The H I Cloud Population in the Lower Halo of the Milky Way. Heather Alyson Ford

Transcription

1 The H I Cloud Population in the Lower Halo of the Milky Way Heather Alyson Ford A dissertation presented in fulfillment of the requirements for the degree of Doctor of Philosophy at Swinburne University Of Technology July 2010

2

3 Abstract To constrain the physical properties and distribution of the population of H I clouds in the lower halo of the Milky Way galaxy, and to provide insight into their origin and nature, the Galactic All-Sky Survey was performed and used to identify and measure properties of H I halo clouds. Two regions of the Galaxy that are symmetricallylocated with respect to the line of sight to the Galactic centre were searched for H I halo clouds located near tangent points. The number of clouds detected within these regions are strikingly different. The physical properties of the clouds are similar, however, suggesting that they belong to the same population and may have originated from similar environments. The cloud-to-cloud velocity dispersions are also similar, despite a factor of two difference in their vertical scale heights. This suggests that the kinematics of the clouds are driven by the same physical processes in each quadrant and that the cloud-to-cloud velocity dispersions are not responsible for the heights the clouds reach. This large, homogeneously selected sample of halo clouds has allowed their spatial distribution to be determined for the first time and has revealed that they are strongly correlated with the spiral structure of the Galaxy. We propose a scenario where the H I halo clouds are related to areas of star formation in the form of superbubbles and gas that has been swept into the halo due to stellar feedback. This proposal was tested by performing three-dimensional hydrodynamic simulations of a superbubble within a realistic clumpy medium. The simulations revealed that H I clouds may form in the disk-halo interface due to the evolution of a superbubble, in the form of disk gas that has been swept into the halo from the walls of chimneys. The resulting clouds have diameters and heights similar to those of observed clouds, but are less dense. The large number of clouds detected in both regions suggest that the clouds are a major component of the Galaxy and would likely be detected throughout it. The H I halo clouds therefore play an important role in the circulation of gas between the disk and halo, and these features likely exist in many external galaxies. iii

4

5 Acknowledgments I have been very fortunate to have had the opportunity to closely collaborate with some of the best researchers in the field of Galactic H I throughout my candidature. The bulk of my supervision was provided by my co-supervisors, Naomi McClure- Griffiths and Jay Lockman, whose guidance was imperative for bringing this thesis to fruition. Their enthusiasm for research, and life in general, made them both a pleasure to work with. Naomi provided advice and assistance throughout my entire candidature, from introducing observations at radio wavelengths and standard software, to providing advice on preparing a presentation, to discussing research ideas, she patiently answered all of my questions and was always easy to relate to. Throughout the latter half of my candidature, Jay was consistently available to answer my many questions and provided excellent advice on how to best interpret results, which continuously helped to hone my research abilities. For these things I am grateful, as well as the useful feedback they provided on drafts of this thesis. It is important to have a high quality mentor, and I was fortunate enough to have two. I have had a succession of supervisors at Swinburne University who, although I have not worked closely with, have all at some point provided guidance for which I am thankful: Brad Gibson, James Murray, Steven Tingay, and Matthew Bailes. I particularly thank Matthew for providing feedback on the chapters of this thesis. Many people have contributed to the collection and reduction of the Galactic All- Sky Survey (GASS). Without these data I could not have performed the analysis presented in this thesis, and I therefore thank all members of the GASS team for their effort and dedication: Naomi McClure-Griffiths, D. J. Pisano, Mark Calabretta, Jay Lockman, Lister Staveley-Smith, Peter Kalberla, Jeremy Bailin, Leonidas Dedes, S. Janowiecki, Brad Gibson, Tara Murphy, H. Nakanishi, and Katherine Newton- McGee. I also thank the staff at Parkes who not only responded quickly when problems arose, but constantly monitored observations to ensure the best observing v

6 experience possible. I am grateful to Ralph Sutherland, who not only made his hydrodynamic model available to me before its release, but also taught me the ropes of performing simulations. I also thank Jarrod Hurley and Craig West for their timely responses to all computing issues I encountered. Much of the work presented in this thesis has been conducted outside of Swinburne University. Many visits to the ATNF sites at Marsfield, Parkes, and Narrabri, as well as visits to the NRAO site in Green Bank were made, and I thank all of the staff at these locations for providing a welcoming environment. I also thank Hugh Couchman and members of the Astronomy group at McMaster University for kindly including me in department activities and providing me with a desk and computing facilities while I finished my thesis from Canada. I thank the following people for useful conversations pertaining to my thesis, stress relief over a pint, support and encouragement, or providing helpful data products: Sharon Bailin, Mike Brown, Bob Bryan, David Champion, Tim Connors, Jo Dawson, Chris Fluke, Jimi Green, George Hobbs, Adam Horvath, Jedd Horvath, Kati Horvath, Annie Hughes, Buell Jannuzi, Peter Kalberla, Mike Keith, Katie Kern, Virginia Kilborn, John Landstreet, Emil Lenc, Glen Mackie, Jenn Morrison, Jenny Peckham, D. J. Pisano, Chris Power, Bill Saxton, Bruce Shawyer, Phil Stooke, Simon Strasser, Chris Thom, Kathryn Vandenberg, and Joris Verbiest. Lastly, I acknowledge my family, who were supportive, encouraging, and patient, especially when I did not have the time to call or visit. In particular, I thank my parents, Ray and Berdina Ford, my brother, Adam Ford, and my grandmother, Jessie Toope. I owe the most gratitude to my husband, Jeremy Bailin, and our son, Heath Bailin. Jeremy provided endless support, either through words of encouragement, by manning the fort while I travelled for work, or by helping during observing runs when we were short on observers or with whatever else came up. More recently he jumped into the role of father with flying colours, doing much more than his share of diaper changes and loads of laundry, spending time with Heath between feedings, and working late at night when we were all asleep, so that I could work as uninterrupted as possible throughout the day. It was these things, and Heath s calm disposition, willingness to sleep throughout the night, and captivating smiles, that enabled me to focus and collate my work into this document.

7 Declaration This thesis contains no material which has been accepted for the award of any other degree or diploma. To the best of my knowledge, this thesis contains no material previously published or written by another person, except where due reference is made in the text of the thesis. All work presented is primarily that of the author. Chapter 3 was published in the Astrophysical Journal, as noted at the start of the text. I authored the majority of the text, receiving limited assistance to refine the text from the coauthors and an anonymous referee. Only minor changes have been made from the published version to avoid repetition and to conform to the structure of the thesis. Naomi McClure-Griffiths, Jay Lockman, Matthew Bailes, and Jeremy Bailin provided comments to refine each chapter of this thesis. This thesis analyses data recorded at the Parkes radio telescope for project P467 during a two year period. Planning, observing and data reduction for P467 were performed by a team of astronomers including myself, Naomi McClure-Griffiths, D. J. Pisano, Mark Calabretta, Jay Lockman, Lister Staveley-Smith, Peter Kalberla, Jeremy Bailin, Leonidas Dedes, S. Janowiecki, Brad Gibson, Tara Murphy, H. Nakanishi, and Katherine Newton-McGee. The hydrodynamic model used in Chapter 6 was written and developed by Ralph Sutherland. I performed all initial condition testing and parameter setup after initial assistance from Dr. Sutherland. Heather Alyson Ford, July 2010 vii

8

9 Contents 1 Introduction Motivation General Structure of the Galaxy The Interstellar Medium Galactic Neutral Hydrogen and H I Clouds H I Clouds and Disk-Halo Interactions Thesis Outline Single-Dish Radio Astronomy and the Galactic All-Sky Survey A Brief History of Radio Astronomy Fundamentals of Single-Dish Radio Astronomy Basic Theory Instrumentation and Observing Fundamentals Galactic Neutral Hydrogen Surveys The 21 cm Neutral Hydrogen Line Galactic H I Surveys The Galactic All-Sky Survey HI Clouds in the Lower Halo: The GASS Pilot Region Introduction The GASS Pilot Region H I Clouds in the GASS Pilot Region Search Method and Criteria Properties of the Entire Cloud Sample Uncertainties in Observed Properties Overview of Observed Properties Selection of a set of Tangent Point Clouds ix

10 3.4.1 Derived Properties Uncertainties in Derived Properties Analysis of the Tangent Point Cloud Population Simulated Halo Cloud Population Cloud-to-Cloud Velocity Dispersion Distance Errors Radial Distribution Vertical Distribution Physical Size and Mass Comparison of GASS Clouds to Lockman Clouds Observed Trends The Origin and Nature of Halo Clouds Kinematics of Halo Cloud Population Halo Clouds, Spiral Structure, and Star Formation Possible Association with High Velocity Cloud Complex L Stability of Halo Clouds Summary HI Clouds in the Lower Halo: Quadrant I Introduction The Quadrant I Region Observed Properties of H I Clouds Derived Properties of H I Clouds Analysis of Quadrant I Cloud Population Simulated Population of Clouds Cloud-to-Cloud Velocity Dispersion Radial Distribution Vertical Distribution Physical Properties and Trends Physical Properties: Quadrant I vs. IV Trends: Quadrant I vs. IV Summary HI Halo Clouds: A Galactic Population Introduction

11 5.2 Comparison of Cloud Properties Cloud-to-Cloud Velocity Dispersion Vertical Distribution of H I Clouds Basic Comparison Comparison with Other Galactic Components Implications for Cloud Evolution The Relationship between h and σ cc Fraction of Galactic H I in Halo Clouds Derived Radial Surface Density Distribution Relation to Spiral Features and Star Formation H I Clouds and Galactic Structure H I Clouds and H II Regions H I Clouds and Methanol Masers H I Clouds and Molecular Gas H I Clouds and Shells The Origin and Nature of H I Halo Clouds Summary Superbubble Simulations and the Production of H I Halo Clouds Introduction The Simulations A Brief Introduction to Supercomputing The Hydrodynamic Model The Stages of Superbubble Evolution Production of H I Clouds in the Lower Halo Implications for the Origin of H I Halo Clouds Additional Physics Summary Conclusions and Future Work Conclusions Future Work References 198 Publications 209

12

13 List of Figures 1.1 Views of the Milky Way Map of the total H I column density of GASS data Area of the Galaxy encompassed by the GASS pilot region Longitude latitude diagram of the GASS pilot region Latitude velocity diagram of the GASS pilot region Spectra from a small sample of H I halo clouds Peak brightness temperature as a function of mean background level V LSR as a function of longitude Latitude distribution with deviation velocity Histogram of peak brightness temperature Histogram of FWHM Histogram of angular size Radial surface density Distribution of deviation velocities of observed and simulated tangent point clouds Relative distance error as a function of deviation velocity Radial distribution of observed and simulated tangent point clouds Histogram of the longitude distribution of tangent point clouds Histogram of the longitude distribution of entire cloud sample Vertical distribution of observed and simulated tangent point clouds Histogram of physical radius of tangent point clouds Histogram of H I mass of tangent point clouds FWHM of velocity profile as a function of height Derived H I mass of clouds as a function of cloud radius H I column density as a function of height xiii

14 3.23 H I mass of the clouds as a function of height H I mass of clouds as a function of deviation velocity Filamentary structures within the GASS pilot region Radial surface density distribution of halo clouds and H II regions with mass surface densities of average Galactic H I and H Longitude latitude diagram encompassing upper portion of GASS pilot region and HVC complex L Longitude latitude diagram of quadrant I region Spectra from a small sample of H I halo clouds in quadrant I Longitude versus V LSR of tangent point clouds Histogram of peak brightness temperature Histogram of FWHM Histogram of angular size Histogram of physical radius Histogram of physical mass Radial surface density of tangent point clouds in quadrant I Vertical distance from the Galactic plane as a function of deviation velocity of observed clouds Histogram of deviation velocities for observed and simulated clouds Distribution of Galactocentric radii of tangent point clouds Longitude distribution of observed, simulated and a uniformly distributed population of clouds Vertical distribution of observed and simulated clouds FWHM as a function of height H I column density as a function of height H I mass of clouds as a function of height H I mass of clouds as a function of deviation velocity Deviation velocity distributions of tangent point clouds Derived vertical distributions of H I clouds in quadrants I and IV Mean H I number density due to halo clouds within the quadrant I region Mean H I number density due to halo clouds within the GASS pilot region

15 5.5 Fraction of extended H I contained in clouds as a function of height within the quadrant I region Fraction of extended H I contained in clouds as a function of height within the GASS pilot region Derived radial surface density distributions of H I halo clouds Milky Way stellar spiral structure with quadrant I and GASS pilot regions overlaid Radial surface density distribution of H I clouds and H II regions within the quadrant I region Radial surface density distribution of H I clouds and H II regions within the GASS pilot region Longitude distribution of H I clouds and methanol masers within the quadrant I region Longitude distribution of H I clouds and methanol masers within the GASS pilot region Derived radial surface density distribution of H I clouds and mass surface density of H 2 within the quadrant I region Derived radial surface density distribution of H I clouds and mass surface density of H 2 within the GASS pilot region Zoomed density, temperature and pressure snapshots of superbubble evolution in Run Density, temperature and pressure snapshots demonstrating evolution of gas above disk in Run H I column density of Run 1 at t = 10 Myr H I column density of Run 2 at t = 10 Myr H I column density of Run 3 at t = 10 Myr H I column density of Run 4 at t = 10 Myr Mass of H I above disk as a function of time Mass of H I above disk as a function of height

16

17 List of Tables 2.1 Summary of GASS Properties Observed Properties of H I Clouds in the GASS Pilot Region Derived Properties of Tangent Point H I Clouds in the GASS Pilot Region Property Summary and Comparison to Lockman Clouds Observed Properties of Tangent Point H I Clouds in the Quadrant I Region Derived Properties of Tangent Point H I Clouds in the Quadrant I Region Property Summary of H I Halo Clouds in GASS Summary of H I Cloud Properties in Quadrant I and IV Summary of Expectations in Various Origin Scenarios Summary of Simulation Properties xvii

18

19 Chapter 1 Introduction 1.1 Motivation To understand the origin and evolution of galaxies it is crucial that we understand the processes that shape them. Spiral galaxies are particularly complex systems in which material is continuously circulated between their disks and hot gaseous haloes, primarily due to violent supernova explosions. The Milky Way is the ideal laboratory for studying these processes: our close proximity to the observed features provides resolution that is unattainable for any external galaxy and, from our position within the disk, we look outwards to the halo, resulting in a view of the environment that is not confused by being projected against the disk General Structure of the Galaxy The Milky Way is a barred spiral galaxy with five major components: the disk, bulge, stellar halo, gaseous halo and dark matter halo. The thin disk has a radius kpc and vertical scale height of 300 pc while the thick disk has a vertical scale height of 900 pc (Jurić et al., 2008). The stellar and dark matter haloes are roughly spherically distributed about the centre of the Galaxy, extending to radii 30 kpc (Binney & Tremaine, 2008), while the shape of the gaseous halo is not well known. Although not tightly constrained, the spiral structure of the Milky Way is thought to be composed of two major arms, which extend from each end of the bar, and two minor arms, which are located between the major arms (see Figure 1.1; Spitzer Space Telescope 2008). The major arms are enhancements in the entire disk due to the excitation of spiral density waves (Binney & Tremaine, 2008), while the 1

20 2 CHAPTER 1. INTRODUCTION minor arms are due to the resonant response of the gas to the spiral density pattern, and are therefore only seen in the gas and associated tracers such as star formation regions (Martos et al., 2004). The Sun is located in the plane of the Galaxy at a distance of R =8.4 ± 0.5 kpc from the centre (Ghez et al., 2008; Reid et al., 2009). We adopt R kpc throughout this thesis as recommended by the International Astronomy Union (IAU; Kerr & Lynden-Bell 1986) The Interstellar Medium The stars of the Galaxy are surrounded by a medium composed of gas, dust, magnetic fields and cosmic rays. These components are dynamically linked and this medium as a whole is known as the interstellar medium (ISM; Binney & Merrifield 1998). Interstellar matter accounts for 10 15% of the total mass of the Galactic disk and is concentrated near the plane and along the spiral arms (Ferrière, 2001). The gas of the ISM can be categorised into five major constituents: 1. Molecular Gas: Molecular gas is extremely cold and dense, with temperatures, T K and number densities, n cm 3, and is concentrated in the plane of the Galaxy (Ferrière, 2001). Molecular hydrogen (H 2 ) is the main molecular component but because it is difficult to observe at any wavelength, this gas is usually traced by CO observations (e.g., see Dame et al. 2001). Clouds of molecular gas are the environment in which stars form. 2. Cold Neutral Medium (CNM): Usually probed by neutral hydrogen (H I) emission and absorption observations, this gas is arranged in sheets, filaments, and other morphologies and is generally confined to the disk of the Galaxy (Heiles & Troland, 2003). It has T K, n cm 3, and a filling factor, f = 0.01, where f represents the fraction of the volume occupied (Heiles, 2001). 3. Warm Neutral Medium (WNM): Also probed by H I emission is the WNM, which has T K, n cm 3 and f 0.5 (Heiles, 2001). While this gas appears to be mostly in the form of sheets and filaments that encompass the denser CNM sheets and filaments (Heiles & Troland, 2003), warm, semi-spherical structures, which we refer to as clouds throughout this thesis, have also been detected in the lower halo of the Galaxy (Lockman, 2002).

21 1.1. MOTIVATION 3 Figure 1.1 Views of the Milky Way. The face-on view of the Milky Way is an artist s conception determined from Galactic Legacy Infrared Mid-Plane Survey Extraordinaire data (top; from NASA/JPL-Caltech/R. Hurt SSC-Caltech) and the edge-on view is the Cosmic Background Explorer/Diffuse Infrared Background Experiment all-sky near-infrared map (bottom; Hauser et al. 1995). The spiral structure is clearly visible in the face-on view while the disk and bulge can be seen in the edge-on view.

22 4 CHAPTER 1. INTRODUCTION 4. Warm Ionised Medium (WIM): The WIM is also known as the Reynold s Layer, and has T 8000 K, n 0.08 cm 3 and a f 0.1 (Heiles, 2001). Because this gas has been ionised by stellar UV photons, it can be probed through the detection of Hα emission. Diffuse Hα has a scale height up to 1 2 kpc (Haffner et al., 1999; Gaensler et al., 2008). 5. Hot Ionised Medium (HIM): The HIM is produced by supernova shocks and is found in supernova remnants and superbubbles. This hot gas (T 10 6 K) rises from the disk into the halo and is detected through X-ray emission and UV absorption. Observations indicate that this gas extends to heights of a few kpc, forming the major component of the gaseous halo (Savage, 1995; Ferrière, 2001), and has an n cm 3 and f 0.5. The latter four components of the ISM are approximately in pressure equilibrium Galactic Neutral Hydrogen and H I Clouds Hydrogen is the most abundant element in the Universe, comprising 75% by mass of all baryonic matter. Neutral atomic hydrogen is ubiquitous throughout the Galaxy with a wide variety of morphologies and kinematics, and exhibits many complex structures, including worms (Koo et al., 1992), sheets and filaments (Heiles, 1967; Dickey & Lockman, 1990), shells (Heiles, 1979; McClure-Griffiths et al., 2002), and clouds (Lockman, 2002). The Galactic H I disk extends to Galactocentric radii, R 30 kpc and its thickness varies from a full-width at half-maximum (FWHM) 100 pc inside R = 3.5 kpc to 3 kpc in the outer Galaxy, with a roughly uniform FWHM of 230 pc between 3.5 kpc and the solar circle (see Ferrière 2001 and references therein). H I is also known to extend far beyond the thin H I disk as a layer into the Galactic halo (Shane, 1967; Lockman, 1984). Early low-resolution observations of the disk-halo interface revealed a handful of H I clouds within this layer that appear to be connected to the Galactic disk (Prata, 1964; Simonson, 1971). More recent high-resolution observations from the Green Bank Telescope (GBT) have revealed that these clouds are more plentiful and the layer is not smooth but is instead composed of a population of co-rotating, discrete H I clouds, with sizes 30 pc and masses 50 M (Lockman, 2002). Confusion may limit the detectability of such clouds at low heights, in which case they may not be confined to the halo (see, e.g., Stil et al. 2006). These clouds follow Galactic rotation

23 1.1. MOTIVATION 5 and are discrete clumps of H I that are localised in space and velocity. Although they are sometimes related to larger structures, for example, embedded in filaments, each cloud appears to be a distinct object. The gross properties of the Milky Way s thick H I layer may in fact be a consequence of the statistical properties of these H I clouds if they constitute a large enough population. Both extraplanar H I and dust have been observed in several nearby spiral galaxies, reaching distances of several kpc from the plane (e.g., Dettmar 1990; Swaters et al. 1997; Howk & Savage 1999; Fraternali et al. 2002; Barbieri et al. 2005). The thick H I disks in these galaxies, once observed with sufficient resolution, may reveal a similar clumpy structure as that within the disk-halo interface of the Milky Way. The presence of H I clouds at heights far from the disk, where the gas should settle due to the flattened potential, is a mystery. If the clouds originated external to the Galaxy it would be easy to understand their large heights, but their corotation clearly indicates that they are related to processes occurring within the disk H I Clouds and Disk-Halo Interactions While the origin of the halo clouds is unknown, one possible explanation is a galactic fountain model, in which hot gas produced by supernovae rises into the halo of the Galaxy and cools and condenses into H I clouds, which then fall back to the plane (Shapiro & Field, 1976; Bregman, 1980). Houck & Bregman (1990) predicted that, from a low temperature fountain, neutral gas in Galactic rotation could be formed at heights close to the plane. This scenario is supported by observations of intermediate velocity clouds (IVCs), such as those of cloud g1, whose location, kinematics and abundances match those expected (Wakker et al., 2008). The abundances in the IV Arch (Richter et al., 2001b) and LLIV Arch (Richter et al., 2001a) also suggest a fountain origin; as their abundances are near solar it is likely that they originated from material enriched from the disk. The location of these IVCs also support this origin, as they are roughly 1 kpc from the disk. Another possibility is that the halo clouds originate in environments in which supernovae and stellar winds disrupt the surrounding medium. Such events can result in the formation of a bubble, and models suggest that with a large enough energy source these bubbles may expand beyond the thickness of the Galactic disk (e.g., Tomisaka & Ikeuchi 1988; Heiles 1990). The bubbles are encompassed by an H I shell as a result of either radiative cooling, which then accumulates more H I

24 6 CHAPTER 1. INTRODUCTION as the bubble continues to expand, or solely the sweeping up of ambient material, depending on the wind speed (Koo & McKee, 1992). The shell remains as a single entity until in some cases Rayleigh-Taylor instabilities cause it to fragment (Mac Low et al., 1989). Once the top of the shell has fragmented, gas that has been shock heated is expelled outwards, mixing material from the disk with that in the halo (Norman & Ikeuchi, 1989) and the remaining fragments of the H I shell may be the observed halo clouds (McClure-Griffiths et al., 2006). It is also possible for the hot gas that has been expelled to cool and recombine, as seen in models of de Avillez (2000), or it may be that energy from increased supernova activity in areas of active star formation has simply pushed disk gas into the halo, forming clumps of H I. Large samples of halo clouds are required to constrain the properties and distribution of the population and provide insight into their origin. As the Lockman (2002) sample has only 38 clouds in a small region of the Galaxy, the properties of the population that constitute the H I layer are currently not well determined. A thorough understanding of the clouds, their physical nature, and their role in the Galaxy is therefore crucial for understanding the circulation of material between the Galactic disk and halo, a critical process in the evolution of galaxies. 1.2 Thesis Outline In this thesis, data from the recently completed Galactic All-Sky Survey (GASS; McClure-Griffiths et al. 2009) are used to identify and measure properties of H I halo clouds to place strong constraints on their origin, nature, and role in the Galaxy, along with implications for the evolution of gas in galaxies. An outline of this thesis is as follows: Chapter 2: The basics of single-dish radio astronomy, including a brief history, fundamental theory, and instrumentation are presented. Details of the 21 cm H I line and Galactic H I surveys, along with a description of GASS, an H I survey that was conducted throughout the duration of this thesis and whose data provide the basis for the majority of this thesis, are also presented. Chapter 3: Results from the GASS pilot region, the initial test region in the fourth Galactic quadrant for the detection and analysis of H I halo clouds, are presented. These include a catalogue of clouds in the GASS pilot region and a

25 1.2. THESIS OUTLINE 7 statistical analysis of their properties and distribution. Implications for the origin of the clouds are briefly discussed. Chapter 4: A complementary region of the Galaxy within the first quadrant was also searched for H I halo clouds and the resulting tangent point cloud sample is presented, together with a discussion of the clouds properties and distribution. Chapter 5: An in-depth comparison between the tangent point cloud samples of the first and fourth quadrant regions is presented, along with a comparison between the clouds distribution and that of Galactic structures and components. Implications for the origin and evolution of the H I halo clouds and their role in the Galaxy are discussed. Chapter 6: Three-dimensional (3D) hydrodynamic simulations of an expanding superbubble in a clumpy medium are presented to test the hypothesis that H I clouds in the lower halo can form as a result of this process. Simulated H I structure is compared to the properties and distribution of the observed clouds. Chapter 7: The major conclusions obtained within this thesis are presented, as well as a description of future work to expand on the results.

26

27 Chapter 2 Single-Dish Radio Astronomy and the Galactic All-Sky Survey Different astrophysical objects emit preferentially in different parts of the electromagnetic spectrum. It is therefore important to observe over a variety of wavelengths to get as complete a picture as possible. Radio wavelengths are a crucial portion of this spectrum spanning roughly 15 MHz to 1300 GHz; such diverse but important sources as molecular gas, neutral hydrogen, pulsars and jets can be observed best at these wavelengths. Due to the numerous radiation mechanisms, the range of surface brightnesses, and the different angular scales at which objects exhibit structure within this portion of the electromagnetic spectrum, many different instruments and observing techniques are required. This chapter focuses on single-dish radio astronomy as the bulk of data presented in this thesis are from the Galactic All-Sky Survey (GASS), the highest angular resolution, highest spectral resolution, complete survey of neutral hydrogen within the Milky Way that has been conducted using the Parkes 64 m single-dish radio telescope. In addition to a description of GASS, a brief history of radio astronomy, basic theory and instrumentation of single-dish telescopes, and observations of Galactic H I are reviewed. 2.1 A Brief History of Radio Astronomy The initial radio detection of an astronomical object was made serendipitously in 1931 by Karl Jansky, an engineer who was asked by Bell Laboratories to determine 9

28 10 CHAPTER 2. SINGLE-DISH RADIO ASTRONOMY AND GASS the source of static that was interfering with a newly implemented transatlantic telephone service. Using an antenna that he built to monitor the radiation at a wavelength of 14.6 m, he determined that while thunderstorms were partially responsible for the interference, there was also a steady hiss that did not occur at the same time every day but instead was linked to the sidereal day. As a result, he determined that the source must be extraterrestrial (Jansky, 1933). Radio engineer Grote Reber was fascinated by this result and subsequently built an antenna in his backyard to study this emission. After several failed attempts, in 1938 he finally confirmed Jansky s result by observing the Milky Way galaxy at three different wavelengths making detections only at 1.87 m. As the flux was not proportional to the square of the frequency, ν, he showed that the emission could not be thermal (Reber, 1940). This advancement in the field of radio astronomy was interrupted by World War II, but the development of radio and radar technology during this war proved to be a boon to radio astronomy, and led to its rapid rise over the subsequent years (Emberson, 1959; Rohlfs & Wilson, 2004). 2.2 Fundamentals of Single-Dish Radio Astronomy The fundamental equations describing radio wavelength detections along with a brief overview of relevant instrumentation are summarised here. For a complete overview of this topic see Rohlfs & Wilson (2004) Basic Theory When a wavelength of electromagnetic radiation is much smaller than the distance traveled, which is always true in astronomy, we can assume that it is composed of rays that travel in straight lines. We can therefore write the power transmitted by the light, dw, as dw = I ν cos θdωdσdν, (2.1) where dw is the infinitesimal power, I ν is brightness, θ is the angle between the normal to the emitting surface and direction of propagation, dω is the infinitesimal solid angle, dσ is the infinitesimal surface area, and dν is the infinitesimal bandwidth. The units of I ν are W m 2 Hz 1 sr 1. Equation 2.1 defines the quantity brightness,

29 2.2. FUNDAMENTALS OF SINGLE-DISH RADIO ASTRONOMY 11 I ν : I ν = dw cos θdω dσ dν, (2.2) which is proportional to the number of photons per unit area per unit time per unit solid angle. The total amount of light received from the source is the flux density: S ν = I ν (θ,ϕ ) cos θdω, (2.3) Ω s where the integral is over the angular extent of the source and units are W m 2 Hz 1 or Janskys, where 1 Jansky (Jy) is Wm 2 Hz 1. The brightness is the same at the source and at the detector and is independent of distance, r, while the flux density depends on the distance to the source, with a dependence of 1/r 2. In a vacuum, I ν would remain constant as the ray propagates. Emission, absorption and scattering can increase or decrease the brightness along the line of sight. These changes are represented in the equation of radiative transfer: di ν ds = ɛ ν κ ν I ν, (2.4) where ds is the distance from the observer along the line of sight, ɛ ν is the emissivity and κ ν is the opacity. The emissivity reflects the amount I ν is increased while the opacity reflects the amount it is decreased, due to emission and absorption between the source and observer, and depends on the properties of the medium it passes through. An object that is in thermodynamic equilibrium emits like a blackbody with surface brightness given by the Planck function: B ν (T )= 2hν3 c 2 1 e hν/kt 1, (2.5) where c is the speed of light, h is Planck s constant, k is the Boltzmann constant, and T is the thermal temperature. As hν/kt << 1 (corresponding to ν/ghz << 20.8T/K, i.e., certainly at centimetre wavelengths and beyond, and even at millimetre wavelengths for all but the coldest sources), the Rayleigh-Jeans limit applies, and it follows that e hν/kt 1+ hν kt + B RJ(ν, T )= 2ν2 kt. (2.6) c2

30 12 CHAPTER 2. SINGLE-DISH RADIO ASTRONOMY AND GASS The brightness temperature, T b, of a source is defined as the temperature of a blackbody with the same surface brightness at a given frequency: T b = c2 1 2k ν B 2 ν = λ2 2k B ν. (2.7) Because this limit does usually apply in the radio regime for thermal sources, for which T b is the same as the actual temperature, it is a common convention in radio astronomy to use T b as the unit of brightness, even in non-thermal cases Instrumentation and Observing Fundamentals Different instruments are required to detect different radio signals. For example, wire antennas such as the Low Frequency Array (LOFAR) are used to detect longer wavelengths (λ > 1 m) while reflector antennas are used for shorter wavelengths (λ < 1 m). Both single-dish telescopes (e.g., the Parkes Radio Telescope and the Green Bank Telescope) and multi-element arrays (interferometers; e.g., the Australia Telescope Compact Array and the Very Large Array) are reflector antennas. As the instrumentation of radio telescopes is a very broad topic and the data in this thesis are primarily from the Parkes Radio Telescope, this overview focuses on single-dish telescopes, and in particular the components related to spectral line observing. In a single-dish telescope, the antenna collects electromagnetic fields over the aperture of the dish at the focus, redirecting them to the feed horn where the fields are added. The primary feed collects the waves that converge at the focus and transmits them along a waveguide to the receiver. Radio receivers measure the power contained in the signal as a function of frequency and convert it into voltage. Although more complex than an individual receiver, multibeam receivers are frequently used, where many receivers in the focal plane are combined and simultaneously used, resulting in faster mapping. Finally, the back-end (or correlator) is used to extract the desired properties of the signal, such as continuum flux, spectral line information or pulse shape. The sensitivity, i.e., the faintest level at which a source can be detected, depends on the parameters of the receiver, and is quantified by the system noise temperature, T sys, which is the rms noise in a 1 Hz channel after 1 s of integration for an ideal receiver, and a lower T sys results in better sensitivity. There are many sources of system noise including, but not limited to, the receiver itself, the ground, and

31 2.2. FUNDAMENTALS OF SINGLE-DISH RADIO ASTRONOMY 13 the atmosphere, and the total T sys is the sum of each of these contributions. The minimum noise obtainable in a system is found from the Nyquist sampling theorem, which states that if the maximum frequency present in a signal is ν max, the signal can be reconstructed completely from samples spaced 1/(2ν max ) apart. A consequence of this is that samples must be spaced at least t =1/ ν apart for their values to be completely independent, where ν is the bandwidth. Over a total time, τ, there are therefore ντ independent samples, each of which has noise T sys. Because random noise falls as the inverse square root of the number of independent samples, the total noise, T is limited by: T = T sys ντ. (2.8) The response of an antenna to incoming signals depends on the direction of the incoming electromagnetic waves. This angular sensitivity is known as the power pattern, P (ϑ,ϕ ), and depends on the observing wavelength, λ. Most often this is normalised so that its maximum is unity and is referred to as the normalised power pattern, P n. Typically the power pattern peaks strongly in one direction, the main beam, but contains smaller peaks in other directions known as sidelobes. The beam solid angle,ω A, is the amount of sky being observed integrated over the full sphere and is given by: Ω A = P n (ϑ,ϕ ) dω, (2.9) 4π while the main beam solid angle,ω MB, is given by: Ω MB = P n (ϑ,ϕ ) dω, (2.10) and is integrated only over the main lobe, i.e., sidelobes are not included. It is important for the beam efficiency, η B, to be a large, as it indicates what fraction of the power comes from the main beam: η B = Ω MB Ω A. (2.11) The effective cross-section of an antenna to incident electromagnetic waves is

32 14 CHAPTER 2. SINGLE-DISH RADIO ASTRONOMY AND GASS known as the effective aperture, A e, and is given by: A e = λ2 Ω A. (2.12) When observing a source with brightness distribution on the sky B ν (ϑ,ϕ ), the power delivered by the antenna, W, is: W = 1 2 A e B ν (ϑ,ϕ )P n (ϑ,ϕ ) dω. (2.13) It is convenient to treat the antenna as a resistor at a finite temperature, and replace the power with the antenna temperature, T A : T A = W k. (2.14) The antenna temperature, which is the quantity that is directly measured, is therefore related to the brightness temperature, T b, by: T A (ϑ 0,ϕ 0 )= Tb (ϑ,ϕ )P n (ϑ ϑ 0,ϕ ϕ 0 ) sin ϑ dϑ dϕ. (2.15) Pn (ϑ,ϕ ) dω Thus, it is the brightness temperature convolved with the power pattern. This expression must be deconvolved using the known P n and the measured T A to determine the astrophysical quantity, T b (equivalent to the thermal temperature for the blackbody source in the Rayleigh-Jeans limit, and otherwise simply a convenient expression for brightness). Because the response pattern contains sidelobes, bright sources can contribute to T A even when they do not fall in the main beam. This spurious signal is referred to as stray radiation, and is a particular problem for low surface brightness features near bright sources. The antenna power pattern also defines telescope resolution. This can be quantified by the half power beam width, HPBW, which is simply the angular distance between the points where P n (ϑ,ϕ ) = 1/2. For a uniformly-illuminated circular aperture of diameter, D, the HPBW in radians is given by: HPBW = 1.02 λ D. (2.16) For a more realistic tapered illumination pattern, HPBW is larger than this. Another

33 2.3. GALACTIC NEUTRAL HYDROGEN SURVEYS 15 commonly used definition of resolution is the Rayleigh criterion, Θ, which is the angular distance from the beam peak to the first null (where P n (ϑ,ϕ ) = 0): Θ = 1.22 λ D (2.17) (e.g., Karttunen et al. 1994). 2.3 Galactic Neutral Hydrogen Surveys The 21 cm Neutral Hydrogen Line In a hydrogen atom the interaction between the magnetic moments of the electron and the proton causes a hyperfine splitting of the ground state into two energy levels. The transition between these energy levels results in H I emission at MHz (λ 21 cm; Kulkarni & Heiles 1988). The probability that an electron in the upper state will drop to a lower state is given by the Einstein A coefficient, A 10, and is s 1 for the H I transition (Burton, 1988). Because the chance of any one hydrogen atom dropping to the lower energy state is so low, a large amount of H I is required to make a detection. By time reversal symmetry, the Einstein A coefficient also gives the probability of an atom in the lower state absorbing an H I photon, so neutral hydrogen is an inefficient absorber and H I observations are often optically thin. H I was first detected by Ewen & Purcell (1951) and this detection was confirmed shortly afterwards by Muller & Oort (1951). This line is a sharp feature, making it easy to determine its velocity via the doppler shift, where the relative change in frequency is equal to the velocity along the line of sight relative to the speed of light. The observed velocity includes the motion of the Earth around the Sun, the peculiar motion of the Sun with respect to the average motion of stars around the Sun (the local standard of rest; Delhaye 1965; Gordon 1976), and the motion of stars in the solar neighbourhood around the Galaxy. H I observations are usually corrected for the first two of these motions, resulting in velocities frequently being reported with respect to the local standard of rest, V LSR. All velocities in this thesis are with respect to the local standard of rest unless otherwise specified. In the case where the hydrogen gas is optically thin, the total number of hydrogen

34 16 CHAPTER 2. SINGLE-DISH RADIO ASTRONOMY AND GASS atoms along a given line of sight, the H I column density, N HI, is: N HI = T b dv cm 2, (2.18) where v is velocity in km s 1 and T b is measured in K (Burton, 1988). The total H I mass of an extended object that occupies a solid angle, Ω, at a distance, d, with mean column density N HI is: M HI =M H N HI Ωd 2, (2.19) where M H is the mass of an hydrogen atom Galactic H I Surveys Numerous observations of Galactic neutral hydrogen have been conducted since the initial H I detections, with the earliest observations focusing on the large-scale structure of H I in the Galaxy. Gradually the focus was shifted to smaller-scale features and the general properties of the H I. Eventually, once it was determined that the H I disk extended twice as far as the stellar disk, it became apparent that the rotation of the outer H I disk could be used to measure the gravitational potential of the Galaxy. The focus therefore shifted to using H I measurements to study the dark matter distribution within the Galaxy (see Burton 1988 and references therein). With advancements in instrumentation and observing techniques, it became increasingly possible to observe larger areas of the sky, culminating in complete coverage of the sky. The most recent complete H I surveys of Galactic emission were the Leiden-Dwingeloo Survey (LDS), covering the sky north of δ = 30 (Hartmann, 1994), and the Instituto Argentino de Radioastronomía Survey (IAR), covering the sky south of δ = 25 (Arnal et al., 2000). The LDS has an angular resolution of 36, is undersampled on a 30 grid, has a spectral resolution of 1 km s 1, a velocity coverage of 450 V LSR +400 km s 1, and 70 mk sensitivity, while the IAR survey has 30 resolution, with comparable sampling and sensitivity. The final data release of these combined surveys, which have been corrected for stray radiation, form the Leiden/Argentine/Bonn Survey (LAB; Kalberla et al. 2005). These surveys were conducted using single-dish telescopes as they are best suited for observing large areas due to their ability to detect large-scale structure, their better surface brightness sensitivity for a given collecting area, and their ability to map

35 2.4. THE GALACTIC ALL-SKY SURVEY 17 large areas relatively quickly. Higher resolution H I surveys have been conducted combining single-dish and aperture synthesis telescopes, such as the Canadian and Southern Galactic Plane surveys (Taylor et al., 2003; McClure-Griffiths, 2001), but such observations are much more time consuming and have therefore been restricted to areas within a few degrees of the Galactic plane. Unfortunately all existent Galactic H I surveys either do not have the spatial resolution necessary to distinguish objects such as the clouds detected at the GBT (Lockman, 2002), or do not span large enough areas to cover the disk-halo interface. These structures went undetected in the H I Parkes All-Sky Survey (HIPASS; Meyer et al. 2004) as well because it was incapable of observing Galactic emission properly due to the data gathering and processing methods used. It has therefore been impossible to detect these features in any of the current all-sky survey data. We have performed the Galactic All-Sky Survey (GASS; McClure-Griffiths et al. 2009), an H I survey of the entire Southern sky whose angular resolution enables the detection of these small H I clouds in the disk-halo interface, and provides the first opportunity to perform a census of and study the detailed properties of a large sample of H I clouds in the Galactic halo. 2.4 The Galactic All-Sky Survey Much of the work involved in this thesis consisted of performing and planning many of the observations for and preliminary reduction of the Galactic All-Sky Survey, which was the source of the observational data presented in this thesis. GASS is a Nyquist sampled Galactic H I survey that covers the entire sky south of declination δ = 1. GASS observations were made at the Parkes Radio Telescope over the course of eight observing sessions that were typically two weeks in length, between 2005 January and 2006 November. GASS data were taken with the 21 cm multibeam receiver (Staveley-Smith et al., 1996) and cover 400 V LSR 500 km s 1. The 21 cm multibeam receiver consists of 13 beams arranged in an hexagonal pattern with seven inner beams and six outer beams. Observations were performed via scans in right ascension and declination, covering the entire sky once with each. With the receiver oriented 19.1 with respect to the scan direction, this arrangement results in equally spaced right ascension and declination scans of the inner seven beams. Data were frequency switched between

36 18 CHAPTER 2. SINGLE-DISH RADIO ASTRONOMY AND GASS Table 2.1. Summary of GASS Properties Parameter Value Spatial Coverage δ< 1 Velocity Coverage 400 V LSR 500 km s 1, Spatial Resolution 16 Sensitivity 57 mk Channel Spacing 0.82 km s 1 Channel Width 1 km s 1 and MHz every 5 s. Spectra were obtained using a special-purpose correlator mode that combines the multibeam and wideband correlators and consisted of 2048 channels over 8 MHz of bandwidth. The spectral resolution of GASS data is 1 km s 1, with channel spacings of 0.82 km s 1, and the angular resolution in the final maps is 16. The properties of GASS are summarised in Table 2.1. Data reduction was performed using the Livedata package, which is part of the ATNF subset of the AIPS++ distribution. The bandpass correction was performed using an algorithm designed specifically for GASS data and was made for each beam using the quotient of each pair of frequency-switched spectra, masking emission before determining a baseline solution. A 15 th order polynomial fit was required to fit the structure of the quotient spectrum accurately, and was applied during an iterative process using both time and channel masks where necessary (McClure- Griffiths et al., 2009). Because all emission was masked during fitting, it is unlikely that the baselines are overfit. Bandpass calibration was applied for each beam and polarisation separately, and baselines are stable over the course of each scan. The baseline structures seen within the data are a result of the frequency dependence of the receiver s front-end gain and are also due to true baseline ripples from the dish of the Parkes Radio Telescope and focus cabin. In regions near the Galactic plane the fit was poorly constrained under the broad emission. An independent comparison to the LAB survey reveals that the baselines are good to within 50 to 70 mk, so any clouds that are detected above this level should be real. Radio frequency interference (RFI) was flagged, and where baseline residuals still remained, a 10 th order polynomial was used to remove the residuals in areas of the sky away from the Galactic plane (b >10 ) while a median level fit was used close to the plane.

37 2.4. THE GALACTIC ALL-SKY SURVEY Figure 2.1 Map of the total H I column density of GASS data, in Galactic coordinates and in units of cm 2. Doppler correction was then applied. Fluxes were calibrated from observations of the standards S6, S8 and S9 (Williams, 1973). The reduced data were gridded into a 3D data cube with voxel dimensions of km s 1 using the Gridzilla package, which is also part of the ATNF subset of the AIPS++ distribution, and was based on the gridding algorithm described in Barnes et al. (2001). The sensitivity is 57 mk. Due to low-level ripples in the bandpass of individual scans, low-level striped artefacts can be seen in some of the gridded data, although typically below the level of the noise. A map of the total H I column density over the entire GASS survey is shown in Figure 2.1. The focus of this thesis is on spatially discrete H I features with an angular size 1. These features are not characteristic of stray radiation, so although GASS data have not been corrected for stray radiation, it is unlikely that any of the H I clouds discussed here are a spurious result of this effect. The final GASS data release will be corrected for stray radiation according to the procedure described in Kalberla et al. (2005).

38

39 Chapter 3 H I Clouds in the Lower Halo: The GASS Pilot Region This chapter has been adapted from a published paper that appeared in the Astrophysical Journal (Ford et al., 2008). It has been slightly modified from the published version to avoid repetition and to conform to the structure of the thesis. We have detected over 400 H I clouds in the lower halo of the Galaxy within the pilot region of the Galactic All-Sky Survey, a region of the fourth quadrant that spans 18 in longitude, 40 in latitude and is centred on the Galactic equator. These clouds have a median peak brightness temperature of 0.6 K, a median velocity width of 12.8 km s 1, and angular sizes 1. The motion of these clouds is dominated by Galactic rotation with a random cloud-to-cloud velocity dispersion of 18 km s 1.A sample of clouds likely to be near tangent points was analysed in detail. These clouds have radii on the order of 30 pc and a median H I mass of 630 M. The population has a vertical scale height of 400 pc and is concentrated in Galactocentric radius, peaking at R = 3.8 kpc. This confined structure suggests that the clouds are linked to spiral features, while morphological evidence that many clouds are aligned with loops and filaments is suggestive of a relationship with star formation. The clouds might result from supernovae and stellar winds in the form of fragmenting shells and gas that has been pushed into the halo rather than from a galactic fountain. 21

40 22 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV 3.1 Introduction To place constraints on the physical properties and distribution of halo clouds, leading to a better understanding of their origin, nature, and role in the structure of the H I layer, it is crucial that we analyse a large sample of clouds. In this chapter we present a catalogue of over 400 H I clouds in the lower halo of the inner Galaxy that we have detected in the Galactic All-Sky Survey pilot region, along with an analysis and discussion of their properties and distribution. We begin with a brief overview of the data in 3.2 and present the observed properties of all clouds in 3.3. In 3.4 we determine the physical properties of a subset of these clouds that can be assumed to be located at tangent points, where the observer s line of sight is tangent to circles of constant Galactocentric radius. An analysis of the cloud properties is presented in 3.5 and implications of these results are discussed in 3.6. We summarise the results in The GASS Pilot Region The GASS pilot region was chosen for a preliminary study on halo clouds to develop techniques to be applied to GASS data. The GASS pilot region is in the fourth quadrant in the inner Galaxy and spans 325 l 343, b 20, and 200 V LSR 70km s 1 (see Figure 3.1). The V LSR range was chosen to include all velocities corresponding to tangent points of the inner Galaxy within the longitude range of the region, allowing for accurate distance determinations of many of the clouds (see 3.4), while not including large fractions of the Galaxy that are unlikely to be near tangent points. The 70 km s 1 boundary satisfies both these criteria, and while there are many interesting features at V LSR > 70 km s 1, analysis of these are beyond the scope of this project. 3.3 H I Clouds in the GASS Pilot Region We detect numerous discrete H I features in the lower halo of the Galaxy with angular sizes 1, similar to the population of halo clouds discovered by Lockman (2002). Samples of these clouds can be seen in Figures 3.2 and 3.3, where we display, respectively, a longitude-latitude image at V LSR = 105 km s 1 and a latitudevelocity image at l = The curved lines at the top and bottom of Figure 3.2

41 3.3. H I CLOUDS IN THE GASS PILOT REGION 23 Figure 3.1 The area of the Galaxy discussed in this chapter is bound by the longitude limits of the GASS pilot region (dotted arrows) and by V LSR = 70 km s 1, which, for a flat rotation curve, gives the area enclosed by the solid line. Dotted circles mark Galactocentric radii of 2.47 and 8.5 kpc and the locus of tangent points is shown by the dashed curve that connects the Sun and the Galactic centre.

42 24 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV represent the boundaries of the region studied in this chapter. The contour in Figure 3.3 represents T b = 3 K. Where the emission is brighter than this it is difficult to distinguish clouds because of confusion, although we do detect some clouds above this threshold. These figures clearly demonstrate the presence of discrete H I clouds at a variety of longitudes, latitudes and velocities, which are seen both close to the disk and into the halo. Some clouds are extended and some compact, and many appear to be related to diffuse and filamentary structures Search Method and Criteria We chose the following selection criteria to generate a homogeneously selected catalogue of halo clouds: (1) The cloud must be within the GASS pilot region, i.e., within l 343.1, b 20 (see Figure 3.2 for exact boundaries), and 200 V LSR 70 km s 1. (2) The cloud must span 4 or more pixels and be clearly visible over three or more channels in the spectra. Most cloud detections were made with T b 5 T b, where T b = 60 mk. (3) The cloud must be distinguishable from unrelated background emission. It was impossible to separate clouds from this emission at low latitudes in the least negative velocity channels, where the emission is particularly complex. We believe that we have identified all obvious clouds that meet the criteria listed above. However, it should be noted that some clouds appear to have double-peaked velocity profiles. In these circumstances, each peak was catalogued as an individual cloud because confusion may be important. If so, confusion may have resulted in the merging of profiles of multiple clouds at similar locations but with distinct velocities, resulting in a double-peaked profile. We will discuss such effects in detail in a subsequent publication. Most of the clouds are not isolated and spherical, but are instead often part of nebulous, filamentary structures and/or sitting in a fluctuating diffuse background. As a result, all automated cloud finding algorithms that we tested had difficulty differentiating clouds from neighbouring clouds and from background emission. Clouds were therefore identified and their properties measured by visual inspection of the data cubes.

43 3.3. H I CLOUDS IN THE GASS PILOT REGION 25 Figure 3.2 GASS pilot region at V LSR = 105 km s 1. Many H I clouds with angular sizes 1 are observed, both near the Galactic plane and into the lower halo. The curved lines at the top and bottom are the boundaries of the region searched.

44 26 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.3 H I clouds are observed at a variety of velocities, as can be seen in this latitude velocity diagram at l = The contour represents T b = 3 K, which is the brightness temperature above which few clouds can be distinguished due to confusion.

45 3.3. H I CLOUDS IN THE GASS PILOT REGION Properties of the Entire Cloud Sample Using the criteria presented in we measured 403 H I clouds in the GASS pilot region. The observed properties of these clouds are presented in Table 3.1. We provide a description of each property and how it was determined below, along with sample spectra in Figure 3.4. An integrated intensity map, which has the summed intensities over the velocity range for a given cloud, was used to aid in the determination of some properties. These maps have had a background subtracted that was the mean flux of unrelated emission in three interactively chosen areas surrounding the cloud. It is apparent from Figure 3.5 that at the low end of the peak brightness temperature distribution ( 0.5 K) there is significant incompleteness in regions of higher background levels. We are confident in our background subtraction because beyond incompleteness there is no trend between the peak brightness temperature and mean background level. We also note that the typical contrast between a cloud s peak brightness temperature and the background emission surrounding it is 1 (see Figure 3.5). However, as GASS data have not been corrected for stray radiation and because there may be many clouds that are smaller and fainter than we are able to detect, this contrast is a lower limit.

46 28 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Table 3.1: Observed Properties of HI Clouds in the GASS Pilot Region l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c Notes (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

47 3.3. H I CLOUDS IN THE GASS PILOT REGION 29 Table 3.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c Notes (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

48 30 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Table 3.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c Notes (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

49 3.3. H I CLOUDS IN THE GASS PILOT REGION 31 Table 3.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c Notes (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

50 32 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Table 3.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c Notes (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

51 3.3. H I CLOUDS IN THE GASS PILOT REGION 33 Table 3.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c Notes (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

52 34 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Table 3.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c Notes (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

53 3.3. H I CLOUDS IN THE GASS PILOT REGION 35 Table 3.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c Notes (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

54 36 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Table 3.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c Notes (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

55 3.3. H I CLOUDS IN THE GASS PILOT REGION 37 Table 3.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c Notes (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

56 38 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Table 3.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c Notes (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

57 3.3. H I CLOUDS IN THE GASS PILOT REGION 39 Table 3.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c Notes (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

58 40 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Table 3.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c Notes (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

59 3.3. H I CLOUDS IN THE GASS PILOT REGION 41 Table 3.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c Notes (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Notes. Descriptions of each property are presented in Clouds that have been catalogued elsewhere in the literature are noted by the following labels: (1) detected by Putman et al. (2002) and (2) detected by Wakker & van Woerden (1991). Although the clouds detected elsewhere do not necessarily have the exact Galactic coordinates and VLSR as listed here, it is likely that they are the same cloud and that the differences are a result of observational constraints. Also, we note that Morras et al. (2000) detected HI in some areas of these clouds but such detections were not identified as individual objects. a Uncertainties in T pk are 0.07 K. b Uncertainties in the maximum angular extents are dominated by background levels surrounding the cloud and are assumed to be 25% of the estimated values. c Mass uncertainties are dominated by the interactive process used in mass determination and are assumed to be 40% of the estimated values.

60 42 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV The columns in Table 3.1 are as follows: Columns 1-2: The Galactic longitude, l, and latitude, b, of the cloud at the position of the peak brightness temperature. Column 3: The cloud velocity with respect to the local standard of rest, V LSR, in km s 1, measured as the velocity of the cloud s peak brightness temperature after background subtraction. The background level was determined by fitting a line between each edge of the cloud velocity profile where it merges with surrounding emission. Column 4: The cloud s peak brightness temperature after background subtraction, T pk, in K. Most clouds that we measured had T b < 3 K prior to background subtraction, as clouds with T b greater than this tended to be found only in areas of high confusion. Column 5: v is the full-width at half-maximum (FWHM) of the velocity profile, determined by inspection after background subtraction, in km s 1. Column 6: The H I column density, N HI, at the cloud centre is T pk v cm 2 in the optically thin limit, an assumption that is reasonable because the emission is faint. Column 7: θ maj and θ min are the maximum and minimum extent of the cloud in arcminutes and were determined by inspection from the integrated intensity map of the cloud. Many of the clouds are unresolved in at least one dimension but we have not deconvolved their angular sizes due to the uncertainties associated with the fluctuating background levels of the integrated intensity maps. These values therefore represent upper limits of the angular extent. Column 8: M HI d 2 represents the H I mass of the cloud in units of M kpc 2, where d is the distance to the cloud in kpc. A background was subtracted from the integrated intensity map, thereby leaving only the flux of the cloud, which was then summed. The background level was highly variable from cloud to cloud being mainly dependent on the latitude and velocity of the cloud. The mass also relies on the assumption that the H I is optically thin. Column 9: Details of prior detections in the literature, if any, are noted Uncertainties in Observed Properties δv LSR : The difference between V LSR and the velocity where the profile decreases from the peak by ( Tb 2 + δt pk 2 )1/2, assuming the profile can be approximated by a

61 3.3. H I CLOUDS IN THE GASS PILOT REGION 43 Figure 3.4 Spectra from a random sample of H I clouds from Table 3.1. Many clouds appear to be sitting on broad spectral wings. Arrows mark the velocity where the brightness temperature peaks (after background subtraction) for each profile.

62 44 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.5 Peak brightness temperature as a function of the mean background subtracted from the integrated intensity map. In regions of high background levels there is a clear decline in the number of clouds with low peak brightness temperatures.

63 3.3. H I CLOUDS IN THE GASS PILOT REGION 45 Gaussian, where δt pk is the error on T pk (see below) and T b is the rms noise (60 mk). δt pk : This error is assumed to be [ Tb 2 + ( T b/ 2) 2 ] 1/2, where the first term represents the error in T pk prior to the background subtraction and the second term represents the error of the mean of the two points defining the background. δ v: Each of the two half maximum points on the profile is defined as the point where the profile, with error T b, crosses (1/2)T pk, with error (1/2)δT pk. We translate these T b errors into velocity errors using the slope of a Gaussian profile. We also add the channel width in quadrature. δn HI : The error on the H I column density, δn HI = N HI [(δt pk /T pk ) 2 +(δ v/ v) 2 ] 1/2. δθ maj and δθ min : The uncertainties due to the interactive nature of the angular extent measurements and the fluctuating background levels of the integrated intensity maps dominate over the nominal statistical error. The uncertainties were therefore calibrated by comparing values obtained during different trials for a randomly selected subset of clouds and are estimated to be approximately 25% of the determined angular extent. δm HI d 2 : The errors introduced by the interactive process of determining the cloud area and subtracting the background dominate the mass uncertainties and are estimated to be 40% of the determined mass, based on the comparison of different trials of randomly selected clouds mentioned above Overview of Observed Properties The V LSR as a function of longitude for all clouds is shown in Figure 3.6, along with a solid line denoting the terminal velocity, V t, which, in the fourth quadrant, is the most negative velocity permitted by Galactic rotation. The clouds are abundant at velocities allowed by Galactic rotation and there is clearly a decline in the number of clouds beyond (more negative than) V t, which demonstrates that the motions of this cloud population are dominated by Galactic rotation. Clouds beyond V t are said to have forbidden velocities and the amount their V LSR differs from V t is the deviation velocity, V dev = V LSR V t, as defined by Wakker (1991). For l , we used V t determined by McClure-Griffiths & Dickey (2007) from H I observations from the Southern Galactic Plane Survey (McClure-Griffiths et al., 2005). For all remaining longitudes we used V t determined by Luna et al. (2006) from the Columbia-Universidad de Chile CO surveys.

64 46 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.6 V LSR as a function of longitude for all clouds in Table 3.1. Crosses represent a subset of clouds with V dev 0.8 km s 1 (the tangent point sample; 3.4) while circles represent all other GASS pilot region clouds. The solid line is the V t curve determined by McClure-Griffiths & Dickey (2007) from H I observations for l and by Luna et al. (2006) from CO observations for all remaining longitudes. As we have searched for all clouds within 200 V LSR 70 km s 1, the decline in the number of clouds at velocities more negative than the terminal velocity demonstrates that the kinematics of the clouds are dominated by Galactic rotation.

65 3.4. SELECTION OF A SET OF TANGENT POINT CLOUDS 47 In Figure 3.7 we display the latitude distribution with deviation velocity. For clouds beyond the terminal velocity (with negative deviation velocities), those at lower latitudes have larger absolute deviation velocities. This is likely an artefact, as the number of clouds would naturally fall off with both more negative deviation velocities and larger latitudes if they were dominated by Galactic rotation. There are, however, some outliers at large positive latitudes with very negative deviation velocities. Histograms of T pk, v and angular size are presented in Figures 3.8, 3.9, and 3.10, respectively. The majority of clouds have T pk 1 K, with the median T pk = 0.6 K. The number of clouds decreases sharply below T pk 5 T b, suggesting that the sample is sensitivity limited. The median FWHM of the velocity profiles is 12.8 km s 1 and very few clouds have linewidths larger than 30 km s 1 (Figure 3.9). All but one of the linewidths are greater than 3.4 km s 1 ; as the velocity resolution of the survey is 0.8 km s 1, most lines are therefore well resolved. The median angular diameter of the clouds is 29, which is approximately twice the beam size. This value and the steep cutoff at small angular sizes seen in Figure 3.10 are most likely due to the spatial resolution limit of the data and suggest that many of the clouds are unresolved. 3.4 Selection of a set of Tangent Point Clouds The largest magnitude velocity from Galactic rotation in the inner Galaxy occurs at the tangent point, defined as the location where the line of sight is perpendicular to a circle of constant Galactocentric radius (see Figure 3.1). Here R t = R 0 sin l and the LSR velocity from Galactic rotation is V t = R 0 [Θ/R t Θ 0 /R 0 ] sin l cos b, where R 0 is the radius of the solar circle, Θ is the circular velocity andθ 0 is the circular velocity at the solar circle. We adopt R kpc andθ km s 1, as recommended by the International Astronomical Union (Kerr & Lynden-Bell, 1986). Clouds in pure Galactic rotation cannot have a circular velocity beyond V t. However, the random motion of a cloud near the tangent point might increase the cloud s V LSR beyond V t. Clouds in Galactic rotation with V LSR V t must therefore lie near the tangent point and thus at a known distance, d t = R 0 cos l/ cos b. While an assumption, it is reasonable to adopt a distance of d t for clouds with V LSR V t given the rapid decline in the number of clouds beyond the terminal velocity, as

66 48 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.7 Latitude distribution with deviation velocity, V dev, for all clouds in Table 3.1, where V dev = V LSR V t, as defined by Wakker (1991). We used V t determined by McClure-Griffiths & Dickey (2007) from H I observations for l and by Luna et al. (2006) from CO observations for all remaining longitudes. Crosses denote a subset of clouds with V dev 0.8 km s 1 (the tangent point sample, 3.4), while circles denote all other clouds in the GASS pilot region. For clouds that are observed at forbidden velocities (V dev < 0 km s 1, denoted by the dashed line), those farther from the plane have smaller absolute deviation velocities. This trend is expected of a population of clouds whose kinematics are dominated by Galactic rotation, as the number of clouds would fall off at more negative velocities and larger latitudes, but there are some outliers at large positive latitudes with more negative deviation velocities.

67 3.4. SELECTION OF A SET OF TANGENT POINT CLOUDS 49 Figure 3.8 Histogram of the peak brightness temperature of the clouds, T pk. The median is T pk =0.6 K and the cutoff at small T pk is likely due to the sensitivity limit. The arrow represents the 5 T b detection level. We assume N errors.

68 50 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.9 Histogram of the FWHM of the velocity profile of the clouds, where the median is 12.8 km s 1. As the spectral resolution is 0.8 km s 1 and all but one linewidth is greater than 3.4 km s 1, the linewidths are well resolved. We assume N errors.

69 3.4. SELECTION OF A SET OF TANGENT POINT CLOUDS 51 Figure 3.10 Histogram of the angular size of clouds, [θ maj θ min ] 1/2, where θ maj and θ min are from Table 3.1. The median angular diameter of the clouds is 29, which is approximately twice the beam size. This median, along with the apparent lack of clouds with small angular sizes, is most likely due to the spatial resolution limit (16, denoted by an arrow) and suggests that many of the clouds are unresolved. We assume N errors.

70 52 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV shown in Figure 3.6. Tangent point clouds constitute a sample uniquely suited for investigating the population s distribution and properties, such as physical size and mass. From the population of clouds detected in the GASS pilot region, we define the tangent point sample as all clouds with V LSR V t +0.8 km s 1, where 0.8 km s 1 is one channel width, and assume that they are at the tangent point; we assess the effect of this assumption in We also assume that Galactic rotation is constant with distance from the plane, i.e.,θ( z) =Θ (0); deviations from cylindrical rotation will be discussed in a subsequent work. As the tangent point clouds in the GASS pilot region are nearly all at the same distance and the latitude boundary of the region corresponds to a constant height of 2.5 kpc (at tangent points), they provide a uniformly selected sample Derived Properties The physical properties and positions of the tangent point clouds are presented in Table 3.2, while descriptions of the derived quantities are presented below.

71 3.4. SELECTION OF A SET OF TANGENT POINT CLOUDS 53 Table 3.2: Derived Properties of Tangent Point HI Clouds in the GASS Pilot Region l b VLSR Vdev d R a z r MHI (deg) (deg) (km s 1 ) (km s 1 ) (kpc) (kpc) (kpc) (pc) (M ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 130 Continued on Next Page...

72 54 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Table 3.2 Continued l b VLSR Vdev d R a z r MHI (deg) (deg) (km s 1 ) (km s 1 ) (kpc) (kpc) (kpc) (pc) (M ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 4 70 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 50 Continued on Next Page...

73 3.4. SELECTION OF A SET OF TANGENT POINT CLOUDS 55 Table 3.2 Continued l b VLSR Vdev d R a z r MHI (deg) (deg) (km s 1 ) (km s 1 ) (kpc) (kpc) (kpc) (pc) (M ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 150 Notes. Derived properties of a subset of the clouds assumed to be located at tangent points (those with Vdev 0.8 km s 1 ). a Along a given line of sight, the smallest Galactocentric radius possible is at the tangent point. If the cloud is not located at the tangent point it must be farther away from the Galactic centre and the error on R must be positive.

74 56 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Columns 1-3: As in Table 3.1. Column 4: The deviation velocity, V dev = V LSR V t (Wakker, 1991), where V t is the most negative velocity expected from Galactic rotation in the fourth quadrant, in km s 1, and was determined by McClure-Griffiths & Dickey (2007) from H I observations for l and by Luna et al. (2006) from CO observations for all remaining longitudes. Column 5: The distance, d, along the line of sight from the Sun to the cloud determined by assuming the cloud is at the tangent point: d = R 0 cos l/ cos b, in kpc. Column 6: The Galactocentric radius, R = R 0 sin l, of the tangent point at the cloud s location, in kpc. Column 7: The height, z, of the cloud from the plane of the Galaxy, determined geometrically to be z = d sin b, in kpc. Column 8: The radius, r, of the cloud in pc, determined by (r maj r min ) 1/2, where r maj = (1/2)dθ maj, r min = (1/2)dθ min, and θ maj and θ min are from Table 3.1. Column 9: The physical mass of H I in the cloud, M HI, determined as described regarding column 8 in but with the tangent point distance assumed, in M Uncertainties in Derived Properties Here we present error estimates for the derived properties in Table 3.2. δv LSR : As in Table 3.1. δv dev : This error is (δv 2 LSR +δv 2 t ) 1/2, where δv t = 3 km s 1 for clouds located at longitudes where V t was determined using H I observations (l ; McClure- Griffiths & Dickey 2007). For all other longitudes, where the terminal velocity was determined using CO observations (Luna et al., 2006), the error is assumed to be 9 km s 1, as suggested from the scatter on Figure 7 of McClure-Griffiths & Dickey (2007). δd: Distance errors, which are inherent to the assumption that the clouds are located at tangent points, were estimated using a simulated population of clouds (see 3.5.3). δr: The error on the Galactocentric radius was determined analogously to δd. Because the closest point to the Galactic centre along a given line of sight is the tangent point, the adopted R is always a lower limit and the error can only be positive.

75 3.5. ANALYSIS OF THE TANGENT POINT CLOUD POPULATION 57 δz: The error on the height is estimated to be δd sin b. δr: The error on the radius of the cloud depends on the uncertainties in the distance and angular size estimates and is δr = r[(δθ/θ) 2 +(δd/d) 2 ] 1/2, where θ = (θ maj θ min ) 1/2 and δθ = (1/2)[δθ 2 min (θ maj /θ min )+δθ 2 maj (θ min /θ maj )] 1/2. δm HI : The error on the mass is M HI [(δm HI d 2 /M HI d 2 ) 2 + (2δd/d) 2 ] 1/2, where M HI d 2 is given in Table Analysis of the Tangent Point Cloud Population Simulated Halo Cloud Population To constrain the spatial and kinematic properties of the observed tangent point cloud population we simulated a population of clouds to which were applied the same l, b and V LSR selection criteria as for the GASS pilot region clouds. The simulated clouds were randomly sampled from the following distribution: ( n(r, z) = Σ(R) exp z ), (3.1) h where Σ(R) is the radial surface density distribution, h is the exponential scale height and R and z are the cylindrical coordinates. Σ(R) is composed of 12 independent radial bins of width 0.25 kpc, spanning R =2.5 to 5.5 kpc. The amplitude of each bin was optimised to best fit the observed longitude distribution of the tangent point clouds by minimising the Kolmogorov-Smirnov (K-S) D statistic (the maximum deviation between the cumulative distributions) using Powell s algorithm (Press et al., 1992). We optimised the fits using three different initial estimates on Σ(R). All converged on a similar solution and we adopted the mean of the three as the best fit to the data, which is shown in Figure Velocities of simulated clouds were based on a flat rotation curve where Θ =Θ 0 with a random velocity component drawn from a Gaussian of dispersion σ cc = 18 km s 1, which is discussed in detail in We generated clouds in a half-galaxy (third and fourth quadrants), of which 2475 were within the Galactic coordinates of the GASS pilot region and also within the defined velocity range of the tangent point cloud sample. We then normalised this distribution to compare

76 58 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.11 Best estimate of the radial surface density, Σ(R), of clouds in the GASS pilot region. Error bars denote the range in amplitude within each bin for the three radial solutions. Clouds were detected at all longitudes within the GASS pilot region, where corresponding tangent point boundaries in R are indicated by arrows. The distribution suggests that the clouds are concentrated in radius, peaking at R =3.8 kpc.

77 3.5. ANALYSIS OF THE TANGENT POINT CLOUD POPULATION 59 directly with the observed distribution. We performed K-S tests to estimate the quality of the fit between the observed and simulated distributions. Based on these tests, we find that the parameters of the simulated population reproduce the distribution of observed tangent point clouds well and are therefore good estimates for those of the intrinsic population. Results of the fits to the distributions are presented in along with tests of other functional forms Cloud-to-Cloud Velocity Dispersion We assume that a random cloud-to-cloud velocity dispersion (σ cc ) is responsible for the presence of clouds at forbidden velocities within the GASS pilot region. Simulations of the global cloud population were required to model the effects of these motions, which can cause clouds that are not located at tangent points to have velocities that are near or beyond V t. The cloud-to-cloud velocity dispersion can provide information on the expected scale height of the distribution and a better understanding of the formation mechanisms of the clouds. We find the V dev distribution of the tangent point sample of clouds to be consistent with that derived from a Gaussian distribution of random velocities whose dispersion is σ cc = 18 km s 1, with a K-S test probability greater than 97% 1. These distributions are presented in Figure 3.12, where the simulated V dev distribution is represented by a dashed line and the observed distribution by a solid line. Velocity dispersions of 16 to 22 km s 1 also provide acceptable fits (K-S test probabilities greater than or equal to 15%), as do fits where the random velocity component is drawn from an exponential distribution rather than a Gaussian distribution (with a K-S test probability of 89% for a scale velocity of 13 km s 1 ). The implied kinematics of the cloud population based on this result are discussed in Estimating the uncertainties on our determined σ cc is extremely difficult. We expect these uncertainties to be coupled to errors introduced by the measurements of V t, which are difficult to untangle due to the possibility that random motions can systematically offset the measured V t. Another possible source of uncertainty is streaming motions associated with spiral features. However, because V t was determined directly from H I and CO measurements rather than from a fitted rotation curve, streaming motions should already be reflected in the adopted V t. 1 For reference, a 97% probability is roughly equivalent in confidence to a 0.04σ detection of a difference, i.e., no detectable difference in the distributions, 15% corresponds to 1.4σ, and 1% corresponds to 2.6σ.

78 60 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.12 Distribution of deviation velocities of the observed tangent point and simulated population of clouds, where the simulated clouds are normalised to the same total number of observed tangent point clouds. Observed clouds are represented by a solid line, while simulated clouds are represented by a dashed line. To reduce the effects of small number statistics, simulations were run with clouds. We assume N errors.

79 3.5. ANALYSIS OF THE TANGENT POINT CLOUD POPULATION 61 We have assumed that there is no dependence of the Galactic rotation curve on distance from the Galactic plane. Other galaxies, however, show evidence for a systematic lag in rotation of km s 1 kpc 1 in both ionised and neutral extraplanar gas (Fraternali et al., 2005; Rand, 2005; Marinacci et al., 2010). Evidence for a similar lag in the Milky Way is scant (Pidopryhora et al., 2007). However, a rough estimate of the effect of neglecting a halo lag on the determination of σ cc can be obtained. Assuming a lag of 15 km s 1 kpc 1, at the median vertical distance of a cloud (560 pc), the expected lag would be 8.4 km s 1. The effect of neglecting lag can be estimated by adding the lag component in quadrature to our estimate of σ cc = 18 km s 1, which gives a revised estimate of σ cc 20 km s 1. This difference is sufficiently small that the assumption of corotation does not affect our conclusions Distance Errors The simulations allow us to estimate the error in our assumption that clouds with V dev 0.8 km s 1 are at the tangent point. We have calculated the fractional distance error of the simulated clouds as a function of deviation velocity within a 5 km s 1 wide V dev bin (Figure 3.13). Clouds at increasingly forbidden velocities have smaller distance errors, i.e., are more likely to be near the tangent point; the degree to which this is true depends on the magnitude of σ cc. We also calculated the fractional distance error as a function of longitude and found that clouds with longitudes corresponding to the Galactocentric radii where few clouds are detected (l 328 ) have larger distance errors (this is expected because few clouds are intrinsically at these radii and therefore a larger fraction of the forbidden velocity clouds at these longitudes are likely interlopers from larger radii). We assume that the errors due to V dev and longitude are independent. The adopted fractional distance error is the product of the fractional distance error due to V dev (relative to the typical error of 0.12), due to longitude (relative to the typical error) and the typical error itself. The fractional distance errors due to V dev and longitude are taken to be the rms error for the simulated clouds within the same bin. We have confirmed that the fractional distance error of the simulated clouds does not significantly depend on their latitude. Based on the relative distance errors, we believe that our assumption that clouds with V dev 0.8 km s 1 are located near their tangent point is reasonable.

80 62 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.13 Relative distance error as a function of deviation velocity (crosses) for a population of simulated clouds. Clouds with a large forbidden velocity (i.e., large negative value of V dev ) are more likely to be at the tangent point and thus have smaller errors in their estimated distances. These points have been placed into bins of width 5 km s 1, denoted by the horizontal bars, and the vertical error bars denote the mean and standard deviation of the distribution within each bin. Approximately one fifth of simulated points are displayed here but the error bars have taken all simulated data into account. The left most error bar is large due to small number statistics.

81 3.5. ANALYSIS OF THE TANGENT POINT CLOUD POPULATION Radial Distribution The adopted radial distribution of the tangent point clouds along with that of the simulated population of clouds is shown in Figure The apparent offset in the distributions results from the assumption that the forbidden velocity clouds are located at tangent points, which are always at the smallest value of R along the line of sight. Another useful quantity that can be extracted from the simulations is the mean radial surface density distribution of clouds within the GASS pilot region (Figure 3.11). Although clouds are observed at all longitudes within the GASS pilot region, the distribution is concentrated in Galactocentric radius and peaks at R = 3.8 kpc. The error bars at R 5 kpc are significantly larger because clouds at these radii must have large random velocities to meet the sample criteria and therefore represent a small tail of the population. With a K-S test probability of 95%, the simulated longitude distribution fits that of the observed distribution of the tangent point sample well (Figure 3.15). The number of clouds in a uniformly distributed population of tangent point clouds should monotonically decrease by a factor of 2 between l = 325 and l = 343, as demonstrated by equation (A2) in Stil et al. (2006), where they show that the line of sight distance effectively surveyed over forbidden velocities is d = 8R0 sin l [σ cc /Θ 0 ] 1/2 sin l. We have overlaid the distribution of a uniform surface density population in Figure 3.15 and it is clear that it is in stark contrast to the centrally peaked longitude distribution we observe. We therefore conclude that the peaked radial distribution is real. In Figure 3.16 we present the longitude distributions of the simulated and observed clouds within the entire pilot region, i.e., at all 200 V LSR 70 km s 1. Even though the vast majority of these clouds were not used to constrain the simulated radial distribution, their longitude distribution is well reproduced. The tangent point sample therefore appears to be a fair representation of the entire GASS pilot region. It is worth investigating whether our assumptions regarding the functional form of the velocity distribution affects the inferred radial distribution, i.e., could a distribution with more velocity outliers and a less strongly peaked radial distribution also fit the data? To test this we have performed the same optimisation of the radial distribution while using an exponential velocity distribution. We find that the resulting radial profile is identical to within the errors, confirming the robustness of

82 64 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.14 Histogram showing the fraction of the sample located at each Galactocentric radius. The solid line denotes the clouds assumed to be located at tangent points and the dashed line denotes the corresponding simulated radial distribution. The shift between the simulated and adopted radial distributions results from systematic errors that stem from the assumption that the forbidden velocity clouds are located at tangent points, i.e., at the minimum possible value of R. Arrows represent the radial boundaries of tangent points within the GASS pilot region. We assume N errors.

83 3.5. ANALYSIS OF THE TANGENT POINT CLOUD POPULATION 65 Figure 3.15 Histogram presenting the longitude distribution of the observed tangent point clouds (solid line) and corresponding simulated distribution (dashed line), along with the distribution expected from a uniform population (dotted line). There is a peak in both the observed and simulated distributions, which is clearly not expected for a uniform population. We assume N errors.

84 66 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.16 Histogram presenting the longitudes of all clouds detected in the GASS pilot region (solid line) along with the corresponding simulated population (dashed line). The distributions are similar, suggesting that the tangent point sample represents the pilot region fairly well. We assume N errors.

85 3.5. ANALYSIS OF THE TANGENT POINT CLOUD POPULATION 67 this result Vertical Distribution The vertical distribution of both the tangent point and simulated cloud population is presented in Figure Clouds have been detected throughout the entire range of latitudes covered by the GASS pilot region, up to corresponding heights of z = 2.5 kpc. However, there are very few clouds at b 2.5 because identification of clouds was extremely difficult close to and within the Galactic plane, except in cases in which clouds were observed at large forbidden velocities. Because of this incompleteness at low latitudes, we compare the latitude distributions of simulated and observed clouds with b > 2 and find that they are consistent for vertical scale heights between 400 and 500 pc, with K-S test probabilities of at least 25%. As incompleteness may still be a problem at b 2, we also compare the height distributions exclusively for clouds at b > 3 and find acceptable fits for scale heights between 300 and 400 pc, both with probabilities greater than 50%. Based on these comparisons we conclude that the population is best represented with a exponential scale height of h = 400 pc. A sech 2 (z/z 0 ) distribution with z 0 = 700 pc provides an equally good fit to the data, which is not surprising given that the exponential and sech 2 distributions differ primarily near z = 0 pc where we cannot constrain the fit. Although incomplete, the combination of surface density, mean mass and scale height of the clouds gives a rough estimate of the vertical distribution of H I contributed by the cloud population. The mid-plane H I number density can be estimated by n(0) = Σ(R) M /(2hM H ), where M is the mean cloud H I mass and M H is the H I atom mass. We find that the clouds are responsible for 5%, by H I number density, of the exponential component of the H I layer in Dickey & Lockman (1990) and have a very similar scale height. We note that there is an asymmetry in the number of observed clouds above and below the disk at z 750 pc and an excess of clouds at large positive latitudes. Possible explanations for these are discussed in Physical Size and Mass Cloud radii, r, vary from 15 to 65 pc, with a median radius of 32 pc, as can be seen in Figure The angular resolution of the telescope sets a lower limit on the

86 68 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.17 Histogram of the fraction of the observed (solid line) and simulated (dashed line) clouds as a function of height. The lack of observed clouds at low z is a selection effect due to confusion. This effect was taken into account during comparisons with the simulated population by omitting all simulated clouds with b 2. The exponential scale height of the distribution is h = 400 pc. We assume N errors.

87 3.5. ANALYSIS OF THE TANGENT POINT CLOUD POPULATION 69 Figure 3.18 Histogram of the physical radius of clouds at tangent points. The median r = 32 pc. The cutoff at small radii is a result of the spatial resolution limit. We assume N errors. observed cloud size. The maximum angular extent of the detected clouds suggests that roughly 80% of the entire sample is unresolved in at least one dimension. Derived radii should therefore be thought of as upper limits. A histogram of the H I mass of the clouds at tangent points is presented in Figure 3.19, demonstrating that the clouds range in size from hundreds to thousands of solar masses, with a median H I mass of 630 M. These values may be overestimates if confusion is significant Comparison of GASS Clouds to Lockman Clouds Based on their angular sizes and location in the lower halo, the clouds detected in the GASS pilot region appear to be similar to those observed by Lockman (2002). We investigate this possibility further by comparing the properties of each distribution summarised in Table 3.3. The median v is strikingly similar in both sets of data, which is not surprising if the clouds are part of the same population. The derived

88 70 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.19 Histogram of log M HI of clouds at tangent points. The 5 T b limit is represented by an arrow for a cloud with the median r = 32 pc. We assume N errors.

89 3.5. ANALYSIS OF THE TANGENT POINT CLOUD POPULATION 71 Table 3.3. Property Summary and Comparison to Lockman Clouds This Sample Lockman (2002) Parameter Median 90% Range Median 90% Range T pk (K) v (km s 1 ) N HI ( cm 2 ) r (pc) 32 < < M HI (M ) z (pc) σ cc (km s 1 ) Note. Median values of the observed halo cloud properties in this sample, where the number of clouds n = 403 for T pk, v, and N HI and n = 81 for remaining properties (of tangent point clouds) and the Lockman (2002) sample, where n = 38. Most properties have a large scatter about the median in both samples, as demonstrated by the 90% range. σ cc are also in agreement. The median z of this sample is smaller than that of Lockman (2002) but this is most likely due to a selection effect, as the areas searched by Lockman (2002) tended to be farther from the Galactic plane to avoid areas of confusion. The most obvious differences in the sample stem from the difference in the angular resolutions of each survey: 15 for GASS data versus 9 for the GBT data. This affects T pk and r, where T pk would naturally be lower for unresolved clouds and r would be larger. If confusion is important, M HI would also be larger. Lockman (2002) estimated that 25% of the clouds were unresolved while we have estimated 80% here. The clouds observed by Lockman (2002) are much less massive, with approximately one third having M HI 30 M, while none of the observed GASS clouds have masses that low. As the only differences in the observed properties are due to differences in the observations, these comparisons reveal that the clouds belong to the same population of clouds as those detected by Lockman (2002).

90 72 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Observed Trends There does not appear to be any correlation between the height of the clouds and V LSR, R, or r. However, as the data in Figure 3.20 suggest, there may be a trend between v and z, where clouds near the plane ( z 1 kpc) have a median FWHM of 10 km s 1 and a large dispersion, whereas those at z > 1 kpc have a median FWHM of 17 km s 1 and a smaller dispersion. One possible explanation for such a trend could be that the clouds at larger heights belong to a different population of clouds than those at lower heights. Another possibility is that if the clouds are in pressure equilibrium, the trend is reflecting pressure variations throughout the halo. Assuming that the internal pressure is dominated by thermal and turbulent motions, P v 2. We can therefore make a very rough estimate of the expected v variations with height using estimates of the total pressure within the Milky Way disk-halo interface from Cox (2005), though these pressures are very poorly constrained. When compared to v at z = 300 pc, v at 1 kpc and 2 kpc would be factors of 1.5 and 2.6 smaller. This is contrary to the trend that we see between v and z, which is roughly the same magnitude, but in the wrong direction. This indicates that either the halo pressure actually increases with z or that the linewidths are not governed by pressure equilibrium. Similar to our results, Lockman (2002) found that V LSR and r are independent of z, and also found evidence that clouds with more narrow linewidths lay closer to the plane. As one might expect if a population of clouds has a narrow range of densities, the larger the radius of the cloud, the more massive it is. This trend is evident in Figure 3.21 and does not appear to be solely due to the H I mass sensitivity limit of the data. This limit is denoted by the curved line and is based on the minimum observable H I column density, N HImin = T bmin v med in cm 2, where T bmin is the minimum observable T b (assumed to be 5 T b ) and v med is the observed median v. Another apparent trend, as seen in Figure 3.22, is that the clouds with higher H I column densities appear to be at lower heights. This could be due to inclusion of unrelated diffuse emission with the clouds at lower heights if the background subtraction was not effective. It could also be explained by a scenario where each cloud was given the same kinetic energy from a formation process or via equipartition in the subsequent evolution of the cloud population. The higher mass clouds, which have higher N HI, would then have preferentially lower velocities and would not reach

91 3.5. ANALYSIS OF THE TANGENT POINT CLOUD POPULATION 73 Figure 3.20 FWHM of the velocity profile, v, of the tangent point clouds as a function of height. Clouds at larger heights may have larger linewidths.

92 74 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.21 Derived H I mass of the clouds as a function of cloud radius (crosses). The curved line represents the lower H I mass limit and is based on the minimum observable N HI. The minimum brightness temperature is assumed to be 5 T b. The larger clouds are more massive.

93 3.6. THE ORIGIN AND NATURE OF HALO CLOUDS 75 heights as large as those reached by clouds with higher velocities. Figure 3.23 provides tentative support for this hypothesis, revealing that at z 1 kpc there are very few clouds with M HI 10 3 M (note, however, that at these heights our statistics are poor). Lockman (2002) did not find a correlation between the H I column density or mass with height, but this could be due to the limited vertical range of his data. Similarly, a correlation between the H I mass of a cloud and its deviation velocity is suggested in Figure 3.24; in particular, the more massive clouds may have lower deviation velocities. If the more massive clouds have lower random motions, they could have lower typical deviation velocities. The relationship between the height achieved by a population and the deviation velocity can be determined by comparing the line of sight component of velocity that results from releasing clouds at different heights and letting them fall in the Kalberla et al. (2007) gravitational potential of the Galaxy. Falling from 2 kpc to 500 pc, a cloud accelerates by 70 km s 1, of which 4 km s 1 is projected along the line of sight, whereas a cloud that falls from 1 kpc accelerates by 50 km s 1 and only attains a 3 km s 1 line of sight velocity. The higher mass clouds having lower random velocities would be consistent with a scenario in which clouds evolve to an equipartition of energy or in which each cloud is given a similar initial kick, for example, from similar supernovae explosions. This could also explain why all of the most massive clouds are seen closer to the plane (Figure 3.23): if they have smaller initial velocities, they would not move as far into the halo before falling back towards the plane. At this stage, however, we cannot exclude the possibility that at lower heights confusion is affecting the determined H I column density and mass of the clouds. We discuss the kinematics of the cloud population further in The Origin and Nature of Halo Clouds Kinematics of Halo Cloud Population The velocity dispersion of the cloud population is likely a remnant of a common formation process, such as violent supernovae explosions. If the population is in equilibrium, the vertical distribution and velocity dispersion are linked via the potential. By using the simulated population of clouds, we found that the vertical distribution is best fit with an exponential scale height of 400 pc ( 3.5.5) and the

94 76 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.22 H I column density as a function of height. All clouds with large column densities are located at lower heights, i.e., closer to the Galactic disk.

95 3.6. THE ORIGIN AND NATURE OF HALO CLOUDS 77 Figure 3.23 H I mass of the clouds as a function of height (crosses). Horizontal lines represent the median M HI per 18 clouds and error bars represent the 25 th to 75 th percentile range. There is no evidence for a trend until z 1 kpc, where there may be a lack of massive clouds.

96 78 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.24 H I mass of the clouds as a function of deviation velocity (crosses). Horizontal lines represent the median M HI per 18 clouds and error bars represent the 25 th to 75 th percentile range. The more massive clouds have lower deviation velocities.

97 3.6. THE ORIGIN AND NATURE OF HALO CLOUDS 79 V dev distribution is best fit with a Gaussian dispersion of σ cc = 18 km s 1 ( 3.5.2). However, assuming a vertical force using the mass model of Kalberla et al. (2007) at R =3.8kpc, we derive an exponential scale height of 90 pc for an isothermal population of clouds with σ cc = 18 km s 1. To produce our observed scale height of 400 pc within this potential, the cloud-to-cloud velocity dispersion would have to be 60 km s 1. If lag is an important factor that we have neglected, we may have underestimated σ cc. However, our calculations of the effect of lag on σ cc in reveal that σ cc is still much too small to account for the large scale heights of the clouds. The difference between the σ cc required and that observed may be due to the lack of clouds observable in the disk; if a large number of clouds within the disk have gone undetected, the scale height could be lower than 400 pc, and could therefore be explained by the observed velocity dispersion. It is also possible that the distribution cannot be explained by a single component, but this is not obvious in the current data. However, the magnitude of this difference suggests that the clouds do not belong to an equilibrium population and their heights must, in part, result from processes that do not increase the velocity dispersion, such as uniformly expanding H I shells, instead of bursts of energy from areas of active star formation generating random kicks. The clouds could have also originated above the disk, as in a galactic fountain model. Similar clouds have been detected at forbidden velocities within the Galactic disk using data from the VLA Galactic Plane Survey, suggesting that the clouds are not restricted to the halo (Stil et al., 2006). Those cloud diameters are much smaller than the ones derived here (they have diameters 10 pc), likely due to the higher spatial resolution of the VLA data. Based on their models, Stil et al. (2006) find that a vertical Gaussian half-width at half-maximum (HWHM) larger than 1 kpc (equivalent to an exponential scale height of 1.2 kpc) best fits their data. Given that they only surveyed within b 1.3, however, this finding is likely to be strongly affected by the lack of coverage at high latitudes, especially because we observe similar clouds up to z =2.5kpcandLockman (2002) has observed clouds up to 1.5 kpc. Stil et al. (2006) find a lower limit to the HWHM of the clouds to be 180 pc (exponential scale height of 216 pc), which is consistent with the value we derive and also inconsistent with the value expected for the derived velocity dispersion. It is worth noting that there is evidence that similar clouds may also be abundant in the outer Galaxy (Stanimirović et al., 2006). It is not yet certain whether they

98 80 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV belong to the same population of clouds, but with larger surveys of H I clouds in the lower halo of the Galaxy, such as those presented here and in the entire inner Galaxy within GASS, along with those underway by the Galactic Arecibo L- Band Focal Plane Array Consortium and other groups, properties such as σ cc and the spatial distribution can be determined more accurately and will therefore help constrain the kinematics and formation mechanisms of the clouds Halo Clouds, Spiral Structure, and Star Formation As discussed in 3.5.4, the surface density of halo clouds is not uniformly distributed but instead peaks at a Galactocentric radius of R =3.8 kpc and the population is concentrated in radius. This confined nature suggests that the halo clouds are related to the spiral structure of the Galaxy. Although distances to spiral arms are currently not well constrained, the location of the peaked radial distribution of the clouds indicates that they may be related to the expanding 3 kpc arm (van Woerden et al., 1957; Rougoor & Oort, 1960), which is tangential to an observer s line of sight at roughly l = 336 (Bronfman 2008; see also Vallée 2008 who argues that this feature lies at l 339 and is the start of the Perseus arm, distinct from the Norma 3 kpc), corresponding to R = 3.5 kpc. Also, it has been suggested that the 3 kpc arm must be confined to an annulus of less than 1 kpc in extent (Lockman, 1981), which corresponds well with the radial concentration of clouds. At this time, however, the possibility that these clouds are instead related to other Galactic structures such as the 5 kpc molecular ring (Jackson et al., 2004) cannot be ruled out. This ring is likely not a coherent structure but instead a complex region where multiple spiral arms originate (Vallée, 2008). The apparent association between many of the H I clouds and filamentary structures within the GASS pilot region suggests that the clouds are related to star formation because such structures are common in areas of significant supernova activity or stellar winds (Dickey & Lockman, 1990). Figure 3.25 displays an integrated intensity map with many clouds that are clearly aligned along filaments and loops, reminiscent of the clouds observed to be associated with the cap of a superbubble by McClure-Griffiths et al. (2006). Recently, proper motion measurements of the molecular cloud NGC 281 West, which is associated with an H I loop, revealed that the cloud is moving away from the Galactic plane at km s 1 (Sato & et al., 2007). This velocity is similar to our observed velocity dispersion and further

99 3.6. THE ORIGIN AND NATURE OF HALO CLOUDS 81 supports the scenario that the clouds are related to expanding shells. In this scenario, violent supernovae and stellar winds may have pushed H I from the disk up into the halo or the clouds may be fragments of H I shells (Mac Low et al., 1989; McClure-Griffiths et al., 2006), rather than a result of a standard galactic fountain. In the galactic fountain model hot gas rises from the disk, cools and condenses, then falls back to the plane (Shapiro & Field, 1976; Bregman, 1980). This would result in a more uniform radial distribution of clouds (Bregman, 1980), while we clearly observe a peak in the radial distribution of the clouds that may be associated with the 3 kpc arm. If the halo clouds are related to star formation, the asymmetry in the number of detected clouds at low heights ( z 750 pc) could be a result of this, as any asymmetry in the structure of the ISM and the location of star forming regions may be reflected in the distribution of clouds, and there appears to be more filaments below than above the Galactic plane in the GASS pilot region. If the clouds are related to spiral structure and star formation then we would expect to see a correlation between the radial surface density distribution of the H I clouds and that of Galactic H II regions. We compare these distributions, along with the mass surface densities of H I and H 2, in Figure The mass surface densities have been averaged over the entire Galaxy and were derived by Dame (1993) using H I data from Dickey & Lockman (1990) and Burton & Gordon (1978), and H 2 data from Bronfman et al. (1988), whereas the halo cloud distribution from this study includes only the GASS pilot region. The H II regions are taken from Paladini et al. (2004). There is no obvious relationship between the H I cloud and H II region distributions or between the H I clouds and the H I and H 2 surface densities, providing conflicting evidence for the relationship between the H I clouds and current star formation. As the evaporation timescale for a cold cloud in a hot medium is expected to be much longer than other timescales associated with cloud evolution (Cowie & McKee, 1977; Nagashima et al., 2006), and if evaporation is the main disruptive mechanism, the clouds are long lived. It is therefore possible that the clouds are tracing past rather than current star formation. With the present data we are only able to make a preliminary study of the relationship between the halo cloud distribution and spiral structure and star formation in the Galaxy.

100 82 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Figure 3.25 Filamentary structures are plentiful within the GASS pilot region, as demonstrated in this longitude latitude integrated intensity map over 135 V LSR 120 km s 1. Many H I clouds in the GASS pilot region are aligned with these loops and filaments, suggesting that they are related to expanding superbubbles or other structures common in areas of star formation. The diagonal striations are instrumental artefacts.

101 3.6. THE ORIGIN AND NATURE OF HALO CLOUDS 83 Figure 3.26 Radial surface density distribution of the simulated GASS pilot region H I clouds (solid histogram) and H II regions (scaled up by a factor of 7 for ease of comparison and shown by the dash-dot histogram; Paladini et al. 2004). Mass surface densities of the average Galactic H I (dotted line) and Galactic H 2 (dashed line) are overlaid and were both determined by Dame (1993) using data from Dickey & Lockman (1990) and Burton & Gordon (1978) for H I and data from Bronfman et al. (1988) for H 2. Arrows indicate limits of R for tangent points in the GASS pilot region. The peaked distribution of halo clouds does not appear to be similar to that of the H II regions nor the H I and H 2 mass surface densities, suggesting that the halo clouds are not directly related to current star formation.

102 84 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Possible Association with High Velocity Cloud Complex L An excess of clouds at large positive heights can be seen in the vertical distribution of the clouds (Figure 3.17); 9% of the clouds at positive heights lie at z 2 kpc while none of the clouds below the plane are seen at such heights, and many of them have unusually large deviation velocities (Figure 3.7). This excess could have several possible origins including infalling gas, increased disk activity on one side of the disk that has resulted in outflowing gas reaching larger heights, or small number statistics. Given the proximity of high velocity cloud complex L to the halo clouds in the upper portion of the GASS pilot region, we compared the H I associated with each population to determine whether or not the presence of complex L could be responsible for the observed excess. Complex L was first described by Wakker & van Woerden (1991) to have velocities in the range 190 V LSR 85 km s 1, longitudes 341 l 348, and latitudes 31 b 41, and they speculated that the clouds were part of a population that was related to a galactic fountain. Since then, Hα distance limits have been determined for complex L, placing it within the Galactic halo at heights of 4 z 12 kpc from the plane and heliocentric distances of 8 to 22 kpc (Weiner et al., 2001; Putman et al., 2003). In Figure 3.27 we display a region of the GASS data that encompasses both the upper GASS pilot region, outlined by the solid black lines, and the lower velocity gas of complex L. We have overlaid circles on high velocity clouds associated with complex L as catalogued by Wakker & van Woerden (1991), regardless of their velocity. The spatial morphology of the H I at velocities at which gas is detected in complex L suggests that there is a connection between the clouds in the GASS pilot region and those in complex L, in the form of a filamentary structure that is contiguous in velocity and connects complex L with the disk. Also, with the assumption that the halo clouds at large positive latitudes are located at tangent points, their heights are in the range 1.5 z 2.5 kpc with distances 8 kpc, which place them in the vicinity of the lower height estimates of complex L. These correlations suggest that complex L may have similar origins as the clouds presented here and it may be responsible for the observed excess of clouds at large positive latitudes.

103 3.6. THE ORIGIN AND NATURE OF HALO CLOUDS 85 Figure 3.27 GASS data encompassing the upper portion of the GASS pilot region, outlined by the solid black lines, along with the high velocity cloud complex L. Circles have been placed at the positions of high velocity clouds within complex L, regardless of their V LSR, as determined by Wakker & van Woerden (1991). The morphology of the H I in the upper GASS pilot region, in complex L, and between them, suggests that complex L may have similar origins as the observed halo clouds and may be responsible for the observed excess of clouds at large positive latitudes.

104 86 CHAPTER 3. H I CLOUDS IN THE LOWER HALO: QUADRANT IV Stability of Halo Clouds The amount of mass required for a spherical cloud to be gravitationally bound is M r v 2 /G, where r is the radius of the cloud, v is the linewidth of its velocity profile and G is the gravitational constant. For a cloud with r = 32 pc and v = 12.8 km s 1, the median observed values, this would require a mass on the order of 10 6 M. As none of the detected clouds have masses this large, they are either pressure confined or transitory. The total thermal plus turbulent pressure is equal to 22 n v 2 and we can therefore place a lower limit on its value, but this does not include additional components such as magnetic and cosmic ray pressures, which are thought to contribute most of the total pressure (Cox, 2005). Pressure changes may be responsible for the observed trend in Figure 3.20, which tentatively shows that clouds farther from the disk of the Galaxy have larger linewidths. If the clouds are in pressure equilibrium, their pressures could provide us with insight into the pressure structure of the halo, which is currently not well understood. Pressure equilibrium would seem unlikely, however, given that the linewidths of the clouds may increase with z. 3.7 Summary We have detected over 400 H I clouds in the lower halo of the Galaxy in the Galactic All-Sky Survey pilot region. These clouds have a median peak brightness temperature of 0.6 K, a median velocity width of 12.8 km s 1, and angular sizes 1. As these clouds follow Galactic rotation, a subset was selected that is likely to be located at tangent points of the inner Galaxy, allowing us to determine their distances and therefore their sizes and masses. The tangent point clouds have radii on the order of 30 pc and a median H I mass of 630 M. The properties of these clouds suggest that they belong to the same population of clouds discovered by Lockman (2002). We simulated the population of clouds to constrain their random cloud-to-cloud velocity dispersion, σ cc, and spatial distribution. We found that σ cc = 18 km s 1, but if the clouds were left to evolve in the Galactic potential without any disruptions, these random motions would produce a scale height of 90 pc, which is inconsistent with our derived scale height of 400 pc. This suggests that the clouds do not belong to an equilibrium population. We detected clouds throughout the entire GASS pilot

105 3.7. SUMMARY 87 region, up to the latitude boundaries ( b 20 ). Few clouds were observed at low latitudes due to confusion, which may have resulted in an underestimate of the number of clouds at low heights, and therefore an overestimate in the derived scale height. Our large, homogeneously selected sample has allowed us to determine the spatial distribution of these halo clouds for the first time and has revealed that although clouds were observed at all longitudes within the GASS pilot region, they do not appear to be uniformly distributed but instead are concentrated in radius, peaking at R = 3.8 kpc. We analysed this distribution and suggest that the clouds are related to the spiral structure of the Galaxy. In particular, the peak in the radial distribution is suggestive of a relation to the 3 kpc arm. This relation to a specific spiral feature remains speculative until further analysis using a larger sample of clouds and better constrained spiral structure models can be performed. It is therefore unlikely that the halo clouds are a result of a standard galactic fountain, as radial enhancements would not be expected in this scenario (Bregman, 1980). Instead, it appears that the clouds may be directly related to areas of active star formation, in the form of fragmenting H I shells and H I gas that has been pushed into the halo. This possibility is further supported by the appearance of numerous clouds related to filaments and loops, whose structures may have resulted from stellar winds and supernovae (Dickey & Lockman, 1990). However, a comparison between the radial surface density distribution of H I clouds and H II regions provides conflicting evidence: if clouds are related to areas of star formation, a relationship between the clouds and H II regions would be expected but is not observed. The morphology of H I at large positive latitudes in the GASS pilot region suggests that some of the clouds may be related to high velocity cloud complex L, whose lower height estimates of 4 kpc and distance estimates of 8 kpc place it in the vicinity of the clouds presented here. If they are related, whatever process is responsible for the halo clouds, e.g., star formation, may also be responsible for complex L. This is further supported by recent observations of IVCs, which suggest that the IVCs are related to energetic events in the Galactic disk and that they are likely linked to spiral structure (Kerton et al., 2006). The majority of those IVCs have V dev 20 km s 1, implying a Galactic origin, and we suggest that they are similar to the cloud population presented here.

106

107 Chapter 4 H I Clouds in the Lower Halo: Quadrant I We have detected 255 H I clouds in the lower Galactic halo that are located near tangent points within a region of the first Galactic quadrant. This region is symmetric to the GASS pilot region of the fourth quadrant about the line of sight to the Galactic centre. We present the observed and derived properties of individual clouds and an analysis of their distribution in the Galaxy. The physical properties of the clouds detected in the first and fourth quadrants are quite similar, suggesting that not only do they belong to the same population, but are likely at similar stages in their evolution. We constrain the spatial distribution of the first quadrant clouds with an analysis of a simulated population, and determine that their cloud-to-cloud velocity dispersion is 14.5 km s 1. The vertical distribution is well represented by an exponential with a scale height of 800 pc, while radially the clouds appear to be somewhat evenly distributed. 4.1 Introduction In Chapter 3 we have shown that H I clouds are abundant in the lower halo of the fourth longitude quadrant of the inner Galaxy. It is important to expand the sample to include halo clouds in other regions of the Galaxy to determine whether their properties and distribution are dependent on location. Also, the larger sample provided by a larger search volume will drastically improve our knowledge of their properties and distribution. Ultimately, through the comparison of data in the 89

108 90 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I different regions, we can determine what environmental factors play a role in their origin and evolution, e.g., areas of star formation, Galactic structures such as spiral arms and superbubbles, or a galactic fountain. We previously focused on clouds within the GASS pilot region, a region of the fourth quadrant of the inner Galaxy, where we detected over 400 clouds, 81 of which are located near tangent points. Here, we complement those data by analysing a region of the first quadrant of the Galaxy that is symmetric to the GASS pilot region about the Galactic centre, covering the same range of radii and heights. Within this first quadrant region, we concentrate solely on clouds that we are confident are located near tangent points. For these, an abundance of information including their physical properties and spatial distribution can be determined, as there is no distance ambiguity when assuming that their motions are governed by pure Galactic rotation. We begin with a discussion of the first quadrant data used ( 4.2). We then present the observed and derived properties of the tangent point clouds detected within the first quadrant region ( 4.3 and 4.4), and include a brief investigation of these properties. This is followed by a discussion of a simulated population of clouds used to better constrain distance errors and to better characterise both the spatial distribution of the clouds and their kinematics ( 4.5). Next, the physical properties of the clouds detected within the first quadrant region and the GASS pilot region (in the fourth quadrant) are compared ( 4.6) and a summary is presented in The Quadrant I Region A region of the first quadrant of the Galaxy that spans 16.9 l 35.3 and b 20 was searched for H I clouds within the lower Galactic halo. This region is completely symmetric about l =0 to the GASS pilot region, i.e., is on the opposite side of the Galaxy and is equidistant from the Galactic centre, spanning the same Galactocentric radii and heights (along the locus of tangent points). A longitude latitude diagram of this region at V LSR = 102 km s 1 is presented in Figure 4.1. As was seen in the fourth quadrant, there are many clouds above and below the plane. The lack of data in the top, left corner of the first quadrant region is due to the declination limit of GASS data (δ < 1 ).

109 4.2. THE QUADRANT I REGION 91 Figure 4.1 Longitude latitude diagram at V LSR = 102 km s 1 of the region of the first Galactic quadrant searched for tangent point clouds. There are many H I clouds in the lower halo within the first quadrant, as there were in the fourth quadrant. This region was chosen as the symmetric counterpart of the GASS pilot region and spans the same Galactocentric radii and heights along the locus of tangent points. The curved lines represent the latitude boundaries of the region searched. The lack of data in the top, left corner is due to the declination limit of GASS; there are no data at declinations higher than δ =1.

110 92 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I 4.3 Observed Properties of H I Clouds We searched for H I clouds located near tangent points within a region of the first quadrant of the Galaxy using the identical selection criteria imposed for the fourth quadrant data: (1) clouds must span 4 or more pixels and be clearly visible over three or more channels in the spectra and (2) clouds must be distinguishable from unrelated background emission. See for further details on the selection criteria and search method. We detect 255 tangent point H I clouds, where tangent point clouds in the first quadrant are defined as those clouds with V dev 0.8 km s 1 (where 0.8 km s 1 accounts for one channel width). The observed properties of the tangent point clouds are presented in Table 4.1 and were determined analogously to those in Table 3.1 (see for property descriptions and for uncertainty descriptions). Sample spectra are shown in Figure 4.2.

111 4.3. OBSERVED PROPERTIES OF H I CLOUDS 93 Table 4.1: Observed Properties of Tangent Point HI Clouds in the Quadrant I Region l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

112 94 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Table 4.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

113 4.3. OBSERVED PROPERTIES OF H I CLOUDS 95 Table 4.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

114 96 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Table 4.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

115 4.3. OBSERVED PROPERTIES OF H I CLOUDS 97 Table 4.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

116 98 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Table 4.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

117 4.3. OBSERVED PROPERTIES OF H I CLOUDS 99 Table 4.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

118 100 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Table 4.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Continued on Next Page...

119 4.3. OBSERVED PROPERTIES OF H I CLOUDS 101 Table 4.1 Continued l b VLSR Tpk a v NHI θmin θmaj b MHId 2 c (deg) (deg) (km s 1 ) (K) (km s 1 ) ( cm 2 ) (arcmin arcmin) (M kpc 2 ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Notes. Observed properties were determined analogously to those described in a Uncertainties in T pk are 0.07 K. b Uncertainties in the maximum angular extents are dominated by background levels surrounding the cloud and are assumed to be 25% of the estimated values. c Mass uncertainties are dominated by the interactive process used in mass determination and are assumed to be 40% of the estimated values.

120 102 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Figure 4.3 presents the longitude vs. V LSR of all clouds from the tangent point sample within the first quadrant region, along with the terminal velocity curve of McClure-Griffiths & Dickey (in preparation) for all l>19 and of Clemens (1985) for all remaining longitudes, where V t was derived from H I and CO observations, respectively. Although we searched for clouds between the terminal velocity curve to V LSR = 300 km s 1, there were no clouds detected at V LSR > 160 km s 1. No cloud has a velocity 50 km s 1 beyond that allowed by Galactic rotation, and as was seen in Figure 3.6 for clouds within the fourth quadrant, there is a sharp cutoff in the number of clouds beyond the terminal velocity (at more positive velocities in the first quadrant), indicating that the motions of these clouds are governed by Galactic rotation. Histograms of the peak brightness temperature, T pk, FWHM of the velocity profile, v, and angular size of the tangent point clouds are presented in Figures 4.4, 4.5, and 4.6. The median values are 0.5 K, 10.6 km s 1, and 25, respectively. The cutoff at low T pk is due to the sensitivity limit of the data. The FWHM of the clouds is broadly peaked around the median value, and most profiles are fully resolved. If the internal motions were entirely thermal, the FWHM would correspond directly to the temperature. It is likely that turbulent motions also play a role, in which case the FWHM provides an upper limit to the thermal temperature, T. For the median v of 10.6 km s 1 this upper limit is T 1230 K. The rapid decrease in the number of clouds towards low angular sizes is likely due to the spatial resolution limit of the data and suggests that many of the clouds are unresolved. We further discuss the properties of the first quadrant clouds in 4.6, particularly in comparison with the properties of halo clouds detected within the fourth quadrant data. 4.4 Derived Properties of H I Clouds In Table 4.2 we present the derived properties that rely on the assumption that the clouds are located at tangent points. These properties were determined analogously to those for the GASS pilot region clouds (as described in 3.4.1). Errors were also determined analogously (see 3.4.2). Here, δv t is estimated to be 3 km s 1 for all clouds at l>19, where the terminal velocities were determined from H I observations by McClure-Griffiths & Dickey (in preparation), and 9 km s 1 for longitudes where the terminal velocities were derived from CO observations by Clemens (1985).

121 4.4. DERIVED PROPERTIES OF H I CLOUDS 103 Figure 4.2 Spectra from a random sample of H I clouds from Table 4.1. Many clouds appear to be sitting on broad spectral wings. Arrows mark the velocity where the brightness temperature peaks (after background subtraction) for each profile.

122 104 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Figure 4.3 Longitude velocity diagram of clouds with V LSR V t (crosses) within the first quadrant region. The terminal velocity curve (solid line) was derived from H I observations by McClure-Griffiths & Dickey (in preparation) for l>19 and from CO observations by Clemens (1985) for the remaining longitudes. There is a sharp cutoff in the number of clouds with velocities beyond the terminal velocity (at more positive velocities), indicating that the motions of these clouds are dominated by Galactic rotation. We searched for all clouds with V LSR 300 km s 1 and find none with V LSR > 160 km s 1.

123 4.4. DERIVED PROPERTIES OF H I CLOUDS 105 Figure 4.4 Histogram of peak brightness temperature of the tangent point clouds. The median T pk =0.5 K while the lower cutoff is due to the sensitivity limit. The arrow represents the 5 T b detection level. We assume N errors.

124 106 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Figure 4.5 Histogram of the FWHM of the velocity profile of the tangent point clouds. The median value is 10.6 km s 1 and most profiles are well resolved. We assume N errors.

125 4.4. DERIVED PROPERTIES OF H I CLOUDS 107 Figure 4.6 Histogram of the angular size of the clouds, (θ maj θ min ) 1/2, where θ maj and θ min are from Table 4.1. The median angular diameter of the clouds is 25. The spatial resolution limit (16, noted with the arrow) of the data results in the rapid decline in the number of clouds with smaller sizes and is represented by the arrow. We assume N errors.

126 108 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I The errors for V t quoted by Clemens (1985) are less than this, so the main source of error is not the observations themselves, but rather the differences between how CO and H I trace the terminal velocity. We assume that these are the same in the first and fourth quadrants, and use the 9 km s 1 calibrated from the comparison between Luna et al. (2006) and McClure-Griffiths & Dickey (2007) data. Where error estimates were dependent on distance, a simulated population of clouds was used to estimate these effects (see for further details).

127 4.4. DERIVED PROPERTIES OF H I CLOUDS 109 Table 4.2: Derived Properties of Tangent Point HI Clouds in the Quadrant I Region l b VLSR Vdev d R a z r MHI (deg) (deg) (km s 1 ) (km s 1 ) (kpc) (kpc) (kpc) (pc) (M ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 150 Continued on Next Page...

128 110 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Table 4.2 Continued l b VLSR Vdev d R a z r MHI (deg) (deg) (km s 1 ) (km s 1 ) (kpc) (kpc) (kpc) (pc) (M ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 5 62 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 4 56 ± ± ± ± ± ± ± ± ± ± ± ± 5 80 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 600 Continued on Next Page...

129 4.4. DERIVED PROPERTIES OF H I CLOUDS 111 Table 4.2 Continued l b VLSR Vdev d R a z r MHI (deg) (deg) (km s 1 ) (km s 1 ) (kpc) (kpc) (kpc) (pc) (M ) ± ± ± ± ± 5 90 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 3 23 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 90 Continued on Next Page...

130 112 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Table 4.2 Continued l b VLSR Vdev d R a z r MHI (deg) (deg) (km s 1 ) (km s 1 ) (kpc) (kpc) (kpc) (pc) (M ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 7 63 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 280 Continued on Next Page...

131 4.4. DERIVED PROPERTIES OF H I CLOUDS 113 Table 4.2 Continued l b VLSR Vdev d R a z r MHI (deg) (deg) (km s 1 ) (km s 1 ) (kpc) (kpc) (kpc) (pc) (M ) ± ± ± ± ± ± ± ± ± ± ± 7 55 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 400 Continued on Next Page...

132 114 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Table 4.2 Continued l b VLSR Vdev d R a z r MHI (deg) (deg) (km s 1 ) (km s 1 ) (kpc) (kpc) (kpc) (pc) (M ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 100 Continued on Next Page...

133 4.4. DERIVED PROPERTIES OF H I CLOUDS 115 Table 4.2 Continued l b VLSR Vdev d R a z r MHI (deg) (deg) (km s 1 ) (km s 1 ) (kpc) (kpc) (kpc) (pc) (M ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 220 Continued on Next Page...

134 116 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Table 4.2 Continued l b VLSR Vdev d R a z r MHI (deg) (deg) (km s 1 ) (km s 1 ) (kpc) (kpc) (kpc) (pc) (M ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 600 Continued on Next Page...

135 4.4. DERIVED PROPERTIES OF H I CLOUDS 117 Table 4.2 Continued l b VLSR Vdev d R a z r MHI (deg) (deg) (km s 1 ) (km s 1 ) (kpc) (kpc) (kpc) (pc) (M ) ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± 600 Notes. Derived properties of the first quadrant HI clouds (those that rely on the assumption that the clouds are located at tangent points). a Along a given line of sight, the smallest Galactocentric radius possible is at the tangent point. If the cloud is not located at the tangent point it must be farther away from the Galactic centre and the error on R must be positive.

136 118 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Histograms of the radius and mass of the clouds are presented in Figures 4.7 and 4.8. The median radius is 28 pc while the median mass is 700 M. It is likely that the radii values are overestimates, as a large fraction of clouds appear to be unresolved (leading to larger measured angular extents) and the cutoff at small radii is likely due to both the spatial resolution limit and effects due to blending. If confusion is important, the background subtraction method, where a background based on the H I emission in the area surrounding each cloud was subtracted from the integrated intensity map, may result in higher M HI estimates. Less massive clouds are less likely to have been identified during cataloguing as they could be obscured by higher mass clouds. It would be interesting to estimate the volume of phase space over which clouds of different masses could be identified. As a rough estimate, we can assume that whenever the positions of two clouds are closer than a typical cloud diameter and their V LSR values are closer than a typical cloud velocity width, the less massive cloud could be missed during the cataloguing process. In fact, the likelihood of being identified is not strictly a function of mass, but rather a function of the peak flux. Clouds of the same mass with narrower line widths and smaller diameters are therefore easier to detect. However, the range of observed v and D values are sufficiently small that we can neglect this complication for the sake of this simple estimate. We determine the cumulative fraction of clouds in the first quadrant catalogue (where we have better statistics) that have V LSR values within the median FWHM of 10.6 km s 1 and angular separations less than the median diameter of 25 arcminutes, as a function of the mass of the larger cloud. This gives an estimate of the total incompleteness as a function of cloud mass, and is roughly 7% for clouds with masses up to 1000 M, rising to 18% for the lowest mass clouds in the sample. We therefore conclude that most clouds at all masses should be detected and that the completeness is not a strong function of mass, so the distribution of cloud number versus mass is not strongly biased above the detection limits. 4.5 Analysis of Quadrant I Cloud Population Simulated Population of Clouds As demonstrated in 3.5 for the fourth quadrant, by using a simulated population of clouds the intrinsic characteristics of an observed population can be determined.

137 4.5. ANALYSIS OF QUADRANT I CLOUD POPULATION 119 Figure 4.7 Histogram of physical radius of the tangent point clouds. The median radius is 28 pc and there is a clear decline in the number of clouds below the spatial resolution limit of 16. We assume N errors.

138 120 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Figure 4.8 Histogram of the H I mass of the tangent point clouds. The median mass is 700 M. We assume N errors.

139 4.5. ANALYSIS OF QUADRANT I CLOUD POPULATION 121 This includes acquiring estimates of the errors introduced by the assumption that the clouds are located at tangent points, how selection effects affect the results, and how to better characterise the distribution. By determining what functions best represent the observed data, an understanding of the cloud population that is otherwise unattainable is achieved. We therefore simulated a population of clouds to represent the observed tangent point population within the first quadrant, applying the same l, b and V LSR selection criteria as for the observed population. A δ 2 limit was also applied to account for the declination limit of GASS. This cutoff does not affect the determined spatial distribution of the clouds and solely affects the total number of tangent point clouds in the given region, as the volume searched is smaller in the first quadrant than in the fourth by 15%. To randomly sample the simulated clouds, the same form for the distribution was used as that for the fourth quadrant: ( n(r, z) = Σ(R) exp z ), (4.1) h where Σ(R) is the radial surface density distribution, h is the exponential scale height and R and z are the cylindrical coordinates. Σ(R) is composed of 16 independent radial bins of width 0.25 kpc, spanning R =2.5 to 6.5 kpc, and is not constrained to be same as Σ(R) determined from the fourth quadrant data. The amplitude of each radial bin was optimised to best fit the observed longitude distribution of the tangent point clouds by minimising the Kolmogorov-Smirnov (K-S) D statistic (the maximum deviation between the cumulative distributions) using Powell s algorithm (Press et al., 1992). The fits were optimised using three different initial estimates on Σ(R), and as they all converged on a similar solution, we adopted the mean of the three as the best fit to the data. The resulting radial surface density distribution is shown in Figure 4.9. The velocities of the simulated clouds were based on a flat rotation curve where Θ =Θ 0 with a random velocity component drawn from a Gaussian clouds were generated in a half-galaxy (first and second quadrants), of which 1954 clouds were within the defined l, b, V LSR, and δ range of the first quadrant clouds. This population was then normalised to compare directly with the observed population and K-S tests were performed to estimate the quality of the fit between the observed and simulated distributions. These tests show that the parameters of the simulated population reproduce the distribution of observed tangent point clouds in the first

140 122 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Figure 4.9 Radial surface density distribution of the simulated population of clouds. The black error bars represent the Poisson error in the number of clouds expected to fall into the observed sample, which determined the ability of the observed data to constrain the simulated distribution, while the grey error bars represent the range in values for the 3 different starting points used in the minimisation. Error bars are large in the right-most bins because very few clouds at these radii have high enough random velocities to be scattered into our sample, so the statistics are particularly poor. Some bins do not have black error bars because no clouds within the observed data are expected to lie within these bins and so the surface densities in these bins are unconstrained by the data. Arrows represent the tangent points at the longitude boundaries of the first quadrant region.

141 4.5. ANALYSIS OF QUADRANT I CLOUD POPULATION 123 quadrant well and are therefore good estimates for those of the intrinsic population. Results of the fits to the distributions are presented in To determine the errors introduced by the assumption that all clouds with V LSR V t are located at tangent points, we calculated the fractional distance error of the simulated clouds as a function of deviation velocity, longitude and latitude, and find that, as for the fourth quadrant data, clouds in the first quadrant with increasingly forbidden velocities have smaller distance errors. There is also a slight dependence on longitude, but no dependence on the latitude for distance errors. Likewise, there is a longitude and deviation velocity dependence for the Galactocentric radius errors. We implemented these errors in Table 4.2 in the same manner discussed in Confusion effects are stronger the closer the cloud is to the bulk of the Galactic disk emission, i.e., at lower heights and lower values of V LSR. This can be seen clearly in Figure 4.1 and manifests itself in Figure 4.10, which shows how the heights of clouds are related to their deviation velocities, as the triangular region near the Galactic plane that is devoid of detected clouds. To account for these effects, we have parameterised the boundaries of the confusion-affected region by the dashed lines shown in Figure 4.10, and apply these cutoffs to both the observed and simulated data to compare the two as directly as possible. This is in contrast to how we addressed these effects with the fourth quadrant data, where we simply omitted all clouds with b 2 ; the number of tangent point clouds in the fourth quadrant was too small to justify a more complicated cutoff, and the morphology of the bulk of the Galactic disk emission and therefore the region where confusion becomes important may be different between the two quadrants. In the following sections we present an analysis of the first quadrant cloud population, deferring a discussion of the implications of these findings along with a detailed comparison between the first and fourth quadrant clouds until Chapter Cloud-to-Cloud Velocity Dispersion The simulated population of clouds that best represents the observed first quadrant population has a random cloud-to-cloud velocity component drawn from a Gaussian of dispersion σ cc = 14.5 km s 1. The resulting histogram of deviation velocities for both the observed and simulated populations is presented in Figure The distributions are quite similar, having a K-S test probability of 51% of both being drawn from the same distribution. Values of σ cc between 14 and 15.5 km s 1 also

142 124 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Figure 4.10 Vertical distance from the Galactic plane as a function of deviation velocity of observed clouds. Dashed lines represent the cutoffs that were applied to the simulations to reflect the effects of confusion, where clouds closer to the disk and clouds with lower deviation velocities were harder to distinguish.

143 4.5. ANALYSIS OF QUADRANT I CLOUD POPULATION 125 Figure 4.11 Histogram of deviation velocities for observed first quadrant (solid line) and simulated (dashed line) clouds, where the simulated population has a cloud-tocloud velocity dispersion of σ cc = 14.5 km s 1. We assume N errors. provide acceptable fits to the deviation velocity distribution of the observed data, having K-S test probabilities 15% Radial Distribution The distribution of Galactocentric radii for the observed and simulated tangent point population of clouds within all l, b, V LSR, and δ cutoffs is shown in Figure The shift in the simulation to higher R is due to the assumption that all clouds with V LSR V t are located at tangent points, as the tangent point occurs at the minimum possible Galactocentric radius along a given line of sight. As selection effects are present, the intrinsic distribution of the observed clouds is best determined by looking at the radial surface density distribution presented in Figure 4.9. The longitude distribution of both the observed and simulated population of clouds, along with that derived from a population of clouds with a uniform sur-

144 126 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Figure 4.12 Distribution of Galactocentric radii of the best-fit simulated (dashed line) and observed (solid line) population of clouds. The offset between the two distributions is due to the assumption that all clouds with V LSR V t are located at tangent points, i.e., at the smallest possible Galactocentric radius along that line of sight. In fact, because of clouds random motions, some clouds at Galactocentric radii larger than that at the tangent point will have V LSR >V t and thus fall in our sample. The simulation can account for this effect. The arrows represent the tangent points at the longitude boundaries of the region searched. We assume N errors.

145 4.5. ANALYSIS OF QUADRANT I CLOUD POPULATION 127 Figure 4.13 Longitude distribution of observed (solid line) and simulated (dashed line) clouds, with a uniform surface density population overlaid (dotted line). The uniformly distributed population of clouds resembles the observed population except at the highest longitudes, suggesting that the tangent point clouds within the first quadrant region are somewhat evenly distributed. We assume N errors. face density is shown in Figure The K-S test probability that the observed and simulated distributions were drawn from the same distribution is 88%. It is not surprising that the simulated distribution has such an unusually high K-S test probability as the parameters of the simulation were fine-tuned to reproduce the observed longitude distribution. The K-S test probability that the observed and uniformly distributed population were drawn from the same distribution was 3%, so the observed population is marginally inconsistent with being drawn from a population of clouds with a uniform surface density, but most of the discrepancy comes from the highest longitude bins. Note that the distribution of longitudes for the uniform population at l 30 declines because of the declination limit imposed.

146 128 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Vertical Distribution The vertical distribution of the first quadrant tangent point clouds is best represented by a simulated distribution with an exponential scale height, h = 800 pc (see Figure 4.14). Because K-S tests are most sensitive to the differences near the median of the distribution, we tested both the b distributions of the populations, which weights clouds closer to the plane more highly, and the b distributions, which gives more weight to the higher latitude clouds. Also, the comparison of the b distributions is sensitive to symmetries about the plane while the comparison of the b distributions is not. For h = 800 pc, the K-S test probability that the latitude distributions of the observed and simulated clouds were drawn from the same distribution is 61% when comparing the b distributions but was not acceptable when comparing the b distributions. For the b distributions, scale heights between 700 and 850 pc were acceptable, while scale heights between 950 and 1100 pc were consistent for the b distributions (with 1000 pc being the best fit, having a 24% probability of being drawn from the same distribution). This discrepancy likely indicates that our confusion cutoff (see Figure 4.10) is not quite conservative enough and results in slightly too many clouds very near the Galactic plane. This has a stronger effect when comparing the distributions of b rather than of b, and we therefore adopt the preferred value from the latter comparison, h = 800 pc. 4.6 Physical Properties and Trends Physical Properties: Quadrant I vs. IV The physical properties of all H I clouds in the lower halo that have been detected within GASS data are summarised in Table 4.3. The clouds in both quadrants of the Galaxy are strikingly similar, which suggests that both samples of GASS clouds are from the same population of clouds and likely have similar origins and evolutionary histories. If they were not similar, it could suggest that their origin, evolutionary stage, or even the very nature of the clouds detected in the two regions were different Trends: Quadrant I vs. IV Here we briefly analyse some of the basic properties of the H I clouds in the first quadrant to search for trends in the data that may provide insight into the origin and

147 4.6. PHYSICAL PROPERTIES AND TRENDS 129 Figure 4.14 Vertical distribution of the observed first quadrant (solid line) and simulated (dashed line) population of clouds. The observed population is well represented by a simulated population with an exponential scale height of 800 pc. Note that inclusion of a simulated confusion cutoff (Figure 4.10) does a reasonable job of reproducing the decrease of clouds near the plane. We assume N errors.

148 130 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Table 4.3. Property Summary of H I Halo Clouds in GASS GASS Pilot Region GASS Quadrant I Region Parameter Median 90% Range Median 90% Range Tpk (K) v (km s 1 ) NHI ( cm 2 ) r (pc) 32 < < MHI (M ) z (pc) σcc (km s 1 ) Note. Median values of the observed halo cloud properties, where the number of clouds, n = 403 for Tpk, v, and NHI and n = 81 for remaining properties (of tangent point clouds) for GASS pilot region data, and n = 255 for clouds in the first quadrant region of GASS. Most properties have a large scatter about the median in all samples, as demonstrated by the 90% range.

149 4.6. PHYSICAL PROPERTIES AND TRENDS 131 Figure 4.15 FWHM of the velocity profile, v, of the first quadrant tangent point clouds as a function of height. Clouds at larger heights have larger line widths. evolution of the clouds. There is strong evidence that clouds in the first quadrant that are farther from the Galactic plane have larger line widths, as seen in Figure This strengthens our suggestion from that the fourth quadrant clouds demonstrated this trend due to our increased sample. Given the continuous nature of the trend, it does not appear to be due to separate populations of clouds (one close to the disk, the other at larger heights), as we hypothesised in 3.5.8, but instead supports our other hypothesis that they may indeed be reflecting pressure variations throughout the halo. However, until higher resolution data are available, which will enable tight constraints to be placed on the cloud densities, we are unable to investigate this trend further. As was seen for the fourth quadrant clouds, we see trends in both N HI and M HI with z, and a trend in M HI with V dev for the first quadrant clouds, where N HI and M HI decrease with z (Figures 4.16 and 4.17), and M HI decreases with V dev (Figure 4.18). The increased sample strengthens our suggestion from that the

150 132 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Figure 4.16 H I column density as a function of height. Clouds at lower heights may have larger H I column densities, but this trend may be a result of confusion effects. fourth quadrant clouds also demonstrated this trend and supports a scenario where each cloud was given the same kinetic energy during their formation process or via equipartition in their evolution, as we described in detail in Again, however, we cannot exclude the possibility that at lower heights confusion is affecting the determined N HI and M HI values, which in turn would affect these trends. 4.7 Summary GASS data were searched for H I clouds in the lower halo that are located at tangent points within a region of the first Galactic quadrant that is directly symmetric with the GASS pilot region of the fourth quadrant. We detected 255 H I clouds, which exhibit properties similar to those of the fourth quadrant, with a median T pk =0.5K, v = 10.6 km s 1, r = 28 pc and M HI = 700 M. To better constrain the distribution of the clouds in the first quadrant, we sim-

151 4.7. SUMMARY 133 Figure 4.17 H I mass of the clouds as a function of height (crosses). The horizontal lines represent the median M HI in bins each containing 25 clouds and the vertical error bars represent the 25 th to 75 th percentile range. As was the case for clouds in the fourth quadrant, there is no evidence for a trend until z 1 kpc, where there may be a lack of more massive clouds at larger heights.

152 134 CHAPTER 4. H I CLOUDS IN THE LOWER HALO: QUADRANT I Figure 4.18 H I mass of clouds as a function of deviation velocity (crosses). The horizontal lines represent the median M HI in bins each containing 25 clouds and the vertical error bars represent the 25 th to 75 th percentile range. The more massive clouds may have lower deviation velocities.

153 4.7. SUMMARY 135 ulated a population of clouds to which we applied the same selection criteria in l, b, V LSR and δ as for the observed sample of tangent point clouds. The simulations reveal that the observed population is well represented by a population of clouds with a random cloud-to-cloud velocity dispersion of 14.5 km s 1. The vertical distribution of the clouds is best fit by an exponential distribution with a scale height of 800 pc, and the population is distributed with a surface density that is roughly constant with Galactocentric radius. The physical properties of the H I clouds from the first and fourth quadrants were compared and it was found that they are strikingly similar, suggesting that the clouds belong to the same population and may have originated from similar environments. In Chapter 5 we will compare the kinematics and spatial distribution of the first and fourth quadrant clouds and explore implications for their origin and evolution in detail.

154

155 Chapter 5 H I Halo Clouds: A Galactic Population Hundreds of H I halo clouds have been detected in two symmetric regions of the Milky Way: the GASS pilot region in the fourth quadrant and its counterpart in the first quadrant, revealing that the clouds are abundant throughout the Galaxy and form a major component of the lower gaseous halo. We directly compare the physical properties and distribution of both cloud samples and find that while the properties of individual clouds are quite similar, those of the entire population differ quite dramatically. This difference in the population s properties is highly dependent on their local environment as well as the evolutionary stage of the population. Although the regions studied are symmetrically placed with respect to the line of sight to the Galactic centre, they reveal a significant asymmetry in the Milky Way. The first quadrant region includes the area where the Scutum-Centaurus arm joins with the near end of the Galactic bar, while the fourth quadrant region covers only the minor Norma arm. The radial surface density distributions of the clouds correspond well with these structures, and therefore suggest a correlation with spiral structure and star formation areas. However, upon comparison with a variety of star formation tracers such as H II regions, methanol masers and CO, we find no evidence of a detailed correlation, suggesting that the clouds may trace past rather than current star formation. We propose a scenario where the H I halo clouds result from fragmenting superbubbles and gas that has been swept into the halo due to stellar winds and supernovae activity, a process that would result in an abundance of clouds throughout the entire Galaxy. 137

156 138 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION 5.1 Introduction Detailed analyses of H I clouds in two separate regions of the Milky Way, one within the first quadrant and the other within the fourth quadrant, were presented in Chapters 3 and 4. The existence of clouds in these entirely different regions of the Galaxy suggests that they form a major component of the Milky Way, and in particular of the disk-halo interface. Comparisons between data in the first and fourth quadrants are usually complicated by the fact that different instruments are typically used to observe them. Our data are unique in that they have the same instrumental properties and selection effects, while also representing independent samples of H I halo clouds located on opposite sides of the Galaxy. We therefore have the unique ability to directly compare these two independent but symmetricallylocated samples, enabling us to learn more about the origin and nature of H I halo clouds and their role in Galactic structure and evolution. This chapter compares the physical properties and distribution of the H I clouds detected in the first and fourth quadrant regions. We discuss implications for their role as a component of the Milky Way along with their relation to other Galactic components and structures. We begin in 5.2 with a brief summary of the major properties of clouds in each Galactic quadrant. This includes a comparison of properties for both individual clouds (e.g., the physical mass and size) and those of the entire cloud population (e.g., the cloud-to-cloud velocity dispersion). In 5.3 we discuss cloud kinematics, followed by an analysis of their vertical distribution in 5.4. Next, we estimate the fraction of the H I layer that is made up of halo clouds ( 5.5). A comparison of the radial surface density distributions of the clouds is presented in 5.6, while the possibility of a relation between the clouds and spiral structures along with areas of star formation is explored in 5.7. We then discuss implications for the origin and nature of the clouds in 5.8 and summarise our results in Comparison of Cloud Properties A summary of the observed and derived properties of H I halo clouds in the first and fourth Galactic quadrants is presented in Table 5.1. The parameters listed above the solid line are the median observed physical properties of individual clouds while the properties below the solid line are those of the population as a whole. Except for the number of tangent point clouds, n tpc, the latter group are derived from simulations.

157 5.2. COMPARISON OF CLOUD PROPERTIES 139 Table 5.1. Summary of H I Cloud Properties in Quadrant I and IV GASS Pilot Region Quadrant I Region Parameter Value Best Fit Value Best Fit Consistent? v (km s 1 ) yes r (pc) yes M HI (M ) yes n tpc no σ cc (km s 1 ) yes h (pc) no Σ(R) peaked roughly uniform no Note. The median values of the observed halo cloud properties in the GASS pilot region and the first quadrant region (above the line). The derived fits for the properties representing the entire populations, as determined via simulations, are presented below the line, along with the number of tangent point H I clouds, n tpc, in each region. Within the fourth quadrant, the number of clouds used to determine the observed properties is n = 403 for v and n = 81 for r and M HI, while n = 255 for all observed properties of the first quadrant clouds. While all physical properties are consistent between quadrants, the number of tangent point clouds is strikingly different, as is the exponential scale height and the radial surface density distribution of the clouds.

158 140 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION As the first and fourth quadrant regions span the same Galactocentric radii and vertical distances from the plane, differing only by being located on opposite sides of the Galactic centre, there is a striking difference in the number of detected tangent point clouds: 255 in the first quadrant compared to only 81 in the fourth quadrant, despite the first quadrant region being 15% smaller. The general properties of individual clouds in each quadrant are similar: in both cases radii are 30 pc, masses are 650 M, and linewidths are 12 km s 1. The properties of the quadrant populations as a whole, however, differ in some aspects; although the cloud-to-cloud velocity dispersions are both consistent with 16 km s 1, the derived best fits for the exponential scale heights of their vertical distributions and their radial surface density distributions vary quite substantially: h = 400 pc with a concentrated radial distribution in the fourth quadrant, while h = 800 pc and the radial distribution resembles that of a uniformly distributed population in the first quadrant. It is these differences that we focus on in the upcoming sections, as they are likely to contain important clues to the origin of H I halo clouds and the role they play in Galactic evolution. 5.3 Cloud-to-Cloud Velocity Dispersion The observed deviation velocity distributions of both the first and fourth quadrant tangent point clouds are presented in Figure 5.1. In previous chapters we used a simulated population of clouds to determine what cloud-to-cloud velocity dispersion would result in a similar deviation velocity distribution of clouds in each quadrant while also providing acceptable fits to the vertical and radial distributions of the clouds. The resulting cloud kinematics in both the first and fourth quadrants are quite similar: clouds in the first quadrant are consistent with those of a population with 14 σ cc 15.5 km s 1 (with σ cc = 14.5 km s 1 having the highest probability) and those in the fourth quadrant with 16 σ cc 22 km s 1 (with σ cc = 18 km s 1 having the highest probability). This similarity, along with the remarkable resemblance between the observed V dev distributions (they have a K-S test probability of 76% of having been drawn from the same distribution), strongly suggests that the kinematics of the clouds are driven by the same physical processes in both quadrants. To put σ cc of the clouds in context with that of other Galactic populations, a

159 5.3. CLOUD-TO-CLOUD VELOCITY DISPERSION 141 Figure 5.1 Observed deviation velocity distributions of first (solid line) and fourth (dashed line) quadrant tangent point clouds. The fourth quadrant values have been negated for ease of comparison. The distributions are similar, with a K-S probability of 76% of having been drawn from the same distribution, and suggest that the cloud-to-cloud velocity dispersion in both quadrants is identical. This conclusion is supported by the simulations, which are consistent with σ cc 16 km s 1 in both quadrants. We assume N errors on the histogram values.

160 142 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION velocity dispersion of 16 km s 1 is lower than that of old disk populations such as red giants, larger than that of young populations like early main sequence stars and supergiants, and quite similar to that of F dwarfs (see Binney & Merrifield 1998 and references within). As there is a general relation between the age of a population and its velocity dispersion, e.g., see Holmberg et al. (2007), this could suggest that the clouds have similar ages to those of F dwarfs (if the clouds originate with the similarly small original velocity dispersion of young stars and are subject to similar kinematic heating processes). However, the clouds are likely to be much younger than the few Gyr age of F dwarfs, particularly if the clouds are related to areas of star formation, as suggested in Vertical Distribution of H I Clouds Basic Comparison The derived vertical distributions of the tangent point clouds in both the first and fourth Galactic quadrants are shown in Figure 5.2. The derived distributions properly account for selection effects, which artificially skew the appearance of the observed distributions, and can be significant close to the Galactic plane. In the first quadrant, the clouds vertical distribution is best represented by an exponential with a scale height, h = 800 pc, while in the fourth quadrant, a more narrow exponential with h = 400 pc best fits the derived vertical distribution. These best fit scale heights differ quite drastically between quadrants, as do the acceptable ranges of the scale height: in the fourth quadrant acceptable fits ranged from h = 300 to 500 pc while in the first quadrant acceptable fits were between h = 700 and 850 pc. Depending on the number of clouds detected at low heights, different confusion limits were applied to each quadrant: in the first quadrant, where more clouds were detected, a velocity-dependent latitude cut was well-defined and applied, while in the sparser fourth quadrant we tested two different latitude cuts ( b > 2 and b > 3 ). These differences in the confusion limit could affect the determined scale heights in the form of much better fits at low z in the first quadrant, where the number of detected clouds is much larger than at similar heights in the fourth quadrant. Also, we compared the observational distributions of both b and b to the simulated distribution in the first quadrant, but only the b distributions in the fourth quadrant. Unlike b, b is sensitive to asymmetries about the plane,

161 5.4. VERTICAL DISTRIBUTION OF H I CLOUDS 143 Figure 5.2 Derived vertical distributions of H I clouds in the first (solid line) and fourth (dashed line) quadrants. Both distributions are well represented by an exponential, but with a scale height of h = 800 pc in the first quadrant and only h = 400 pc in the fourth quadrant. but is less sensitive to differences in the median height. Despite these cavils, the inconsistency of the cloud scale heights between quadrants is a robust result: there is no overlap in the acceptable h ranges, there are few clouds detected at lower heights in either quadrant because of confusion, and we do not expect the symmetry of the distributions to greatly affect h Comparison with Other Galactic Components The relation between H I halo clouds and other Galactic components might be elucidated by comparing their vertical distributions. Gaseous components have scale heights that range from a few parsecs for the largest molecular clouds (Stark & Lee, 2005) to a few tens of parsecs for the smooth molecular gas component (Malhotra, 1994), to 110 pc for the smooth H I disk (Malhotra 1995, see also 5.5), and up to

162 144 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION 1 2 kpc for the warm ionised medium (Haffner et al., 1999; Gaensler et al., 2008). The clouds, with scale heights of 800 pc and 400 pc in the first and fourth quadrants respectively, lie between these latter two components, being more extended than the overall H I distribution, but not quite extended as the ionised gas. Most importantly, we compare the vertical scale height of the H I halo clouds to the thick H I layer, which has three components, including an exponential with a scale height of 403 pc (Dickey & Lockman, 1990) that likely contains blended H I clouds discussed here. This scale height is identical to that of the clouds in the fourth quadrant but half of that of the clouds in the first quadrant. Interestingly, the cloud scale heights are similar to those of the stellar thin ( 300 pc) and thick ( 900 pc) disks (Jurić et al., 2008), but it is likely that this is coincidental and does not give insight into the cloud distribution, which has very different physics Implications for Cloud Evolution There are a variety of scenarios that may explain the existence of clouds in the lower halo: 1. the clouds are formed in situ, either by condensing from hot gas within the lower halo or as a result of Rayleigh-Taylor instabilities of a supershell that is no longer expanding, and fall back to the plane, 2. the clouds are launched from the disk due to stellar feedback, as part of superbubbles, or if they contain dust, from radiation pressure on dust grains (Franco et al., 1991), they continuously circulate between the disk and halo, and are in equilibrium, 3. the clouds are accelerated from the disk via feedback, superbubbles, or radiation pressure on dust grains, are not in equilibrium, and dissipate, 4. the clouds are clumps of disk gas that are lifted into the halo as a result of expanding superbubbles, stellar winds, or radiation pressure on dust grains, and fall back to the plane, or 5. the clouds are neither rising nor falling and somehow maintain their individual vertical locations.

163 5.4. VERTICAL DISTRIBUTION OF H I CLOUDS 145 Table 5.2. Summary of Expectations in Various Origin Scenarios Description Origin Initial V z Lifetime n( z ) n evolution 1: formed in situ disk-halo 0 t ff rising decreasing z 2: lifted, in equilibrium disk +V z t ff falling constant 3: lifted, dissipates disk +V z t ff falling increasing z 4: lifted, falls disk +V z t ff peaks peaks rise/fall 5: stationary disk-halo 0?? constant Note. V z represents the vertical velocity of the clouds, where +V z is moving away from the disk and V z is falling towards the disk. t ff represents the free-fall time from 1 kpc. n( z ) is the vertical cloud distribution and n evolution represents the shape of the distribution with time. These scenarios can be distinguished by three main parameters: (a) their formation height, (b) the sign of their initial vertical velocity, and (c) their lifetime. For example, scenarios 2, 3, and 4 differ only in the cloud lifetimes but in each case they form in the disk with a positive vertical velocity, while in scenario 1 the clouds form far from the disk with zero initial velocity. A summary of expectations for each scenario is presented in Table 5.2. In we noted that the clouds are either pressure confined or transitory features, as they are not massive enough to be gravitationally bound. Without higher resolution data than those presented here, we are unable to constrain the pressure of the clouds, so cannot comment further on the first possibility. If they are transitory features, however, their lifetimes must be significantly less than the thermal evaporation timescale, which is 100 Myr (Stanimirović et al., 2006). This is longer than the free-fall time from 1 kpc ( 30 Myr, based on the vertical potential of Benjamin & Danly 1997), which is the timescale on which the vertical locations of the clouds and their vertical distribution, if not in equilibrium, change. The clouds lifetimes must be much shorter than the free-fall time in scenario 3, similar to it in scenarios 1 and 4, and much longer in scenario 2 (in scenario 5, free-fall is presumably unimportant). The implied vertical cloud distribution, n(z) for scenario 1 is biased towards higher z, where the clouds form, and the shape of the distribution falls in z with time. For scenario 2, n(z) is biased toward lower

164 146 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION z and determined entirely by the vertical velocity dispersion, σ z, and the vertical potential, and is in equilibrium. n(z) for scenario 3 is also biased towards lower z, and the distribution rises in z with time. In scenario 4, n(z) is peaked near the height of the top of the superbubble. If there are few superbubbles then there would be a few characteristic peaks whose location could either rise or fall with time, while the distribution would asymptote toward an equilibrium low- z bias if there are many superbubbles or regions of swept-up gas. Lastly, n(z) could have any form for scenario 5. It is not obvious why the exponential scale heights of the vertical distributions of clouds in separate quadrants differ by a factor of two. This difference is most likely a result of clouds in different quadrants being at different stages of their evolution. For example, in scenario 3, where the clouds have been accelerated from the disk into the halo and eventually dissipate, the population of clouds with the larger scale height (those in the first quadrant) would represent a later stage of this evolution process. Alternatively in scenario 1, where the clouds form from material above the disk that eventually rains back down (e.g., fragments of a superbubble), then the cloud population with the larger scale height (again, those in the first quadrant) would represent an earlier stage of the clouds lifetime. The clouds may even have been propelled ballistically up from the disk and eventually fall back (scenario 4), in which case the cloud population with the larger scale height is in the middle of its evolution while the other population (those in the fourth quadrant, with a smaller scale height) is either at an early or late stage. To fully decipher what is happening, more information would be required to break the degeneracy, e.g., in scenario 3 the total number of clouds decreases with time, so the population with more clouds would be earlier in its evolution, whereas if the clouds are a result of a standard galactic fountain (scenario 1), the population with more clouds would represent a later stage of the evolution process The Relationship between h and σ cc In we showed that clouds in the fourth quadrant could not reach their observed scale height of 400 pc with a vertical velocity dispersion, σ z σ cc but instead would require a vertical velocity at least three times larger. As the scale height in the first quadrant is more than twice as large as that in the fourth quadrant, yet σ cc is similar, not only is it impossible for the derived vertical scale height of clouds in

165 5.4. VERTICAL DISTRIBUTION OF H I CLOUDS 147 the first quadrant to be explained by σ cc, but the cloud-to-cloud velocity dispersion cannot be connected to the observed scale height. Therefore, either σ z σ cc, i.e., the clouds random velocity is not isotropic but is much greater vertically than along the line of sight and is more anisotropic in the first quadrant, or the clouds are not in an equilibrium vertical distribution. In the latter case, if we were to again observe the same population of clouds a few million years from now, either the vertical distribution of the clouds would be different, the newly observed clouds would be different from those we currently have detected, or a combination of these possibilities. How do clouds reach such large distances from the plane if the cloud-to-cloud velocity dispersion cannot result in such heights? If the clouds are a result of scenario 1, where the clouds form far from the disk, it is not surprising that they are detected at fairly large heights, but if they form as in scenarios 2-4, where they are ejected from the disk, how do they reach z > 2 kpc? Blasts from supernovae may lift or push the clouds to large heights, and the original location of the supernovae may play a role in this. As the mean gaseous layer of molecular clouds is not always at b =0, as can be seen in Dame et al. (2001), and as supernovae may be displaced from their parent molecular cloud, the energy injection may occur away from the midplane and thus increase the probability that H I clouds reach larger heights, especially if there is a path where the hot gas can easily escape. If the cluster of stars was particularly large or energetic this could also result in material reaching large heights. Observationally, there are several examples of superbubbles whose shells reach z > 1.5 kpc, such as in McClure-Griffiths et al. (2006) and Pidopryhora et al. (2007). It is important to note that in both quadrants n(z) is roughly exponential, i.e., there are more clouds close to the plane than far from it and the clouds appear as if they are in a stable, isotropic distribution. According to Houck & Bregman (1990), n(z) of a rain from a galactic fountain is peaked at large heights with a small tail to lower z (see their Figures 6 and 7), in stark contrast to what we derive for the observed clouds. We may therefore rule out scenario 1 based on the form of the vertical distribution. Our suggestions that the clouds are fragments of superbubbles or gas that has been swept into the halo, which falls under scenario 4, is consistent with the derived n(z) as long as there are a sufficient number of superbubbles or regions of swept-up gas. Finally, for a sufficiently short cloud lifetime, the clouds may not be dynamic,

166 148 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION individual objects, but simply transitory high-density fluctuations in a turbulent medium, a form of scenario 5, and σ cc, which would reflect the turbulent velocity dispersion, would be completely decoupled from n(z), which would reflect the distribution of the underlying medium. By using data with higher spatial resolution it may be possible to determine if clouds are always associated with turbulent structure or if any are truly isolated from the background, which would rule out this idea. 5.5 Fraction of Galactic H I in Halo Clouds Here we explore questions about the relationship of halo clouds to the H I layer. Do the scale heights of H I correspond to the derived scale heights of the clouds? Is a fixed fraction of Galactic H I in the form of halo clouds, regardless of environment? And how much of the H I is confined within these clouds? With GASS data we can determine the amount of H I contained in clouds. However, stray radiation can produce a halo of spurious emission around bright H I features, and as a result, some of the tails of emission that the H I cloud profiles appear to be within may be spurious features. The current version of GASS can therefore only provide an upper limit to the diffuse material. This possibility of stray radiation, along with a data reduction artefact that results in negative mean fluxes at high z in quadrant IV, prevents us from using GASS data to determine the global properties of the H I layer and so we must use previous estimates instead. Figures 5.3 and 5.4 represent the vertical profile of the mean H I number density contained in the tangent point cloud sample within the first and fourth quadrant regions, where the amount of mass that each cloud contributes at its height is from Tables 3.2 and 4.2. We assume that the line of sight depth that is probed is as shown in Figure 5.8 and imposed the declination limit of GASS when calculating the mean H I density. We have fit an exponential to the determined profile. While performing this fit, we excluded a region between 0.2 < z < 0.35 kpc in the first quadrant and z < 0.35 kpc in the fourth quadrant as there is little information at these heights due to confusion. The exponential scale height, z exp, of H I in detected halo clouds is 370 pc in the first quadrant and 460 pc in the fourth quadrant. This suggests that although there are three times as many clouds in the first quadrant, the shape of the vertical distribution of the mass is similar.

167 5.5. FRACTION OF GALACTIC H I IN HALO CLOUDS 149 Figure 5.3 Mean H I number density due to tangent point halo clouds within the quadrant I region. The profile has been fit by an exponential (dotted line), with a scale height of 370 pc. A region between 0.2 < z < 0.35 kpc has been excluded for the fit as there is little information at these heights due to confusion. To determine the error bars, we propagated forward the 40% estimated error in each mass measurement.

168 150 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION Figure 5.4 As in Figure 5.3, but for the GASS pilot region. The profile has been fit by an exponential (dotted line), with a scale height of 460 pc. A region between z < 0.35 kpc has been excluded for the fit as there is little information at these heights due to confusion. To determine the error bars, we propagated forward the 40% estimated error in each mass measurement.

169 5.5. FRACTION OF GALACTIC H I IN HALO CLOUDS 151 There is more H I mass in halo clouds within the first quadrant region than in the fourth quadrant as the median mass of the clouds is the same for both quadrants but yet there are three times as many clouds in the first quadrant (255 vs. 81 tangent point clouds). This can also be seen by comparing the normalisation of the exponential fits in Figures 5.3 and 5.4. The fraction of extended H I emission that is contained within the tangent point clouds as a function of z is presented in Figures 5.5 and 5.6, where the density profile of the H I layer is based on the exponential component of the Dickey & Lockman (1990) curve. The dips visible at low heights in these figures result from a lack of information due to confusion. Within the first quadrant region, the fraction of the mass within halo clouds is typically 5%, while in the fourth quadrant region the fraction is typically 1%, and in both cases is approximately consistent with height. This is expected from the similarity of the exponential scale heights of the H I mass in clouds and of the global H I layer. This lends support to the idea that the layer s properties are related to the aggregate emission of the clouds, although they appear to not be responsible for the majority of its mass. As the properties of the global H I layer in Dickey & Lockman (1990) are averages over regions with large fluctuations, it may be that the density of the global layer varies between the quadrants rather than the fraction contained in clouds. As the current version of GASS data has not been corrected for stray radiation, we cannot do a direct comparison of the clouds with the line wings to determine their filling factor. We therefore use previous estimates of the volume filling factor of H I in the Milky Way to estimate the volume filling factor of the H I halo clouds. According to Heiles (2001), the filling factor for the warm neutral medium is f 0.5. Based on Figures 5.5 and 5.6, the clouds account for roughly 5% of the H I in the disk-halo interface. The filling factor of the halo clouds is therefore We note that it is quite likely that there is a population of clouds below the GASS angular resolution and sensitivity limit, which may appear as part of the background and result in an underestimate of the filling factor of the cloud population. To estimate the magnitude of this population, we have extrapolated the cloud mass spectrum to masses M HI < M using a power law fit to the mass function over the range 2.4 logm HI /M 3.4; as discussed in 4.4, selection effects are unlikely to strongly bias the mass function. This extrapolation indicates that only 7% of the total H I mass in clouds may be missing. The scale height of the clouds vertical distribution is similar to the scale height

170 152 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION Figure 5.5 Fraction of extended H I contained within tangent point clouds in the first quadrant region as a function of height assuming the total H I is represented by the form of the exponential component given in Dickey & Lockman (1990). The fraction of mass contained in the halo clouds is typically 5% and does not vary systematically with height. The dip near the Galactic plane arises from our inability to identify clouds due to confusion. To determine the error bars, we propagated forward the 40% estimated error in each mass measurement.

171 5.5. FRACTION OF GALACTIC H I IN HALO CLOUDS 153 Figure 5.6 As in Figure 5.5, but for the GASS pilot region. The fraction of mass contained in clouds is typically 1% and does not vary systematically with height. To determine the error bars, we propagated forward the 40% estimated error in each mass measurement.

172 154 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION of the mass distribution of the clouds in the fourth quadrant (h = 400 pc and z exp = 460 pc), while in the first quadrant these properties differ by a factor of two (h = 800 pc and z exp = 370 pc). This implies that the mass of the clouds is a systematic function of height in the first quadrant, where clouds are substantially more massive at lower z, but not in the fourth quadrant. This strengthens the suggestion made in Chapters 3 and 4 for an anti-correlation between cloud mass and height in both quadrants based on an entirely independent analysis. One possible explanation for this anti-correlation is that the clouds are pushed or swept-up into the lower halo from the disk in large clumps that eventually break into smaller pieces as they get farther from the disk. It is puzzling, however, that we only see this anti-correlation in the first quadrant and not in the fourth quadrant from the present analysis. Another possibility is that we are seeing the effects of blending. In confused regions, the number of clouds is underestimated and mass of each cloud is overestimated, but their product, the total mass in clouds, is relatively robust. The most confused region would be at low z in the first quadrant, where we detect the most clouds. If the number of clouds was underestimated there, the derived vertical scale height would be overestimated. In it was noted that σ cc, which is similar in both quadrants, could not be related to the scale height, which is dramatically different. The cloud-to-cloud velocity dispersion may instead be related to the vertical mass distribution, which is similar, rather than the vertical distribution of the number of clouds. 5.6 Derived Radial Surface Density Distribution The derived radial surface density distributions, Σ(R), of H I clouds within both the first and fourth quadrant regions are presented in Figure 5.7. As a reminder, simulated populations of clouds were used to determine the best fit to the observed distributions, and the amplitude of each radial bin was optimised to best fit the observed longitude distribution of the tangent point clouds by minimising the K-S D statistic using Powell s algorithm (Press et al., 1992). For each quadrant, the fits were then optimised using three different initial estimates for Σ(R), and we adopted the mean of the three as the best fit to the data. As expected, given the dramatic difference in the absolute number of clouds detected within separate quadrants, the surface density of clouds is found to be

173 5.6. DERIVED RADIAL SURFACE DENSITY DISTRIBUTION 155 Figure 5.7 Derived radial surface density distribution of H I halo clouds within the GASS pilot region (dashed-dot line) and the first quadrant region (solid line). More clouds were detected within the first quadrant than in the fourth quadrant, and their distribution marginally resembles that of a uniformly distributed population, as opposed to the peaked distribution that is evident in the fourth quadrant, where beyond roughly R =4.2 kpc there is a rapid decline in the number of clouds. The arrows represent the minimum and maximum Galactocentric radii along the locus of tangent points for the regions searched, and represent the radii where the surface densities are best constrained (as indicated by the error bars). Error bars denote the Poisson error in the number of clouds expected to fall into the observed sample.

174 156 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION much larger in the first quadrant than in the fourth quadrant. The surface density in the first quadrant is also roughly uniform, in contrast to the surface density in the fourth quadrant, which is peaked and declines rapidly at R>4.2kpc. This significant difference is seen within the well-constrained area of the radial surface density distribution (denoted by arrows), and is therefore a robust result. As these features of the radial surface densities differ so dramatically, it is likely that they are related to Galactic structures, e.g., spiral arms, as was previously suggested in We expand on the discussion of this possibility in Relation to Spiral Features and Star Formation In Chapter 3 we proposed a relation between the H I halo clouds and areas of star formation and spiral features, based on the peaked radial surface density distribution of the clouds in the fourth quadrant and the presence of many loops and filaments, with which many of the clouds appeared to be associated. If such a relation exists, the differences between the radial surface density distributions of each quadrant can be used to probe the relation further. For example, the large surface densities of clouds may be due to increased star formation rates, and structure in the radial profile may represent the structure of the Galaxy. In this section we consider these possibilities, as well as compare the distributions of H I clouds to tracers of star formation, to determine whether or not it is likely that such a relation exists, and if it does, to better understand it H I Clouds and Galactic Structure It is puzzling that there are three times as many clouds in the first Galactic quadrant as in the fourth and that their surface densities differ so dramatically. A hint to what may be happening can be seen in Figure 5.8, which contains an artist s conception of the spiral structure of the Milky Way (based on a spiral arm model by Benjamin et al. 2003), with the outline of the regions of the first and fourth quadrants in which we detected tangent point clouds overlaid. The regions span the longitude limits of both the GASS pilot region and the first quadrant region, and cover velocities within ±18 km s 1 of the terminal velocity. This velocity range visually reflects the

175 5.7. RELATION TO SPIRAL FEATURES AND STAR FORMATION 157 likelihood that the clouds are not located directly along the locus of tangent points (dashed circle), but instead are in a region whose circular velocities are close enough to the terminal velocity that the random velocity dispersion pushes the clouds beyond the terminal velocity. This depiction of the Galaxy incorporates recent findings from the Galactic Legacy Infrared Mid-Plane Survey Extraordinaire (GLIMPSE; Benjamin et al. 2003), a survey that was conducted using the Spitzer Space Telescope. These GLIMPSE data suggest that the spiral structure of the Galaxy, which had originally been thought to be dominated by four major arms, is dominated by two arms (the Scutum- Centaurus and Perseus arms) that extend from each end of the central bar. There are also two minor arms (the Norma and Sagittarius arms), which are located between the major arms (Spitzer Space Telescope, 2008). Although this model has not yet been through peer review, we consider it to be the most appropriate model to date because it uses a combination of kinematic information from gas and infrared star counts from GLIMPSE, resulting in a more complete picture of the Milky Way s spiral structure. The fourth quadrant region spans a sparse section of the Galaxy, through which only the minor Norma spiral arm passes, while the first quadrant region contains a much denser portion of the Galaxy where the near end of the Galactic bar merges with the beginning of the major Scutum-Centaurus arm. The radial distribution of the total stellar density in the fourth quadrant region peaks at just one radius (corresponding to the Norma arm), while the total stellar density is more evenly distributed in the first quadrant region (due to the complexity of the region). These distributions are strikingly similar to those of the tangent point cloud sample, suggesting that there is a strong relation between the stellar spiral structure of the Galaxy and the H I halo clouds. This physical alignment of spiral features and H I clouds suggest that the clouds are related to areas of star formation, as within the Milky Way the most massive molecular clouds that contain most forming stars are strongly associated with spiral arms (see McKee & Ostriker 2007 and references therein). This relation is further supported by observational evidence that there has been a significant burst of star formation near where the Scutum-Centaurus arm meets the end of the bar, in the form of observations of multiple red supergiant clusters in the area (e.g., see Figer et al. 2006; Davies et al. 2007; Alexander et al. 2009; Clark et al. 2009). External galaxies have been observed to exhibit similar behaviour, having increased star formation occurring where spiral arms meet the bar

176 158 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION Figure 5.8 The spiral structure of the Milky Way as determined from GLIMPSE data, with outlines of the first and fourth quadrant regions overlaid, spanning velocities within ±18 km s 1 of the terminal velocity (assuming a flat rotation curve). The longitude boundaries of the regions are denoted by the dotted arrows while the locus of tangent points is represented by the dashed circle. The Galaxy has two major arms that extend from either end of the central bar, along with minor arms between the major arms. The radial distribution of the total stellar density in the fourth quadrant region peaks at just one radius (corresponding to the Norma arm), while it is more evenly distributed in the first quadrant region. This is strikingly similar to the distribution of H I halo clouds, and is therefore quite suggestive of a relation between the stellar spiral structure of the Milky Way and the clouds. The artist s conception image is from NASA/JPL-Caltech/R. Hurt (SSC-Caltech).

177 5.7. RELATION TO SPIRAL FEATURES AND STAR FORMATION 159 ends (Phillips, 1993). We observe a large number of H I clouds in the region where the Scutum-Centaurus arm extends from the bar end in the first quadrant, where the clouds are much more numerous than in the fourth quadrant region, suggesting that the number of clouds is proportional to the amount of star formation. Also, as the Scutum-Centaurus arm is a major arm while the Norma arm is a minor arm, less star formation would be expected in the Norma arm. If there was a relation between star formation and the H I halo clouds, this would result in fewer clouds in the fourth quadrant, as is consistent with our observations. In Chapter 3 we suggested that the clouds could be a result of superbubbles and material that has been pushed up from the disk, as the presence of many loops and filaments were clearly visible within the GASS pilot region data and many clouds appeared to be associated with these structures. These features are quite common in areas of star formation (Kalberla & Kerp, 2009), so the presence of such structures would also be expected in the first quadrant region. The Ophiuchus superbubble is an excellent example of such a structure: it is an old superbubble ( 30 Myr) that lies within the first quadrant region and is above a section of the Galaxy containing many H II regions, including W43 (Pidopryhora et al., 2007). This superbubble is at a distance of 7 kpc and is capped by a plume of H I at 3.4 kpc above the plane. Many H I features have been observed to be affiliated with this superbubble at velocities near those expected at tangent points, including both whiskers, i.e., gas that has been swept-up sideways from the disk, and clouds (Pidopryhora et al., 2007, 2009), and it is highly likely that many of the tangent point clouds in the first quadrant region are associated with this superbubble. It is interesting to note that extraplanar gas in some external spiral galaxies, such as NGC 4559, appears to be spatially related to star formation activity (Barbieri et al., 2005). Also, the presence of extraplanar dust in external galaxies implies that if extraplanar gas is a result of feedback, the processes transporting the material from the disk must be gentle and likely have low velocities for the dust grains to survive (Howk, 2005) H I Clouds and H II Regions If there is indeed a correlation between H I clouds and areas of star formation, a correlation might also be expected between H I clouds and H II regions. In we compared the radial surface density distribution of the GASS pilot region tangent

178 160 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION Figure 5.9 Radial surface density distribution of H I clouds (solid line) and H II regions (dotted line) in the longitude range of the first quadrant region. The H II regions are from Paladini et al. (2004) and have been scaled by a factor of 50 for ease of comparison. There is no correlation between the H I cloud distribution and that of the H II regions. Arrows represent the radial boundaries of the region searched (at the tangent point). point clouds to that of all H II regions within the entire Galactocentric radii range of the tangent point clouds, regardless of longitude, and found no evidence of a correlation. In Figures 5.9 and 5.10 we present the radial surface density distributions of the H I clouds with that of Galactic H II regions overlaid, for both the first and fourth quadrants, respectively, this time plotting only those H II regions within the longitude ranges of the GASS pilot region and the first quadrant region including only those whose distances place them within ±30 km s 1 of the terminal velocity. The catalogue of H II regions, including distances, are from Paladini et al. (2004). There does not appear to be any correlation between the H I clouds and H II regions in either quadrant.

179 5.7. RELATION TO SPIRAL FEATURES AND STAR FORMATION 161 Figure 5.10 Radial surface density distribution of H I clouds (solid line) and H II regions (dotted line) in the longitude range of the GASS pilot region. The H II regions are from Paladini et al. (2004) and have been scaled by a factor of 50 for ease of comparison. There is no correlation between the H I cloud distribution and that of the H II regions. Arrows represent the radial boundaries of the region searched (at the tangent point).

180 162 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION H I Clouds and Methanol Masers 6.7 GHz methanol masers are known tracers of areas of high-mass star formation (Xu et al., 2008), and some trace particularly early protostellar cores (Minier et al., 2005). A correlation between the distribution of methanol masers and that of H I clouds would therefore be expected if clouds are associated with areas of active star formation. As such, we compare the distribution of halo clouds with methanol masers that have been detected by Pestalozzi et al. 2005; Pandian et al. 2007; Ellingsen 2007; Xu et al and Cyganowski et al within ±30 km s 1 of V t. This comparison is shown in Figures 5.11 and There is a factor of 2.2 excess of masers in the quadrant I region compared with the GASS pilot region, which is not as large as the difference between the number of H I clouds, but larger than in any other species of which we are aware. There are peaks in the maser distribution in both quadrants: at l 30 in the first quadrant and l 331 in the fourth quadrant. These distributions are different from the tangent point H I cloud distributions, which contain no peak in the first quadrant and a peak at l 335 in the fourth quadrant. This peak does not appear to be correlated with the peak at l 331 of methanol masers. The lack of an obvious correlation between the clouds and masers could be explained if the masers are tracing recent star formation while the H I clouds are remnants of past star formation H I Clouds and Molecular Gas As environments containing dense regions of molecular gas are prominent sites of star formation (McKee & Ostriker, 2007), the distribution of H I halo clouds should also be compared to that of molecular hydrogen, H 2. However, due to the lack of a dipole moment in H 2 molecules, molecular hydrogen is usually traced through the detection of carbon monoxide, CO. We present the derived radial surface density distribution of H I clouds along with the mass surface density distribution of H 2 as derived by Bronfman et al. (1988) in Figures 5.13 and 5.14 for first and fourth quadrant data, respectively. The H 2 data have been scaled to R 0 =8.5 kpc to allow for direct comparison with the H I halo cloud data, and have been derived from CO observations that span 12 l 60 and 300 l 348. There does not appear to be any correlation between the distribution of H I clouds and H 2 in either the first or fourth quadrant, contrary to what would be expected if the clouds were related to areas of star formation. However, the H 2 data represent the average characteristics

181 5.7. RELATION TO SPIRAL FEATURES AND STAR FORMATION 163 Figure 5.11 Longitude distribution of H I clouds (solid line) and methanol masers (dashed line) in the longitude range of the first quadrant region. The maser distribution has been scaled down by a factor of four for ease of comparison. There appears to be no correlation between the H I cloud distribution and that of the methanol masers.

182 164 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION Figure 5.12 Longitude distribution of H I clouds (solid line) and methanol masers (dashed line) in the longitude range of the GASS pilot region. The maser distribution has been scaled down by a factor of four for ease of comparison. There appears to be no correlation between the H I cloud distribution and that of the methanol masers.

183 5.7. RELATION TO SPIRAL FEATURES AND STAR FORMATION 165 Figure 5.13 Derived radial surface density distribution of H I clouds in the first quadrant region (solid line) with the mass surface density distribution of H 2 for first quadrant data (dot-dashed line) overlaid and plotted against the right-hand vertical axis. The H 2 data are from Bronfman et al. (1988) and have been scaled to R 0 =8.5 kpc. Arrows indicate limits of R for tangent points in the quadrant I region. There does not appear to be any correlation between the distribution of H I clouds and H 2 but this may be due to the averaging of CO data in the determination of the H 2 distribution, as those data were not restricted to the tangent points. of molecular gas on scales of half a kiloparsec or more (Bronfman et al., 1988), and include gas at all locations, not just that at the tangent points, so it is quite likely that the averaging may have removed any structure in the distribution that would be visible over smaller ranges of longitudes and Galactocentric radii H I Clouds and Shells It would be very interesting to compare the distribution of the clouds with known shells and supershells within the Milky Way. However, due to the difficulty in detecting such shells within the inner Galaxy, there is not a complete catalogue of

184 166 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION Figure 5.14 Derived radial surface density distribution of H I clouds in the GASS pilot region of the fourth quadrant (solid line) with the mass surface density distribution of H 2 for fourth quadrant data (dot-dashed line) overlaid and plotted against the right-hand y-axis. The H 2 data are from Bronfman et al. (1988) and have been scaled to R 0 =8.5 kpc. Arrows indicate limits of R for tangent points in the GASS pilot region. There does not appear to be any correlation between the distribution of H I clouds and H 2 but this may be due to the averaging of CO data in the determination of the H 2 distribution, as those data were not restricted to the tangent points.

185 5.8. THE ORIGIN AND NATURE OF H I HALO CLOUDS 167 such structures in these regions. This is apparent in Figure 18 of McClure-Griffiths et al. (2002), which includes shells from both their catalogue and that of Heiles (1984). Most catalogued shells in their figure are in the outer Galaxy. Within their catalogues there are less than 5 shells within each of the longitude ranges of the first and fourth quadrant regions we have studied, and none of those shells are at velocities similar to those of the clouds discussed here. We are therefore unable to compare the distribution of clouds with that of shells and supershells directly. 5.8 The Origin and Nature of H I Halo Clouds H I halo clouds are abundant in both regions of the Galaxy searched, which span many kpc 3, revealing that the clouds are not an isolated phenomenon and suggesting that these clouds would be detected in many more regions of the Milky Way, forming a major component of the Galaxy. It is therefore quite likely that they also exist in external galaxies, though current resolution limits would prevent their detection. Just as there are similarities between the samples detected in the first and fourth Galactic quadrants, particularly in the basic cloud properties and cloud-to-cloud velocity dispersion, there are also notable differences. In particular, the large scale distribution of the clouds is unusually discrepant for the Milky Way, whose components are generally quite regular; most components of the Galaxy are found equally prominent in the first and fourth quadrants. These clouds are therefore ideal probes of the evolution of the gas in the disk-halo interface. The observations that provide the most insight into the nature of the H I halo clouds are as follows: 1. The radial distribution of the clouds closely mirrors the location of spiral arms: the GASS pilot region of the fourth quadrant contains only the minor Norma arm; the halo clouds within the GASS pilot region are few and their radial distribution peaks near the radius of the Norma arm. In contrast, the first quadrant region contains the merging of the major Scutum-Centaurus arm and the near end of the Galactic bar; the halo clouds in the first quadrant region are plentiful and are spread fairly evenly over the large area. 2. There are three times as many tangent point clouds in the observed region of the first quadrant, where the star formation rate is expected to be quite high,

186 168 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION as in the observed region of the fourth quadrant, where the star formation rate is expected to be much lower. 3. There is no clear correlation between the locations of clouds and sites of current star formation, as traced by H II regions, methanol masers, and H The observed cloud-to-cloud velocity dispersion is both insufficient to raise the clouds to their observed heights and is uncorrelated with the scale height of the cloud population, which differs by a factor of two between quadrants. 5. Superbubbles, which are powered by bursts of star formation, evolve for 20 to 30 Myr (McClure-Griffiths et al., 2006; Pidopryhora et al., 2007) and push H I into the lower halo without inflating the internal velocity dispersion. 6. The clouds are associated with many loops and filaments. Based on these items, we propose the following scenario for the origin and evolution of H I halo clouds: the clouds are related to areas of star formation, where stellar winds and violent supernovae activity sweep and push gas from the disk into the lower halo. It is through this mechanism that clouds are lifted into the lower halo, which sometimes occurs in the form of superbubbles, whose shells eventually fragment. As star formation is abundant in spiral arms, if the clouds are produced by episodes of star formation, they would naturally be found to be highly correlated with the spiral structure of the Galaxy and would appear to be associated with many loops and filaments, as observed. The number of clouds would also be expected to correlate with the star formation rates in those regions of the Galaxy, as is also observed. The timescales of superbubble evolution are fairly long and therefore clouds that are produced by this mechanism may no longer be at the same locations as the sites where high-mass stars are forming at the present day, as traced by masers and H II regions. This could explain why no correlation is seen between the cloud distribution and that of these star formation tracers. This lack of correlation may also be due to the tracer samples being restrictive, either by being averaged over quite a large area (as in the case of the molecular gas, with which there also appears to be no correlation), or by small number statistics within the regions of interest (as in the case of both the methanol masers and H II regions). The scale heights of the cloud population are reasonable if the gas is brought into the lower halo by superbubbles, as a high velocity dispersion is not required. The difference in the scale heights in

187 5.8. THE ORIGIN AND NATURE OF H I HALO CLOUDS 169 the two regions could either be due to more energetic star formation activity in the first quadrant where the star formation rate is higher, causing a larger scale height (note that the higher star formation rate must result in more energetic individual events on average, i.e., larger superbubbles or more forceful winds that sweep up material, and not simply increase the total amount of swept-up gas, for this explanation to be viable), or due to clouds at different evolutionary stages, where clouds leaving the disk is the earliest stage and returning to the disk is the latest, while those at larger heights are likely in the early to middle stages of their evolution. It is important to contrast this scenario with that of a standard galactic fountain model, which proposes that the clouds are formed by the cooling and condensing of hot gas that has been expelled from the disk, which then falls back towards the plane (Shapiro & Field, 1976; Bregman, 1980). While this model has similarities to our proposed scenario, an important difference is that the distribution of clouds would not be expected to have small-scale features, such as a peaked radial distribution, or a dramatic difference in the number of clouds between different regions of the Galaxy at similar radii, as the hot gas from which they condense is expected to be much more uniform (Bregman, 1980). Such features are clearly present in the H I halo cloud distributions, which argues for a scenario where clouds are produced more directly by events occurring within spiral arms. There is also no reason to expect the clouds to be associated with loops and filaments if related to a galactic fountain, but these structures are clearly prominent and associated with many clouds. Other possible, but less likely, explanations for the origin of H I halo clouds are that they are created through tidal stripping or infalling primordial gas, but in both of these cases there is again no reason to expect any correlation between the distribution of the clouds and the spiral structure of the Galaxy, and the concentration of the clouds to the Galactic plane and the dominance of Galactic rotation in their motions appear to definitively rule out such external origins. Using 3D hydrodynamic simulations in Chapter 6, we test our hypothesis that H I clouds can form due to the expansion of a superbubble, and determine what heights they reach and how much H I mass is in the lower halo as a result of such an event.

188 170 CHAPTER 5. H I HALO CLOUDS: A GALACTIC POPULATION 5.9 Summary The large coverage of GASS, combined with its excellent spectral and angular resolutions, enabled us to study hundreds of tangent point H I clouds within the lower halo of the inner Galaxy. By comparing two regions of the Galaxy that span the same Galactocentric radii and heights, but on opposite sides of the Galactic centre, using data that was obtained using the same instrument, we were able to learn more about the origin and nature of H I halo clouds than would otherwise be possible. The number of tangent point clouds within the first and fourth quadrants is strikingly different: 255 clouds were detected in the first quadrant and only 81 over a slightly larger volume in the fourth (due to the declination limit of GASS). These clouds exhibit similar physical properties however, and even their cloud-tocloud velocity dispersions are similar ( 16 km s 1 ) despite a large difference in the exponential scale height of their vertical distributions (400 pc vs. 800 pc). This is quite suggestive that the kinematics of the clouds are driven by the same physical processes in each quadrant and that the cloud-to-cloud velocity dispersions are not responsible for the heights these clouds reach. The vertical profile of the mean H I density of clouds in each quadrant is similar, and is also similar to that of the extended H I layer. As the median values of the H I mass of individual clouds are similar in both quadrants, but there are three times as many clouds in the first quadrant, there is more H I mass in halo clouds in the first quadrant than in the fourth. In the first quadrant, 5% of the average mass of the H I layer is composed of clouds while in the fourth quadrant 1% of the mass of H I is in clouds. The scale height of the number of clouds is similar to that of the mass distribution of clouds in the fourth quadrant, while n(z) is twice as large as the scale height of the mass distribution in the first quadrant. Another remarkable difference between the quadrants can be seen in the radial surface density distributions, where in the first quadrant the sample resembles that of a uniform population of clouds while in the fourth quadrant the sample clearly peaks near R =3.8 kpc and rapidly declines beyond R>4.2 kpc. The comparison of these structures with the spiral structure of the Galaxy reveals an alignment of the minor Norma arm with the peak in the fourth quadrant, and that the near end of the bar and beginning of the major Scutum-Centaurus arm is contained within the first quadrant region, strongly suggesting a correlation between the radial distribution of halo clouds and the spiral structure of the Milky Way. The difference in the

189 5.9. SUMMARY 171 number of clouds detected in each quadrant also strongly supports this hypothesis, as many more clouds would be expected near the bar-end and beginning of a major arm than near a minor arm if clouds are related to the amount of star formation activity occurring within these regions. The comparison of the distribution of H I clouds and that of H II regions, methanol masers and molecular gas reveals no evidence of a correlation between the clouds and these star formation tracers. This may be the result of small number statistics in the case of the H II regions and methanol masers, and the averaging of the molecular gas that would likely remove any structures in the distribution. The lack of a correlation could also be a result of the H I clouds tracing past, rather than current, star formation. We propose a scenario where the H I halo clouds are related to areas of star formation in the form of superbubbles and gas that has been swept into the halo due to stellar winds and supernovae. These events occur frequently within spiral arms, so the correlation with the spiral arm structure of the Milky Way supports this proposal, as does the appearance of many loops and filaments with which many clouds are associated. Many of the clouds are likely pushed from the disk, while others leave the disk as part of an expanding shell, which eventually can fragment and cause the clouds to rain back towards the disk. Under such circumstances, σ cc would not be expected to be responsible for the height distribution, and different stages in the evolution of the clouds that are in different regions of the Galaxy could explain the differences in their vertical scale heights. The mass distributions in both quadrants are roughly the same and σ cc may be related to the mass distribution rather than n(z). The large number of clouds detected in both regions suggest that the clouds are a major component of the Galaxy and would likely be detected throughout it. The H I halo clouds therefore play an important role in Galaxy evolution and the circulation of gas between the disk and halo, and are likely prominent features in many external galaxies.

190

191 Chapter 6 Superbubble Simulations and the Production of H I Halo Clouds Three-dimensional hydrodynamic simulations of a superbubble within an inhomogeneous medium were performed to test the hypothesis that H I clouds can form in the lower halo as a result of this process. Energy is injected into the simulated galaxy disk, resulting in the formation of a bubble. The bubble is not smooth but instead develops scalloped edges, and eventually expands to a size so large that it breaks out of the disk vertically. Hot gas within the bubble escapes via chimneys and mixes with the halo gas, and as this hot gas leaves the disk via the chimneys, the edges of the chimney walls are stripped. This stripped gas results in many clumps of H I in the lower halo, with diameters 15 pc, and reaching heights up to 1.4 kpc over the course of 20 Myr. Supplementary simulations were performed with slight modifications to the initial conditions to see how the ISM structure, density and location of the energy source affect the results. In all simulations clumps of H I were seen above the disk. It is therefore possible for H I clouds to form in the disk-halo interface due to the evolution of a superbubble. These clouds form from disk gas that has been swept into the halo from the walls of chimneys, and have both similar diameters and heights of the observed H I halo clouds, but lower densities, which are likely due to the clouds being spatially unresolved. The results of these simulations are therefore consistent with a superbubble origin for H I halo clouds, but resolution limits prohibit us from concluding that they are the same population of H I clouds as those observed. 173

192 174 CHAPTER 6. SUPERBUBBLE SIMULATIONS AND H I HALO CLOUDS 6.1 Introduction Throughout the previous chapters we have presented an in-depth study of the properties and distribution of H I halo clouds that have been detected within GASS data. This analysis has provided strong evidence that the H I clouds are related to areas of star formation, in the form of fragmenting shells and disk gas that has been pushed into the halo. Observations of multiple superbubbles also support this hypothesis, such as in GSH , where McClure-Griffiths et al. (2006) observed fragments of H I along the cap of the supershell, and the Ophiuchus superbubble, where Pidopryhora et al. (2007) detected over 600 H I features within the system, including many clouds. Previous simulations of superbubbles (e.g., Mac Low et al. 1989; Tomisaka 1998; Korpi et al. 1999; Breitschwerdt & de Avillez 2006; Melioli et al. 2008) and simulations of the global ISM that include superbubbles (e.g., Rosen & Bregman 1995; de Avillez & Berry 2001; de Avillez & Breitschwerdt 2005) have been performed with ever-increasing resolution and inclusion of more accurate physical processes and initial conditions. Initially two-dimensional simulations (2D; Mac Low et al. 1989; Rosen & Bregman 1995) have given way to full three-dimensional (3D) simulations, and pure hydrodynamic simulations have been supplemented with magnetohydrodynamic (MHD) models (Tomisaka, 1998; Korpi et al., 1999; de Avillez & Breitschwerdt, 2005). Rather than an idealised smooth ambient medium, some simulations now include a more realistic inhomogeneous or turbulent environment for the explosion (Korpi et al., 1999; de Avillez & Berry, 2001; Breitschwerdt & de Avillez, 2006). No previous simulation has been performed with the aim of reproducing the population of co-rotating H I clouds that we have studied in the earlier chapters of this thesis, but some of these simulations have formed clouds through a variety of processes. In particular, some authors have noted the superbubble shell fragmenting via Rayleigh-Taylor instabilities into clouds, although they have remained within the disk rather than rising to the disk-halo interface (e.g. Mac Low et al. 1989; de Avillez & Berry 2001; Breitschwerdt & de Avillez 2006). Others have found clouds formed from the collisions of supernova shells (Korpi et al., 1999), or from thermal instabilities (de Avillez & Berry, 2001; Melioli et al., 2008). Perhaps the most intriguing clouds are those found in Melioli et al. (2008), which form at z = 2 kpc from hot disk gas that has been lifted into the halo, and fall back to the disk with velocities, v = km s 1 ; however, the unrealistic smooth ISM

193 6.2. THE SIMULATIONS 175 model that they use is likely to have a significant effect on the existence of small cloud-like objects. To test whether it is theoretically possible to produce H I halo clouds in the lower halo as a direct result of the evolution of a superbubble, high-resolution 3D hydrodynamic simulations of a superbubble expanding within a realistic inhomogeneous medium are performed. The region above the simulated superbubble is examined to determine if clouds analogous to the observed H I halo clouds form, and if so, how they are produced. In 6.2 a brief overview of supercomputing is presented, followed by a description of the hydrodynamic model and initial conditions used for our simulations. The evolutionary stages of the simulated superbubble are briefly described ( 6.3) and halo clouds that appear as a result are discussed ( 6.4). Implications for the origin of H I halo clouds are given in 6.5, while a discussion of the impact of additional physical processes that are not included in the model is given in 6.6. Lastly, a summary of findings is given in The Simulations A Brief Introduction to Supercomputing The amount of computational power required to run 3D hydrodynamic simulations is large. As an example, it would take over CPU (Central Processing Unit) core hours to run one of the simulations presented in this chapter. It is therefore a necessity to parallelise the algorithm, i.e., to break the problem into several pieces that can be worked on mostly independently, and run on multiple CPUs simultaneously. There are two different ways that an algorithm can be parallelised: the first, shared memory, is a scheme where the memory is shared amongst the processors for the entire problem, while the second, distributed memory, is when communications are made via messages between processors and each individual processor works on its piece of the problem in its own local memory. The shared memory architecture is faster because each process has access to all information in its local memory (there is no need to send extra information from one process to another since all processors simultaneously work on the same data space), while the distributed memory architecture is much more flexible because all processors do not need to be on the same machine, making it much easier to run on a larger number of processors. The simulations in this chapter were run on a distributed memory

194 176 CHAPTER 6. SUPERBUBBLE SIMULATIONS AND H I HALO CLOUDS architecture, using the Message Passing Interface (MPI), the most commonly-used and well-supported interface for communicating between different parallel processes. Individual processors usually work on a specific volume within the data and communicate with neighbouring processors, ensuring the boundary conditions are met. Hydrodynamic grid codes are particularly easy to parallelise as the total volume is divided into cubes and information is communicated between processors about the edges of each cube. The simulations in this thesis were performed using the Green Machine, an energy efficient supercomputer that is owned and operated by the Centre for Astrophysics and Supercomputing at Swinburne University of Technology. The Green Machine is composed of 160 nodes, each with two quad-core Intel Clovertown 64-bit processors operating at 2.33 GHz, for a total of 8 cores per node resulting in 1280 cores that can each operate simultaneously. Each node has 16 GB RAM, 1 TB disk space, and they are connected via gigabit ethernet, resulting in a well-equipped supercomputing resource 1. Jobs can be run using a shared memory model on the 8 cores of a single node, or via message-passing on an arbitrary number of nodes The Hydrodynamic Model Simulations were run using Fyris Alpha (Sutherland, 2010), a 3D hydrodynamic grid code that solves the equations of fluid dynamics using a piecewise parabolic method (PPM) and uses Godunov s method to solve the Riemann problem for the jump conditions at the cell boundaries. Fyris Alpha is parallelised by subdividing the computational domain into cubic zones. The main simulation presented, hereafter referred to as Run 1, was executed in parallel on the Green Machine using 64 cores. The resulting data contain cells, where one cell has a linear dimension of 2.8 pc. The final data set therefore spans ±800 pc along the x and y axes and 200 pc to 1400 pc along the z axes (see Table 6.1). Periodic boundary conditions were imposed for the x and y axes. Run 1 evolved for 20 Myr, with full output every 0.5 Myr and 2D snapshots of the central y-z slice every 0.1 Myr. Simulations consisted of a cool, clumpy disk centred on the x-y plane, surrounded by a hot halo. Within the disk, the initial conditions consisted of, in the mean, a particle number density, n =1.0 cm 3, temperature, T = 1280 K, and a pressure, P = dyne cm 2, while the hot halo had an initial T = K, n = 1

195 6.3. THE STAGES OF SUPERBUBBLE EVOLUTION cm 3, and P = dyne cm 2. The cool component was given an initially turbulent velocity field with a magnitude twice that of the adiabatic sound speed. The initial density structure was generated using a Kolmogorov structure function and can be seen in the top panels of Figure 6.1. An energy source representing a cluster with mass of M was used to drive the superbubble, which is at the peak of the cluster mass function in Kroupa & Boily (2002). The energy was computed using Starburst05 using a Miller-Scalo Initial Mass Function (IMF) and the Geneva evolutionary tracks (see Leitherer et al. 1999; Vázquez & Leitherer 2005 and references therein), and exhausted after 12.1 Myr. The cooling function is derived from Mappings III (Sutherland & Dopita, 1993), with additional molecular cooling from Neufeld & Kaufman (1993), Shapiro & Kang (1987), and Le Bourlot et al. (1999). We used an external gravitational potential derived observationally at the solar circle by Holmberg & Flynn (2004). A unique feature of Fyris Alpha is that the equation of state allows the mean molecular weight to vary with temperature, resulting in more accurate pressures. There is no selfgravity, which is a safe assumption as the halo clouds are not gravitationally bound. Further simulations were performed to test the effects of a homogeneous medium (Run 2), different positioning of the cluster energy within the ISM (Run 3), and an initial surface density that is five times larger and a cluster ten times more massive than in Run 1 (Run 4). These supplementary simulations encompass cells, have the same resolution of 2.8 pc, and span ±400 pc in x and y and 200 to 600 pc in z. The surface density, Σ, of the simulated H I disk in Runs 1-3 is 5 M pc 2 and for Run 4 is 25 M pc 2, which brackets the observed range of surface densities of the inner Galaxy (Misiriotis et al., 2006). The disk has a half-height at half-intensity of 115 pc, which is also comparable to observed values (Malhotra, 1994, 1995; Dickey & Lockman, 1990). A summary of all simulations are given in Table The Stages of Superbubble Evolution Simulations of a superbubble were performed to test the hypothesis that H I clouds can form in the lower halo as a result of this process, whether in the form of Rayleigh- Taylor fragments, condensing hot gas that has risen from the disk, gas that has been swept from the disk into the halo, or by another scenario. In Figure 6.1,

196 178 CHAPTER 6. SUPERBUBBLE SIMULATIONS AND H I HALO CLOUDS Table 6.1. Summary of Simulation Properties ID Grid Size Resolution x y z Σ Mcl (pixel) (pc) (kpc) (kpc) (kpc) (M pc 2 ) (M ) Run ±0.8 ± Run ±0.4 ± Run ±0.4 ± Run ±0.4 ± Note. Summary of simulations, where Run 1 is the main simulation and Runs 2 4 are smaller versions with slight modifications: Run 2 has a homogeneous medium instead of a realistic structure, Run 3 has an initial cluster that is offset within the disk, and Run 4 has a larger initial surface density and a cluster that is ten times more massive. x, y, and z represent the range of each of these axes, Σ represents the mean surface density of the disk, and Mcl is the mass of the cluster powering the energy source.

197 6.4. PRODUCTION OF H I CLOUDS IN THE LOWER HALO 179 2D snapshots of a zoomed region of Run 1 present the temperature, density and pressure of the central y-z slice at t = 0, 2, 4, and 6 Myr, illustrating various stages in the superbubble s development; t = 0 Myr represents the initial conditions of the undisturbed, clumpy medium. Initially, an energy source representing a cluster with mass of M injects energy within the disk, causing the surrounding disk material to be shocked and then pushed outwards from the energy source, forming a bubble filled with hot gas within the clumpy medium. This bubble is confined within the disk and as the energy continues to be expelled, the surrounding material is pushed outwards. The carved bubble is not smooth but instead appears to have scalloped edges, due to the denser knots within the disk that are not disrupted as easily as the more diffuse material (see the second row of Figure 6.1). These features are reminiscent of those observed by McClure-Griffiths et al. (2003) and Dawson et al. (2009), where denser clumps form finger-like structures pointing towards the energy source. After 4 Myr, enough energy has been expelled from the cluster that the hot gas inside the bubble has pushed its way through the disk, breaking out of the disk vertically, forming a superbubble with chimneys. Multiple chimneys from which the hot gas escapes the disk and mixes with material in the hot halo are visible in the third row of Figure 6.1. At t = 6 Myr, energy is continuously expelled from the cluster, which causes the bubble to expand farther and the hot gas to continue to escape through the chimneys that have formed. Snapshots of the entire y-z range of Run 1 are shown in Figure 6.2 at t = 4, 6, 8, and 10 Myr. At 4 Myr the bubble has already broken out of the disk and hot gas can been seen rising into the halo. A shock is visible along the top edge of the superbubble. At later time steps, the hotter gas is joined by cooler, denser gas, that is pushed or swept into the halo via the edges of the chimneys, reaching far above the disk. The walls of the chimneys continue to be stripped of gas while the cluster releases energy, resulting in more gas being swept into the halo with time. This continues for a few Myr after the energy source dissipates at t = 12.1 Myr. 6.4 Production of H I Clouds in the Lower Halo As seen in Figures 6.1 and 6.2, during the evolution of the bubble, chimneys are formed allowing hot gas to escape the bubble. Some of the gas along the walls of

198 180 CHAPTER 6. SUPERBUBBLE SIMULATIONS AND H I HALO CLOUDS Figure 6.1 Two-dimensional snapshots of a zoomed region of Run 1 at x = 0 of the density (left), temperature (middle) and pressure (right) at t = 0, 2, 4, and 6 Myr. The initial conditions of the inhomogeneous disk are shown at t = 0 Myr, while the confined bubble can be seen at t = 2 Myr. This bubble is not smooth and its shape is highly dependent on the initial structure of the disk. By t = 4 Myr the bubble has broken out of the disk vertically via chimneys, and at t = 6 Myr hot gas continues to be expelled from the disk.

199 6.4. PRODUCTION OF H I CLOUDS IN THE LOWER HALO 181 Figure 6.2 Two-dimensional snapshots of Run 1 at x = 0 of the density (left), temperature (middle) and pressure (right) at 4, 6, 8, and 10 Myr. By t = 4 Myr hot gas is expelled into the halo via chimneys. The superbubble continues to expand, expelling more hot gas into the halo, the edges of the chimneys are stripped, and gas is pushed farther into the halo. A shock along the upper edge of the bubble can be seen.

200 182 CHAPTER 6. SUPERBUBBLE SIMULATIONS AND H I HALO CLOUDS chimneys is ionised, while some neutral gas survives the blast of energy, getting swept into the halo along with the hot gas. This neutral gas is in the form of clumps and filaments and is seen at a variety of heights above the disk (Figure 6.3). This is the most significant result of these simulations: not only are clumps of gas produced in the lower halo, but these clumps contain H I. The simulated clouds typically have diameters of 15 pc, with a lot of variation, and typical H I number densities of cm 3. These number densities are lower limits because of the resolution limits; the clouds are not well-resolved, being only a few pixels across any given dimension. At higher resolution, densities would be higher because more smallscale structure would be resolved, which would enhance the cooling and allow more neutral gas to survive, but it is impossible to estimate the magnitude of this effect without running higher resolution simulations 2. One of our existing simulations took 250 hours (16000 CPU core hours) on 64 cores and consumed 8 GB of memory total (1 GB per node). To increase the resolution by an order of magnitude, the execution time and memory would have both increased by a factor of 1000, and thus was beyond the resources available. Many clouds are associated with filamentary structures while some appear to be isolated. Vertical velocities of the clouds are generally 30 to 60 km s 1. The H I clumps in the halo have lower pressures than the gas surrounding them, and are found particularly at strong pressure gradients within the halo, i.e., at the edges of high pressure regions. This suggests that the clouds are pressure confined and that they owe their clumpy existence to pressure variations within the breakout region. The clumps of H I are not a unique result of the initial conditions of Run 1; the H I column density images at t = 10 Myr are presented for Runs 2 4 in Figures A simulation with a homogeneous disk using the same initial conditions as in Run 1 (Run 2) still resulted in breakout but over a longer timescale, and also produced clouds, but mostly in the form of large clumps of gas instead of small, individual features. By changing the position of the cluster within the disk (Run 3), the structure of gas above the disk changes. The initial structure within the disk is therefore an important component of the simulations as it affects how much gas is above the disk at later time steps as well as its distribution. The simulation with an initial surface density of 25 M pc 2 and a cluster mass ten times larger (Run 4) also resulted in breakout and clouds above the disk. These clouds are 20 pc 2 Note that this is in contrast to radio observations, where the maximum density that could be observed at a higher resolution is directly calculable from lower resolution observations.

201 6.4. PRODUCTION OF H I CLOUDS IN THE LOWER HALO 183 Figure 6.3 H I column density of Run 1 at t = 10 Myr, projected along the x-axis for gas with T<12000 K and n HI > 10 3 cm 3. Run 1 spans the largest volume of all simulations (576 3 cells), allowing for a more realistic view of the vertical distribution of the clouds. Numerous clumps of H I within the lower halo can be seen, the highest reaching 800 pc at this time step, which is beyond the upper z limit of Runs 2 4.

202 184 CHAPTER 6. SUPERBUBBLE SIMULATIONS AND H I HALO CLOUDS Figure 6.4 H I column density of Run 2 at t = 10 Myr, which only includes gas with T<12000 K and n HI > 10 3 cm 3. As the disk is homogeneous, there are no easy routes of less dense gas by which the hot gas of the bubble can escape, resulting in breakout at a later stage. The gas that eventually rises above the disk is also smoother and less clumpy than that of the inhomogeneous runs, but there are still some clouds within the lower halo, up to heights of 400 pc at this time step. and are more dense than those of Run 1, with typical n HI 0.05 cm 3 (it is not surprising that the increase in cloud density is even greater than the increase in the initial disk density as the cooling rate scales as the square of density). The mass of H I above the edge of the initial disk (z = 256 pc) as a function of time is presented for all four runs in Figure 6.7. Although the resolution of the simulations is not high enough to accurately determine the structure of each clump, and therefore the mass of each clump, the relative mass in each run is directly comparable. As for the number densities, our measured H I masses are lower limits. Breakout of the bubble occurs at roughly the same time in all runs except that with the homogeneous medium, which occurs on timescales roughly twice as long. Placing the cluster at a different location in the disk changes the resulting amount

203 6.4. PRODUCTION OF H I CLOUDS IN THE LOWER HALO 185 Figure 6.5 H I column density of Run 3 at t = 10 Myr, which only includes gas with T<12000 K and n HI > 10 3 cm 3. The initial position of the cluster within the disk is different from that of Run 1, resulting in a different expansion of the superbubble and a different path from which the hot gas escapes the disk. Numerous clouds are present within the lower halo, up to the maximum z within this run (600 pc).

204 186 CHAPTER 6. SUPERBUBBLE SIMULATIONS AND H I HALO CLOUDS Figure 6.6 H I column density of Run 4 at t = 10 Myr, in which the surface density of the disk is five times denser than that of Run 1. Only gas with T < K and n HI > 10 3 cm 3 is shown. There are many clouds above the disk, with larger densities than those seen in the other runs, extending up to the maximum z of this run (600 pc).