 0.9±0.3 1.5 1570±470

C I]b 2478.6 2484.0±0.1 650±10 2.3±0.6 3.7 2580±510 81

Table 3.3 (cont’d)

c d f g h λlab λmean rest ∆v Rel. Flux EW FWHM

Linea (A)˚ (A)˚ (km s−1) [100 ×F/F(Lyα)] (A)˚ (km s−1)

Al II] 2669.9 2674.7±0.4 540±50 0.8±0.6 1.3 1050±430

Mg II 2797.9 ··· ··· 31.4±6.6 58.0 · · ·

..narrow 2797.9 2801.6±0.3 390±30 11.2±2.2 20.7 1680±60

..broad 2797.9 2798.3±0.4 40±40 20.2±6.5 37.4 4620±1090

O IIIe 2960.6 2957.4±1.7 –320±170 0.3±0.2 0.9 850±600

Fe IIe 2964.3 2966.7±3.6 240±360 0.1±0.1 0.2 830±900

O IVe 2982.5 2981.9±2.5 –60±250 0.3±0.3 0.9 820±610

Ne III]e 2986.9 2990.0±0.1 310±10 0.2±0.1 0.5 830±680

F Ve 3109.0 3109.5±0.4 50±40 0.6±0.1 2.0 550±10

O IIIe 3122.5 3122.9±1.3 40±120 0.8±0.4 2.7 910±370

O IIIe 3133.7 3134.2±1.3 50±120 1.8±0.4 6.0 940±210

C IIe 3166.7 3165.3±0.6 –130±60 0.8±0.3 2.7 760±390

He I 3188.6 3190.6±1.7 190±160 0.7±0.1 1.6 990±230

O IVb 3216.8 3215.7±0.5 –100±50 0.5±0.1 1.1 830±120

Ne IIb 3230.6 3230.9±0.8 30±70 0.9±0.2 2.1 940±260

Fe I Opt91 3261.2 3260.8±0.7 –40±60 1.0±0.1 2.5 840±130

Fe II Opt1 3281.2 3281.6±0.5 40±40 1.2±0.1 3.0 840±70

[Fe III]b 3308.5 3307.8±2.4 –60±220 0.7±0.1 1.9 830±110

[Ne V] 3346.8 3345.3±0.2 –130±20 0.5±0.3 1.2 790±260

[Ne V] 3426.8 3425.3±0.2 –130±20 1.2±0.3 3.3 790±260

Fe II]b 3495.6 3496.2±0.5 50±40 1.5±0.1 4.4 880±100 82

Table 3.3 (cont’d)

c d f g h λlab λmean rest ∆v Rel. Flux EW FWHM

Linea (A)˚ (A)˚ (km s−1) [100 ×F/F(Lyα)] (A)˚ (km s−1)

[Fe VII] 3587.1 3584.1±1.0 –250±80 0.6±0.3 1.8 1080±180

O II]b 3621.0 3620.6±1.0 –30±80 0.6±0.1 1.8 1280±310

N III]b 3685.9 3686.3±0.8 30±60 0.5±0.3 1.5 1170±110

[O II] 3727.1 3727.7±0.2 50±20 1.0±0.1 3.3 610±70

[O II] 3729.8 3730.3±0.2 50±20 0.1±0.1 0.2 610±70

[Ne II] 3745.7 3749.0±0.1 290±20 0.4±0.1 1.0 870±250

[Fe VII] 3759.7 3758.3±0.6 –110±60 0.5±0.2 1.4 790±350

[Fe IV] 3769.7 3769.4±1.2 –30±100 0.3±0.1 1.1 790±140

Fe I 3785.3 3787.2±1.8 150±140 1.7±0.1 5.6 3620±190

[Ne III] 3870.1 3868.3±0.3 –140±20 1.1±0.1 4.0 680±190

He I 3889.7 3890.5±2.5 60±190 0.7±0.2 2.6 920± 410

O III] 3938.7 3939.2±1.0 40±70 0.4±0.1 1.5 830±190

[Ne III] 3968.9 3966.9±0.3 –150±20 0.3±0.1 1.2 660±190

O II]b 3969.6 3971.3±0.5 130±40 1.1±0.1 4.0 830±70

[S II] 4069.7 4069.5±0.8 –20±60 0.2±0.1 0.8 600±270

[S II] 4077.5 4077.3±0.7 –20±50 0.1±0.1 0.3 600±260

Hδ 4102.9 4105.6±0.7 200±50 3.7±0.1 14.9 2240±160

Fe III]b 4180.0 4181.7±0.8 120±60 1.4±0.1 5.8 1220±220

Fe II 4234.3 4239.2±1.0 350±70 1.3±0.1 5.7 1010±150

Hγ 4341.6 4344.7±0.2 210±10 7.3±0.2 26.3 1070±50

[O III] 4364.4 4367.4±0.1 210±10 0.7±0.1 2.4 270±60 83

Table 3.3 (cont’d)

c d f g h λlab λmean rest ∆v Rel. Flux EW FWHM

Linea (A)˚ (A)˚ (km s−1) [100 ×F/F(Lyα)] (A)˚ (km s−1)

He Ib 4472.7 4477.2±3.1 300±200 0.6±0.2 2.2 1270±280

He II 4687.0 4688.6±0.9 100±50 4.9±0.1 19.0 2890±80

Hβ 4862.6 ··· ··· 20.5±1.1 81.5 · · ·

..narrow 4862.6 4864.6±0.1 120±10 17.5±0.3 69.6 1020±10

..broad 4862.6 4864.5±9.9 120±610 3.0±1.1 11.9 6890±200

[O III] 4960.2 4963.4±0.1 40±10 2.3±0.1 9.0 260±10

[O III] 5008.2 5008.5±0.1 40±10 8.0±0.2 31.3 260±10

[Fe VII] 5159.8 5162.1±1.4 –130±130 0.9±0.3 3.8 1080±150

[Fe VI] 5177.5 5180.2±1.9 –240±190 1.2±0.3 4.8 860±250

[N I] 5200.5 5203.5±2.9 130±280 0.8±0.8 3.3 870±820

[Fe VII] 5277.3 5278.3±1.4 –130±130 1.8±0.3 7.4 1080±150

O IVb 5318.7 5318.2±0.7 –30±40 1.3±0.2 5.3 750±120

He I 5877.3 ··· ··· 2.3±0.1 11.1 · · ·

..narrow 5877.3 5878.4±1.1 60±50 1.5±0.1 7.5 970±70

..broad 5877.3 5881.4±9.1 210±460 0.7±0.1 3.6 5010±550

[Fe VII] 6087.9 6088.5±1.3 30±60 0.3±0.5 1.4 870±340

[Fe V] 6088.5 6088.8±1.5 20±70 0.4±0.6 2.0 870±290

[O I] 6302.1 6304.5±0.8 110±40 1.2±0.1 6.4 1150±150

[O I] 6365.5 6367.5±0.5 90±20 0.4±0.1 2.2 1140±150

[Fe X] 6376.3 6373.5±0.6 –130±30 1.2±0.1 6.5 810±90

[N II] 6548.8 6549.5±0.9 30±40 1.8±0.6 10.1 760±70 84

Table 3.3 (cont’d)

c d f g h λlab λmean rest ∆v Rel. Flux EW FWHM

Linea (A)˚ (A)˚ (km s−1) [100 ×F/F(Lyα)] (A)˚ (km s−1)

Hα 6564.6 ··· ··· 61.0±1.4 344.2 · · ·

..narrow 6564.6 6565.9±0.1 60±5 57.8±1.0 325.8 760±20

..broad 6564.6 6566.7±0.4 90±20 3.3±1.2 18.4 5500±1140

[N II] 6585.3 6585.9±0.9 30±40 6.1±0.6 34.3 760±70

aEmission lines are measured from the average composite spectrum unless otherwise noted bUncertain identiﬁcations

cvacuum wavelengths, both below and above 2200A˚

dgiven by the centroid of the Lorentzian fit f The line fluxes are normalized to the Lyα flux to ease comparison with published results for other AGN/quasar samples eLines measured from the median composite

gThe errors in EWs are not quoted as the relative errors are the same as for the ﬂuxes hIntrinsic width of the emission line; an instrumental broadening of 230 km s−1, the approximate FOS spectral resolution, high-resolution gratings, was assumed

For most features, the line emission is modeled using single Lorentzian proﬁles. However, the most prominent lines required two components, a narrow and a broad Lorentzian, the broad one being in general blueshifted relative to the narrow component. The broad and prominent Fe emission surrounding and redward of Mg IIλ2798 < < is not measured; thus, line measurements for 2000 A˚ ∼λ ∼4250A˚ may include some Fe contamination, and therefore, a high level of uncertainty. For the optical wavelength range, redward of Hγ, we use in the ﬁtting process the empirical Fe II template obtained by Boroson & Green (1992) from the I Zw1 spectrum. No broadening of the 85

iron template was necessary in order to match the line width of the NLS1 composite spectra. In fitting forbidden line doublets, the line pairs are assigned to have common velocity widths and offsets, and their flux ratios are constrained to values determined by branching ratios, when appropriate. Because the composites were constructed using redshifts based on the position of a single emission-line, C IV (see Section 3.2), a check for systematic velocity offsets for other lines can be performed. Such line shifts, which are definitely present for many of the lines listed in Table 3.3, have been detected previously in other AGN samples (Gaskell 1982; Espey et al. 1989; Tytler & Fan 1992; McIntosh et al. 1999; Vanden Berk et al. 2001). Note that most of the lines are redshifted relative to C IV, the reference line, and also, the largest values of the recorded velocity shifts correspond to the lowest ionization states (e.g., Si II, C II,O I). Figure 3.6 shows the velocity shifts versus the ionization potentials, compiled for all emission lines that are > stronger than C IIλ1335, i.e., EW ∼4.0 A,˚ and that have well defined non-blended peaks. We treat the permitted and the forbidden lines separately since they may have very different origins. The values for the ionization potentials are chosen such that, for the recombination lines of H and He, they express the energy necessary to ionize the respective state for later recombination, while for the collisionally excited lines from heavy elements, they represent the energy that is needed to create the ionization state. For the most prominent features, which are fitted with two components, we consider only the velocity shift measurements given by their narrow components (they are identified by different symbols in Figure 3.6, upper panel); this treatment is justified by the fact that the reference wavelength is chosen based on the location of the peak of C IV, which is measured using the narrow component fit5. This approach is consistent with measurements of velocity offsets performed in previous similar studies of other AGN samples (e.g. Vanden Berk et al. 2001). The results presented in Figure 3.6 provide evidence for an anticorrelation between the velocity shifts and the degree of ionization of the emitting species. The trend is

5 < −1 Single Lorentzian fits for these features result in line shifts that differ by ∼100 km s from the ones given by the narrow component only; this is due to the fact that the cores of these relatively narrow lines account for the bulk of the flux. 86

definitely present in both categories of emission features. For the forbidden lines, significant velocity shifts, usually blueshifts relative to the systemic velocity, have been detected in the past for the high ionization transitions (e.g., Penston et al. 1984; Appenzeller & Wagner 1991); however, the evidence for a correlation similar to that presented in this study has only recently been revealed by measurements from the high S/N Sloan Digital Sky Survey quasar composite spectrum (Vanden Berk et al. 2001). In both types of lines measured in the NLS1 sources, the amplitude of this correlation appears similar to that present in other, more heterogeneous Sy1/quasar samples6. Although the ionization potential seems to govern the magnitudes of the velocity offsets, there is a significant amount of scatter in this empirical relationship, in both Sy1/quasars and NLS1 objects, suggesting the influence of other parameters as well. The scatter in the NLS1 trend may be amplified by wavelength zero-point calibration uncertainties, particularly for the FOS spectra. The origin of the velocity offsets between the AGN emission lines is not well understood, but a promising means of interpreting this behavior is in terms of the disk-wind model for the broad-line region (BLR). In this picture, the high ionization lines are produced in outflowing winds, accelerated by radiative line driving, that arise from the accretion disk, which is the base for the low-ionization emission (e.g., Murray & Chiang 1998; Proga, Stone & Kallman 2000; Leighly 2001). Additionally, the disk and/or the radiative outflow itself, believed to proceed from near the plane of the disk, is assumed to be optically thick, so the emission from the receding wind is obscured. The fact that NLS1s and other broad-line AGNs show similar line shifts provides additional constraints for the physical models that best characterize the NLS1s. The preferred explanations for the primary drivers of the NLS1’s extreme measured properties postulate either higher ratios of their luminosity to the Eddington luminosity (e.g., Pounds et al. 1995), or a more pole-on view of an assumed disk-shaped BLR (e.g., Brandt & Gallagher 2000). The velocity offsets between lines reported here, and their similarity to those found in less restricted AGN samples, may present a

6The velocity shifts may be unusually large in I Zw 1; see Laor (2000). 87

Figure 3.6 Emission-line velocity offsets, relative to the rest frame (defined by Civ λ1549), as a function of ionization potential for selected emission lines (see text). Error bars show the 1 σ uncertainty in the velocity measurement. Permitted and semi-forbidden lines are shown in the top panel, and forbidden lines are shown in the bottom panel. Only one measurement is shown for each line, and the point type indicates the adopted profile. Overlapping points are slightly offset horizontally from each other for clarity. The points are labeled by ion. 88

challenge to either NLS1 scenario, if the velocity differences stem from a disk-wind phenomenon or something similar. For the orientation model, it would be surprising if the typical line-of-sight differences in line velocity remain essentially unchanged when the orientation is such that the line widths are diminished. For the high Eddington ratio picture, the offsets ultimately trace the disk outflow velocity, which is expected to scale with the local orbital speed, hence large velocity differences combined with small widths are again unexpected. The solution in either case may ultimately involve the fact that the narrow line widths characterize only a subset of emission features in NLS1s, notably including the Hβ feature, while other lines (particularly in the UV) still reveal high-velocity gas components. Indeed, as discussed by Wills et al. (2000), and elaborated by Shang et al. (2003), the UV line widths do not correlate in a simple way with Hβ width or with other NLS1 defining properties.

3.4 Are NLS1s the analogues of high z QSOs?

Based on the apparent similarity of some of the emission properties of NLS1s > and z ∼4 QSOs, Mathur (2000) suggested a connection between these two classes of AGNs, in the sense that the NLS1s are the low redshift, low luminosity analogues of the high z quasars. In the proposed picture, both categories of sources are in an early evolutionary phase, in which accretion proceeds at or near the Eddington limit. This scenario has several appealing aspects for explaining NLS1 phenomena, but additional tests are desirable to verify this idea, and especially to gauge its applicability to the high z sources. One of the key arguments employed in support of a NLS1 – high z QSO association > was the initial report of narrow UV lines in z ∼ 4 quasars by Shields & Hamann (1997). While the presence of an enhanced low-velocity component in the high redshift quasars was later confirmed by Constantin et al. (2002), the UV line profiles do not necessarily provide an adequate basis for linking the high z QSOs to a NLS1 classification, as this is based on optical lines. In fact, as noted above (Section 3.3.5), the UV and optical emission-line properties are substantially independent (Shang et al. 2003). 89

Another possible similarity between high z quasars and NLS1 galaxies is a high > metallicity (Z) in their emitting gas. In quasars at z ∼3, super-solar heavy element enrichments have been derived in specific abundance studies based on both emission (Dietrich et al. 1999; Dietrich & Wilhelm-Erkens 2000; Dietrich et al. 2003) and absorption properties (e.g., Hamann 1997). The high gas-phase metallicity is believed to derive from a recent episode of vigorous star formation. For the NLS1s, several lines of argument have suggested the presence of enhanced metallicity. One of the potential indicators of high abundances in these objects is their characteristically strong Fe II emission (Collin & Joly 2000). However, the excitation of this emission is complicated and its strength does not necessarily map in a simple way to the Fe abundance (Verner et al. 1996). Moreover, if the Fe is largely produced in Type-Ia supernovae, as would be expected, the evolution of the progenitor population and the > resulting enrichment should occur on a timescale ∼1Gyr; consequently, a high iron enrichment is not a strong indicator of youth, and therefore, would not support the hypothesis of NLS1s being galaxies in the making. A better tracer of abundances in NLS1 galaxies is provided by nitrogen. This element is of particular interest since its production is believed to be dominated by secondary enrichment, which translates into N/H ∝ Z2. Nagao et al. (2002) have recently discussed the forbidden line spectra of NLS1s, including [N II] λ6583, and argued that the observed line ratios suggest higher metallicities than in the Sy1s; as the authors also point out, however, this statement is very model-dependent. More robust diagnostics are obtained from studies of the N V λ1240 feature, and in particular, measurements of its strength relative to C IV λ1549 and He II λ1640 (Hamann et al. 2002, and references therein). Evidence for systematic trends of stronger N V and weaker C IV with increasing EV1 (which seems to trace NLS1 behavior, see Section 1) has been reported by Wills et al. (1999). Based on the same flux ratios, Shemmer & Netzer (2002) showed that the NLS1 sources deviate significantly from the well known relationship between Z and luminosity (L) in AGNs (Hamann & Ferland 1993), by exhibiting higher Z at a given L. Interestingly, several of the nine extreme NLS1s used in their study have line ratios as high as those 90 measured in the most luminous high z QSOs. Measurements for larger samples of NLS1s are clearly desirable to verify these findings. > In this section, we investigate the NLS1 – z ∼ 4 QSO analogy by directly comparing their emission-line properties. As most of the line emission studies of the NLS1 objects > are in the optical regions, and the corresponding (rest-frame) data for z∼ 4 are almost nonexistent, the UV spectral observations are the only accessible tool to exploit in such a comparison. We have available both NLS1s and high z quasar UV data, for samples that permit an adequate statistical analysis of their emission characteristics. Our study employs both comparisons of their composite spectra, and an eigenvector analysis that determines the degree to which the spectral variances throughout the samples share common properties. The NLS1 objects and composites used here are > described in Sections 3.2 and 3.3.1. For the z ∼ 4 QSO sample, we made use of 44 spectra of non-BAL quasars, presented by Constantin et al. (2002) and Schneider, Schmidt & Gunn (1991), which span the 1100 – 1700A˚ rest-frame range.

3.4.1 Direct comparison of composite spectra

Figure 3.7 shows the NLS1 average composite (Section 3.3.1) overplotted on the > z ∼ 4 QSO average composite (Constantin et al. 2002). In a first approximation, the spectra agree well: similar continuum shape and the same strong emission lines, > with comparable profiles. The pronounced Lyα forest in the z ∼4 composites does not reflect an intrinsic difference in the nature of the emission sources, but the expected increase in the opacity of the intergalactic medium at larger redshifts. The most evident distinction between the NLS1 and high z quasar composite spectra is in the strength of the principal emission features: Lyα, Si IV+O IV], C IV and He II. This difference in the line strengths is a clear manifestation of the Baldwin Effect (the inverse equivalent width – luminosity relationship, Baldwin 1977). The two samples differ significantly in their average luminosity, with the lower luminosity objects, the NLS1s, exhibiting stronger lines. Figure 3.8 shows the individual measurements of the rest-frame continuum luminosity, Lν(1450A),˚ and the strength (EW) of the C IV line, plotted for both the NLS1s and high z quasars. The Baldwin Effect 91

> Figure 3.7 Direct comparison between NLS1 and z ∼ 4 QSO average composite spectra. The same normalization as in Figure 3.2 is used.

correlation in C IV is clearly present over the combined sample; the segregation in luminosity of the two samples is also evident 7. Besides the basic Baldwin trend, the comparison in Figure 3.7 shows also the known trend with ionization: steeper Baldwin Eﬀect for more highly ionized species (e.g., Dietrich et al. 2002). The diﬀerence in the strength of the emission lines in the NLS1 and high z QSO composites is likewise most obvious in Lyα and the high ionization features (Si IV+O IV], C IV and He II), and almost absent in the low ionization lines (O I, C II). As found in previous studies of the Baldwin relationships, the N Vλ1240 emission proves to be the exception for which the line strength remains nearly independent of luminosity. Another important issue that can be examined via comparison of the composite spectra is the degree of similarity in the chemical enrichment for the two source > types. As noted before, the broad line region enrichments in both z ∼ 4 QSOs and NLS1s have been evaluated based on the N V/C IV and N V/He II line ratios,

7The ﬂattening of the Baldwin correlation at low luminosities has been reported in several previous studies; see Osmer & Shields (1999) for discussion and references. 92

Figure 3.8 Rest-frame EW of the Civ emission line, in A,˚ as a function of 1450A˚ luminosity, −1 −1 > in ergs s Hz , for individual NLS1 and z ∼ 4 QSO spectra. The Baldwin relation found > by Osmer, Porter & Green (1994) for a sample of 186 luminous quasars [logLν (1450) ∼ 29, 0 8 in the NLS1 spectra than in those of the z ∼ 4 QSOs . These trends result in lower N V/C IV and N V/He II line ratios, and presumably lower average metallicities, in the NLS1 sources than in the high z quasars, and thus do not support a strong connection between the two classes of objects. The results of this comparison differ from those of Shemmer & Netzer (2002) who argued that the line strengths suggest comparable metallicity for the NLS1s and the high luminosity QSOs. However, their conclusion was based on a very small number of NLS1s; also, the high abundances in these sources were supported by a relatively large N V/C IV ratio, but not clearly confirmed by N V/He II. The discrepancy between their findings and ours may be related to the details of the samples used in each case. With the larger sample size employed here, we tentatively conclude that the abundances in NLS1s and high z QSOs are in fact characteristically different; further study with larger sets of object spectra would clearly be desirable.

3.4.2 Principal Component Analysis

> Additional information on the NLS1 – z ∼4 QSO relationship can be performed by the means of a Principal Component Analysis (PCA), a mathematical decomposition of a set of properties describing the sample into a smaller number of eigenvectors that can account for the bulk of the total variance present in the data. Due to the fact that quantitative measurements of the spectral properties are in general subjective to the chosen parametrization, we chose to apply the PCA method directly to the observed spectra (Francis et al. 1992). We proceed with this analysis for the two samples separately, and for the combined set as well. This approach should allow for further identiﬁcation of potential common behaviors and their distribution among the two object classes.

8The same results are obtained when median composites are used for comparison. 94

For the NLS1 data, the spectral range with the most extensive coverage among the sample objects (21 out of 22) spans the region around the C IV line, and this bandpass is thus optimal for detailed statistical investigation. In this wavelength range, > the z ∼ 4 QSO sample is well represented by all 44 object spectra. Therefore, we conduct the PC analysis on this interval, where the highest S/N is expected. Figure 3.9 shows the mean spectra, the RMS spectra, and the first five principal components (PCs), ordered by the fraction of the sample variance for which each accounts, for the NLS1s, high z QSOs, and the combined sample. The range of properties exhibited by these objects is illustrated by the PCs and by their contribution to the total spectral variation. Principal Component 1 (PC1) is dominated by line-core modulations (compared with the emission in the average composites), and therefore, it can be considered primarily as a measure of the strength of the lines. This is a similar result to what Francis et al. (1992) obtained with PC analysis for a larger and more heterogeneous sample (232 objects), from the Large Bright Quasar Survey (LBQS). The first principal component is very similar for the three samples. Its profile differs −1 > in the two individual samples (FWHM(NLS1s) = 2050 km s , FWHM(z ∼ 4 QSOs) = 3100 km s−1); as might be expected, the NLS1 core contribution is narrower than that in the high z QSOs. The second component, PC2, accounts for spectrum-to- spectrum variations present in the wings of the lines. This is again exhibited by both samples but in a much larger proportion by the NLS1s. Also, the line-core and the line-wing modulations show opposite trends, suggesting that objects with prominent wings possess weak cores. Higher-order components continue to display, in different proportions, the core and the wing variations, but their intricacies become difficult to characterize and compare. Associating physical explanations to the properties exhibited by PCs is difficult, but additional information can be gained from the statistics of the PCA results. Table 3.4 lists the proportions (fractions of total variance) contributed by the first five PCs for each individual sample and for the combined one. The numbers indicate > that the NLS1 sample can be better represented than the z ∼ 4 quasar sample by a low number of principal eigenvectors. The difference in the PCs’ individual and 95

Figure 3.9 Mean and standard deviation spectrum and the ﬁrst 5 principal components given by PCA, performed for the Civ region only, applied to the NLS1 sample (dotted > line), the z ∼ 4 QSOs (dashed line), and the combined sample (continuous line). The principal components show similar modulations in the spectral variation in both categories of objects; the amplitude diﬀers. 96

Table 3.4. Proportions of PCs in the C iv emission region

> Component NLS1 z ∼4 QSO Combined sample

(21 obj.) (44 obj.) (65 obj.)

PC1 0.358 0.272 0.296

PC2 0.139 0.077 0.091

PC3 0.097 0.071 0.061

PC4 0.064 0.053 0.051

PC5 0.046 0.044 0.041

cumulative proportions indicate that NLS1 sources comprise a more spectroscopically compact sample than the high z quasars. Another basis for evaluating these results is a statistical comparison of the distributions of the weights of the ﬁrst two principal > components for the NLS1 and z ∼4 QSO objects. In the PCA run for the combined sample, i.e., for a common set of eigenvectors representing the NLS1s and the high z QSOs, the Kolmogorov-Smirnov (KS) test shows that the possibility of the two data sets being drawn from the same parent population is not excluded, but the evidence for this is weak: the KS probabilities are 0.047 and 0.188 for weights of PC1 and PC2 respectively. As a consistency test, we also performed this analysis for a much larger spectral range, covering Lyα to He II bandwidth, but with a lower number of NLS1s (only 16 objects), and therefore lower statistics; the main results remained unchanged. NLS1s > and z ∼4 QSOs are probably not close spectroscopic or physical analogues. 97 3.5 Conclusions

In this paper we present an analysis of all publicly available spectra for NLS1 galaxies in the HST archive. The resulting sample of 22 NLS1s with spectra spanning the UV-blue wavelength range is the largest that has been used, to date, in emission line studies of these objects. We employed these data to construct composite spectra (average and median), and thus, to characterize in detail the typical spectral properties of the NLS1 class. The resulting composite spectra are used to estimate the strengths of a large number of emission lines, and to quantify the continuum shape over a broad bandpass. Power-law fits to the continuum indicate a discrepancy from the results obtained from more general AGN composite studies: NLS1s have steeper UV-blue spectra. Possible explanations include intrinsic reddening and a trend of the low redshift sources to have intrinsically softer continua. A relation between the continuum shape and the redshift is not readily evident in this sample. A significant correlation is however observed between the spectral slope and luminosity, indicating that the redness of the NLS1s is related to the low luminosity of these objects. Moreover, the apparent connection between the UV resonance absorption lines and luminosity suggests that the steep slopes measured in these objects are due at least partially to reddening. The ionization state of the absorbing material and its relationship to the accretion source are not well determined, however. The NLS1 composites additionally allow us to quantify emission-line velocity offsets. The correlation between the velocity shifts and the degree of ionization that is found in normal broad-line AGNs is also present in the NLS1 sources, in both permitted and forbidden lines. This result may be of interest for comparison with NLS1 model predictions. The NLS1 data permit further investigation into the proposed analogy between > these sources and the z ∼ 4 quasars. Previous work suggested that both may be described by a high accretion rate and super-solar metallicities. The comparative study that we conduct based on the NLS1 and high z quasar average UV emission properties reveals a significant contrast between their spectral characteristics. The 98

composite spectra exhibit primarily the anticipated differences associated with the Baldwin Effect, i.e., relatively stronger lines in the less luminous sources. Nonetheless, the comparison makes it evident that the metal enrichment of the surrounding gas in the high z QSOs is higher than that in the NLS1 objects. This result does not support the hypothesis that these two types of objects are similar in their detailed physical characteristics or evolutionary phase. Additional confirmation of the differences exhibited by the two types of objects comes from a principal component decomposition applied directly to the spectra. > The spectral PC analysis indicates that the z ∼4 sources are statistically likely to share only part of their emission properties with the NLS1 galaxies; in particular, the low-velocity component, although enhanced in the high z QSOs, is less prominent > than in the NLS1 objects. The statistics show that the z ∼ 4 QSO phenomenon is controlled by a broader, more heterogeneous family of properties than those governing the NLS1s; the NLS1 spectroscopic features can be reconstructed from a smaller > number of parameters (eigenvectors) than that necessary in the case of z ∼4 sources. In conclusion, NLS1s and high z quasars are spectroscopically disparate, and it is therefore doubtful that a close physical connection exists between these source types. 99

Chapter 4 Dust Reddening and the AGN Spectral Energy Distribution

To be submitted as:

Constantin, A., & Shields, J. C.

4.1 Introduction

More than half of the radiative power of an unobscured Active Galactic Nucleus (AGN) is typically emitted from 1 µm to shortward of 1000A,˚ and its origin is believed to be in an accretion disk. The nature of this accretion disk is currently not well understood, as models still do not provide a satisfactory explanation of the optical-UV continuum observed in AGNs. Although to a first approximation the various members of the quasar family show a remarkable unity in their spectral energy distribution at α these wavelengths, which is represented in general by a power-law (Fν ∝ ν ), a closer look reveals a considerable diversity. An important question is the extent to which this variation stems from properties intrinsic to the accretion source, or alternatively from absorption due to intervening dust in the circumnuclear environment. Several previous studies have produced evidence for various correlations between the quasar optical-UV slope α and (monochromatic) luminosity L or redshift z. Among the first claims for a strong dependence on luminosity are those of Wan- del (1987) and Mushotzky & Wandel (1989), who find that, for a sample of 193 luminous quasars (B band luminosity > 1044 ergs s−1), the 4200 – 7500 A˚ continuum shape is systematically flatter (bluer) as source luminosity increases; although the redshift span is considerable, 0

finds a similar correlation between optical continuum shape and luminosity, but he argues that this is only present in the low luminosity objects. Zheng & Malkan (1993) point out that their measured UV-optical (1250 – 6500 A)˚ slopes correlate likewise with the absolute monochromatic magnitude at 4800A,˚ in a sample of 145 QSOs and Seyfert 1s. When only the UV range is measured, claims are however for stronger trends with z than with L (O’Brian, Gondhalekar & Wilson 1988; Cheng, Gaskell & Koratkar 1991). More recent data (Kuhn et al. 2001) are interpreted in favor of a weak correlation between α and the UV luminosity L(1285A),˚ but only for the low redshift sources (z ∼ 0.1), with no statistically significant link between the quasar SEDs and their optical luminosity [L(4200A)˚ or L(4750A)].˚ Another interesting result that may have bearings on these phenomena is found by e.g., Edelson, Pike & Krolik (1990) and Giveon et al. (1999), who show that variable AGNs have bluer continua in their higher luminosity states. Collectively, these studies show significant evidence of a relationship between the optical-UV continuum shape and luminosity, such that lower L sources are redder. Understanding the origin of this α − L trend is potentially important for understanding several aspects of AGN physics. Interpretations of this correlation vary however. Zheng & Malkan (1993) associate it with a shift of the big blue bump from UV to optical wavelengths in the more luminous objects, which could reflect a variation in the temperature structure of the accretion disk. They also invoke this correlation as the principal cause for the Baldwin effect (the weakening of the emission line equivalent width with increasing luminosity), and the fact that it is stronger in the high-ionization lines (i.e., C IV). Mushotzky & Wandel (1989) suggest this slope – luminosity relation is due to a luminosity dependence in the strength of the blue bump, presumably tracing the accretion disk, relative to an IR power-law of probable non-thermal origin. Neugebauer et al. (1987) and Kriss (1988) attribute the observed correlation mostly to starlight contamination in the less luminous sources. Most of these studies do not favor reddening by dust as an important contributor to the color-luminosity dependence. 101

More recent work suggests that intrinsic reddening might actually play an important role in shaping the quasar continuum, and introduce a luminosity dependence. One of the first hints for a luminosity-dependent reddening, that might determine a continuum steepening in the weaker nuclei, comes from Constantin & Shields (2003). They find that, in their sample of 22 narrow-line Seyfert 1s (NLS1) and similar objects, lower L sources show steeper continua and also preferentially display UV resonance line absorption. The upper panel of Figure 4.1 shows explicitly these results (a more detailed discussion is presented in Constantin & Shields 2003). Based on the idea that the dust is expected to accompany the absorbing gas, it is proposed in this study that a luminosity dependence of the solid angle covered by the dust as seen by the central source might explain the α − L trend. Motivated by these findings, we put together a similar plot for z < 0.5 Palomar-Green (PG) QSOs (defined by Schmidt & Green 1983, but see Boroson & Green 1992 for details on their properties), shown in the lower panel of Figure 4.1. The data are limited to sources for which archival HST spectra are available, which we use to identify the presence of line absorption. The PG quasars exhibit a significant slope-luminosity correlation, and confirm our result for the NLS1s, that AGNs with UV line absorption tend to have steeper optical slopes. Similar behavior has been reported by Hopkins et al. (2004) for a large sample of SDSS quasars. The presence of dust along our line of sight or located close to the central engine is suggested by other observations in some cases. The infrared spectral morphology indicates that thermal dust emission may be the principal source of radiation at these wavelengths (Rieke 1978; Sanders et al. 1989; Barvainis 1990; Haas et al. 2000; Polletta et al. 2000). In variable sources, the response of the 2 µm emission to changes in the UV continuum implies that dust resides on relatively small scales, probably just outside the broad line region (Puschell 1981; Rieke & Lebofsky 1979; Stein & Soifer 1983). Recent analysis of large, statistically significant X-ray data (e.g., Ueda et. al 2003) confirm that the large amounts of obscuration are common in AGNs, and that, moreover, the fraction of absorbed QSOs decreases with increasing luminosity. 102

Figure 4.1 Upper panel: Spectral indices measured in the HST NLS1 sample plotted vs. −1 −1 1450A˚ luminosity, with Lν expressed in erg s Hz . Lower panel: Optical spectral indices

(measured by Neugebauer et al. 1987) vs. continuum luminosity Lν(1450A),˚ for the PG quasars for which HST-FOS spectra are available in the archive. The filled circles and the asterisks denote objects with and without UV resonance line absorption respectively. The α − L correlation is definitely present in both samples (the Spearman rank coefficients and the probabilities of the correlations to happen by chance are indicated in each plot), along with a trend for the low L objects to be more prone to gas absorption. 103

A more quantitative investigation of what determines which objects have significant amounts of dust, and where and how the dust modiﬁes the QSO’s intrinsic behavior, is clearly desired. We present and discuss here the results of simulations designed to study the possible role of circumnuclear obscuring matter in producing the observed luminosity dependence. This project investigates in more depth the idea proposed by Constantin & Shields (2003) that a luminosity-dependent reddening process may account for the observed α − L trend. Monte Carlo methods are employed in a scheme in which the inner radius of the absorbing material scales with the intrinsic luminosity of the central source, consistent with expectations that grains will evaporate above a certain radiation density. We show that this simple model results in a statistical dependence of reddening, in general accord with the observed correlation between quasar luminosity and continuum shape. We employ this scenario in further studying the importance of orientation and absorption in modifying the shape of the quasar SED, and possibly in constraining the physical and geometrical properties of the obscuring material. Furthermore, because gas associated with the dust would also be expected to absorb X-rays, this model is used in estimating the fraction of the obscured total accretion power in the universe. The numbers of absorbed sources predicted by this picture appear consistent with recent X-ray observations of the extragalactic point source population.

4.2 The model: receding torus

Considering a synthetic population of quasars, with a given distribution of intrinsic properties, we test here whether their observed total power output correlates in any manner with the slopes of the power-laws that best ﬁt their reddened observed SEDs. A very simple scenario is invoked: a central nuclear source emits isotropically α a continuum radiation modeled as a power-law (Fν ∝ ν ), which is absorbed and reddened by obscuring material (i.e., dust). The extinction is assumed to happen mostly around the nucleus; for the sake of simplicity, the dusty medium is given the shape of a torus. A key ingredient in the proposed investigation is the assumption that the inner radius of the torus scales with the square root of the source luminosity, 104

as its value is basically given by the distance at which the radiation ﬁeld of the quasar ceases to sublimate the dust. In this picture, the ‘receding torus’ model (Lawrence 1991), the more luminous sources would be less obscured than the weaker ones, since in the latter objects, the opening angle of the torus, and therefore the covering factor of the obscuring material, are larger. A schematic representation of this geometry is illustrated in Figure 4.2. Some of the quantities used in the model, and explained in detail in the next section, are also indicated. The entire ensemble of quasars is assumed to have similar torus properties (e.g., particle density, reddening law). The distributions of quasar properties (intrinsic luminosity and power-law slope, viewing angle, etc.) are built using Monte Carlo techniques, and are based on, or further tested against the most recent observational results based on large statistical samples of AGNs.

4.2.1 Simulation parameters

The quasar intrinsic bolometric luminosities, continuum slopes, orientation relative to our line of sight (the viewing angles), and all other parameters involved in this model (e.g., h the half-height of the torus) are drawn from probability distribution functions that are described below. In building these functions we used simple Monte Carlo methods, in which random numbers are associated with cumulative probability distributions in order to generate the desired non-uniform distribution of values.

1. The assumed incident continuum energy distribution is shaped as a power-law. In this project only the UV-optical (1000A-˚ 6000A)˚ emission range is modeled, therefore, the slope that best ﬁts the continuum in this range is referred to

αo,UV.

2. The individual power-law slopes that deﬁne the intrinsic continuum shape for each quasar in the simulated sample follow a Gaussian distribution around

α0 = −0.5, with a dispersion σα = 0.3. This choice is based on observed distributions of spectral indices derived for Sloan Digital Sky Survey luminous quasars (magnitudes brighter than i = 19) by Ivezic et al. (2002). 105

Figure 4.2 Schematic representation of the receding torus model. The predicted behavior of the geometry of the absorbing material is presented for the low-luminosity (top) and the high-luminosity (bottom) objects: a larger inner radius (RS) of the torus produces an

increased opening angle (θc) for the more luminous sources, provided the half-height of the torus (h) remains unchanged. 106

3. The intrinsic bolometric luminosity distribution is approximated by a two component broken power-law function, which is a form that describes well the present-day (z ≈ 0) hard X-ray luminosity function (e.g., Ueda et. al 2003):

dΦ(L,z=0) L γ1 L γ2 −1 p(L)= dlogL ∝ [( L∗ ) +( L∗ ) ] , with γ1 =0.86 and γ2 =2.23.

40 −1 The range of values for luminosities is set to span the Lmin = 10 erg s 50 −1 to Lmax = 10 erg s interval, which includes the general observed range.

The break in the power-law distribution function, corresponding to L∗, is at approximately 1044 erg s−1. We chose this parametrization based on the idea that the X-ray luminosity function is closest in form to the intrinsic bolometric distribution, due to the limited sensitivity of the X-rays to absorption.

4. For each quasar an inclination angle, which is the angle between our line of sight and the symmetry axis of the individual coordinate system, is ascribed. The viewing angle probability distribution is given by

p(θ) ∝ sin θ

which is equivalent to a uniform distribution weighted by the area element of the sphere cut by the solid angle delimited by the respective viewing direction. This translates into a random orientation in space, with the high inclinations being more probable. Because the problem is highly symmetric, the calculations are run only for emitting angles θ = [0,π/2].

5. The radial distance at which the dust grains start to exist, at which the radiation ﬁeld becomes weak enough not to sublimate them, gives the inner radius of the

torus, Rs, which scales with the intrinsic luminosity as

1/2 Rs =0.06 × L45 pc (Laor & Draine 1993),

45 −1 where L45 is the bolometric luminosity in units of 10 ergs s .

6. The covering factor is given by 107

Ω fc = 4π = cos θc, where tan θc = Rs/h,

with h being the half-height of the torus. If h is set to be independent (or not

a strong function) of the total AGN power output, fc remains a function of

luminosity, through Rs. Figure 4.3 illustrates quantitatively this dependence for diﬀerent alternative values of h, expressed in parsec (pc). The range of the values considered for half-height of the torus is chosen to comply with observational constraints obtained from visual and infrared extinction measurements in quasars or in our Galaxy (e.g., Malhotra 1994; Malhotra 1995; Hill, Goodrich & DePoy 1996): 2h ≈ 0.5 − 1 pc. However, dust lanes with 2h ≈ 100pc have also been measured in galaxies (Dalcanton, Yoachim & Bernstein 2004). Simp- son (2003) indicates that the receding torus model produces diﬀerences between the quasar and radio galaxy luminosities that are closer to the observed values when a log-normal distribution of 1 dex variation in h is employed, rather than a constant value. Thus, for the tests presented here we adopt this choice as

well, with the distribution maximum located at h0 =0.5 pc. Because the range of h for which the covering factor still permits some of the low L objects to be unabsorbed is quite narrow (Figure 4.3), we chose a variation of only 0.25 dex.

7. For simplicity, an inﬁnite horizontal extent of the dust is assumed. The interac- tion of the dust with the radiation emitted by the central source is considered to dominate over the gas absorption (which is believed to be the case for media with high ionization parameters), and the scattering eﬀects are neglected. The extinction of the incident radiation through the dust is then equivalent with an attenuation by a factor of e−τdustkλ , as given by the radiative transfer equa-

tion. Here, kλ = A(λ)/A(V ) is the reddening law, assumed for now to have the Galactic shape as given by Seaton (1979). We test the eﬀects of diﬀerent

reddening laws in Section 4.3. τdust is the optical depth of the dusty material, and is given by:

dust τdust = nH σ l, 108

Figure 4.3 Covering factor as a function of intrinsic luminosity for diﬀerent parameter values of the half-height of the torus (h), expressed in parsecs. Note that, with the chosen parametrization, in this simple picture, most often h ≈ 0.5, and objects less luminous than 45 −1 > 53 −1 10 erg s are highly obscured, while luminous sources with L ∼10 erg s are barely aﬀected by dust absorption. 109

dust where nH is the particle density, σ is the average photon’s cross section for dust absorption and isotropic scattering per hydrogen atom at 5500A,˚ and l is the path length through the dust.

We experimented with different values of σdust which reflect different dust to gas ratios, and adopt as a representative value the one given by Netzer & Laor (1993): σdust ' 1.4 × 10−21 cm2.

The path length through the dust l is a geometrical result of the considered model:

h Rs −1 Rs l = cos θ − sin θ , for θ> tan h , and zero otherwise.

Its scale is therefore determined by h, by the absolute intrinsic luminosity range

through its dependence on Rs, and by the viewing angle.

8. The particle density is a parameter that is currently poorly known; its order 3 4 −3 of magnitude varies observationally within nH ' 10 − 10 cm (Dopita et al. 1998; Hill, Goodrich & DePoy 1996), with theoretical estimates as high as 109.4 cm−3 (Netzer & Laor 1993) for the very central regions. Laor & Draine (1993) give an even broader range of densities which, for four diﬀerent grain size −1/2 distribution models of “optically thin” dust, can be nH (rin) ' 6−12000×L46 cm−3; this luminosity dependence, also adopted by Netzer & Laor (1993) in −3/2 the form nH ∝ R , suggests also that the absorbing dust is less dense in the more luminous objects (making thus the weak AGNs even more prone to reddening relative to their more powerful counterparts).

For the sake of simplicity, we consider for now that the density has the same value throughout the whole torus, with no radial dependence for a given source. However, for each quasar, the density is set to depend on the central source’s luminosity as:

−3/2 nH ∝ Rs .

For the chosen range in bolometric L, and with the deﬁnition of Rs, appropriate 110

values of density result with a proportionality constant of unity when nH is −3 7 −3 expressed in cm , and Rs in pc. The density reaches the value of 10 cm for the lowest L objects, which is in good agreement with the range predicted by Laor & Draine (1993). The normalization factor in this relation plays however an important role, as its order of magnitude may govern the shape (slope), and even the existence of the α − L trend. We examine the results for two diﬀerent values of this factor, in the following section.

4.3 Results

After running the simulation of dust reddening of a sample of quasars whose parameters are described above, we “observe” these objects, i.e., ﬁt again power-laws to their emergent radiation ﬁeld, and calculate the slopes and the monochromatic

luminosities (Lν ), in order to compare them with the observed values. We plot the re-

sulting slopes of the continuum fits against their absorbed Lν (1450A)˚ and Lν (4800A),˚ for value ranges that are generally found in AGN samples (e.g., for comparison with Figure 4.1, and with Zheng & Malkan 1993, their Fig. 1, respectively). Figure 4.4 illustrates these results obtained for two different sets of values for the −3/2 particle density of the obscuring torus: (I) nH = Rs , the filled symbols, and (II) 3/2 −3/2 −3 nH = 10 × Rs , the open symbols, with nH expressed in cm , and Rs in pc.

The emergent speciﬁc monochromatic luminosities Lν are obtained by attenuating

−τdustkλ by a factor of e the unabsorbed Lν . Following Laor (1998), we assume a representative AGN SED such that

˚ 1 νLν (3000 A) = 8.3 Lbolometric,

where Lbolometric is the bolometric luminosity. In general, then,

1 1 ν α Lν = 8.3 Lbolometric ν(3000) ( ν(3000) ) , where ν(3000) = c/3000A,˚ and c is the speed of light. The main result presented in Figure 4.4 is the readily apparent correlation between the observed optical-UV slopes and the attenuated luminosities, measured both at 111

1450A˚ and 4800A.˚ The slope-luminosity trend is easily recovered for the parameter choices determined by existing observational constraints presented in Section 4.2.1. A reasonable quantitative agreement with the observed α − L trends displayed in

Figure 4.1 and Zheng & Malkan (1993) is found for the ﬁrst choice of nH , while the agreement is marginal for the second one. As expected, the extinction is more pronounced at UV wavelengths (1450A),˚ and therefore the slope of the correlation

is shallower because of its eﬀect on the reddened Lν . The trend is weaker at high luminosities, where sources are subject to little reddening. The detailed characteristics of this correlation depend of course on the selection of quasar and dust properties. While this model is a very simple representation of what the central region of a quasar may look like, it is yet possible to constrain some of the QSO’s intrinsic attributes like the intrinsic luminosity range, or the dimensions (e.g., vertical scale), and the density of the absorbing material. A direct confrontation of the distribution of the reddened luminosities and slopes predicted by this simulation with the respective observed values in actual AGN surveys is an easy and valuable test for the range of values within which the model’s variable parameters should vary:

1. The range of intrinsic bolometric luminosity. To identify the degree to which intrinsic L inﬂuences the observed range of αo,UV and reddened monochromatic power, we scaled each data point in Figure 4.4 by its intrinsic luminosity. It can be seen that more intrinsically luminous objects are, in general, less obscured, as expected from the assumptions of the model we used: the spread of α values is much less for high L quasars than among their weaker counterparts (i.e.,

the diﬀerence αintrinsic − αreddened is more pronounced for the low L sources). For a given observed luminosity, the reddest objects (i.e., steepest continua) are invariably those that intrinsically are more luminous. Equivalently steep continua in low L nuclei would translate into very low ﬂuxes, making them thus impossible to be observed.

2. The particle density. It is clear from the plots shown here that denser tori steepen the correlation, for a given range in luminosity. Note however that in this approximation the tori are inﬁnitely extended in the horizontal direction, 112

Figure 4.4 The observed (reddened) power-law slopes plotted against the quasar absorbed monochromatic absolute luminosities measured at 1450A˚ [Lν(1450A),˚ top] and at 4800A˚

[Lν(4800A),˚ bottom]. The results are presented for two particle density value sets: nH = −3/2 3/2 −3/2 Rs (black ﬁlled symbols), and nH = 10 × Rs (orange open symbols). The symbol size is proportional to the quasar intrinsic bolometric luminosity. 113

which means the absorption is overestimated for the highly inclined objects. For a ﬁnite radial length of the obscuring material the slope of the correlation would ﬂatten, making possible to have even denser absorbing material, and still recover the observed α−L relationship for the assumed range of intrinsic luminosities. A real radial dependence of the particle density would also decrease the amount of

reddening, allowing for even larger nH values at the inner edge of the absorbing torus.

3. The intrinsic SED shape. As Figure 4.4 suggests, almost all quasars suffer from dust absorption to a certain degree (i.e., there are very few objects that escape the dust obscuration). For our choice of intrinsic slope distribution, the few objects that have observed SED slopes of ≈ 1 or even higher (e.g., see Zheng & Malkan 1993, their Fig. 1) are hardly present in our results. Plausible explanations are either that the absorption is too strong and omnipresent, or that the intrinsic continua are bluer than what we think they are. We tested how the range of the intrinsic continuum slopes influences the characteristics of the α − L correlation, and illustrated the results in Figure 4.5. Bluer intrinsic quasar continua, as represented by a Gaussian distribution centered on <α>= +0.5, recover a significant number of luminous blue objects, while in the lower luminosity ranges, the α − L trend is indistinguishable from that obtained for a

Gaussian distribution in αintrinsic centered on −0.5. The overall diﬀerences are thus not signiﬁcant, which means the intrinsic quasar continua might actually be harder than what is usually assumed.

4. The inclination. Although the majority of the unknowns are strongly related with the total incident power output, the viewing angle plays an important role as well. It is interesting to see how samples characterized by narrower ranges of inclination angles would aﬀect the slope-luminosity correlation. Figure 4.6 illustrates the previous results by displaying with diﬀerent symbol/color codes the objects viewed under θ < 30◦ (more pole-on), and those for which θ > 70◦ (more edge-on). A comparison with Figure 4.4 shows that the upper envelope of the trend contains mainly intrinsically low L, which means, these weak sources 114

Figure 4.5 Same α − L plot as in Figure 4.4, for luminosities measured at 1450A˚ only, −3/2 calculated with absorption through dust with nH = Rs , for two diﬀerent sets of intrinsic

SED shapes: open, black symbols correspond to a Gaussian distribution of αintrinsic around

α0 = −0.5, while for the ﬁlled, blue symbols α0 = +0.5. 115

Figure 4.6 Same α − L plot as in Figure 4.4, for luminosities measured at 1450A˚ only, −3/2 calculated with absorption through dust with nH = Rs , where different symbol/colors are assigned to different quasar orientation ranges: blue, double-lined symbols for the more pole-on objects (θ = 0◦ − 30◦), and filled, orange symbols for the more edge-on sources (θ = 70◦ − 90◦).

become observable only if they are viewed more pole-on. Even in these cases, the weak AGN do not escape the absorption totally, as the covering factor is still high.

5. The extinction law. There is a significant possibility that the the dust in the circumnuclear region of AGNs has different properties than in the Galactic diffuse interstellar medium. For example, Maiolino et al. (2000) argue that the extinction curve is almost always flatter than Galactic, and does not show the prominent features at 9.7 µm and at 2175A.˚ We experimented thus with other extinction laws. Figure 4.7 presents the power-law slopes measured in the reddened quasar SEDs and the observed specific monochromatic luminosities ( at 1450A˚ and 4800A)˚ obtained in this receding-torus model, where the reddening 116

is governed by three different functions: 1. the Small Magellanic Cloud (SMC)- −1.2 like extinction law (Prevot et al. 1984), in the form kλ = 1.39λ , with λ in µm, which, in particular, lacks the 2175A˚ feature, 2. the shallower than Galactic Calzetti law obtained for starburst galaxies (Calzetti et al. 2000), and 3. the reddening law for AGNs, as recently derived by Gaskell et al. (2004) from a large and heterogeneous sample of quasars, which is flatter in the UV than any other extinction curve. The results are somewhat as expected: a greyer extinction law modifies the slope of the α − L correlation by flattening it, however the trend itself remains significant.

6. Selection effects/Redshift dependence. The distribution of points in the α − L plots should be sensitive to selection effects, which are not considered in the tests presented here. For example, the weak and distant AGNs would not be detected in a flux/magnitude limited survey, which means they would vanish from plots like those in Figure 4.4. Their more luminous counterparts would also be hardly obscured, and thus not show a luminosity dependent SED shape. This means that at high z, the α−L correlation might not be significant enough to be detected. For nearby AGN samples, where the high luminosity nuclei are hardly observed, and where, in this scenario, a large majority of sources are expected to be reddened to some extent by dust, a more significant α − L correlation is produced. This is consistent with the observational studies that claim more often such trends in the low-L quasar samples.

Another interesting feature of these plots presented in Figure 4.4 is an increase in the dispersion in the data points in the α − L correlation at lower luminosities and steeper slopes, which is very similar to what is observed in about all the cases such plots are presented. Such behavior is invariably interpreted in the literature as being due to changes of the AGN continua caused by starlight in their host galaxies, even if stellar features signaling such contamination are not evident or if the measurements are made in the rest-frame UV bandpass where starlight contamination is less likely. The results presented here suggest that further consideration should be given to the importance of interstellar reddening. 117

Figure 4.7 Same α − L plot as in Figure 4.4, for luminosities measured at 1450A˚ (ﬁlled symbols) and at 4800A˚ (open symbols), for three diﬀerent extinction laws, as indicated in each panel: Small Magelanic Cloud type (top), starburst galaxy-like (middle), and AGN- −3/2 type (bottom). The particle density values are given by nH = Rs , in all three cases. 118

Whether the eﬀects of strong starlight emission or dust absorption on the measured quasar SED are dominant in producing the α−L trend, can potentially be determined. If interstellar reddening is the primary factor, dust absorption should be accompanied by atomic absorption by the associated gas, which should be visible in UV resonance lines or X-ray continuum absorption. As the plots in Figure 4.4 suggest, if the softest continua are measured in sources that show stronger obscuration, then the slope – Luminosity trend is most probably due to dust obscuration.

4.3.1 A note of caution

Overall, the details of the results of this model remain diﬃcult to interpret, as the relatively large number of parameters involved in the calculations create a degeneracy that is not easily breakable. Of special concern is the fact that, among the “observed” sources, there are barely any systems that totally escape absorption, which otherwise seem to be detected in canonical AGN surveys. In the simple receding torus picture that we employ here, objects with logL ≤ 45 are in almost all cases subject to absorption (see Figure 4.3). This result is mainly due to the choice of the absorbing geometry, expressed in terms of Rs and h, for which the assumption that h is constant for all radii makes the torus opening angles extremely small for the objects in this particular luminosity range. Consequently, in these cases, the covering factor ceases to play an important role in the observed α−L correlation. The presence of this correlation for logL ≤ 45 can then be attributed

to our selection of the luminosity dependent particle density nH , which, through −3 the definition of Rs, scales as nH ∝ L . Future investigation would benefit from exploring the dependence of the results on the adopted radial density function. A more natural spatial configuration for the obscuring matter would probably em-

ploy a tapered disk shape, with a small h near the accreting sources (i.e., at Rs), that increases radially outward. In this new picture, even low luminosity sources would have a signiﬁcant chance of being seen directly, and therefore to appear unabsorbed. At the same time however, the upper side of the generally linear trend recovered by the simple receding torus (e.g., Figure 4.4) would be more populated, giving rise to 119

a more triangular spread of points in these plots. The details of such a distribution of points in the α − L plane would continue to depend on the physical parameters involved in the absorption process, and on additional geometrical parameters introduced by the particular shape of such a tapering structure. Further constraints provided by the behavior of the continuum slopes and luminosities in large samples of AGNs are required for a full investigation of these issues.

4.3.2 The fraction of obscured accretion power

This study provides also a possible framework for estimating the demographics of the AGNs for which the obscuration is large enough to eﬀectively hide any optical signature of an active nucleus (e.g., featureless power-law continua, broad emission lines), and thus prevent the inclusion of these sources in optical-UV surveys. By employing this receding torus-like picture, we found that only a fraction of ∼ 30−40% of the starting sample of objects have emergent luminosities and slopes that fall within the limits observed in canonical AGN surveys: αo UV = –6 to 2, and log Lν(1450A)˚ = 26 – 31 erg s−1 Hz−1, or the magnitude M(4800A)˚ = –16 to –28. This result predicts that 60 – 70% of all AGNs might be obscured. X-ray background measurements and counts of X-ray source populations show that indeed, there is much obscured accretion in the universe, and that populations of type 2 AGNs and distant Compton-thick sources (with column densities that are > −1 24 equal or higher than the inverse of the Thomson cross-section, NH − σT ' 1.5 × 10 cm−2) are very common. Matt et al. (2000) for example, suggest that the space density of heavily obscured AGNs could outnumber unobscured or mildly absorbed AGNs by at least one order of magnitude based on 1. a comparison of number estimates of relatively unabsorbed AGNs from a soft X-ray luminosity function (Miyaji et al. 1998) in the very local universe (within 4 Mpc), with the actual counts of 3 nearest AGNs (Circinius galaxy, NGC 4945, and Centaurus A), which are all heavily obscured, and 2. a comparison of the local infrared luminosity function with the X-ray luminosity function where the infrared sources are expected to include Compton-thick nuclei that are usually missed from the X-ray samples. 120

The observed fraction of absorbed sources (with column densities larger than 1022) in diﬀerent X-ray surveys is not as high as the above arguments predict, but still add up to: ∼ 40%, for a study of bright AGNs sampled with HEAO1-A2 by Piccinotti et al. (1982), ∼ 30% for the ASCA Large Sky Survey conducted by Akiyama et al. (2000), ∼ 20 – 30% among the XMM observations of Piconcelli et al. (2002), and ∼ 50 – 60% among the AGNs studied with Chandra by Brusa (2004). By employing a combination of surveys conducted at photon energies above 2 keV with HEAO 1, ASCA, and Chandra, e.g., Ueda et. al (2003) ﬁnds that the fraction of AGNs with 22 24 NH = 10 − 10 to the total Compton-thin sources is ' 0.57, which translates into a total obscured fraction of ' 36%. The fraction of obscured or elusive AGNs obtained with this receding-torus model may be somewhat higher than what is predicted and observed. However, our pre- diction is at present a rough estimate, and its range varies within 20% depending on the initial parameter space of the simulations (e.g., the particle density, the half height of the torus). With more detailed simulations we might be able to obtain such theoretical demographics as a function of redshift, and through comparisons with the observational results which continue to improve, constrain further the physical characteristics of the absorbing material in quasars.

4.4 Summary & Conclusions

We present here a simulation of the eﬀects of dust reddening on quasar continua by employing the receding torus model. We investigated the implications of this scenario for the observed SED as a function of luminosity. The results show promise for explaining the observed correlation between optical-UV continuum slope and L. It is found that this trend is potentially caused by a luminosity-dependent intrinsic dust extinction. The general empirical behavior of AGN samples is recovered for many reasonable parameter choices, which brings additional support for the receding torus scenario. The model may be potentially used in the future to constrain some of the most important physical characteristics of quasars and their circumnuclear dust, after extending the model to incorporate a quasar redshift distribution, selection eﬀects, a 121

radial dependence of the particle density, and an outer edge limit on the torus (which may soon be constrained by Spitzer data). Disagreements in observational studies of the α − L trend may stem from a combination of a luminosity dependence of the inner edge of the absorbing material with the selection effects that are invariably present in the QSO surveys. The α − L correlation is significantly stronger, and thus more easily noticed in the low-z samples of quasars because the small probability of including in the analysis very luminous objects which, in this scenario, are less obscured. Conversely, the more distant and powerful objects have slim chances of showing a luminosity dependent reddening at a statistical level. We found that the red SED often measured in the weak AGNs is not, as previously believed, necessarily due to contamination by starlight in their host galaxy. It is illustrated here that a similar global softening of the continuum can be well obtained by means of intrinsic dust absorption only. One method for distinguishing between these possibilities would be to examine the α − L relationship using small aperture, rest-frame UV spectra which minimize circumnuclear starlight and provide the opportunity to detect atomic absorption. Despite the crude approximations of the quasar intrinsic properties, the results produced by this simple receding-torus model are in good agreement with predictions of the fraction of obscured sources obtained from modeling and observations of the X-ray cosmic background and other multiwavelength AGN surveys. Future investigations and testing should be able to offer more stringent conclusions regarding the nature, strength and shape of the α−L correlation, and consequently a clearer picture of the relationship between the intrinsic and the observed AGN phenomena. 122

Chapter 5 The Power Sources of the Low Luminosity Emission-Line Galaxy Nuclei

To be submitted as:

Constantin, A., Shields, J. C., Ho, L. C., Barth, A., & Filippenko, A. V.

5.1 Introduction

Galaxy nuclei that show emission-line activity are very common in the nearby universe. For a signiﬁcant fraction (at least 20%) of these sources, their emission is believed to derive from non-stellar processes, that in many cases can be associated with accretion of matter onto massive black holes (Ho et al. 1997a), similar to Seyferts and quasars. The physical nature of the large majority (about two-thirds) of the nearby emission-line nebulae remains however controversial. In the emission- line diagnostic diagrams that are used in classifying nuclei (e.g., Baldwin, Phillips, & Terlevich 1981; Veilleux & Osterbrock 1987), these ambiguous objects show high ionization lines that are weaker than those of the accretion-type sources (Seyferts), and forbidden lines that are much stronger than those usually present in H ii regions (believed to be powered by hot, young stars). These peculiar properties characterize the Low Ionization Nuclear Emission Regions (LINERs, Heckman 1980), in particular those lacking broad lines (LINER 2s, by analogy with Seyfert 2s), and the LINER/H II “transition” sources (Ho et al. 1993). Understanding the underlying emission source(s) of these low luminosity emission- line nuclei is a key ingredient in constraining the faint end of the AGN luminosity function, and therefore in ﬁnding the true incidence rate of massive black holes in 123

the present universe. Moreover, the ubiquity of these nuclei may trace a potentially very common process of black hole growth which, if identified, should be able to offer the missing links in the evolution between AGNs and “normal” galaxies. Therefore, it is important to understand whether and how these particular objects fit into the standard AGN paradigm, and, if they are accretion powered, what is the nature of the central accretion structure. Several excitation scenarios have been proposed for these nuclei. Mechanisms alternative to accretion onto a black hole, like photoionization by hot, young stars (Filippenko & Terlevich 1992; Shields 1992; Barth & Shields 2000), clusters of planetary nebula nuclei (Taniguchi, Shioya, & Murayama 2000), or shocks (Dopita & Sutherland 1995), have proved moderately successful in explaining the optical spectra of these sources. These models however require often exceptional physical conditions or fine-tuned parameters, which are not consistent with the statistical trends present in large samples of such nuclei (e.g., the Palomar survey of nearby galaxies, Filippenko & Sargent, 1985; Ho, Filippenko & Sargent, 2003, and references therein). In particular, the applicability of scenarios that invoke excitation by young, hot stars are at odds with the recent evidence that the nearby galaxy nuclei contain predominantly old stellar populations (Sarzi et al. 2004). Particularly ambiguous in their interpretation are the transition-type sources whose emission-line ratios are intermediate between those of LINERs and Hii regions. A seemingly natural explanation for these nuclei is the so-called composite picture, in which a central accretion-type source (a weak AGN, maybe a LINER) is surrounded by star-forming regions; such structure would generally remain unresolved in the typical ground-based aperture. This scenario is supported by the similarity in the statistical traits –both in their nuclear properties and those of their home galaxies– of LINERs and Transition nuclei, and the indication that, in the latter class, star formation might be more enhanced and that potential projection effects would lead to greater contamination of the nuclear emission (Ho et al. 2003). Véron, Gon¸calves, & Véron-Cetty (1997) and Gon¸calves Véron-Cetty, & Véron (1999) also argue that the composite nature is revealed in the emission-line profiles. They found through a 124

Gaussian decomposition that the emission separates into very narrow (∼ 200 − 270 km s−1) and relatively broad (∼ 400 − 700 km s−1) components, with H ii and AGN (Seyfert 2)-like flux ratios, respectively. We present here a simple, expedient, yet powerful method to investigate the central nebula and its excitation structure, and in particular this composite model, in a large sample of nearby galaxy nuclei: a quantitative comparison of nebular fluxes and line ratios measured in small apertures with the Hubble Space Telescope (HST, ∼ 000.2) and large apertures from the ground (∼ 200). In this way, the high angular resolution isolates the spectrum from the central few tens of parsecs, while the degree of central concentration is constrained by comparison with the larger aperture data. If the composite picture for the transition sources is correct, then we expect to see a difference in the nebular line emission in the two apertures, with a more AGN-like behavior in the small aperture spectra. Moreover, when an active nucleus is at work, the ionizing radiation emerges from a central source and falls off in density as r−2; if the gas density gradients are small, the degree of ionization of the emitting gas is expected to diminish with radius. Thus, radial variations in nebular line ratios should reflect this behavior as such gradients are not expected for distributed sources of ionization (e.g., shocks, hot stars, turbulent mixing layers). A clear signature of an accretion-type nucleus would be the presence of broad line features. A lack of broad emission in these ambiguous objects, can arise because the broad line region is simply absent, or because it is below the detection threshold, which is set mostly by the degree of contamination by starlight in the host galaxy. Ground- based observations (e.g., the Palomar survey) employ apertures that span typically many hundreds of pc in size, and thus can intercept significant continuum and narrow line emission from circumnuclear regions. Thus, a significant improvement to the sensitivity to broad components should come from observing the nuclear emission through smaller slit sizes. Spectra acquired through apertures that are an order of magnitude smaller would be expected to include significantly less surrounding starlight, as the metric scale would be ten times smaller, and thus to produce a greater contrast that would allow for a more sensitive measure of the nuclear emission. 125

A pilot study that uses a statistically complete (distance-limited) sample of 23 nuclei (the Survey of Nearby Nuclei with STIS, SUNNS, Shields et al. 2004b) that span a range of spectral classification (as defined by ground-based measurements) has been presented by Shields et al. (2004a). The initial results are surprising in that the emission-line spectrum is often aperture-independent, suggesting that the ambiguous emission-line categories do not necessarily harbor an accreting-type nucleus. The significance of these findings is nevertheless limited by the small number of objects used in that analysis. In this paper, we expand this work by employing a much larger sample to investigate the physical properties of emission nuclei.

5.2 The nuclei sample & data processing

5.2.1 The data

The data used in this study comprise HST–STIS non-proprietary archival observations of galaxy nuclei for which a ground based spectral classiﬁcation and line ﬂux measurements are provided by Ho et al. (1997a), based on their Palomar Survey. These STIS spectra, most of which were acquired for purposes of deriving central black hole masses, cover the Hα and adjacent strong nebular features that are used in classifying the nuclear ionization mechanism. In all cases, the grating employed was G750M centered at λ6581 or λ6768, which makes available for measurements [N II] λλ 6548, 6583, [S II] λλ 6716, 6731, and at the λ6581 setting, [O I] λ 6300. As of March 2004, such spectra were available for 93 objects. Details on the HST- STIS observations are listed in Table 5.1. We excluded from analysis 15 objects from program IDs 82281 (5 H ii, 4 Transition, and 1 Seyfert) and 8607 (5 Transition) due to extremely low signal-to-noise ratio. The remaining 78 sources consist of 21 H ii, 21 Transition (T), 12 LINER 2 (L2), 12 LINER 1 (L1), and 12 Seyfert (S, both type 1 and 2) nuclei. Our sample includes the SUNNS data, but these objects are not listed in the table as they are presented in detail in Shields et al. (2004b).

1For 3 out of these 10 objects (2 H ii, NGC 2748 and NGC 3949 and 1 T, NGC 5055), Hughes et al. (2003) show the 2-D spectra only, and characterize them as displaying weak extended emission 126

Table 5.1. Nuclei Sample, HST observations

Object Prop. ID z Spectral aperture central λ plate Notes

name typea size (A)˚ scaleb

NGC315 8236 0.016485 L1.9 000.25× 000.1 6768 0.05 (1)

NGC1052 7403 0.004903 L1.9 000.25× 000.2 6581 0.05 (1)

NGC1961 9106 0.01312 L2 000.25× 000.2 6581 0.10 (2)

NGC2685 8607 0.002945 S2/T2 000.25× 000.2 6581 0.10

NGC2748 8228 0.0049 H 000.25× 000.2 6768 0.05 (3)

NGC2903 8228 0.0019 H 000.25× 000.2 6768 0.10

NGC2911 7354 0.010617 L2 000.25× 000.1 6768 0.05 (2)

NGC2964 8228 0.0044 H 000.25× 000.2 6768 0.05

NGC2976 8591 0.000010 H 000.25× 000.1 6581 0.05

NGC3003 8228 0.0049 H 000.25× 000.2 6768 0.10

NGC3031 7351 -0.00011 S1.5/L1 000.25× 000.1 6581 0.05 (1),M81

NGC3162 8228 0.0043 H 000.25× 000.2 6768 0.10

NGC3227 7403 0.003859 S1.5 000.25× 000.2 6581 0.05 (1)

NGC3245 7403 0.004530 T2: 000.25× 000.2 6581 0.05 (1)

NGC3254 8228 0.0042 S2 000.25× 000.2 6768 0.05 (3)

NGC3310 8228 0.0033 H 000.25× 000.2 6768 0.10

NGC3521 8228 0.0027 H/L2:: 000.25× 000.2 6768 0.10 (3) 127

Table 5.1 (cont’d)

Object Prop. ID z Spectral aperture central λ plate Notes

name typea size (A)˚ scaleb

NGC3627 8607 0.002425 T2/S2 000.25× 000.2 6581 0.10

NGC3642 8228 0.0053 L1.9 000.25× 000.2 6768 0.05 (1)

NGC3675 8607 0.002568 T2 000.25× 000.2 6581 0.10

NGC3684 8228 0.0039 H 000.25× 000.2 6768 0.10 (3)

NGC3686 8228 0.0038 H 000.25× 000.2 6768 0.05 (2)

NGC3705 8607 0.003396 T2 000.25× 000.2 6581 0.10 (3)

NGC3756 8228 0.0043 H 000.25× 000.2 6768 0.05 (3)

NGC3917 8607 0.0 T2 000.25× 000.2 6581 0.10 (3)

NGC3949 8228 0.0027 H 000.25× 000.2 6768 0.10 (3)

NGC3953 8228 0.0035 T2 000.25× 000.2 6768 0.10

NGC3953 8607 0.003510 T2 000.25× 000.2 6581 0.10

NGC3998 7354 0.003469 L1.9 000.25× 000.1 6768 0.05 (1)

NGC4036 7403 0.004820 L1.9 000.25× 000.2 6581 0.05 (1)

NGC4041 8228 0.0041 H 000.25× 000.1 6768 0.05 (2)

NGC4051 8228 0.0024 S1.2 000.25× 000.2 6768 0.10 (1)

NGC4088 8228 0.0025 H 000.25× 000.2 6768 0.05

NGC4100 8228 0.0036 H 000.25× 000.2 6768 0.10 128

Table 5.1 (cont’d)

Object Prop. ID z Spectral aperture central λ plate Notes

name typea size (A)˚ scaleb

NGC4150 8607 0.000754 T2 000.25× 000.2 6581 0.10

NGC4212 8228 -0.0003 H 000.25× 000.2 6768 0.05

NGC4258 8228 0.0015 S1.9 000.25× 000.2 6768 0.05 (1)

NGC4258 8591 0.00149 S1.9 000.25× 000.1 6581 0.05 (1)

NGC4261 8236 0.007465 L2 000.25× 000.1 6768 0.05

NGC4278 7403 0.002165 L1.9 000.25× 000.2 6581 0.05 (1)

NGC4303 8228 0.0052 H 000.25× 000.2 6768 0.10

NGC4321 8228 0.0052 T2 000.25× 000.2 6768 0.05 in SUNNS

NGC4414 8607 0.002388 T2: 000.25× 000.2 6581 0.10 (3)

NGC4429 8607 0.003689 T2 000.25× 000.2 6581 0.10 (1)

NGC4435 9068 0.002672 T2/H 000.25× 000.2 6581 0.05

NGC4526 9163 0.001494 H 000.25× 000.2 6768 0.05

NGC4527 8228 0.0058 T2 000.25× 000.2 6768 0.05 (3)

NGC4527 8607 0.005791 T2 000.25× 000.2 6581 0.10

NGC4536 8228 0.0060 H 000.25× 000.2 6768 0.10

NGC4569 8607 -0.000784 T2 000.25× 000.2 6581 0.10 (3)

NGC4579 7403 0.005067 S1.9 000.25× 000.2 6581 0.05 (1) 129

Table 5.1 (cont’d)

Object Prop. ID z Spectral aperture central λ plate Notes

name typea size (A)˚ scaleb

NGC4594 7354 0.003416 L2 000.25× 000.1 6768 0.05 (1)

NGC4736 8591 0.001027 L2 000.25× 000.1 6581 0.05 (1)

NGC4826 8591 0.001361 T2 000.25× 000.1 6581 0.05

NGC4826 8607 0.001361 T2 000.25× 000.2 6581 0.10

NGC5005 8228 0.0032 L1.9 000.25× 000.2 6768 0.05 (1)

NGC5055 8228 0.0017 T2 000.25× 000.2 6768 0.10 (3)

NGC5077 7354 0.009396 L1.9 000.25× 000.1 6768 0.05 (1)

NGC5248 8228 0.0038 H 000.25× 000.2 6768 0.05

NGC5364 8228 0.0041 H 000.25× 000.2 6768 0.10 (3)

NGC5879 8228 0.0026 T2/L2 000.25× 000.2 6768 0.10

NGC5879 8607 0.002575 T2/L2 000.25× 000.2 6581 0.10

NGC5905 9177 0.011308 H 000.25× 000.1 6581 0.05

NGC5921 8228 0.0049 T2 000.25× 000.2 6768 0.05 (1)

NGC6500 7354 0.010017 L2 000.25× 000.1 6768 0.05 (2)

NGC6503 8607 0.000200 T2/S2 000.25× 000.2 6581 0.10

NGC6951 8228 0.0047 S2 000.25× 000.2 6768 0.05

NGC7331 8228 0.0027 T2 000.25× 000.2 6768 0.05 (3) 130

Table 5.1 (cont’d)

Object Prop. ID z Spectral aperture central λ plate Notes

name typea size (A)˚ scaleb

NGC7331 8607 0.002722 T2 000.25× 000.2 6581 0.10

NGC7626 8236 0.011358 L2:: 000.25× 000.2 6768 0.10 (2)

Note. — (1): broad Hα component is required to ﬁt the Hα + [N II] emission complex. (2): broad wings are apparent in the Hα + [N II] blend, however best ﬁts obtained with models of 2 Gaussians for the [S II] lines do not require any additional broad component for Hα. (3): poor quality HST-STIS spectra afrom ground-based spectroscopy, Ho et al. (1997a).

bin arcsec/pixel.

We retrieved from the archive all on-the-fly calibrated frames with the intention to use the fully rectified and flux- and wavelength-calibrated 2-D images to further extract the 1-D spectra. For the majority of objects the signal-to-noise ratio is low, and/or the cosmic-rays (CRs) are abundant. For these cases we obtained the best results by performing our own CR correction to the archival flat-fielded images before feeding these frames back into the HST pipeline software (the x2d task under the stsdas package, IRAF2) to generate the calibrated images. The removal of incident CRs and negative bad pixels was done by employing the ccdclip algorithm under the imcombine task in IRAF; this task rejects pixels with

2The Image Reduction and Analysis Facility (IRAF) is distributed by the National Optical As- tronomy Observatory, which is operated by the Association of Universities for Research in Astronomy Inc. (AURA), under cooperative agreement with the National Science Foundation. 131

values outside of a range specified based on comparisons with the median of all corresponding pixels in all the combined images (after they are realligned to account for their different offsets parallel to the slit). The parameters involved in this procedure (e.g., hsigma, mclip, nkeep) were fine-tuned for each individual case after close comparisons of the central 10 rows (of interest for the 1-D extraction) in both the initial and the corrected image spectra, to ensure that valid emission features were not damaged. For the observations that consist of one nuclear image spectrum only, and which have a high CR density, we interpolated over affected pixels using an au- tomated detection algorithm for the strongest events, and manual intervention for weaker events comparable in strength to real emission features. The 1-D object spectra were extracted by using the apall procedure. As indicated in Table 5.1, the majority of objects were observed through a slit with a 000.2 width. In these cases, we extracted the central 5 rows (pixels) for most of the objects, where the spectra were not rebinned in the spatial direction (plate scale of 0.05 arcsec/pixel), and the central 2.5 rows (pixels) for the cases where the spectra were rebinned to binsize=2, resulting in a plate scale of 0.1 arcsec/pixel (see Table 5.1). In this way, the angular size is preserved. However, because these galaxies are at different distances/redshifts, the metric scale mapped out by these observations is not conserved. For the sake of simplicity, the object spectra acquired with the 000.1 slit width (which were all unbinned spatially) were also generated using 5-pixel-wide extractions.

5.2.2 Measurements of emission-lines

The final 1-D extracted spectra were measured with specfit (Kriss 1994), a method that employs line and continuum spectral fitting via an interactive χ2 minimization. The continuum was approximated by a linear fit, while the emission features were modeled by Gaussian profiles. The [S II] empirical line profile was used as a template in fitting the [N II] lines and the narrow component of Hα. When the [N II] + Hα blend presented visibly strong broad wings, the [S II], and therefore all the other narrow components, were modeled with multiple Gaussians, in order to determine 132

the necessity of an additional broad component for the Balmer Hα. This procedure is in principle able to account for potential asymmetries, velocity blendings or multiple velocity peaks in the narrow lines, and wings in their bases (see also Section 5.2.3). The fitting process was conducted by assigning a common velocity offset for all measured narrow emission features. Line doublets were forced to share the same velocity widths. The flux ratio for the [N II] doublet was constrained to the value determined by the branching ratio. As in Ho et al. (1997a), the measurements of the line flux ratios involving Hα employed the narrow component only. The results are recorded in Table 5.2. The HST-STIS aperture is two orders of magnitude smaller in area than that used in typical ground-based measurements of galaxy nuclei, and hence the corresponding contamination by the host galaxy starlight is significantly reduced. However, the stellar component is not negligible and we need to correct for it, as it sometimes intro- duces additional structure in the continuum, and thus might influence the appearance of the emission features. Fitting a template galaxy with pure starlight emission that would eventually be subtracted from the nuclear spectrum is impossible for most of our data because of the modest signal-to-noise ratio and the relatively small λ coverage. The strongest effect of starlight contamination is expected to be in the Hα emission, due to the presence of the Balmer absorption line. We therefore chose the alternative of adopting reasonable values for the strength of the Hα absorption, using information offered by simple stellar population synthesis models, and correct for it. A typical value of the equivalent width (EW) of the Hα absorption feature was estimated from measuring all the templates contained in the Bruzual & Charlot (2003) library of synthetic galaxy spectra. For three different models of star formation history (instantaneous starburst, constant star formation and exponentially decaying star forming history), each equally represented for three different metallicity values (sub-, supersolar, and solar, with Z = 0.004, 0.05, 0.02 respectively, where Z is the total fraction of metals), and ages that range from 0.005 Gyr to 12 Gyr, the strength (EW) of the Hα absorption line varies between 2A˚ and 6A,˚ with most of the cases concentrating around ∼ 4A.˚ Recent studies of the properties of the nuclear stellar 133

populations (Sarzi et al. 2004) show however that, in a large majority of nuclei (∼ 80%), the age is at least 5 Gyr and the chemical enhancement of the emitting gas is at least solar, narrowing down the possible range in Hα absorption EW to 2.5 – 4.5 A.˚ Therefore, for the corrections used in the final Hα line flux and the line flux ratio comparison involving this feature, we used for the EW of the Hα absorption line the value of 3 A˚ with a typical error of 1.5 A.˚ This uncertainty translates into a typical uncertainy of ∼ 10% in the resulting line flux ratios, but with considerably larger error bars for the weak-lined objects. To understand the consequences of our adopted absorption correction, we repeated the correction with different values of the absorption equivalent width ranging from 2 − 4 A.˚ We found that the distributions of line ratios overall and within object subclasses did not change significantly, and the resulting conclusions were similarly robust. The procedure described above does not account for the starlight contamination of other features considered in this study (e.g., the [S II] doublet), or for its effect on the shape of the Hα emission feature. In these particular data however, these effects are in general expected to be much smaller than in typical ground-based observations of these nuclei. Figure 5.1 provides an illustration of the degree to which the contrast between the nuclear emission and the (stellar) continuum is increased in the HST observations. Here, the ratio of the EWs of the total Hα line emission in the two apertures is shown separately for all individual spectral types, and it is readily apparent that this line is clearly stronger in the HST spectra for all Seyferts and LINERs, and for the majority of the Transition objects. Concerning the rest of the T sources and the H ii nuclei, the errors introduced by the above presented correction for starlight in the Hα emission, albeit significant, are most probably swamped by the large uncertainties generated by the generally limited signal-to-noise ratio, and thus would not affect the final conclusions of this study. 134

Figure 5.1 Ratio of EW of total (narrow+broad) Hα line emission in HST and Palomar apertures, on a logarithmic scale, as a function of nuclear spectral class, as deﬁned by Ho et al. (1997a). The size of each data point is scaled by the respective distance to the host galaxy. The SUNNS data are not included here. 135

Table 5.2. HST Fluxes of Strong Emission Lines

Object [O I] Hα [N II] [S II] [S II]

name λ6300 (narrow) λ6583 λ6716 λ6731

NGC 315 ··· 207±2 520±7 109±8 122±7

NGC 1052 3505±27 4801±26 4088±75 1930±20 2727±21

NGC1961 53±2 165±3 377±7 117±4 141±6

NGC2685 11±3 18±2 86±3 6±2 24±3

NGC 2748 ··· ··· ··· ··· ···

NGC 2903 ··· 54±5 72±6 32±5 18±4

NGC 2911 ··· 44±8 119±8 39±9 50±8

NGC 2964 ··· 83±5 89±5 25±4 26±4

NGC 2976 ··· 166±2 82±2 28±4 9±3

NGC 3003 ··· 345±11 87±6 25±4 19±4

NGC 3031 6440±170 6360±190 13440±160 1980±70 3130±70

NGC 3162 ··· 141±7 53±5 18±4 13±4

NGC3227 637±19 2032±24 3351±13 749±18 777±17

NGC3245 62±7 84±7 217±11 57±4 66±4

NGC 3254 ··· ··· ··· ··· ···

NGC 3310 ··· 3539±60 2139±43 273±23 296±23

NGC 3521 ··· ··· ··· ··· ··· 136

Table 5.2 (cont’d)

Object [O I] Hα [N II] [S II] [S II]

name λ6300 (narrow) λ6583 λ6716 λ6731

NGC3627 14±3 172±3 209±3 67±3 71±3

NGC 3642 ··· 271±27 95±12 77±8 73±8

NGC3675 12±5 22±3 94±3 11±3 23±3

NGC 3684 ··· ··· ··· ··· ···

NGC 3686 ··· 850±20 516±16 31±9 47±10

NGC 3705 ··· ··· ··· ··· ···

NGC 3756 ··· ··· ··· ··· ···

NGC 3917 ··· ··· ··· ··· ···

NGC 3949 ··· ··· ··· ··· ···

NGC 3953 ··· 28±6 54±7 18±6 30±7

NGC 3953 ··· 21±2 30±2 6±2 10±2

NGC 3998 ··· 6475±2 6466±5 1585±8 1984±6

NGC 4036 135±6 161±9 479±12 158±4 171±4

NGC 4041 ··· 371±26 199±28 66±11 70±12

NGC 4051 ··· 5790±410 2510±610 614±68 628±94

NGC 4088 ··· 31±5 23±4 6±3 11±3

NGC 4100 ··· 286±10 144±7 37±6 26±5 137

Table 5.2 (cont’d)

Object [O I] Hα [N II] [S II] [S II]

name λ6300 (narrow) λ6583 λ6716 λ6731

NGC 4150 ··· <15.65 15±1 6±1 5±1

NGC 4212 ··· 39±8 53±8 27±7 11±6

NGC 4258 ··· 1274±19 1052±61 544±17 651±18

NGC4258 750±9 1811±13 1591±4 325±30 376±29

NGC 4261 ··· 15±2 56±3 26±3 22±3

NGC 4278 201±14 462±29 622±37 205±29 273±34

NGC 4303 ··· 392±16 415±16 81±12 106±13

NGC 4321 ··· 632±17 628±15 83±9 103±10

NGC 4414 ··· ··· ··· ··· ···

NGC4429 36±6 44±5 289±8 31±5 45±5

NGC 4435 ··· 23±2 69±3 9±2 15±3

NGC 4526 ··· 31±2 227±6 21±8 45±35

NGC4527 4±5 7±2 28±2 6±2 9±3

NGC 4527 ··· ··· ··· ··· ···

NGC 4536 ··· 34±7 77±9 36±10 42±11

NGC 4569 ··· ··· ··· ··· ···

NGC 4579 1518±21 912±14 3014±36 935±17 1047±18 138

Table 5.2 (cont’d)

Object [O I] Hα [N II] [S II] [S II]

name λ6300 (narrow) λ6583 λ6716 λ6731

NGC 4594 ··· 806±19 2466±22 886±28 886±20

NGC4736 53±6 49±3 185±8 21±5 17±5

NGC 4826 ··· 5±2 77±3 19±3 10±2

NGC4826 11±8 7.4±4 137±5 30±5 37±5

NGC 5005 ··· 278±28 1160±36 342±14 380±14

NGC 5055 ··· ··· ··· ··· ···

NGC 5077 ··· 418±18 522±2 247±7 286±11

NGC 5248 ··· 25±4 77±5 5±3 12±2

NGC 5364 ··· ··· ··· ··· ···

NGC 5879 ··· 29±3 52±4 17±3 17±3

NGC5879 8±3 44±2 50±2 17±2 22±1

NGC 5905 ··· 144±5 174±5 26±6 38±7

NGC 5921 ··· 181±12 307±12 95±8 104±9

NGC 6500 ··· 271±13 241±6 171±15 154±24

NGC6503 2±3 <23.8 17±2 9±3 9±3

NGC 6951 ··· 145±7 572±11 85±7 89±6

NGC 7331 ··· ··· ··· ··· ··· 139

Table 5.2 (cont’d)

Object [O I] Hα [N II] [S II] [S II]

name λ6300 (narrow) λ6583 λ6716 λ6731

NGC 7331 ··· ··· ··· ··· ···

NGC 7626 ··· 87±4 299±10 65±6 71±6

Note. — The ﬂuxes are in units of 10−17 erg s−1 cm−2, and represent the observed values, not corrected for redden- 1/2 ing. The upper limits are listed as (2π) σλ(3σc), where σλ

is the width of the N II line, and σc is the rms uncertainty per pixel in the local continuum. Objects are in the same order as in Table 5.1.

5.2.3 New evidence for broad emission

One of the clearest indications of accretion-powered activity is the presence of broad permitted lines (full width at half-maximum [FWHM] of a few thousand kilo- meters per second). The strongest such line at optical wavelengths is expected to be Hα, and thus detection of a broad Hα component is a promising method of constraining the origin of the nuclear activity. The results of the Palomar survey suggest that broad Hα is quite common in local Seyferts and LINERs, and is found in at least 10% of all nearby luminous galaxies (Ho et al. 1997b). The true incidence of broad-lined sources must however be higher, as, especially in the weakly active nuclei, ground-based observations are unable to distinguish a possibly diluted broad line region from the structure present in the continuum due to starlight. HST spectroscopy 140 that employs small aperture observations has the power to eliminate a significant fraction of contaminating starlight, and thus to increase the sensitivity to broad wings of line emission. An important question to address is thus whether, at HST resolution, the fraction of nuclei classified as broad-lined is larger than that emerging from ground-based data. Ho et al. (1997b) describe the set of objects that show definite or probable evidence of broad Hα emission in their survey. Our sample contains 14 of the nuclei with unambiguous broad Hα in the Palomar spectra, and the HST-STIS observations confirm the presence of the broad Balmer feature in these sources (Table 5.1). In Table 5.3 we summarize for each of these nuclei several parameters concerning the broad Hα component. The fractional contribution of the broad component of Hα to the entire Hα+ [N II] complex is given in col (2), while col (3) indicates its fractional contribution to the whole Hα (narrow + broad) emission. The FWHM of broad Hα and its velocity shift relative to the narrow component are recorded in columns (4) and (5). The last two columns give the observed (not corrected for reddening) flux and luminosity of broad Hα in units of erg s−1 cm−2, and erg s−1 respectively, where we used the distances given by Ho et al. (1997a). Of the objects that are ambiguous in terms of the presence/detection of broad Hα in the Palomar survey, two nuclei are included in this study. While one of them (NGC 4501, a Seyfert included in the SUNNS sample) is confirmed as lacking the broad emission in the HST-STIS aperture (Shields et al. 2004b), the second one, NGC 3245 (viewed from the ground as a narrow-lined T nucleus), requires for a good fit of the Hα+ [N II] doublet the presence of a broad component, with a contribution to the total blend of fblend ≈ 47%. The detection significance is tested by fitting the feature with models based on both one- and two-Gaussian component templates for the [S II] lines; fblend remains basically unchanged in both cases. The existence of the broad Balmer feature in this nucleus has been also suggested by Barth et al. (2001), based on this same dataset. Figure 5.2 shows the individual Gaussian components, the broad Hα feature and the final spectral fit for both the Hα+ [N II] line blend and the [S II] λλ6716, 6731 doublet. The residuals show that the strength of the broad 141

Hα is well constrained, and only weakly dependent on the quality (i.e., S/N ratio) of the data. In the same Ho et al. (1997b) study that presents the search for “dwarf” nuclei with broad emission, the authors also list the sources where the total Hα+ [N II] blend hints at a probable presence of weak broad wings of Hα, for which however a careful profile fitting analysis indicates null detections. Among those objects, 7 are included in our study. For six of them, NGC 4314, 4321, 4698 (included in the SUNNS sample), and NGC 1961, 2911, 6500, the line fits of the HST spectra do not require any broad component. The seventh object is the now widely studied NGC 4594. Analysis of the nebular emission of this nucleus is based mostly on HST-FOS observations, that employ a 0.86-arcsec diameter circular aperture. In these data, the Hα+ [N II] complex shows evidence for broad wings, however attributing them to a broad Balmer component remains a matter of debate (Kormendy et al. 1996; Nicholson et al. 1998). This nucleus shows indeed a rather complicated velocity structure in the STIS slit (0.100), which is almost one order of magnitude narrower than that used in the FOS observation. The optical Hα+ [N II] emission remains difficult to interpret, however we find that a good fit can be obtained only by adding

a moderately strong broad component (fblend ≈ 29%, Figure 5.2). Interestingly, besides these nuclei whose ground-based spectra hinted toward a possible presence of the high-velocity plasma, there are 3 new cases for which the broad emission is detected for the first time in the STIS observations: 1 LINER 2 (NGC 4736) and 2 transition objects (NGC 4429, 5921). The decomposition based on two-Gaussian models of [S II] of their Hα+ [N II] to [S II] spectral region is also presented in Figure 5.2. The fractional contribution of the broad component to the total flux of the blend is 57%, 36% and 20% respectively, and it is not significantly sensitive to the fitting model employed. 142

Figure 5.2 [N ii]+Hα to [S ii] region in NGC3245, NGC4429, NGC4594, NGC4736 and NGC5921. The thin continuous lines show the individual Gaussian components (2 per narrow line), the dashed line represents the broad Hα feature, and the dotted line is the final fit to the observed spectrum, which is presented in thick continuous line. The residuals after the subtraction of the model fit are shown at the bottom of each panel. 143

Table 5.3. Galaxies with Broad Hα Emission

Object fblend fHα FWHM ∆v log F (Hα) log L(Hα) Notes

(1) (2) (3) (4) (5) (6) (7) (8)

NGC315 0.25 0.59 2870 1190 -14.52 39.19

NGC1052 0.34 0.54 2800 10 -13.25 39.33

NGC3031 0.72 0.90 3680 360 -12.15 38.22 a

NGC3227 0.20 0.44 2680 0 -11.78 40.93

NGC3245 0.47 0.79 4300 420 -14.39 38.38

NGC3642 0.79 0.85 1650 450 -13.81 39.15 a

NGC3998 0.59 0.77 6320 320 -12.65 40.09 a

NGC4036 0.24 0.61 2440 730 -14.59 38.27

NGC4051 0.89 0.96 6500 0 -12.40 40.14 a

NGC4258 0.50 0.70 1470 90 -13.45 38.29

NGC4258 0.25 0.41 1310 -40 -13.86 37.88

NGC4278 0.41 0.67 2880 50 -14.03 38.02

NGC4429 0.28 0.75 2650 820 -14.60 37.93

NGC4579 0.70 0.92 6590 1280 -12.92 39.61 a

NGC4594 0.29 0.77 4000 910 -13.59 39.09

NGC4736 0.57 0.89 1570 230 -14.36 36.98

NGC5005 0.17 0.58 2610 5 -14.41 38.32

NGC5077 0.34 0.58 2570 550 -14.24 39.05 144

Properties of the broad Hα emission in these 5 galaxies for which HST spectra exhibit for the first time broad-lined nuclei are also included in Table 5.3. It is interesting to note that in all of these cases there is a significant redward velocity offset of the broad Hα component with respect to the narrow component, whether this is fitted with single or multiple Gaussians. This trend is present sporadically among the Palomar findings, in only 5 out of 34 confident detections. Ho et al. (1997b) suggested that these redshifts may be caused by systematic errors in determining the continuum level or from assumptions about the profiles of the narrow lines. For the cases identified here however, the velocity shifts are relatively larger than those measured by Palomar, and thus unlikely to originate in errors in the line and continuum fitting. Another plausible explanation for these redshifts, also invoked by Ho et al. (1997b) by analogy with the results of spectropolarimetry of NGC 1068 by Antonucci & Miller (1985) and Miller et al. (1991), is scattering of the broad line photons by outflowing electrons in the narrow-line region.

5.2.4 General nuclear properties of the sample

The sources used in this study do not form a complete sample in any statistical sense. They comprise however a large number of local galaxy nuclei, that span quite homogeneously the whole range of spectral morphology of the emission-line galactic nuclei. As noted above, the fraction of the galaxy area observed for each object de- pends, because of the fixed slit size, on the distance to the source. This aperture effect may influence the comparisons of flux ratios, and thus the conclusions of this study; for example, Storchi-Bergmann (1991) reported a negative correlation between [N II]/Hα and distance for a sample of (distant) Seyfert 2s and LINERs that can be attributed to inclusion of increasing amounts of circumnuclear H ii region emission in more distant sources. We investigate the potential influence of this effect in our sample of nearby nuclei in Figure 5.3, where this line flux ratio as measured with HST is shown as a function of galaxy distance. The different spectral types (as defined by ground-based observations) are illustrated by different symbol types/colors (the 145

Table 5.3 (cont’d)

Object fblend fHα FWHM ∆v log F (Hα) log L(Hα) Notes

(1) (2) (3) (4) (5) (6) (7) (8)

NGC5921 0.20 0.45 2280 40 -14.82 38.06

Note. — Col.(1): Galaxy name. When the objects are observed in different programs, the rows showing the results are listed in increasing order of the PID number. Col.(2): Fractional contribution of the broad component of Hα to the Hα+[N II] blend. Col.(3): Fractional contribution of the broad Hα to the total (narrow + broad) Hα emission. Col.(4): FWHM of broad Hα in units of km s−1. Col.(5): Velocity shift of the broad Hα component relative to 6563A,˚ the accepted laboratory wavelength of Hα; the units are km s−1. Col.(6): Observed (not corrected for reddening) flux of broad Hα, in units of erg s−1 cm−2. Col.(7): Observed (not corrected for reddening) luminosity of broad Hα, in units of erg s−1, assuming distances from Ho et al. (1997a). Col.(8): a. non-gaussian profile of the broad component, FWHM and velocity shift are flux weighted averages of the respective values in all Gaussians used to fit the broad Hα. 146

Seyferts include both type 1 and type 2). The Hα flux is corrected for the absorption from the starlight in the host galaxies (see Section 5.2.2). The distribution in distances (and consequently, in the metric scales that the HST nuclear observations encompass), is heavily concentrated around ∼ 16 Mpc ( corresponding to an equivalent aperture radius of ∼ 9 pc, for an area of 000.2 × 000.25), with outliers extending to 65 Mpc. The LINER 2s are the only class that shows a large dispersion in distance (6 – 55 Mpc). No obvious trend in line ratio with source distance is present in either the full sample, or the subgroups of specific nuclear spectral types, suggesting that distance-dependent aperture effects are not a major concern. Of great interest is the degree to which the AGN phenomenon extends to luminosities fainter than those of the well established accretion-like structures, i.e., Seyferts and LINER 1s. Figure 5.4 illustrates the range in narrow Hα luminosity spanned by these nearby nuclei, as resolved with STIS-HST, separately for each spectral category. The scaling of the symbol size with the galaxy distance shows a clear tendency for the more distant objects to be more luminous. This effect is most pronounced in the Transition, the LINER 2, and the H ii sources, that span a wider range in the (narrow) Hα luminosity than the more powerful Seyferts and LINER 1s. It can be seen that the ambiguous objects (T and L2) are, on average, less luminous than the S and L1 sources; the H ii nuclei cover a wide range of observed power. Combined with their distribution in distances, this trend makes Ts and L2s appear weaker than < −15 −1 −2 Ss and L1s in Hα fluxes (∼3 × 10 erg s cm ); the H ii objects are scattered over a large range in the narrow Hα flux. Because the broad-lined objects are of special interest, especially in terms of potential statistical trends toward finding (or missing) them among the nearby nuclei, we indicate the new STIS-HST detections (see Section 5.2.3 for details) by big circles, in all figures throughout the paper. Also marked (by big squares) are those sources for which the narrow emission lines seem to have contributions from multiple kine- matically distinct regions (i.e., multiple Gaussians are needed to fit their line profiles) that are interpreted by Véron, Gon¸calves, & Véron-Cetty (1997) and Gon¸calves Véron-Cetty, & Véron (1999) as evidence for composite S(LINER) + Hii emission. 147

Figure 5.3 The STIS-HST [N ii] λ 6583/Hα line flux ratio plotted against the galaxy distance. The data points are symbol (and color) coded based on the ground-based classification of their nuclear emission into: H ii systems (red diamonds), transition nuclei (T, green triangles), narrow-lined LINERs (L2, blue squares), broad-line LINERs (L1, cyan crosses), and the Seyferts (S, magenta filled dots). The Transition and L2 objects for which HST- STIS observations suggest the presence of broad Hα features are indicated by big circles, while the big squares indicate the T, L2 and H ii nuclei with narrow features that show broad wings. 148

Figure 5.4 The narrow Hα luminosity – measured in the STIS-HST aperture – as a function of nuclear spectral class, as defined by Ho et al. (1997a). The size of each data point is scaled by the respective distance to the host galaxy. The objects with possible broad Hα features are indicated by big circles (L1s are all broad-lined but are not indicated as big circles). The big squares indicate the nuclei with narrow features that show broad wings, and whose profiles are fitted by multiple Gaussian components. 149

In Figure 5.4 we indicate these 2 particular types of objects among the Seyferts as well. Note that, with only one exception (the L2 NGC 4736, see discussion later) the nuclei that potentially signal the presence of an accretion source are among the most luminous ones. In general, the objects are not labeled by their names in the figures. In the plots where the abscissa represents the nuclear spectral type, each individual object is assigned a different x-value, which is preserved for all figures; thus, it becomes handy to interrelate the different properties of each object/class – as revealed by the different line flux ratios presented in these plots. For the 4 Transition sources (NGC 3953, 4321, 4826, 5879) and the 1 Seyfert galaxy (NGC 4258) that are observed in two distinct programs (see Table 5.1 for the Prop IDs), the Hα fluxes differ by ∼ 0.2 dex when different aperture sizes are employed (000.2 versus 000.1), and are consistent within the errors when the same aperture size is used. The measurements from both programs are included in all plots; these cases are identified by vertical bars, or the same value on the x-axis. The differences in the Hα fluxes do not affect the general trend revealed by the aperture comparison (see Section 5.3); the nebular line flux ratios measured in these particular sources from both program spectra are all consistent with each other within the uncertainties, regardless of the size of the observing aperture. None of these objects with repeated observations show broad Balmer emission.

5.3 Is the nuclear emission resolved?

An important issue in this comparative study is the degree to which the high spatial resolution HST observations are able to resolve the nuclear emission in these nuclei. Spectra obtained through a smaller aperture should be characterized by a reduction in the nebular flux. In particular, for extended sources of uniform surface brightness, the decrease in flux is expected to match the simple ratio of the aperture areas (∼ −2.2 dex); less significant changes in the emission fluxes would indicate some degree of central concentration. Figure 5.5 presents the ratios of Hα flux, in the narrow component only, measured in the HST spectra relative to those given by the Palomar survey. The results 150

are sorted according to the ground-based nuclear spectral classification by Ho et al. (1997a). For almost all sample objects, the Hα flux is smaller by at least 0.5 dex in the high resolution HST aperture, which means that, even for the most distant objects in the sample, the nucleus is resolved. The symbol size scales with the galaxy distance, and it can be seen that the ratios do not show any significant trend with the sampled metric aperture, i.e., the more nearby objects are not necessarily better resolved spatially than the more distant sources. It is readily apparent that the smallest change in the flux ratio is present in the nuclei that are unambiguously accretion-powered, the Seyfert and the LINER 1 objects, indicating as expected the highest degree of central concentration of emission. The H ii, transition, and LINER 2 nuclei show lower values of the flux ratio, suggesting a shallower gradient in the surface brightness distribution, and thus a more distributed ionization source. The degree of central concentration also seems to be higher in the L2 and T objects than in the H ii nuclei. The behavior of the transition and the LINER 2 sources is very similar; this means the T nuclei are probably not more prone to contamination by extranuclear star-forming emission than LINER 2s, contrary to what has been suggested by Ho et al. (2003) based on indications that Ts might reside in more inclined host galaxies than those hosting LINERs. It is thus doubtful that the projection effects, if present, play a major role in distinguishing between these two types of emission-line nuclei. The trends revealed in the H ii nuclei are particularly noteworthy: there are a number of sources (5 out of 21) for which the HST aperture spectra show Hα emission at levels that are even lower than those predicted by scaling down the Palomar value by the aperture ratio. These cases show that “nuclear” star formation can evidently spread over a significant range of radial scales, but that it may not continue into the actual center of many galaxies. 151

Figure 5.5 Ratio of Hα ﬂux measured with STIS-HST and by Palomar, as a function of nuclear spectral class, as deﬁned by Ho et al. (1997a). The size of each data point is scaled by the respective distance to the host galaxy. The T and L2 objects with possible broad Hα features are indicated by big circles. The big squares indicate the ambiguous nuclei with narrow features that show broad wings. 152 5.4 The nebular excitation and the central engines

The central emission structure and its energetics can be further tested by simple comparisons of the nebular line flux ratios between the HST-STIS and Palomar spectra. With the present data, various diagnostics are possible. 1. Ratios of collisionally- excited forbidden lines relative to a Balmer recombination line (i.e., [N II]/Hα) are sensitive to the nebular temperature, and play a key role in the standard spectroscopic classification diagrams. 2. The [S II] λλ 6716/6731 ratio provides information on the range of the particle density (ne) for the emitting clouds. 3. Particularly strong tests of nonstellar ionization are supplied by [O I] λλ6300, 6364, as these features are very weak in normal H ii regions. Figures 5.6, 5.7, and 5.8 present “ratio of ratios”, that compare R = [N II]/Hα, [S II] λλ 6716, 6731/Hα, and [O I] λ6300/Hα respectively, measured from the HST data to those obtained from the ground-based spectra in the Palomar survey. There is a general pattern that can be seen in all of these plots: for the majority of the L1 and S nuclei, these ratios show little scatter around unity, while H ii, T and L2s reveal a wide range of values for these ratios. The L1 and S nuclei show a systematic decrease in [S II]/Hα in the small aperture. For the H ii and L2 nuclei, the majority of sources have larger [N II]/Hα, [S II]/Hα, and possibly [O I]/Hα, in the small aperture observations; there is no net trend in the Transition objects. As expected from the results shown in Figure 5.3, there is no obvious correlation with distance to the host galaxy in any of these ratios. The aperture effects on the ratio [S II] λλ 6716/6731, that serves as a density indicator (e.g., Osterbrock 1989), are investigated in Figures 5.9 and 5.10. These plots show, respectively, the behavior of the ratio of the HST values to those of Palomar, separately for each nuclear spectral type, and a direct comparison of the values of the [S II] λλ 6716/6731 line ratio in the two apertures. It is clear from both of these representations that in the more central nuclear regions, as mapped by the STIS observations, the gas densities are larger, i.e., smaller [S II] λλ 6716/6731 flux ratios, in all types of objects. A few exceptions are among the H ii nuclei, however the large error bars for these values do not necessarily exclude them from the general 153

Figure 5.6 [N ii]/Hα from HST spectra ratioed to the Palomar value, as a function of nuclear spectral class (ground based classiﬁcation). Same symbol code as in Figure 5.5. 154

Figure 5.7 [S ii] λλ 6716, 6731/Hα from HST spectra ratioed to the Palomar value, as a function of nuclear spectral class (ground based classification). Same symbol code as in Figure 5.5. trend. The Transition sources show a particularly pronounced difference in the gas density measured in the two apertures. The general tendency of H ii and L2s to exhibit stronger forbidden lines relative to Hα in the “more nuclear” spectra may stem from a variety of reasons. The most attractive one is that the emission in the small aperture is dominated to a greater degree by a central accretion-powered source. The enhancement in these ratios can however also be simply a direct consequence of the trend toward larger electron densities at small radii, as revealed by the patterns in [S II] doublet ratio, for virtually all nucleus types. Because large densities may lead to suppression of the infrared fine 155

Figure 5.8 [O i]/Hα from HST spectra ratioed to the Palomar value, as a function of nuclear spectral class (ground based classification). Same symbol code as in Figure 5.5. structure line cooling, a boost may result in the optical forbidden lines like [N II], [S II] and [O I]. Evidently even higher densities are present in the L1s and Seyferts, resulting in less efficient emission of S II on small scales. It is noteworthy that the differences between the Palomar and HST ratios of forbidden to Hα lines are in general small. Although as described in Section 5.3, the flux sampled in the HST apertures is typically a small fraction of that contributing to the Palomar spectra, and the spatial scale of the observed regions differs by at least an order of magnitude, there are very few sources that might indicate a change in their classification from ambiguous (narrow-lined) nuclei to a more secure AGN-like behavior. Consistent with this result is the relatively small fraction of the objects 156 with newly detected broad Balmer lines; as illustrated in the plots, these objects, with the exception of NGC 4736 (discussed in more detail in Section 5.5), are also among those that show the larger narrow line ratios in the spectra acquired at smaller scales. Thus, at least for these objects, the scenario in which the small aperture emission may be dominated to a greater degree by a central accretion source seems applicable. Interestingly, the nuclei whose ground-based spectra indicate a potential mix of velocity fields (i.e., distinct multiple Gaussian profiles required to fit the narrow emission), behave similarly in both apertures, as their spectra remain consistent with very little or no change in the nebular line ratios; they are also among the most distant targets, which might mean, if the “composite” picture holds, their HST nuclear observations could still include a significant amount of light from circumnuclear star- forming regions. The sources clearly hosting an accretion-type source in their centers, the S and L1 nuclei, show the weakest evidence for radial gradients in ionization or excitation. A plausible interpretation of this trend may be that the ionization, which scales with the ratio of ionizing photons to particle densities, shows a weaker change than predicted from simple r−2 dilution of the radiation field, due to a radial decrease in density, as indicated in Figures 5.9 and 5.10. It is then interesting to investigate to what degree this effect might contribute to the lack of change in the Ts and L2s as well. It is noticeable that, among Ts and L2s, the sources with new detections of broad components show the smallest differences in the particle densities. Thus, for these particular objects the potential counteracting changes in ne are probably not very strong, and a significant aperture dependence is seen. On the other hand, the rest of the Ts, with the greatest changes in ne among the entire sample, would be expected to show weak gradients, in accord with the observations. This pattern does not seem to apply for L2s as well, but it may also be the case that the wider range in distances to these galaxies may skew the aperture comparison and reduce the expected trends. Thus, these unexpected results of similar ionization outcomes in the small and large apertures might be accounted for, if only in part, by possible differences in the physical conditions of the more central regions, as opposed to those 157

Figure 5.9 [S ii] 6716/6731 from HST spectra ratioed to the Palomar value, as a function of nuclear spectral class (ground based classiﬁcation). Same symbol code as in Figure 5.5. 158

inferred from nebular environs that include more circumnuclear light. Even if the radial variations in the nebular line emission are potentially weakened by the gradients in the particle density, the great number of T nuclei that are consistent with no diﬀerence in the measured line ratios remains a big puzzle. In particular, the fact that there is no net majority of Ts that show a rise in the strength of the nebular emission at smaller scales challenges the proposed composite picture of these nuclei, and questions its applicability to the majority of the members of this class: the intermediate line ratios that persist throughout the nebula imply that there is no spatial segregation, especially of the kind in which a central unambiguous LINERs is surrounded by H ii regions. An alternative hypothesis that may be more consistent with these observed trends would be that the central source is not the compact, accreting-type, but rather in some sort of distributed form(s) (i.e., turbulent mixing layers, stellar phenomena).

5.5 Discussion

The results emerging from the comparisons of ratio-of-ratios in Section 5.4 are complex, and not easy to understand. In the attempt to better interpret them, we measured for several pairs of spectral types the probabilities that their distributions in Hα fluxes and various line ratios are consistent with having the same parent population using Kolmogorov-Smirnov (KS) tests. Tables 5.4 and 5.5 present the resulting statistics for HST and Palomar data respectively, with the most significant probabilities (higher than 10%) shown in boldface. There are a few noteworthy facts that are apparent from these tables: 1. the L1s and L2s are not as similar in their HST nebular properties as they appear in the Palomar data; 2. there is a great similarity between Ts and the accretion-type objects (L1s and Seyferts) in the HST aperture, while they are almost totally distinct in the Palomar spectra; and 3. the H ii nuclei remain separated from the other spectral types in all of their nuclear properties except the narrow Hα flux. That the L1s and L2s appear to show more discrepancy in their characteristics in the more central regions is intriguing, and brings back the question whether these 159

Figure 5.10 [S ii] 6716/6731 from HST spectra compared to the same ratio from the Palomar measurements. Same color/symbol code as in Figure 5.3. 160

two spectral types are fundamentally the same or have real intrinsic differences. As presented in Sections 5.2.4 and 5.3, these two categories of nuclei span different ranges in narrow Hα luminosities and fluxes as measured with STIS. Figures 5.6 and 5.7 indicate that the global/mean values of their ratio of ratios in [N II]/Hα, [S II]/Hα seem to follow different trends: they are scattered in quite opposite directions around the unit value, with LINER 1s showing particularly low [S II]/Hα values in the HST spectra relative to those of Palomar. At least for the sample presented here, the L1s show a tendency to higher densities than those indicated in L2s in both ground-based 3 3 and space-based observations: median ne(L1) ≈ 8.5 × 10 , ne(L2) ≈ 5 × 10 in the 3 3 HST spectra, and ne(L1) ≈ 1.5 × 10 , ne(L2) ≈ 0.6 × 10 in Palomar data. An impediment to understanding these differences is that the L2s cover a wider distribution in galaxy distances than the L1s, Ts or Seyferts, which are more concentrated near 20 Mpc. We address this issue by making the same KS comparison tests on the line ratios of the narrow and broad-lined LINERs (as defined from the ground), where we consider several subsamples of L2s that exclude some or all of the 4 most distant L2s (NGC 1961, 2911, 6500, and 7626), in an attempt to match the two samples of LINERs in both mean/median and distribution of distances. In all cases, the subsamples of L2s manifest the same behavior as the whole sample of L2 nuclei, which means that differences in the metric aperture do not influence these results that suggest L1s and L2s might be intrinsically distinct galaxy nuclei. Another remarkable aspect of these KS tests is the similarity in the distributions of measured properties with HST for Ts and L1/S nuclei. A first naive interpretation of these statistics would be that the Transition nuclei are more AGN-like in the small aperture data. However, it is important to note that [N II]/Hα alone is a weak discriminant between signatures of starlight and an accretion powered spectrum, and that the [S II]/Hα ratios are strongly influenced by densities, which, as shown previously, generally go up in all types of objects discussed in this project, with a stronger effect for the type 1 LINERs and the Seyferts. Also, measurements of the [O I] feature, which would be a far more stronger probe of potential differences between these types of nuclei, are insufficient. The crucial test that would tell whether Ts and L1s 161 or Seyferts are indeed close in their behavior would be given by comparisons of [O III] λ 5007/Hα. Such data however are not available, and thus a definitive conclusion regarding this interesting matter is not yet possible. The data used in this project show that [O I] λ6300 might possess additional unexpected qualities as tracer of accretion. Although very few measurements of this feature are available for this aperture comparison study, it is readily apparent that the nuclei with the strongest gradient in [O I] almost always signal the presence of a broad Balmer feature, and therefore the presence of an accreting source. Interest- ingly, among all ratio-of-ratios measured for the newly found broad-lined NGC4736, that of [O I]/Hα is the only one that shows a clear increase in the HST aperture. Future observations should be able to determine whether this trend is present at the statistical level necessary to be used as a secondary characteristic of an AGN-type nucleus, and to possibly be used in spotting non-stellar emission in the low spatial resolution data.

5.5.1 The remarkable nucleus of NGC 4736

As noted at almost every step in this comparison of nebular characteristics in small and large apertures, the nearly face-on spiral NGC 4736, a LINER 2 as defined from the ground, is an exception in many ways; it seems to deviate from all of the trends the other nuclei follow, especially from those manifested by the objects that reveal broad Balmer emission: the HST spectrum of this nucleus presents a weak increase in [N II]/Hα, and, unlike the majority of the L2s, a clear decrease in [S II]/Hα. This object resides in one of the closest galaxies, and its narrow Hα flux and luminosity are among the lowest in the whole sample. This object shows the lowest [S II]/Hα ratio, and possibly the lowest particle density among Ts and L2s. The width of its broad Hα component is among the smallest ones measured in other cases in this study, as well as in the whole Palomar survey. The interesting nature of NGC 4735 becomes even more difficult to understand when its appearance in other wavelengths is considered: Chandra observations (Era- cleous et al. 2002) reveal nuclear emission that is dominated by a plethora of discrete 162

Table 5.4. HST data, KS statistics

type-pair F(Hα)a [N II]/Hα [S II]/Hα [S II] ratio

L1–L2 0.002 0.100 0.551 0.029

L1–T 1.1×10−5 0.308 0.974 0.436

L2 – T 0.138 0.431 0.394 0.004

S–T 1.6×10−4 0.589 0.209 0.308

HII–T 0.231 1.6×10−3 0.004 0.202

HII – L2 0.905 1.3×10−3 0.010 0.727

aObserved ﬂux of the narrow Hα line, not corrected for reddening 163

Table 5.5. Palomar data, KS statistics

type-pair [O I]/Hα [N II]/Hα [S II]/Hα [S II] ratio

L1–L2 0.002 0.996 0.032 0.459

L1–T 5.2×10−7 0.009 2.5×10−6 0.051

L2–T 1.1×10−6 0.004 2.0×10−4 0.571

S–T 6.6×10−5 0.066 0.006 2.4×10−4

HII–T 1.9×10−8 7.8×10−8 1.4×10−7 0.389

HII–L2 8.2×10−7 8.2×10−7 8.2×10−7 0.930

X-ray sources (most likely X-ray binaries) embedded in a diﬀuse but clumpy halo, rather than by a single central source. Its UV image (Maoz et al. 1995) reveals a compact central nuclear source, while narrow band Hα and O III images with HST that resolve its nuclear emission region down to scales of less than 10 pc (Pogge et al. 2000) show complex ﬁlamentary structure. Ground-based Hα (Pogge 1989) and radio continuum images show that this nucleus is surrounded by a star-forming ring of diameter of 2 kpc. The spatial distribution of the X-ray point sources in the Chandra data is not uniform with distance from the center, and indicates one concentration of sources in the central 1000, and a separate concentration at the radius of the star-forming ring (∼ 4500 − 5000). This distribution of X-ray sources does not follow the host galaxy starlight (i.e., infrared light), but rather the emission-line light (Smith et al. 1994), suggesting a connection with star formation. All these characteristics put into question a possible single, compact accreting supermassive black hole as responsible for ionizing the nuclear nebula in this object. Nevertheless, a smaller black hole could still produce the (relatively narrow) broad 164

Balmer component measured in the STIS spectrum, and in this case, it would be interesting to ﬁnd out which of the four brightest X-ray nuclear sources revealed by Chandra would be its counterpart. Since in both weak and powerful AGNs the strength of the hydrogen recombination lines generally scales with the X-ray luminosity (e.g., Ho et al. 2001), it is interesting to see whether any of these 4 central X-ray point-like sources -assuming their Hα luminosity is what we measure in the STIS spectrum- follow such trend. NGC 4763 behaves peculiarly in this case too: for any of the X-ray nuclear sources, the ratio LX (2-10 keV)/L(Hα)(narrow+broad) is well above the value predicted for AGNs, which is completely surprising given that, in general, the narrow-lined nuclei (L2s and Ts) fall below the predicted trend. A good alternative explanation for the properties of this object remains a starburst, which might still be ongoing to some degree in the circumnuclear star-forming ring. In this picture, the broad Hα could be attributed to SNe or SN remnants, whose typical X-ray luminosities could be well matched with the values measured for the point sources detected in NGC 4736. To explore this idea in more detail is however diﬃcult (and beyond the scope of this chapter), as optical spectra of SNe, even if only the type 2s, are widely heterogenous, making a spectral comparison rather cumbersome.

5.6 Summary & Conclusions

In this paper we have presented the results of an analysis of the nebular emission in a large sample of nearby galactic nuclei, with the purpose of understanding the underlying energy sources in these common objects. Our investigations employs a comparison of the nebular line emission sampled by large (ground-based) and small (HST) spectroscopic apertures. With this approach, we have examined spatial properties of emission-line surface brightness and ionization structure, to ﬁnd signatures of accretion or other power sources. The conclusions of this analysis are rather unexpected: 1. For a large majority of nearby emission-line galaxy nuclei, the origin of their nebular excitation is not an accreting black hole, but rather a more distributed power 165

source. For many of the objects included in our sample, the expected increasing domination on small scales by a compact, accreting-type nucleus (i.e., an increase in the flux line ratios) is not perceived, in spite of the enhanced sensitivity offered by the HST data. 2. The densities are almost always higher in the more central regions, for all types of nuclei. This phenomenon explains in principle why the accretion-type nuclei, L1s and Seyferts, do not show a change in their nebular line ratios, as the aperture effect is probably counteracted by the different physical conditions that arise into more dense environments, which conspire toward reducing the expected gradients. 3. Nuclear emission is clearly resolved by the STIS aperture, as the narrow Hα flux is generally less than that measured in the Palomar spectra by a factor at least as large as the simple ratio of aperture areas. The difference in flux between the two apertures is at least as strong in the Ts as in the LINERs, suggesting that the geometry of the obscuring/starforming material in these spectral categories do not differ substantially. 4. For 3 Transition and 2 Liner 2 nuclei, the HST spectra reveal for the first time the presence of a broad Balmer component. With the exception of the peculiar L2 NGC 4736, this is interpreted as a clear dominance of an AGN-type ionizing source. 5. The data suggest that strong [O I] λ6300 might correlate with the presence of the broad Hα. This correlation, if proved statistically significant in larger datasets, might be able to pinpoint the presence of an accreting supermassive black hole, i.e., an AGN, as the dominant nuclear source power. 6. A close look at the nuclear properties of the different categories of nuclei disclose a possibly intrinsically different nature for the L1s and L2s. A closer resemblance between Ts and the accreting-type objects (L1s and Seyferts) is apparent from these data, however the stronger tests to prove such similarity remain unavailable. The H ii nuclei remain separated from the other types of sources in almost all of their nebular properites. 166

7. The STIS-HST spectrum of the NGC 4736 nucleus reveals a markedly uncon- ventional behavior of its nebular emission. Although it exhibits a broad Hα component, this object is doubtly associated with an accreting supermassive black hole. Its attributes, as revealed by other wavelengths, conform reasonably well with those of a SN or a SNR, however further studies are necessary to decide on its true nature. 167

Chapter 6 Summary

”Life is short, Art long, Occasion sudden and dangerous, Experience deceitful, and Judgment diﬃcult.” (Hippocrates)

6.1 Basic Results

6.1.1 Nuclear activity at the highest redshifts

High redshift QSOs provide important opportunities for advancing our understanding of the astrophysics of galaxy formation and evolution. Although a growing number of these sources are now known to exist at redshifts beyond 4, systematic studies of the emission-line properties of these objects have been quite limited. Chapter 2 presents results of a program of high signal-to-noise spectroscopy for 44 QSOs using the MMT and Keck observatories. The quasar spectra cover 1100 – 1700 A˚ in the rest frame for sources spanning a luminosity range of ∼ 2 orders of magnitude. Com- parisons between these data and spectra of low redshift quasars reveal a high degree of similarity, although differences are present in the profiles and the strengths of some emission features. An examination of the luminosity dependence of the emission lines > reveals evidence for a weak or absent Baldwin effect among z ∼ 4 QSOs. We compare measurements for objects in our sample with results from other high redshift surveys characterized by different selection techniques. Distributions of equivalent widths for 168

these different ensembles are consistent with a common parent population, suggesting that our sample is not strongly biased, or in any case, subject to selection effects that are not significantly different from other surveys, including the Sloan Digital Sky Survey. Based on this comparison, we tentatively conclude that the trends identified here are representative of high z QSOs. In particular, the data bolster indications of supersolar metallicities in these luminous, high-z sources, which support scenarios that assume substantial star formation at epochs preceding or concurrent with the QSO phenomena.

6.1.2 Getting closer to home: the NLS1 galaxies

Narrow Line Seyfert 1 galaxies are remarkable for their extreme continuum and emission line properties which are not well understood. New results bearing on the spectroscopic characteristics of these objects are presented in Chapter 3, with the aim of establishing their typical UV and optical spectral behavior. The study employs HST observations of 22 NLS1s and similar objects, which represent a substantial improvement over previous work in terms of data quality and sample size. High signal-to-noise NLS1 composite spectra are constructed, allowing accurate measurements of the continuum shape and the strengths, ratios, and widths for emission lines, including weak features which are barely identiﬁable in other AGN composites. NLS1 sources reveal redder UV-blue continua than those typically measured in other quasars and Seyferts. Objects with UV line absorption show redder spectra, suggesting that dust is important in modifying the continuum shapes. The data also permit > a detailed investigation of the previously proposed link between NLS1s and z ∼ 4 quasars. A direct comparison of their composite spectra, as well as a Principal Com- ponent Analysis, suggest that high-z QSOs do not show a strong preference toward NLS1 behavior.

6.1.3 Is dust shaping the quasar SED?

Chapter 4 describes a synthetic observational analysis of a numerically simulated sample of AGNs designed to explore the role of dust absorption in modifying their 169

spectral energy distribution, and in particular, possible luminosity dependent eﬀects. Monte Carlo methods are employed in a scheme in which the inner radius of the absorbing material scales with the intrinsic luminosity of the central source (a variant of the receding torus picture), consistent with expectations that grains will evaporate above a certain radiation density. We show that this simple model results in a statistical dependence of reddening in general accord with the observed correlation between quasar luminosity and continuum slope. The resulting numbers of absorbed sources predicted in this framework appear consistent with recent modeling/observations of the X-ray background light and extragalactic point source population.

6.1.4 Emission-line nuclei in the nearby universe

Low-luminosity AGNs greatly outnumber the more luminous quasars and may be the most common mechanism for cosmic black hole growth. A significant number of these objects, although labeled as AGNs, do not exhibit the traditional hallmarks of an accretion-powered system. Chapter 5 delineates the results from a sensitive search for accretion signatures in a large sample of nearby emission-line galaxy nuclei, by employing a quantitative comparison of the nebular line flux ratios in small (HST) and large (ground-based) apertures. It is found that some objects show emission-line trends consistent with increasing domination by a non-stellar central power source on small scales. However in most cases, the nebular line flux ratio measured within the central tens of pc is not much different from that averaged over an order of magnitude larger sizes, containing at least ten times as much emission. This suggests that many nearby emission-line nuclei are not necessarily powered by accretion onto a compact central object, and that the composite model suggested for the LINER/H II transition nuclei (that assumes a geometry in which a central accreting-type nucleus is surrounded by star-forming regions) is not generally supported. Although the nuclear emission is resolved by the HST data and the contrast with the surrounding starlight is significantly enhanced, only 5 narrow-lined objects reveal broad Balmer Hα features, changing thus their spectral classification from type 2 to type 1. We find that the strength of the [O I] λ6300 emission may be related to the presence 170

of the broad Hα, and some evidence that the type 1 and type 2 LINERs might be intrinsically diﬀerent. We discuss in more detail the emission characteristics of NGC 4736 which, although exhibiting a broad Balmer component, is more likely to be a supernova-type source than an AGN-type nucleus.

6.2 Future Directions

This dissertation began with a series of questions concerning the accretion process and the relationship to luminosity and redshift. Although this work has led to signiﬁcant new insights regarding the nature of AGNs, other questions have been raised, and it seems only natural to list them here, along with suggestions for future observations and techniques that might resolve them.

1. If the majority of the transition objects and maybe LINER 2s are not accretion powered, what then is the nature of their energy source?

Comparing the line measurements in the small and large apertures has been a sensible first step in resolving the emission structure in the nearby nuclei. However, to properly trace out any radial trends we would wish to be able to look at more resolution elements between 000.2 and 200. The existing data however, do not permit such investigation for any individual object, as often outside of the central few pixels the signal fades away drastically. Luis Ho, one of my collaborators on the project presented in Chapter 5, proposed that composite spectra might offer sufficient photons to analyse an “average” radial profile for an “average” object for some designated class. This approach seems promising for future investigations.

2. What can we tell about the black holes in the ﬁve newly found broad-lined objects? To what degree are they diﬀerent from the super massive dark objects lurking in the nuclei of the nearby bright galaxies? What is powering the nucleus of NGC 4736?

Stringent limits on the mass of the central dark object would be of particular interest for gauging whether this parameter relates to any of these sources’ emission 171 properties. Estimating the black hole masses (MBH) for these nuclei may be possible through various techniques, some more expensive than others, especially in terms of observing time. Relatively direct methods employ follow-up HST observations, that would provide either rotation curves of the optical emission lines from the gas (assuming there is no strong influence of non-gravitational forces), or spatially resolved stellar absorption lines whose kinematics could be modelled. Indirect ways to weigh the central compact dark object rely on the now calibrated relationships between the black hole mass and other host galaxy properties, e.g., the MBH − σ correlation (σ being the stellar velocity dispersion), or reverberation mapping results combined with measurements of the line profile widths. In the latter method however, an estimate of the central source’s luminosity is necessary, and it may be problematic in these low power objects. A fundamental test for potential peculiarities of the accretion mechanism in these objects remains the comparison of their nuclear radio-to-X-ray SEDs, obtained at arcsecond resolution at most wavelengths, with the SEDs of their more luminous counterparts. Our understanding of NGC 4736 would particularly benefit from such an approach. The intrinsic weakness of the low-luminosity nuclei poses great practical challenges in obtaining such data. However, it seems increasingly possible to put together such observations, hence to answer whether these sources are a result of low accretion rate, low efficiency, or low black hole mass.

3. Is the receding torus model the right prescription for the geometry of the circumnuclear dust in AGNs?

Investigating the eﬀects of circumnuclear dust obscuration on our view of the AGN’s intrinsic properties is not a trivial matter, however, as indicated in Chap- ter 4, important insights may be obtained by combining simulations with statistical observational results. Although the receding torus model seems to account for various observational results, there are still tests to run to assess whether this geometry applies to AGNs in general. A ﬁrst step in further examining this receding torus picture, through its luminosity dependent behavior, is modifying the model presented in Chapter 4 to obtain a more 172

realistic representation of the absorption process (e.g., putting a limit on the outer edge of the torus, implementing a radial dependence of the particle density, and using radiative transfer treatment that further accounts for scattering eﬀects), and incorporate a more complete ensemble of AGN intrinsic and observed characteristics (e.g., redshift distribution, selection eﬀects). On the observational front, a statistical study of the emission and absorption properties in large samples of Seyferts and quasars that span simultaneously wide wavelength ranges, and provide the smallest possible contamination from the host galaxies, is a key approach to a variety of issues regarding the AGN intrinsic characteristics. HST UV data is crucial for quantifying the relationship between gas absorption and dust obscuration, in assessing the role played by the starlight contamination in modifying the continuum shape, and thus in constraining further the proposed picture of a luminosity-dependent absorption. Such a dataset would also allow for a systematic analysis of independent reddening indicators in Seyferts and quasars.

4. What happens to the fraction of the nuclear luminosity that is intercepted by the dusty torus in the highly obscured AGNs?

This is by no means a new question and various studies have been dedicated to ﬁnding the answer. Predictions are that the light absorbed by the obscuring torus is re-radiated into the far infrared, however, this process remains little understood. Spitzer observations will be very important for investigating this topic. The receding torus model presented here could potentially help in interpreting the Spitzer data, by extending the calculations of the dust absorption (and scattering) to IR wavelengths and adding a reasonable prescription for dust reprocessing of the radiation. By employing this type of simulation, it may also be possible to obtain statistical estimates of the fraction of the AGN power that contributes to the IR background. Such analysis would help assess the incompleteness in quasar surveys in limited bandpasses, and thereby help in our knowledge of AGN evolution. 173

Bibliography

Akiyama, M., et al., 2000, AJ, 532, 700

Antonucci, R. R. J., & Miller, J. S., 1985, ApJ, 297, 621

Appenzeller, I. & Wagner, S. J., 1991, A&A, 250, 57

Baldwin, J.A. 1977, ApJ, 214, 769

Baldwin, J. A. Phillips, M. M, & Terlevich, R., 1981, PASP, 93, 5

Baldwin, J.A., et al. 1996, ApJ, 461, 664

Barvainis, R., 1990, ApJ, 353, 419

Barth, A. J., Reichert, G. A., Filippenko, A. V., Ho, L. C., Shields, J. C., Mushotzky, R. F., & Puchnarewicz, E. M., 1996, AJ, 112, 1829

Barth, A. J., Reichert, G. A., Ho, L. C., Shields, J. C., Filippenko, A. V., & Puchnarewicz, E. M., 1997, AJ, 114, 2313

Barth, A. J., Filippenko, A. V., & Moran, E. C., 1999a, ApJ, 515, 61L

Barth, A. J., Filippenko, A. V., & Moran, E. C., 1999b, ApJ, 525, 673

Barth, A., J., Sarzi, M., Rix, H.-W., Ho, L. C., Filippenko, A. V., & Sargent, W. L. W., 2001, ApJ, 555, 685

Barth, A., & Shields, C. J., 2000, PASP, 112, 753

Becker, R. H., Helfand, David J., & White, R. L. 1992, AJ, 104, 531 174

Boller, Th., Truemper, J., Molendi, S., Fink, H., Schaeidt, S., Caulet, A., & Den- nefeld, M., 1993, A&A, 279, 53

Boller, Th., Brandt. W. N., & Fink, H., 1996, A&A, 305, 53

Boroson, T. A. & Green, R. F., 1992, ApJS, 80, 109

Boroson, T. A., 2002, ApJ, 565, 78

Brandt, W. N., & Gallagher S. C., 2000, NewA Rev., 44, 461

Brotherton, M. S., Wills, B. J., Francis, P. J., & Steidel, C. C. 1994, ApJ, 430, 495

Brotherton, M. S., Tran, Hien D., Becker, R. H., Gregg, Michael D., Laurent- Muehleisen, S. A., & White, R. L. 2001, ApJ, 546, 775

Brusa, M., 2004, PhD thesis, Bologna University

Bruzual, G., & Charlot, S., 2003, MNRAS, 344, 1000

Calzetti, D., Armus, L., Bohlin, R. C., Kinney, A. L., Koornneef, J., & Storchi- Bergmann, T., 2000, ApJ, 533, 682

Cardelli, J. A., Clayton, G. C., & Mathis, J. S. 1989, ApJ, 345, 245

Cheng, F. H., Gaskell, C. M., & Koratkar, A. P., 1991, ApJ, 370, 487

Collin, S. & Joly, M., 2000, NewA Rev., 44, 531

Constantin, A., Shields, J. C., Hamann, F., Foltz, G. B., & Chaﬀee, F. H., 2002, ApJ, 565, 50

Constantin, A., & Shields, J. C., 2003, PASP, 115, 592

Crenshaw, D. M., Kraemer, S. B., Boggess, A., Maran, S. P., Mushotzky, R. F., & Wu, C., 1999, ApJ, 516, 750

Crenshaw, D. M. & Kraemer, S. B., 2001, ApJ, 562, 29 175

Crenshaw, D. M., et al., 2002, ApJ, 566, 187

Dalcanton, J. J., Yoachim, P., & Bernstein, R. A., 2004, 608, 189

Dickinson, M., & McCarthy, P. J. 1987, BAAS, 19, 1125

Dietrich, M., et al., 1999, A&A, 352, L1

Dietrich, M., & Wilhelm-Erkens, U. 2000, A&A, 354, 17

Dietrich, M., Hamann., F., Shields, J. C., Constantin, A., Vestergaard, M., Chaﬀee, F. H., 2002, Foltz, G. B., & Junkkarinen, V. T., 2002, ApJ, 581, 912

Dietrich, M., Hamann, F., Shields, J. C., Constantin, A., Heidt, J. J¨ager, K., Vester- gaard, M., & Wagner, S. J., ApJ, 589, 722

Dopita, M. A. & Sutherland, R. S. 1995, ApJ, 455, 468

Dopita, M. A., Heisler, C., Lumsden, S., & Bailey, J., 1998, ApJ, 489, 570

Dumont, A. M., & Mathez, G. 1981, A, &A 102, 1

Edelson, R. A., Pike, G. F., & Krolik, J. H., 1990, ApJ, 359, 86

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

Espey, B. R., Carswell, R. F., Bailey, J. A., Smith, M. G., & Ward, M. J. 1989, ApJ, 342, 666

Fan, X., et. al 1999, AJ, 118, 1

Fan, X., et al. 2000, AJ, 119, 1

Fan, X., et al. 2001, AJ, 121, 54

Ferland, G. J., & Netzer, H., 1983, ApJ, 274, 105

Ferland, G. J., et al. 1996, ApJ, 461, 664

Ferrarese, L., & Merritt, D. 2000, ApJ, 539, L9 176

Filippenko, A. V. 1982, PASP, 94, 715

Filippenko, A. V., & Sargent, W. L. W., 1985, ApJS, 57, 503

Filippenko, A. V., & Terlevich, R., 1992, ApJ, 397, 79L

Forster, K., Paul J. G., Aldcroft T. L., Vestergaard M., Foltz G. B., & Hewett P. C. 2001, ApJS, in press

Francis, P. J., Hewett, P. C., Foltz, C. B., Chaﬀee, F. H., Weymann, R. J., & Morris, S. L. 1991, ApJ, 373, 465

Francis, P. J., Hewett, P. C., Foltz, C. B., & Chaﬀee, F. H. 1992, ApJ, 398, 476F

Francis, P. J. 1993, ApJ, 407, 519

Francis, P. J., Hooper, E. J., & Impey, C. D. 1993, AJ, 106, 417

Gabel, J. R., Bruhweiler, F. C., Crenshaw, D. M., Kraemer, S. B., & Miskey, C. L., 2000, ApJ, 532, 883

Gaskell, C. M. 1982, ApJ, 263, 79

Gaskell et al., 2004, ApJ submitted (astro-ph/0309595)

Gebhardt, K., et al. 2000, ApJ, 539, L13

Ghez, A. M., et al., 2003, ApJ, 586, 127

Giveon, U., Maoz, D., Kaspi, S., Netzer, H., & Smith, P. S., 1999, MNRAS, 306, 637

Gnedin, N. Y., & Ostriker, J. P. 1997, ApJ, 486, 581

Gon¸calves, A. C., V´eron-Cetty, M.-P., & V´eron, P., A&AS, 135, 437

Goodrich, R. W., 1989, ApJ, 342, 224

Grupe, D., Beuermann, K., Mannheim, K., & Thomas, H.-C., 1998, A&A, 350, 805 177

Grupe, D., Beuermann, K., Thomas, H.-C., Mannheim, K., & Fink, H. H., 1999, A&A, 330, 25

Haas, M., M¨uller, S. A. H., Chini, R., Meisenheimer, K., Klaas, U., Lemke, D., Kreysa, E., & Camenzind, M., 2000, A&A, 354, 453

Haehnelt, M. G., & Rees, M, J. 1993, MNRAS, 263, 168

Haiman, Z., & Loeb, A. 2001, in press

Halpern & Steiner, 1983, ApJ, 269, 37

Hamann, F., 1997, ApJS, 109, 279

Hamann, F., & Ferland, G. 1992, ApJ, 391, L53

Hamann, F., & Ferland, G. 1993, ApJ, 418, 11

Hamann, F., & Ferland, G. 1999 ARA&A, 37, 487

Hamann, F., Korista, K. T., Ferland, G. J., Warner, G., & Baldwin, J., 2002, ApJ, 564, 592

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

Henry, J. P., et al. 1994, AJ, 107, 127

Hill, G. J., Goodrich R. W., & DePoy, D. L., 1996, ApJ, 462, 163

Ho, L. C., Filippenko, A. V., & Sargent, W. L. W., 1993, ApJ, 417, 63

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

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

Ho, L. C., Filippenko, A. V., & Sargent, W. L. W., 2003, ApJ, 583, 159

Ho, L. C., Feigelson, E., D., Townsley, L. K., Sambruna, R. M., Garmire, G. P., Brandt, W. N., Filippenko, A. V., Griﬃths, R. E., Ptak, A. F., & Sargent, Wallace L. W. 2001, ApJ, 549, L51 178

Hopkins et al., 2004, AJ, in press (astro-ph/0406293)

Hughes, M. A., et al. 2003, AJ, 126, 742

Irwin, M., McMahon, R. G., & Hazard, C. 1991, in ASP Conf. Ser. 21, The Space Distribution of Quasars, ed. D. Crampton (San Francisco: ASP), 117

Ivezic, Z., et al., 2002, AJ, 124, 2364

Isaak K. G., McMahon, R. G., Hills, R. E., & Withington, S. 1994, MNRAS, 269, 28

Katz, N., Quinn, T., Bertschinger, E., & Gelb, J. M. 1994, MNRAS, 270, L71

Kauﬀmann, G., & Haehnelt, M. 2000, MNRAS, 311, 576

Kenneﬁck, J. D., et al. 1995a, AJ, 110, 78

Kenneﬁck, J. D., Djorgovski, S. G., & de Carvalho, R. R., 1995b, AJ, 110, 2553

Kenneﬁck, J. D., Djorgovski, S. G., & Meylan, G., 1996, AJ, 111, 1816

Kennicutt R. C., Jr., Keel, W. C., Blaha, C. A., 1989, AJ, 97, 1022

Komossa, S., & Greiner, J., 1999, in ASP Conf. Ser. 161, High Energy Processes in Accreting Black Holes, ed. J. Poutanen & R. Svensson (San Francisco: ASP), 228

Korista, K., Baldwin, J., & Ferland, G. 1998, ApJ, 507, 24

Kormendy et al., 1996, ApJ, 473, L91

Kraemer, S. B., George, I. M., Turner, T. J., & Crenshaw, D. M., 2000, ApJ, 535, 53

Kriss, G. A., 1988, ApJ, 324, 809

Kriss G., 1994, in ASP Conf. Ser. 61, Astronomical Data Analysis Software and Systems III, ed. Crabtree, D. R., Hanisch, R.J., & Barnes, J., 437

Kuhn, O., Elvis, M., Bechtold, J., & Elston, R., 2001, ApJS, 136, 225

Kuraszkiewicz, J., & Wilkes, B. J., 2000, ApJ, 542, 692 179

Laor, A. & Draine, B. T., 1993, ApJ, 402, 441

Laor, A., Jannuzi, B. T., Green, R. F., & Boroson, T. A., 1997, ApJ, 489, 656

Laor, A. 1998, ApJ, 505L, 83L

Laor, A., 2000, NewA Rev., 44, 503

Lawrence, A., 1991, MNRAS, 252, 586

Leighly, K. M., 1999, ApJS, 125, 317

Leighly, K. M., 2001, in ASP Conf. Ser. 224, Probing the Physics of Active Galactic Nuclei, ed. Peterson, B. M., Pogge, R. W., & Polidan, R. S. (San Francisco: ASP), 293

Lynden-Bell, D., 1969, Nature, 223, 690

Magorrian, J., et al. 1998, AJ, 115, 2285

Maiolino, R., et al., 2000, A&A, 355, 47

Malhotra, S., 1994, ApJ, 433, 687

Malhotra, S., 1995, ApJ, 448, 138

Maoz, Dan, Filippenko, A. V., Ho, L. C., Rix, H.-W., Bahcall, J. N., Schneider, D. P., Macchetto, F. D., 1995, ApJ, 440, 115

Mathur S. 2000, MNRAS, 314, 17

McIntosh, D. H., Rix, H.-W., Rieke, M. J., & Foltz, C. B. 1999, ApJ, 517, 73

Matt G., Fabian, A. C., Guainazzi, M., Iwasawa, K., Bassani, L., & Malaguti, G., 2000, MNRAS, 318, 173

Miller, J. S., Goodrich, R. W., & Mathews, W. G., 1991, ApJ, 378, 47

Miyaji, T., Ishisaki, Y., Ogasaka, Y., Ueda, Y., Freyberg, M. J., Hasinger, G., & Tanaka, Y., 1998, A&A, 334, 13 180

Moran, E. C., Halpern, J. P., & Helfand, D. J., 1996, ApJS, 106, 341

Murray, N., & Chiang, J., 1998, ApJ, 494, 125

Mushotzky, R. F., & Wandel, A., 1989, ApJ, 399, 674

Nagao, T., Murayama, T., Shioya, Y., & Taniguchi, Y., 2002, ApJ, 575, 721

Netzer, H. 1990, in Active Galactic Nuclei, ed. T. J.-L. Courvoisier, & M. Mayor (Berlin: Springer), 57

Netzer, H. & Laor, A., 1993, ApJ, 404, 51

Netzer, H., & Penston, M. V. 1976, MNRAS, 174, 319

Neugebauer G., Green, R. F., Matthews, K., Schmidt, M., Soifer, B. T., and Bennett, J., 1987, ApJS, 63, 615

Nicholson, K. L., Reichert, G. A., Mason, K. O., Puchnarewicz, E. M., Ho, L. C., Shields, J. C., & Filippenko, A. V., 1998, MNRAS, 300, 893

O’Brian, P. T., Gondhalekar, P. M., & Wilson, R., 1988, MNRAS, 233, 801

Oke, J. B., & Lauer, T. R., 1979 ApJ, 230, 360

Oke, J. B., et al. 1995, PASP, 107, 375

Osmer, P. S., Porter, A. C., & Green, R.F., 1994, ApJ, 436, 678

Osmer, P. S., & Shields, J. C. 1999, ASP Conf. Ser.162, Quasars and Cosmology, ed. G. Ferland, & J. Baldwin (San Francisco: ASP), 235

Osterbrock, D. E., 1989, Astrophysics of Gaseus Nebulae and Active Galactic Nuclei, University Science Books

Osterbrock, D. E. & Pogge, R. W., 1985, ApJ, 297, 166

Osterbrock, D. E. & Shaw, R. A., 1988, ApJ, 327, 89 181

Osterbrock, D. E. & Martel, A., 1993, ApJ, 414, 552

Penston, M. V., Fosbury, R. A. E., Boksenberg, A., Ward, M. J., & Wilson, A. S., 1984, MNRAS, 208, 347

Peterson, B. M., 1997, An Introduction to Active Galactic Nuclei, Cambridge Univer- sity Press

Piccinotti, G., Mushotzky, R. F., Boldt, E. A., Holt, S. S., Marshall, F. E., Serlemit- sos, P. J., & Shafer, R. A., 1982, ApJ, 253, 485

Piconcelli, E., Cappi, M., Bassani, L., Fiore, F., Di Cocco, G., & Stephen, J. B., 2002, A&A, 394, 835

Polletta, M., Courvoisier, T. J.-L., Hooper, E. J., & Wilkes, B. J., 2000, A&A, 362, 75

Pogge, R. W., 1989, ApJS, 71, 433

Pogge, R. W., Maoz, D., Ho, L. C., & Eracleous, M., 2000, ApJ, 532, 323

Pounds, K., Done, C., & Osbourne, J., 1995, MNRAS, 277, 5

Prevot, M. L., Lequeux, J., Prevot, L., Maurice, E., & Rocca-Volmerange, B., 1984, A&A, 132, 389

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

Ptak, a., Terashima, Y., Ho, L.C., & Quataert, E., 2004, ApJ, 606, 173

Puchnarewicz, E. M., Mason, K. O., Siemiginowska, A., & Pounds, K. A., 1995, MNRAS, 276, 20

Puschell, J. J., 1981, AJ, 86, 16

Quataert, E., di Matteo, T., Narayan, R., Ho, L. C., 1999, ApJ, 525, 89L

Richstone, D. O., et al., 1998, Nature, 395, 14 182

Rieke, G. H., 1978, ApJ, 226, 550

Rieke, G. H., & Lebofsky, M. J., 1979, ARA&A, 17, 477

Rodriguez-Pascual, P. M., Mas-Hesse, J. M., & Santos-Lleo, M., 1997, A&A, 327, 72

Rodriguez-Pascual, P. M., Pastoriza, M. G., & Donzelli, C. J., 2000, ApJS, 126, 63

Sanders, D. B., Phinney, E. S., Neugebauer, G., Soifer, B. T., & Matthews, K., 1989, ApJ, 347, 29

Sarzi et al., 2004, ApJ, submitted

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

Schmidt, M., & Green, R. F., 1983, ApJ, 269, 352

Schmidt, G. D., Weymann, R. J., & Foltz, C. B., 1989, PASP, 101, 713

Schmidt, M., Schneider, D. P., & Gunn, J. E., 1987, ApJ, 316, L1

Schmidt, M., Schneider, D. P., & Gunn, J. E., 1995, AJ, 110, 68S

Schneider, D. P., Schmidt, M., & Gunn, J. E., 1989, AJ, 98, 1507

Schneider, D. P., Schmidt, M., & Gunn, J. E., 1991, AJ, 101, 2004

Schneider, D. P., Schmidt, M., & Gunn, J. E., 1997, AJ, 114, 36

Schneider, D. P., et al., 2001, AJ, 121, 1232S

Sch¨odel, R., Ott, T., Genzel, R., Eckart, A., Mouawad, N., Alexander, T., 2003, ApJ, 596, 1015

Seaton, M. J., 1979, MNRAS, 187, 73

Shang Z., Wills, B. J., Robinson, E. L., Wills, D., Laor, A., Xie, B., & Yuan, J., 2003, ApJ, in press

Shemmer, O., & Netzer, H. 2002, ApJ, 567, 19 183

Shields, J. C., 1992, ApJ, 399, 27L

Shields, J. C., & Hamann, F., 1997, RevMexAA (Serie de Conf.), 6, 221

Shields, J. C., et al., 2004a, in Carnegie Observatories As- trophysics Series, Vol. 1: Coevolution of Black Holes and Galaxies, ed. L. C. Ho (Pasadena: Carnegie Observatories) (http://www.ociw.edu/ociw/symposia/series/symposium1/proceedings.html)

Shields, J. C., et al., 2004b, ApJ, submitted

Simpson, C., 2003, New Astron. Rev., 47, 211

Smith, B. J., Harvey, P. M., Colome, C., Zhang, C. Y., Difrancesco, J., & Pogge, R. W., 1994, ApJ, 425, 91

Smith, J. D., Thompson, D., & Djorgovski, S. G., 1994a, AJ, 107, 24

Smith, J. D., et al., 1994b, AJ, 108, 1147

Stein, W. A., & Soifer, B. T., 1983, ARA&A, 21, 177

Stirpe, G. M., 1990, A&AS, 85, 1049

Storchi-Bergmann, T., 1991, MNRAS, 249, 404

Storri-Lombardi, L. J., McMahon, R. G., Irwin, M. J., & Hazard, C., 1996, ApJ, 468, 121S

Storri-Lombardi, L. J., Irwin, M. J., McMahon, R. G., & Hook, I. M., 2001, MNRAS, 322, 933S

Sulentic, J. W., Marziani, P., Zamanov, R., Bachev, R., Calvani, M., & Dultzin- Hacyan, D., 2002, ApJ, 566, 71

Taniguchi, Y., Shioya, Y., & Murayama, T., 2000, AJ, 120, 1265

Telfer, R. C., Zheng, W., Kriss, G. A., & Davidsen, A. F., 2002, ApJ, 565, 773 184

Turner, E. L., 1991, AJ, 101, 1

Tytler, D., & Fan, X. M. 1992, ApJS, 79, 1

Ueda, Y., Akiyama, M., Ohta, K., & Miyaji, T., 2003, ApJ, 598, 886

van Groningen, E., 1993, A&A, 272, 25

Vanden Berk, D. E., et al., 2001, AJ, 122, 549

Veilleux, S., & Osterbrock, D. E., 1987, ApJS, 63, 295

Verner, D. A., Verner, E. M., & Ferland, G. J., 1996, Atomic Data Nucl. Data Tables, 64, 1

V´eron, P., Gon¸calves, A. C., & V´eron-Cetty, M.-P., A&A, 319, 52

V´eron-Cetty, M. -P., V´eron, P., & Gon¸calves, A. C., 2001, A&A, 372, 730

Vestergaard, M., & Wilkes, B. J., 2001, ApJS, 134, 1

Wandel, A., 1987, ApJ, 316, 55

Wang, T., Brinkmann, W., & Bergeron, J., 1996, A&A, 309, 81

Wills, B. J., Brotherton, M. S., Fang, D., Steidel, C. C., & Sargent, W. L. W., 1993, ApJ, 415, 563

Wills, B. J., Brotherton, M. S., Laor, A., Wills, D., Wilkes, B. J., Ferland, G., & Shang, Z., 1999, in ASP Conf. Ser. 162, Quasars and Cosmology, ed. G. Ferland & J. Baldwin, (San Francisco: ASP), 161

Wills, B. J., Shang, Z., & Yuan J. M., 2000, New Astronomy Reviews, 44, 511

Winkler, H., 1992, MNRAS, 257, 677

Winkler, H., Stirpe, G. M., & Sekiguchi, K., 1992, A&AS, 94, 103

Zheng W., & Malkan, M. A., 1993, ApJ, 415, 517 185

Zheng, W., Kriss, G. A., Telfer, R. C., Grimes, J. P., & Davidsen, A. F., 1997, AJ, 475, 469