The Globular Cluster System of NGC 4365

Christina Blom

Presented in fulﬁllment of the requirements of the degree of Doctor of Philosophy

May 2013

Faculty of Information and Communication Technology Swinburne University

i Abstract

This thesis presents a study of the globular cluster (GC) system of NGC 4365 and the GCs associated with the W 0 group of galaxies, of which NGC 4365 is the dominant galaxy. We use the analysis of these GC systems to investigate the evolutionary history of NGC 4365 as well as the ongoing interactions within the W 0 group. To analyse the photometric properties of NGC 4365’s GC system we combined three filter imaging from the SuprimeCam instrument on the 8m Subaru telescope with eight, two filter, pointings from the Advanced Camera for Surveys on the Hubble Space Tele- scope. To analyse the kinematic properties of NGC 4365’s GC system we obtained spectra for over 250 GCs around NGC 4365 from the DEep Imaging Multi-Object Spectrograph (DEIMOS) on the Keck II telescope. The photometric properties of the group GCs were analysed with square degree, three filter imaging from the MegaCam instrument on the Canada-France-Hawaii Telescope. We confirmed that NGC 4365 hosts three GC subpopulations, the usual blue and red GC subpopulations plus an additional subpopulation at intermediate colour: the green subpopulation. Photometric analysis showed that the three subpopulations have distinct radial and azimuthal distributions, different median sizes and mass distributions drawn from different populations. Using recession velocities calculated from GC spectra we also determined that each GC subpopulation rotates about the galaxy in a different direction. Analysis of the spatial distribution of GCs around NGC 4365 in the wider W 0 group environment uncovered a stream overdensity of GCs between NGC 4365 and a nearby small lenticular galaxy NGC 4342, extending South West beyond NGC 4342. This GC stream is spatially coincident with a stellar stream recently presented in the literature. We found that the recession velocities of the stream GC are consistent with the recession velocity measured from NGC 4342’s starlight. We conclude that NGC 4365 formed two GC subpopulations during two separate in situ, dissipative star formation episodes (the green GCs forming before the red GCs) and accreted the blue GCs from smaller galaxies throughout its evolutionary history. We also conclude that NGC 4365 is currently accreting blue GCs and stars from the small lenticular galaxy NGC 4342. As predicted by the hierarchical merging models of ΛCDM, we have observed evidence of ongoing growth of the giant elliptical galaxy NGC 4365 in the nearby Universe. ii iii Acknowledgements

I begin by acknowledging my parents for their love, guidance, motivation and patience. I have always known that I am loved by you and that knowledge let me be the exploring and questioning person that started this thesis. You have guided me through the maze of getting educated with my curiosity and joy of discovery intact. I am especially grateful for the constant words of encouragement to dream big, to work hard, to keep my eyes on the goal and to take one step forward after the other. Without my Ph.D. supervisor Prof. Duncan A. Forbes, I would not have had access to the data that contained the discoveries presented in this thesis. More importantly I would not have had the skills or insight to see what was to be uncovered. Thank you for your patient teaching, daily availability for questions and discussions, openness to explore new ideas and consistent reminders of short-term goals. To my co-supervisors; thank you Dr Lee R. Spitler for the recurrent question ‘What does it mean?’ and thank you Prof. Alister Graham for encouraging me to pay attention to every detail. I am also indebted to my collaborators Prof. Jean P. Brodie, Dr Caroline Foster, Dr Aaron J. Romanowsky and Dr Jay Strader for their support and critical assessment of my work. I would also like to thank the staﬀ and students at the Centre for Astrophysics and Supercomputing, Swinburne University for the friendship, community and advice over the years. In particular I would like to thank Dr. Lina Levin, Vincenzo Pota, Christopher Usher, Sreeja Kartha and Nicola Pastorello for your individual contributions to life in our corner of the oﬃce and my Ph.D. journey. Swinburne University has generously supported me through my Ph.D. with scholar- ships as well as an administrative backbone aiming to give students the best resources for their degree. Thank you, my beloved Michael D. Smith, my husband and best friend. You have been steady, patient and unfailingly kind and generous. I am privileged to be adventuring through this world with you and I need to tell you that I would not have made it through this challenge without you. My family and friends scattered throughout the globe have been for me ‘God’s love with skin on’ during my Ph.D. I would be lost without His Light and I feel honoured to have had the opportunity to discover parts of His Universe. iv v Declaration

The work presented in this thesis has been carried out in the Centre for Astrophysics & Supercomputing at Swinburne University of Technology between 2009 and 2013. This thesis contains no material that 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 author, except where due reference is made in the text of the thesis. The content of the chapters listed below has appeared in refereed journals. Minor alterations have been made to the published papers in order to maintain argument continuity and consistency of spelling and style.

• Chapter 2, Chapter 3, Appendices A and B have been published together as: “Wide-ﬁeld imaging of NGC 4365’s globular cluster system: the third subpopulation revisisted” Blom, C., Spitler, L. R. and Forbes, D. A., 2012, MNRAS, 420, 37

• Chapter 4 and Appendix C have been published together as: “The SLUGGS survey: globular cluster system kinematics and substructure in NGC 4365” Blom, C., Forbes, D. A., Brodie, J. P., Foster, C., Romanowsky, A. J., Spitler, L. R. and Strader, J., 2012, MNRAS, 426, 1959

• Chapter 5 has been submitted to MNRAS as: “The SLUGGS survey: New evidence for a tidal interaction between the early type galaxies NGC 4365 and NGC 4342” Blom, C., Forbes, D. A., Foster, C., Romanowsky, A. J. and Brodie, J. P., 2013

Contributions to the papers from coauthors are included in this work for clarity and continuity. All work is my own, unless a Section or piece of work is speciﬁcally attributed to a coauthor by name.

Christina Blom Melbourne, Victoria, Australia 2013 vi Contents

Abstract i

Acknowledgements iii

Declaration v

List of Figures xi

List of Tables xv

1 Introduction 1 1.1 Globular clusters ...... 2 1.2 Globular cluster systems ...... 2 1.2.1 Bimodality ...... 3 1.2.2 Trends within GC systems ...... 4 1.2.3 GC system trends ...... 5 1.3 Early-type galaxies ...... 6 1.4 Using globular cluster systems to study galaxy evolution ...... 7 1.5 Thesis outline ...... 8

2 Observing extragalactic globular clusters 9 2.1 Introduction ...... 9 2.2 Subaru/Suprime Cam Data ...... 11 2.2.1 Observations and data reduction ...... 11 2.2.2 Photometry and calibration ...... 11 2.3 HST/ACS Data ...... 13 2.3.1 Observations and object measurement ...... 13 2.4 GC candidate Selection ...... 14 2.4.1 HST/ACS GCs ...... 14 2.4.2 Subaru/SuprimeCam GCs ...... 16 2.4.3 Contamination and completeness ...... 19 2.4.4 Overview of GC selection ...... 21

3 Wide-ﬁeld imaging of NGC 4365’s globular cluster system 23 3.1 Introduction ...... 23 3.2 Analysis of the galaxy light ...... 25

vii viii Contents

3.2.1 Parameters of the isophotal ﬁts ...... 25 3.3 Analysis of the GC System ...... 27 3.3.1 Characterising the total GC System ...... 27 3.3.2 Radial colour gradients ...... 34 3.3.3 Colour-magnitude trends ...... 36 3.3.4 Quantifying the GC subpopulations ...... 38 3.3.5 Characterising the GC system subpopulations ...... 42 3.3.6 Comparison with galaxy surface brightness ...... 53 3.4 Discussion ...... 53 3.5 Summary and Conclusions ...... 57

4 Globular cluster system kinematics and substructure in NGC 4365 59 4.1 Introduction ...... 59 4.2 Spectroscopic sample ...... 63 4.2.1 Data acquisition ...... 63 4.2.2 Obtaining line-of-sight velocities ...... 65 4.2.3 Low velocity GCs ...... 69 4.3 Kinematics of the GC subpopulations ...... 71 4.3.1 Kinematic model description ...... 71 4.3.2 Kinematics as a function of colour ...... 73 4.3.3 Dividing the sample into three subpopulations ...... 77 4.3.4 Radial kinematics for three subpopulations ...... 79 4.4 Discussion ...... 84 4.5 Conclusions ...... 87

5 Tidal interaction between NGC 4365 and NGC 4342 91 5.1 Introduction ...... 91 5.2 Photometric Analysis ...... 93 5.2.1 Imaging Data ...... 93 5.2.2 Globular cluster candidate selection ...... 93 5.2.3 Spatial distribution ...... 97 5.2.4 Colour distribution ...... 101 5.3 Spectroscopic Analysis ...... 106 5.3.1 Spectroscopic Data ...... 106 5.3.2 Velocity selection ...... 108 5.4 Kinematic properties of NGC 4342’s GC system ...... 111 Contents ix

5.5 Dark matter constraints ...... 113 5.6 Discussion ...... 116 5.7 Conclusions ...... 118

6 Conclusions and future work 119 6.1 Chapter Summary ...... 119 6.2 Avenues for Further Research ...... 122 6.2.1 Substructure within galaxy groups and clusters ...... 122 6.2.2 Searching for the causes of signiﬁcant kinematic misalignment . . . . 123 6.3 Broad impact of this work ...... 126

Bibliography 129

A Statistical tests for trimodality 139 A.1 Kaye’s Mixture Model algorithm ...... 139 A.2 The Kolmogorov-Smirnov test ...... 142 A.3 Chi-Squared minimization ...... 143

B Surface density data 145 B.1 Surface density data ...... 145

C Additional supopulation kinematic fits 147 C.1 Total subpopulation kinematic fits ...... 147 C.2 Radial kinematic fits with fixed position angle ...... 150

List of Figures

2.1 Three filter (g0, r0 and i0) image stack of NGC 4365 from the Subaru Tele- scope SuprimeCam instrument...... 10 2.2 Calibration of Subaru SuprimeCam photometry to Sloan Digital Sky Survey filters...... 12 2.3 Comparison of ACS sizes and magnitudes published in Jordánet al. (2009) with those measured in this work...... 14 2.4 Size in parsec vs. g − z colour distribution for GC candidates...... 15 2.5 Colour magnitude diagram for GC candidates selected from HST/ACS imaging...... 15 2.6 Diagram to show selection of point sources in the S-Cam imaging...... 17 2.7 Colour selection of GCs in the S-Cam field-of-view...... 18 2.8 Colour magnitude diagram for GC candidates selected from S-Cam imaging. 19 2.9 Estimate of contamination in the S-Cam data...... 20 2.10 Estimate of the completeness of the S-Cam photometric sample...... 21

3.1 Isophotal parameters of NGC 4365 for g0, r0 and i0 ﬁlters...... 26 3.2 Surface density of GC candidates plotted against galactocentric radius. . . . 28 3.3 Surface density of GC candidates in two dimensions from S-Cam data. . . . 30 3.4 Histogram of the azimuthal distribution of GC candidates...... 31 3.5 GC density with colour and galactocentric radius for the ACS data. . . . . 32 3.6 GC density with colour and galactocentric radius for the S-Cam data. . . . 33 3.7 Rolling peak colour values plotted against galactocentric radius...... 34 3.8 The colour-magnitude relation for blue GC candidates...... 37 3.9 Colour distribution of GC candidates at all radii...... 40 3.10 The normalised distribution of ACS GCs with colour in radial bins...... 41 3.11 The normalised distribution of S-Cam GCs with colour in radial bins. . . . 43 3.12 GC radial surface density for three subpopulations incorporating ACS and S-Cam GCs...... 44 3.13 The percentage of each subpopulation plotted against galactocentric radius. 46 3.14 The half light radius of bright HST/ACS GC candidates against galactocentric radius...... 47 3.15 The normalised mass function for GC candidates from the ACS catalogue. . 49 3.16 Azimuthal distributions of high probability GC subpopulations...... 51 3.17 Comparison of galaxy light surface brightness with GC surface density. . . . 52

xi xii List of Figures

4.1 The positions of the six DEIMOS slitmasks plotted on the S-Cam image. . 60 4.2 Colour distribution of GCs with radial velocity measurements compared with the total photometric sample...... 63 4.3 Line-of-sight velocity distribution for GCs in the kinematic sample...... 66 4.4 Line-of-sight velocity of GCs plotted against galactocentric radius...... 67 4.5 GC surface density with colour and galactocentric radius...... 68 4.6 Spatial distribution of GCs showing individual radial velocities...... 70 4.7 Kinematics for NGC 4365 GCs as a function of g0 − i0 colour...... 73 4.8 Observed line-of-sight velocity plotted against position angle for the three GC subpopulations...... 77 4.9 Kinematics as a function of radius for three GC subpopulations ...... 80 4.10 Kinematic parameters of GCs as a function of galactocentric radius. . . . . 82 4.11 Schematic representation of the NGC 4365 stellar light and GC system. . . 88

5.1 Spatial alignment of u and g ﬁlter imaging from the CFHT...... 94 5.2 Colour selection of globular cluster (GC) candidates...... 95 5.3 Spatial distribution of GC candidates around NGC 4342 and NGC 4365. . . 96 5.4 Radial surface density of GCs around NGC 4342...... 98 5.5 Azimuthal distribution of GCs around NGC 4342...... 100 5.6 Simulated spatial distribution of the GCs around NGC 4365 and NGC 4342.102 5.7 Calculated stream overdensities for 1000 simulated spatial distributions. . . 103 5.8 Scaling between Subaru/SuprimeCam and CFHT/MegaCam colours. . . . . 104 5.9 Colour distribution of NGC 4342 and stream GCs...... 105 5.10 Comparison of independent velocity measurements...... 107 5.11 Phase-space diagram of GC velocity as a function of distance from NGC 4365.109 5.12 GC system velocity dispersion in three radial bins...... 112 5.13 Imaging of NGC 4365 and other nearby galaxies in the W 0 group...... 113 5.14 Circular velocity proﬁle of NGC 4342...... 115

6.1 Reproduction of Figures 5.13 and 5.3...... 124 6.2 Reproduction of Figure 8a from Krajnovi´cet al. (2011)...... 125

A.1 Bimodal and trimodal ﬁts to the Epanechnikov kernel smoothing of the ACS GC colour distribution...... 144 A.2 Bimodal and trimodal ﬁts to the Epanechnikov kernel smoothing of the S-Cam GC colour distribution...... 144 List of Figures xiii

C.1 Kinematics as a function of radius for the three GC subpopulations. . . . . 151

List of Tables

2.1 Imaging obtained from the Hubble Legacy Archive. The pointings are numbered to correspond with the numbers in Figure 2.1 and HST ID refers to the HST proposal identification number where the principle investigators are as follows: 9401 - Patrick Côté,10584 - Gregory Sivakoff, 9488 - Kavan Ratnatunga...... 13

3.1 Values for g, r and i Sérsicprofiles fitted to galaxy light...... 25 3.2 Fitted values of the Sérsicfit to the GC surface density...... 29 3.3 GC colour gradients for the linear fit to colour with radius in arcmin. . . . 35 3.4 GC colour gradients for the logarithmic fit to colour with normalised radius. 35 3.5 Metallicity gradients for the logarithmic fit to colour with normalised radius. 36 3.6 Gaussian values for the blue and red GC distributions...... 39 3.7 Sérsicprofile fits to the radial surface density of GC subpopulations. . . . . 45 3.8 Estimated ellipticity for blue, green and red GC subpopulations...... 52

4.1 Summary of kinematic ﬁts to various GC subpopulation divisions...... 75 4.2 Summary of the NGC 4365 galaxy system properties...... 83

5.1 Recession velocities for GCs around NGC 4342 and in the stream...... 110

A.1 Results from the extended KMM test for signiﬁcance of m modes...... 141 A.2 KS test results comparing galaxies in the ACS Virgo Cluster Survey. . . . . 142 A.3 KS test results comparing simulated distributions with NGC 4365’s colour distribution...... 142

B.1 GC radial surface density data for the total GC system and each subpopulation...... 146

C.1 Kinematic ﬁts to various divisions of the three GC subpopulations...... 149 C.2 Minimisation parameter divided by degrees of freedom for kinematic ﬁts. . . 150

1 Introduction

The heavens declare the glory of God; the skies proclaim the work of his hands. Day after day they pour forth speech; night after night they reveal knowledge. They have no speech, they use no words; no sound is heard from them. Yet their voice goes out into all the earth, their words to the ends of the world. —Psalm 19:1-4a

Since the discovery that the Universe contains galaxies separate from our own (Hubble, 1926), scientists have been fascinated by the question of how they came to be. Once it was discovered that galaxies are not all spiral in structure, like our own Milky Way and the Andromeda galaxy, but that some have elliptical shapes, like M32 and NGC 404, the question of whether they formed differently or evolved into different types of galaxies, was also raised. Spiral (late-type) and elliptical (early-type) galaxies have many very different properties, but large galaxies (both spiral and elliptical) all host globular clusters (GCs). In this thesis we use GCs and the properties of a GC system to answer some of the current questions of galaxy evolution. We will consider the case of an intriguing giant elliptical galaxy, NGC 4365. Here we introduce the properties of individual GCs, current understanding of GC systems, salient aspects of elliptical galaxies and then discuss the ways in which GC systems can be used to study questions of galaxy formation and evolution.

1 2 Chapter 1. Introduction

1.1 Globular clusters

Globular clusters are very dense clusters of stars, gravitationally bound to within 2 - 10 pc in radius (van den Bergh et al., 1991; Harris, 1991). Unlike other stellar systems GCs contain very little gas and no dark matter. The discovery of GCs began in the mid 17th Century when Abraham Ihle spotted the perfectly round nebula-like M22, but he could not resolve the GC into stars. Only in the 1780’s when William Herschel was doing a survey of many of these round nebulae were they first termed globular clusters (Herschel, 1786, 1789). The internal properties of Milky Way GCs have been studied in great detail since those first discoveries. Studies of the photometric colours and chemical properties of individual stars in GCs have shown that, to first order, the stars within a single GC formed at the same time, from the same original gas cloud. Using resolved stellar photometry and an understanding of stellar evolution, the horizontal giant branch turnoff (Hesser et al., 1987) or main sequence turnoff (Gratton, 1985) have been used to age date most of the GCs in the Milky Way. It has been found that most of the GCs in the Milky Way are > 10 Gyr old (Mar´ın-Franch et al., 2009). Spectroscopic studies have confirmed that the mean [Fe/H] metallicity of the stars in a Milky Way GC range from 0.0 to -2.3 (Harris, 2010). The picture of GCs as old stellar systems that were formed early in the evolution of the Universe from low-metallicity gas, has been confirmed by detailed studies of GCs in other galaxies (see Strader et al., 2005, and references therein). Very recent studies of resolved stars in Milky Way GCs have revealed that some GCs actually host multiple stellar populations (Bedin et al., 2004; Piotto et al., 2007; Milone et al., 2008). This was determined from the presence of multiple turnoff points for the main sequence of stellar evolution. However, for the GCs where this has been seen the age difference between the stellar populations is < 1 Gyr (Gratton et al., 2012). There is debate as to the formation processes for multiple stellar populations within GCs (e.g. Decressin et al., 2007; Conroy & Spergel, 2011; Valcarce & Catelan, 2011) but for GCs where the stars cannot be individually resolved (i.e. for galaxies outside the Local Group) it is reasonable to view GCs as single stellar population systems that have survived from the early Universe.

1.2 Globular cluster systems

In 1932 Edwin Hubble presented a catalogue of ‘Nebulous objects in Messier 31 provision- ally identiﬁed as globular clusters’ identiﬁed from photographic plates (Hubble, 1932). 1.2. Globular cluster systems 3

This marked the start of GC system studies outside the Milky Way. That ﬁrst study used many of the photometric techniques we use to study extragalactic GC systems today (e.g. measures of the radial distribution and most common GC magnitude), however many new techniques have also been added to the ﬁeld over the last 80 years. We present a summary of the salient discoveries of GC system research since the publication of Edwin Hubble’s catalogue.

1.2.1 Bimodality

Zepf & Ashman (1993) discovered that the GC systems of the giant elliptical galaxies NGC 5128 and NGC 4472 are both bimodal in optical colours. Within a decade it was clear that the GC systems of most giant elliptical galaxies are bimodal and that GC bimodality is common in large spiral galaxies too (Forbes et al., 1997; Larsen et al., 2001; Kundu & Whitmore, 2001). Given the degeneracy between age and metallicity in optical colours is was not initially clear whether GC systems were bimodal in metallicity or in age. Observations of Milky Way GCs had shown them to be > 10 Gyr old and if both modes/subpopulations had the same metallicity a 13 Gyr age gap would be required to explain the observed gap in colour between the two subpopulations (Forbes et al., 1997). It was generally assumed that all the GCs in a galaxy were > 10 Gyr old, roughly coeval and with different intrinsic metallicity, referring to a blue metal-poor subpopulation and a red metal-rich subpopulation. This paradigm was challenged by Yoon et al. (2006) who postulated that while GCs are likely old and roughly coeval they are not bimodal in metallicity. Rather, that there is a continuous distribution of GCs from metal-poor to metal-rich and the observed bimodality in optical colours is due to a highly non-linear relationship between intrinsic GC metallicity and observed GC colour (see also Cantiello & Blakeslee, 2007; Blakeslee et al., 2010). Recent observations of GC metallicity, using various spectroscopic indices, have shown that many GC systems are bimodal in metallicity and that the relationship between GC colour and GC metallicity is well approximated with a linear relation (Strader et al., 2007; Woodley et al., 2010; Alves-Brito et al., 2011; Usher et al., 2012). Since the discovery of GC system bimodality various people have researched the different properties of each GC subpopulation to investigate their possible origins. They found that while there is a large scatter in GC sizes, the mean size of blue GCs is larger than the mean size of red GCs (Larsen & Brodie, 2003; Jordán,2004). It was also determined, through fitting de Vaucouleurs and Sérsicprofiles to the radial density of GC subpopulations, that blue GC subpopulations are generally flatter and extend further from the 4 Chapter 1. Introduction galaxy centre than red GC subpopulations (Harris, 2009b; Pota et al., 2013). It is also possible to do kinematic studies of GC systems. These studies require secure recession velocities for a large number of GCs in each subpopulation and the sample of galaxies where this analysis has been possible is still growing. Early results show that the red GC subpopulation of large galaxies generally have similar kinematic properties to the galaxy stellar light, while the blue GC subpopulations either have kinematic properties that are decoupled from that of the red subpopulations or are dominated by random motions (e.g. Côtéet al., 2001, 2003; Lee et al., 2010; Arnold et al., 2011; Foster et al., 2011; Pota et al., 2013). Bimodality in GC systems implies that there are significant differences in the formation of blue and red GCs. There are various interpretations of the formation differences; perhaps blue GCs formed at a different time to red GCs (Forbes et al., 1997), perhaps blue and red GCs were formed via different physical processes (Ashman & Zepf, 1992) or perhaps they were formed in very different environments (Côtéet al., 1998). It is thought that GCs formed during intense periods of general star formation over the history of galaxy evolution (Larson, 1990) and consequently the conclusions we draw regarding the differences in formation of blue and red GC subpopulations extend to our understanding of the formation of their host galaxies.

1.2.2 Trends within GC systems

The study of GC systems has revealed several relationships between the properties of GCs within each galaxy. For example, the mean GC size (usually measured from the half- light radius of the GCs) increases with increasing galactocentric radius. This size-radius relationship has been measured in several galaxies (the Milky Way: van den Bergh et al. 1991, the brightest eight Virgo ellipticals: Jordánet al. 2005, NGC 4594: Spitler et al. 2006a, NGC 5128: Gómez& Woodley 2007, and a sample of 6 galaxies together: Harris 2009a) and with slopes varying from 0.05 to 0.36 dex dex−1. It has been suggested that this apparent relationship arises due to a combination of two properties of the blue and red GC subpopulations, namely that blue GCs tend to be larger in size than red GCs and blue GC subpopulations tend to be radially more extended than red GC subpopulations (Jordán et al., 2005; Masters et al., 2010). The argument being that smaller, red GCs dominated the central regions and larger, blue GCs dominated the outer regions of galaxies, while others (e.g. Harris, 2009a) argued that the radial trend of increasing GC size was due to differences in the GCs interaction with the galaxy potential well at different radii. 1.2. Globular cluster systems 5

Spitler et al. (2006a) also discovered a relationship between the magnitude and the colour of GCs in the blue GC subpopulation of the Sombrero galaxy (NGC 4594) and called it a ‘blue-tilt’. It was seen soon after in several other galaxies (Strader et al., 2006; Mieske et al., 2006; Harris et al., 2006). This colour-magnitude, or mass-metallicity relationship has been attributed to the self-enrichment of GCs, whereby larger GCs are better able to retain the gas ejected by fast lived stars to create more metal rich stars (Strader & Smith, 2008; Bailin & Harris, 2009). Several galaxies show a blueward trend of mean colour with increasing galactocentric radius of one or both GC subpopulations (see Forbes et al., 2011, and references therein). One or both GC subpopulations of NGC 1407 (Forbes et al., 2011), NGC 4472 (Geisler et al., 1996), NGC 1399 (Bassino et al., 2006), M87 (Harris, 2009b), NGC 3268, NGC 4696 and NCG 7626 (Harris, 2009a) have mean metallicities that are more metal rich in the outer parts than the inner parts. The cause of this trend is not yet known but Forbes et al. (2011) suggest that it arises due to dissipative processes when stars and GCs are formed in situ.

1.2.3 GC system trends

To understand the connection between GC systems and their host galaxies it is important to compare the GC system properties of various different galaxies and identify trends. It became clear relatively early that the luminosity function of the GCs is almost identical across all the observed GC systems (Hanes, 1977; Hanes & Whittaker, 1987). The log- normal distribution of GC luminosity peaks at Mz = −8.4 mag (Jordánet al., 2007) in every GC system that has been studied and the universality of this value has been used to determine galaxy distances in several cases (see the review by Jacoby et al. 1992 and a modification by Ashman et al. 1995). The specific frequency, a measurement of the number of GCs normalised by the luminosity of the host galaxy (Harris & van den Bergh,

1981), does vary with host galaxy property. SN plotted against host galaxy luminosity forms a horse-shoe shape (see Forbes, 2005; Peng et al., 2008), low for intermediate mass galaxies and increasing with both decreasing and increasing luminosity. The relationship is highly scattered at low luminosities, as these galaxies are diﬃcult to observe and their GC numbers are low, but the relationship of increasing SN with increasing galaxy luminosity is well deﬁned. In general, larger/brighter galaxies host more GCs per unit luminosity than fainter/smaller galaxies. Several researchers over the last decade have found that the properties of the red GC subpopulation correlate well with the properties of the host galaxy. Both the colour and number of red GCs in a GC system show a trend with host galaxy luminosity (Peng et al., 6 Chapter 1. Introduction

2006) and the mean colour of red GCs is often similar to the colour of the galaxy starlight (Forbes & Forte, 2001). These results support the hypothesis that the red subpopulation of GCs are associated with the bulge stars of galaxies (Forbes et al., 1997; Forbes & Forte, 2001), while the blue GC subpopulation is likely associated with the halo of galaxies. Recent investigation has found that for a number of galaxies the blue GC radial surface density is similar to the hot gas density proﬁle as measured from X-ray emission (Forbes et al., 2012).

1.3 Early-type galaxies

The primary feature of early-type (lenticular and elliptical) galaxies is that their current rate of star formation is generally very low. Viewed in optical colours they appear red because they do not contain many young blue stars, but mainly older yellow and red stars. They also generally do not host significant amounts of neutral hydrogen nor do giant ellipticals have strong rotation profiles, although it seems low luminosity ellipticals rotate more significantly (Emsellem et al., 2007). Elliptical galaxies are often found in the very centre of the largest clusters of galaxies and the very largest observed galaxies are elliptical (e.g. M87 and NGC 1399). There has been significant debate about whether spiral galaxies become lenticular or elliptical galaxies when they are starved of gas and/or merge with other galaxies (Kauffmann et al., 1993; Cole et al., 1994; Baugh et al., 1996; Navarro et al., 1997), or if elliptical galaxies were formed without coherent rotation and have since just been starved of the gas used for original star formation (Eggen et al., 1962; Larson, 1974; Carlberg, 1984; Arimoto & Yoshii, 1987). The lack of neutral hydrogen presents a significant hurdle in attempts to study the internal structure or dark matter distribution of these galaxies and thus it is very difficult to disentangle the evolutionary history of elliptical or lenticular galaxies. Daddi et al. (2005) discovered superdense or ‘red nugget’ galaxies in the early universe and as a result a new scenario for elliptical galaxy formation has emerged. It is suggested that elliptical galaxies evolve from the red nuggets in the early universe, likely formed through dissipative collapse into dark matter overdensities. To evolve into the giant passive galaxies seen today the ‘red nuggets’ accrete many small galaxies, gaining a little bit of mass but growing significantly in radial extent due to the frequent, minor tidal events (van Dokkum et al., 2008). This two-phase growth of galaxies, with early dissipative collapse and subsequent minor mergers, is also supported by the results of simulations such as Naab et al. (2009). 1.4. Using globular cluster systems to study galaxy evolution 7

1.4 Using globular cluster systems to study galaxy evolution

Perhaps the first example of using GC systems to study the fundamental parameters of galaxies is the work of Harlow Shapley using GCs in the Milky Way to determine the size and centre of our own Galaxy (Shapley, 1918). He mapped the positions and distances of all the known Galactic GCs to form a 3D model, finding that the GCs roughly filled a sphere and that the sphere was not centred on our own solar system but a point roughly 15 kpc away from us. He concluded that the sun is not at the centre of the Milky Way and that our Galaxy is much larger than other measurements at the time predicted.

With modern observing techniques it is possible to investigate intrinsic GC colours or metallicities as well the kinematic properties of the GC subpopulations for many galaxies and compare the results (e.g. Usher et al., 2012; Pota et al., 2013). We can use these detailed studies of GC system properties to disentangle the evolutionary history of their host galaxies. Additionally, because GC systems are much more extended than the underlying galaxy light, GC systems are often used to trace the dark matter distribution around elliptical galaxies (e.g. Romanowsky et al., 2009).

The discovery of universal GC system bimodality has had a major impact on the study of galaxy evolution. Because the blue and red GC subpopulations have been shown to be distinct in metallicity space (Usher et al., 2012) it follows that they have formed separately. If GC formation is associated with episodes of major star formation (Larson, 1990), universal GC bimodality suggests that the stars seen in galaxies were overwhelmingly formed during two episodes. This seems to contradict the continual star formation that is predicted by the ΛCDM model (Blumenthal et al., 1984; Davis et al., 1985; Navarro et al., 1997). However, more subtle implementations of ΛCDM modelling with GC formation included (e.g. Muratov & Gnedin, 2010; Tonini, 2013) indicate that GC system metallicity bimodality can be reproduced by hierarchical clustering. Muratov & Gnedin (2010) and Tonini (2013) diﬀer in the processes of red and blue GC formation, using galaxy mergers to trigger GC formation, and using the measured peak of cosmic star formation as an indicator for GC formation, respectively. There are still many questions to be answered about the formation of elliptical galaxies and their GC systems.

Here, we look at the speciﬁc case of a giant elliptical galaxy where a third GC subpopulation is suspected. This investigation could provide the key to understanding the causes of universal bimodality, by identifying the origin of the additional GC subpopulation in NGC 4365. 8 Chapter 1. Introduction

1.5 Thesis outline

Chapter Two presents the photometric data, the starting point for the investigation of the ‘Globular Cluster System of NGC 4365’. We present techniques to discover GCs at a large distance from the observer and contrast the process of identifying GCs with ground based and with space based imaging. The data analysis and discussion are presented in Chapter Three, characterising the total GC system in Section 3.3.1 followed by an argument for the division of the GC system into three separate subpopulations in Sections 3.3.2 - 3.3.4 and analysis of the properties of each GC subpopulation in Section 3.3.5. We analyse the kinematic properties of the GCs of NGC 4365 in Chapter Four. Section 4.2 presents the spectroscopic data and recession velocity distribution while Section 4.3 contains the kinematic analysis of each GC subpopulation in NGC 4365. The analysis and discussion in Chapter Five illuminate the process of tidal stripping that is happening in the galaxy group around NGC 4365. This is done using both photometric (Section 5.2) and spectroscopic (Sections 5.3 and 5.4) studies. In Chapter Six we summarise the main results and arguments from Chapters Two to Five, discuss possible avenues for future research and conclude with a comment on the broader impact of the discoveries presented in the thesis. 2 Observing extragalactic globular clusters

A scientist in his laboratory is not a mere technician: he is also a child confronting natural phenomena that impress him as though they were fairy tales. —Prof. Marie Skodowska-Curie

2.1 Introduction

The observation of globular clusters (GCs) in our own galaxy is fundamentally different from studies of GCs in external galaxies at the distance of the Virgo Cluster. With modern instruments a large percentage of the stars in Galactic GCs can be individually resolved for photometric and spectroscopic measurement. This means that galactic GCs can be unambiguously identified as such, without risking confusion between GCs, stars, star- forming regions or background galaxies. Probing out to the distance of the Virgo Cluster it is no longer possible to measure internal chemical, dynamical or evolutionary properties for each GC by resolving and analysing each star. It is however possible to obtain information about the GC from the integrated properties of all its stars combined in one observation. The difficulty with observing GCs when individual stars cannot be resolved starts with the problem of unambiguously determining if the object is actually a GC. We will discuss the photometric methods currently used to select a sample of objects that is dominated by GCs. These techniques are different depending on whether ground-based observations or space based observations are available. The data set obtained for the giant elliptical galaxy NGC 4365 (a member of the W 0 group behind the Virgo Cluster) contains imaging from the Suprime-Cam instrument on the ground-based 8m Subaru telescope (Subaru/S- Cam) as well as from the Advanced Camera for Surveys on the Hubble Space Telescope

9 10 Chapter 2. Observing extragalactic globular clusters

Figure 2.1: The central 34 × 25 arcmin section of the Subaru/S-Cam g0, r0 and i0 ﬁlter combined image (trimmed from 35 × 27 arcmin for cosmetic purposes) and the footprints of the 8 HST/ACS pointings are shown. The scale of the ACS pointings is shown in the bottom left corner. Image is centred on α=12:24:26.824; δ=+07:19:03.52 (J2000.0). At a distance of 23.1 0.8 Mpc (Blakeslee et al., 2009) 1arcsec = 0.112 kpc.

(HST/ACS).

While observations of the individual GC properties are diﬃcult to do for a galaxy at the distance of NGC 4365, observations of the global properties of its GC system are more easily done because it is a giant elliptical galaxy external to the Milky Way. We can see the whole of NGC 4365, as apposed to missing all the Milky Way GCs hidden by the dense centre of our Galaxy. The other beneﬁt of studying the GC systems of elliptical galaxies is that they generally contain little or no dust, that obscures many GCs on the far side of the Andromeda galaxy for example.

In Section 2.2 and 2.3 we overview the data acquisition and reduction for the Subaru/S- Cam and HST/ACS photometry respectively. We describe the GC candidate selection criteria and detail the photometric properties of the GC candidate samples in Section 2.4. 2.2. Subaru/Suprime Cam Data 11

2.2 Subaru/Suprime Cam Data

2.2.1 Observations and data reduction

On 2008, April 1st we obtained 35 × 27 arcmin three-filter imaging of NGC 4365 using Subaru/S-Cam. Conditions were good and the worst seeing was ∼ 0.8 arcsec, exposure times were 5 × 130 s, 5 × 70 s and 5 × 60 s for g0, r0 and i0 filters respectively. The images were bias subtracted, flat field corrected and stacked using the SDFRED package (Yagi et al., 2002; Ouchi et al., 2004) and put onto the USNO-B2 astrometric system. Figure 2.1 shows a combination of the g0, r0 and i0 filter images. The S-Cam instrument has a pixel scale of 0.202 arcsec. The NGC 4365 galaxy light was modelled and subtracted in each of the three filters using the IRAF task ELLIPSE before an object detection algorithm was employed. This was done to increase the success with which the DAOFIND detection algorithm finds faint objects in the central regions of the galaxy (it is unsuccessful at finding objects in areas in which the background surface brightness varies). The standard deviation of the background, after galaxy subtraction, was σ = 8.64 counts in g0, σ = 11.16 counts in r0 and σ = 13.15 counts in i0. Detection thresholds of 2.7σ, 3.0σ and 2.1σ in g0, r0 and i0 respectively were used so as to probe as deep as possible in all areas of the images, with the assurance that the selection criteria employed later (see Section 2.4.2) would remove most spurious detections (due to cosmic rays, background fluctuations and galaxy subtraction artifacts).

2.2.2 Photometry and calibration

The point-like objects (see Section 2.4.2 for details on point source determination) brighter than i0 = 22 were used to define an optimum aperture for which to extract the photometry. The photometry was extracted at a radius of 3.5 pixels in g0 and 3.0 pixels in r0 and i0 (the seeing was ∼ 0.6 in r0 and i0 but ∼ 0.8 in the g0 filter). The radius of extraction was chosen as a trade-off between increasing error from background noise and increasing uncertainty in the correction for light outside the extraction aperture (i.e. an aperture correction). The aperture correction was −0.382 0.016 mag in g0, −0.263 0.012 mag in r0 and −0.249 0.011 mag in i0. The sky value was calculated as the mode of all pixels in an annulus between 15 and 20 pixels from the centre of the object. The standard deviation of the pixel values in the sky subtraction annulus was included in the calculation of the error for each individual source. Standard stars were not observed with NGC 4365, instead the photometry was cali- 12 Chapter 2. Observing extragalactic globular clusters

Figure 2.2: Calibration of S-Cam photometry to SDSS filters. The difference between zeropoint corrected S-Cam and SDSS magnitudes for g0, r0 and i0 filters are plotted from top to bottom. (Left) The residuals are plotted against the (g0 − i0) colour and the colour range is restricted to that expected for GCs. All three filters show asymmetry in their distributions, scattering to brighter S-Cam magnitudes. (Right) The residuals are shown against SDSS magnitude. brated using bright point-like objects (18 < i0 < 22) detected in the NGC 4365 field that were also found in the Sloan Digital Sky Survey (SDSS) catalogue. Both PSF and model derived SDSS magnitudes were used for this and no systematic difference was found in the results. There were 1822, 2174 and 2813 cross matched objects for calibration in g0, r0 and i0 filters respectively and the photometric calibration zeropoints were determined from a best fit linear relation between the SDSS catalogue magnitudes and the S-Cam instrumental magnitudes. These were found to be zpg0 = 27.64 0.03, zpr0 = 27.76 0.08 and zpi0 = 27.72 0.10 on the AB photometric system. The errors in the zeropoints were determined by adding in quadrature the errors from the SDSS photometry and the errors from the S-Cam photometry and taking the mean. Across all magnitudes the root mean square of the differences between SDSS and S-Cam photometry are 0.36, 0.37 and 0.38 for the g0, r0 and i0 filters respectively. The zeropoint corrected residuals of point-like objects in each filter are shown in Figure 2.2, plotted against g0 − i0 colour (left) and filter magnitude (right). The SDSS and S-Cam g0, r0 and i0 filters show good agreement with no obvious systematic colour or magnitude trend.

It was necessary to correct for foreground dust extinction as such a correction was not applied to the SDSS calibration objects. We used the Schlegel et al. (1998) dust maps to calculate the extinction correction in each filter. The values were compared at different positions across the field and not found to vary significantly. For g0, r0 and i0 filters the extinction correction values used were Ag0 = 0.081, Ar0 = 0.060 and Ai0 = 0.045 mag. 2.3. HST/ACS Data 13

Central Exposures HST R.A. Dec. Filter Time (s) ID F475W 750 1 12:24:27.0 07:19:20.8 F850LP 1210 9401 F475W 680 2 12:24:17.5 07:21:09.5 F850LP 1170 10582 F475W 680 3 12:24:16.0 07:17:37.1 F850LP 1170 10582 F475W 680 4 12:24:26.3 07:15:44.8 F850LP 1170 10582 F475W 680 5 12:24:36.4 07:17:25.1 F850LP 1170 10582 F475W 680 6 12:24:41.3 07:20:18.2 F850LP 1170 10582 F475W 680 7 12:24:31.5 07:22:20.4 F850LP 1170 10582 F475W 1744 8 12:24:07.2 07:12:11.3 F775W 1624 9488

Table 2.1: Imaging obtained from the Hubble Legacy Archive. The pointings are numbered to correspond with the numbers in Figure 2.1 and HST ID refers to the HST proposal identification number where the principle investigators are as follows: 9401 - Patrick Côté, 10584 - Gregory Sivakoff, 9488 - Kavan Ratnatunga.

Hereafter we quote extinction corrected magnitudes and colours.

2.3 HST/ACS Data

2.3.1 Observations and object measurement

We obtained, from the Hubble Legacy Archive, g and z (g and i for one pointing) filter imaging for eight separate pointings of the HST/ACS instrument around NGC 4365. Table 2.1 summarises these observations and Figure 2.1 shows the ACS pointing footprint on the S-Cam image. These archival data probe down to z = 25.2, at a 50 per cent completeness level (Jordánet al., 2007). The resolution of HST/ACS data (pixel scale is 0.05 arcsec) partially resolves GCs at the distance of NGC 4365, 23.1 Mpc away (Blakeslee et al., 2009). The eight individual ACS fields were analysed using a custom built pipeline (see e.g. Strader et al. 2006 and Spitler et al. 2006b) to find small, round objects and measure their magnitudes and half light radii (henceforward referred to as object size). The pipeline was run by co-supervisor Dr Lee Spitler. For details on the methods used by the pipeline 14 Chapter 2. Observing extragalactic globular clusters

Figure 2.3: Comparison of ACS sizes and magnitudes published in Jordánet al. (2009) with those measured in this work. The difference between the g (blue) and z (red) magnitudes is plotted in the left panel and the differences between the sizes (in arcseconds) is plotted in the right panel. including point spread function determination see Strader et al. (2006).The object lists from each field were collated, with the arithmetic mean taken of the sizes and magnitudes in the field overlap areas, and associated errors adjusted to reflect the more accurate measurement. The median standard deviation between the measured magnitudes in overlap areas is 0.04 mag and the median standard deviation in the measured sizes is 0.25 pixels (1.4 parsec). In Figure 2.3 the measured magnitudes and sizes are compared with those published for objects in the central field (Jordánet al., 2009). The magnitude measurements show no evidence of a statistically-significant systematic offset from the published data. The scatter increases to fainter objects but is on the order of 0.1 magnitudes. The size measurements in this work show a small ∼ 0.005 arcsec systematic offset to smaller sizes attributed to differences in size measurement techniques.

2.4 GC candidate Selection

2.4.1 HST/ACS GCs

GC candidates were selected from the HST/ACS images based on their size, colour and magnitude. The GC candidate distribution is selected to have colours 0.7 < g − z < 1.6 and sizes 0.1 pc < rh < 20 pc. The choice of g − z colour upper and lower bounds was based on the colour size diagram shown in Figure 2.4. There is a clear drop oﬀ in the density of objects bluewards of 0.75 and a similar but less clear drop oﬀ redwards of 1.6 2.4. GC candidate Selection 15

Figure 2.4: Size in parsec vs. g − z colour distribution for GC candidates brighter than the turnover magnitude of z = 23.4, detected in the HST/ACS ﬁelds. The GC candidate selection criteria is marked with a box. The blue GC candidates tend to have larger average sizes and the red GC candidates have a larger dispersion in colour.

Figure 2.5: Colour magnitude diagram for GC candidates selected from HST/ACS imaging. 16 Chapter 2. Observing extragalactic globular clusters

(−2.4 < [Fe/H] < 0.18 using the empirical transformation of Peng et al. (2006)). While most of the GC candidates in Figure 2.4 are resolved, the redder GC candidates show a size distribution that overlaps with objects of zero size (objects indistinguishable from stellar point sources). A size cut at 0.1 pc (10−4 arcsec) was made because at smaller sizes the objects did not show a distribution that was centrally concentrated on the galaxy and therefore does not sample mainly GCs. A generous upper bound on size of 20 pc (0.2 arcsec) was employed to exclude background galaxies but include possible Ultra Compact

Dwarfs (UCDs) in the system. Lastly, only objects fainter than z = 19 (Mz = −12.8 mag) are considered GC candidates of NGC 4365, following the convention of the ACS VCS published catalogue (Jord´anet al., 2009). The catalogue contains no objects fainter than z = 26.2 mag, see Figure 2.5. Objects brighter than z = 19 are most likely to be stars.

2.4.2 Subaru/SuprimeCam GCs

Point source determination

At the distance of NGC 4365, GCs are not resolved by our Subaru observations and consequently they cannot be separated from stars in our Galaxy via a size distinction. By selecting only objects that are unresolved in the S-Cam imaging background galaxies are excluded from the analysis of NGC 4365’s GC system. Here the distinction between a point-like object (mostly stars and NGC 4365 GCs) and an extended object (galaxies) is determined by a measure of the flux difference between two aperture radii. Shown in Figure 2.6 are the objects that were identified to be point-like in all three filters. Magnitudes were extracted for two apertures different in radius by a half or full pixel (filter dependent) both centred on the object. Extended objects have extra light in the larger aperture compared to point-like objects. The lists of point-like objects for each filter were cross-matched in position and only objects classed as point-like in all three filters were kept in the sample. This was done using the IRAF task TMATCH with a tolerance in positional offset between images of 1.26 arcsec (this was determined by the uncertainty in astrometry at the image edges). Because the photometry is deeper in the i0 filter than either r0 or g0 filters this procedure likely excluded genuine faint GCs from the analysis that were detected in i0 but not in either of the other filters. Using a cross-matching technique spurious detections (due to cosmic rays etc.) in any of the three filters were eliminated and only objects with photometry in all three filters selected. 2.4. GC candidate Selection 17

Figure 2.6: The difference in flux between two aperture radii (the smaller of which is the extraction aperture) vs. aperture corrected magnitude at the extraction aperture for all objects found by the DAOFIND detection algorithm. Objects determined to be point-like in all three S-Cam filters are shown in black dots, overlapping the total object detection in the g0 filter (grey dots). The dotted line shows the cutoff between point-like objects and extended objects in the g0 filter. Bright point-like objects have aperture differences of ∼ 0.1 magnitudes but fainter objects show significant scatter in measured magnitudes. A difference in magnitude between the inner and outer aperture of up to ∼ 0.6 mag is required to include fainter point-like objects. 18 Chapter 2. Observing extragalactic globular clusters

Figure 2.7: Colour selection of GCs in the S-Cam field-of-view based on the r0 − i0 vs. g0 −r0 colour of matched objects in the HST/ACS data. The objects shown are all brighter than i0 = 22 mag and have magnitude errors smaller than 0.02 for visual clarity. S-Cam detected point-like objects are plotted in black plusses and HST/ACS matched objects are overplotted as red dots. The HST/ACS matched objects define a very tight sequence (black line) that is used to define the area from which S-Cam objects are selected as GC candidates (red box).

Colour-colour and magnitude selection

As shown in Figure 2.4 the HST/ACS is able to partially resolve GCs at the distance of NGC 4365 and can therefore directly distinguish between unresolved foreground stars, GCs and resolved background galaxies. We used the central HST/ACS pointing to determine where on a S-Cam r0 − i0 vs. g0 − r0 diagram the GC sequence lies by matching S-Cam point source detections with ACS GC candidates. A line was fitted to the brightest S-Cam objects found to be GC candidates in the ACS field and used to make a box, 2σ in each direction from the best fit GC sequence line. Objects were considered to be consistent with the GC colour-colour definition if the standard error in their photometry placed them within the box (see Figure 2.7). The selected distribution contains objects with colours 0.6 < g0 − i0 < 1.4, corresponding to metallicities of −2.1 < [Fe/H] < 0.7 when the Lee, Park & Hwang (2010) empirical transformation is used. The blue (g0 − i0 ∼ 0.8) GCs lie nearer the bottom left and red (g0 − i0 ∼ 1.2) GCs lie nearer the top right of the box in Figure 2.7. Although blue GCs are generally more concentrated than red GCs this does not affect the range in g0 − i0 of the selection box that should be used when the central 2.4. GC candidate Selection 19

Figure 2.8: Colour magnitude diagram for GC candidates selected from S-Cam imaging.

GCs define the box, but this does affect the distribution of GCs in the box at different radii. For example, the HST/ACS selected GCs in Figure 2.7 extend to the bluest colours but are very sparsely distributed there, whereas the S-Cam selected GC candidates much more densely populate the blue end. This is because the S-Cam imaging extends to much larger radii. 0 An upper limit on the magnitude of any GC candidate was set to i = 19 (Mi0 = −12.8 mag) by the brightest GC found by HST/ACS in the central field of NGC 4365. A limit of 0.1 mag in photometric error resulted in a lower limit of i0 ≈ 25 mag on the blue end of the colour-colour selection and i0 ≈ 24.5 mag on the red end of the colour-colour selection. The resulting colour-magnitude diagram (CMD) is seen in Figure 2.8.

2.4.3 Contamination and completeness

At galactocentric radii smaller than 0.9 arcmin S-Cam detection numbers drop due to galaxy subtraction artifacts and at radii greater than 3.4 arcmin ACS detections are no longer spatially complete due to the tiling pattern of HST/ACS pointings. Therefore only in an annulus between 0.9 and 3.4 arcmin from the centre of NGC 4365 is it possible to compare the object detections and photometry for the S-Cam and ACS imaging. This was done to determine the level of contamination in the S-Cam GC sample (contamination is deﬁned as non GC objects that meet the colour, size and magnitude conditions set for the S-Cam photometry) as well as to determine the photometric completeness of our S-Cam 20 Chapter 2. Observing extragalactic globular clusters

Figure 2.9: Estimate of contamination in the S-Cam data by comparison with HST/ACS matched objects. The solid line shows the number of S-Cam GCs in each magnitude bin, the dashed line shows the number of those objects that are determined to be contaminants by HST/ACS imaging and the dotted line shows the percentage contamination of the S- Cam GC sample as a function of magnitude. imaging. This analysis is relative to the HST/ACS sample but the ACS imaging is almost perfectly complete several magnitudes deeper than our Subaru/S-Cam imaging (Jord´an et al., 2009).

The contamination percentage of the sample in S-Cam was determined by keeping record of the GC candidates detected in the ACS imaging but removed by the conditions outlined in Section 2.4.1. At a magnitude of z = 24.2, 50 per cent of S-Cam GC candidates are determined to be contamination and at z = 24.5 mag essentially all of the S-Cam GC candidates are contaminants, see Figure 2.9. Across the whole GC candidate sample the contamination determined in this way is 4.14 0.35 objects arcmin−2 and if the sample is restricted to only objects brighter than z = 23.4 mag this value drops to 1.23 0.19 objects arcmin−2. The luminosity function of GCs can be well described by a Gaussian distribution peaked at the turnover magnitude of Mz ≈ −8.4 mag (Jord´anet al., 2007), which corresponds to z = 23.4 mag at the distance of NGC 4365 and all GC candidates described in further analysis are brighter than this turnover magnitude.

The completeness of the S-Cam imaging was determined by keeping record of how many of the HST/ACS detected objects were also detected by the Subaru/S-Cam imaging at diﬀerent magnitudes. The S-Cam imaging is found to be 50 per cent complete at 2.4. GC candidate Selection 21

Figure 2.10: Estimate of the completeness of the S-Cam photometric sample by comparison with HST/ACS photometry. The solid line shows the number of HST/ACS GCs in each magnitude bin, the dashed line shows the number of those objects that were also detected in the S-Cam imaging and the dotted line shows the percentage completeness of the S-Cam photometry as a function of magnitude. i0 = 23.8 mag and 70 per cent complete at i = 23.6 mag as shown in Figure 2.10. In the very central regions (radii < 0.325 arcmin) of NGC 4365 the HST/ACS detection of GC candidates is inhibited by the high surface brightness of the galaxy. In Jord´an et al. (2007) the completeness of objects is tabulated as a function of object size, object magnitude and surface brightness of the galaxy at the location of the GC candidate. For this analysis it is only necessary to consider completeness as a function of radius. We measured the average surface brightness of the galaxy in 0.1 arcmin annuli on the ACS imaging. We compared that to the completeness fraction for an object of mean size = 0.039 arcsec (4.3 parsec) and z = 23.4 mag, which is representative of the GC candidate sample. At distances of 0.2 and 0.325 arcmin from the galaxy centre the ACS imaging is 93.3 and 98.9 per cent complete.

2.4.4 Overview of GC selection

The selection of GC candidates from the Subaru/S-Cam photometry was done based on object size, magnitude and locus in colour-colour space compared to GC candidates obtained from HST/ACS photometry. Point-like objects were selected in each S-Cam filter using a limit on the flux difference between two apertures (see Section 2.4.2). The 22 Chapter 2. Observing extragalactic globular clusters astrometry of the ACS GC catalogue was shifted to that of the S-Cam catalogue. The positions of objects classed as point-like in all S-Cam filters were compared with the positions of GC candidates obtained from the ACS photometry (see Section 2.4.1 for ACS selection criteria; colours 0.7 < g − z < 1.6, sizes 0.1 pc < rh < 20 pc and magnitudes 26.2 ≤ z ≤ 19.0). The matched S-Cam point-like objects were used to determine the locus of GCs on a r0 −i0 vs. g0 −r0 diagram. No contaminating objects could be visually identified on inspection of the GC candidates brighter than i = 21 mag. Finally, estimates of the contamination and completeness of the S-Cam catalogue were determined by comparison with the ACS VCS catalogue (see Section 2.4.3). The ACS catalogue determined here contains 2254 GC candidates with colours ranging from 0.7 to 1.6 in g − z. The bright cutoff for GC candidates in the ACS catalogue is z = 19 mag and the faintest GC candidate has z = 26.2 mag. The S-Cam catalogue contains 5623 GC candidates (1327 of which are also found in the ACS catalogue) of which 2615 are brighter than i0 = 23.6. The g0 − i0 colours of GC candidates in the S- Cam catalogue range from 0.5 to 1.45. The bright cutoff for GC candidates in the S-Cam catalogue is i0 = 19 mag and the faintest, reliable GC candidated has i0 = 24.5 mag. 3 Wide-field imaging of NGC 4365’s globular cluster system: The third subpopulation revisited

The most beautiful gift of nature is that it gives one pleasure to look around and try to comprehend what we see. —Prof. Albert Einstein

3.1 Introduction

Here we study NGC 4365, a giant elliptical (E3) galaxy (Ferrarese et al., 2006) on the far edge of the Virgo Cluster, 23.1 Mpc away (Blakeslee et al., 2009) with MB = −21.3 mag (Ferrarese et al., 2006). There is debate as to whether the apparent Kinematically Distinct Core (KDC) of the galaxy is an actual kinematic property or the observed effect of triaxial orbits (e.g. Davies et al., 2001; van den Bosch et al., 2008). NGC 4365 is also one of few galaxies that rotate about the photometric minor rather than major axis (Davies et al., 2001). Tal et al. (2009) claim evidence of a faint fan feature in the SW of the galaxy. Forbes (1996) and Forbes et al. (1996) studied the GC system properties of NGC 4365 and other giant elliptical galaxies with KDCs. Bimodality was not suspected at the time in most GC systems and they found no significant differences between NGC 4365 and other giant elliptical galaxies with KDCs. Two independent groups working with HST/WFPC2 imaging, Larsen et al. (2001) and Kundu & Whitmore (2001), found that the GC system of NGC 4365 was best fit by a unimodal distribution, whereas most of the elliptical galaxies in their samples were best fit by bimodal distributions. Soon afterwards, Puzia et al. (2002) analysed the GC system of NGC 4365, using the combination of HST/WFPC2 optical and VLT/ISAAC K-band photometry to break the age-metallicity degeneracy,

23 24 Chapter 3. Wide-ﬁeld imaging of NGC 4365’s globular cluster system which is a problem inherent to photometric analysis of GCs. In their analysis they found what appeared to be a slightly younger and very metal rich subpopulation in addition to two old subpopulations. When Larsen et al. (2003) published Lick indices (obtained from Keck/LRIS) of 14 GCs in NGC 4365 it still looked likely that the galaxy hosted some old GCs and some very metal rich, young GCs that “conspire to produce the single broad distribution observed in optical colors” (Larsen et al., 2003).

However, when Brodie et al. (2005) extended the sample of NGC 4365 GCs with Lick indices by adding 19 new objects (also re-observing 3 objects from Larsen et al., 2003) they found a uniform old age for all the GCs in the sample and suggested that the third GC subpopulation in the system, also found at intermediate optical colours, was due to a population with intermediate metallicity not young age. Hempel & Kissler-Patig (2004) published “better age determinations” for the intermediate GC subpopulation by including observations in another infrared filter and Kundu et al. (2005) approached the problem with deeper HST observations in the H-band. Larsen et al. (2005) confirmed the presence of intermediate optical colour GCs at small galactocentric radii (from an analysis of optical HST/ACS photometry) and compared photometric age determinations with spectroscopic ones to assess the accuracy of photometrically determined ages. This work cast some doubt on the accuracy with which ages and metallicities can be measured using a combination of infrared and optical photometry. Later, Hempel et al. (2007) published a photometric comparison of elliptical galaxies in the centres of clusters and those in smaller groups, claiming (from optical and near-infrared photometry) that many group elliptical galaxies (including NGC 4365) host intermediate age GCs. Using g, z and K filter photometry, Chies-Santos et al. (2011) showed that NGC 4365’s GCs all have similar ages (much like the other large ellipticals in their sample) but that the distribution in the g-z direction was significantly different to other large ellipticals.

While there is still debate on the nature (in age, colour and metallicity) of a third GC subpopulation in NGC 4365, there is consensus that the GC system of NGC 4365 is diﬀerent to other galaxies of similar luminosity. Most analyses of NGC 4365’s GC system subpopulations to date have been done with tens or hundreds of GCs. We revisit this issue in colour, using 6550 GC candidates from optical photometry over most of the spatial extent of the galaxy, to analyse whether there is statistically signiﬁcant indication of three colour subpopulations in its GC system.

Section 3.2 describes the properties of the galaxy light and we proceed to the analysis of the GC system properties in Section 3.3, ﬁrst in terms of the properties of the total GC system and then in terms of the subpopulations and their characteristics. We discuss 3.2. Analysis of the galaxy light 25

Filter µe n Re (mag arcsec−2) (arcmin) g 21.87 0.11 6.02 0.21 2.40 0.14 r 21.46 0.12 5.92 0.23 2.06 0.12 i 21.20 0.11 5.97 0.24 2.10 0.11

Table 3.1: Values for g, r and i Sérsicprofiles fitted to galaxy light. results in Section 3.4 before concluding. Appendices A and B contain the statistical analysis of NGC 4365’s GC subpopulations and GC surface density data respectively.

3.2 Analysis of the galaxy light

3.2.1 Parameters of the isophotal ﬁts

We used the IRAF task ELLIPSE to model and subtract the galaxy light from the Subaru/S-Cam photometry. From the model of the galaxy light we obtain the surface brightness, ellipticity, position angle and higher order Fourier terms (S3, S4, C3, C4) in g0, r0 and i0 filters for NGC 4365. Results are plotted in Figure 3.1 and briefly discussed here. We have surface brightness information for all three filters between ∼ 0.1 and 5 arcmin. We fit Sérsicprofiles (Graham & Driver, 2005), i.e.

" 1 # 2.5bn R n µ(R) = µe + − 1 (3.1) ln10 Re to the three surface brightness profiles, where Re is the effective radius of the galaxy, µe is the surface brightness at that radius, n is the shape parameter of the Sérsicprofile and bn = 1.9992n − 0.3271. The fitted values are tabulated in Table 3.1. The quoted value for NGC 4365’s effective radius (Goudfrooij et al., 1994; Bender et al., 1992) is 1.1 arcmin based on a de Vaucouleurs profile (equivalent to a Sérsicprofile with the n parameter set to 4) fitted by Burstein et al. (1987). When a de Vaucouleurs profile is fit to the data in this work the effective radius is found to be 1.32 0.03 arcmin but we find this to be a significantly poorer fit than a general Sérsicprofile. For all further analysis we use our Re value of 2.1 0.1 arcmin (14.1 0.7 kpc), with an n value of 6.0 0.2. This is close to the value of 1.6 0.1 arcmin (10.9 0.7 kpc) with n = 5.8 in z that Chen et al. (2010) found and slightly smaller than the value of 3.07 0.22 arcmin (20.6 1.5 kpc) with n = 7.1 0.4 in B that Kormendy et al. (2009) found. Results for the ellipticity, position angle and higher order Fourier terms in this work 26 Chapter 3. Wide-field imaging of NGC 4365’s globular cluster system

Figure 3.1: Isophotal parameters for g0, r0 and i0 filters from Subaru/S-cam photometry of NGC 4365. From left to right and top to bottom the parameters plotted are: surface brightness of the galaxy light in g0 (crosses), r0 (stars) and i0 (plusses) filters; galaxy colour; ellipticity of the fitted isophotes in all three filters as labelled before; position angle of the fitted isophotes in all three filters as labelled before; higher order Fourier coefficients to the sine (S3 and S4) and cosine (C3 and C4) terms. All parameters are plotted against equivalent galactocentric radius, calculated as the geometric mean of the semi-major and semi-minor axis lengths of the fitted ellipse. The fitted values become unstable beyond 3 arcmin. 3.3. Analysis of the GC System 27 agree with the values Goudfrooij et al. (1994) found in the B,V and I filters and extend

2 arcmin further, to ∼ 1.5Re. The ellipticity measured in this work varies between 0.22 at 0.15 arcmin and 0.28 at 2.5 arcmin and is consistent with that seen by Goudfrooij et al. (1994) interior to 1 arcmin. The position angle shows a small twist of ≤ 5◦beyond the radial range explored by Goudfrooij et al. (1994). The S3 parameter (coefficient to the sine term in the fit) shows an indication of increase at the edge of their radial range (1 arcmin) which is confirmed in this work (between 1 and 2 arcmin) and correspondingly the S4 parameter increases between 1 and 2 arcmin. There is no significant difference in C3 and C4 parameters between Goudfrooij et al. (1994) and this work. The C4 disky/boxy parameter remains at values between -0.01 and -0.02 out to 2 arcmin, indicating that NGC 4365 has boxy isophotal structure to large radii. The galaxy colour becomes bluer with increasing galactocentric radius, changing by ∼ 0.1 mag from ∼ 0.1 to 2.5 arcmin for both g − i and g − r colour indices. A similar trend, implying a negative metallicity gradient, is seen in the Goudfrooij et al. (1994) B–I colour index.

3.3 Analysis of the GC System

3.3.1 Characterising the total GC System

Surface Density

By combining the wide field imaging from S-Cam with the spatially resolved imaging from ACS and further restricting the GC candidate sample to objects brighter than the turnover magnitude we can derive a very accurate radial surface density profile over a factor of 100 in radius. To derive surface density profiles, the number of GC candidates in each radial bin were divided by the total area in that radial bin. The results for both the S-Cam and ACS GC candidate samples are plotted on Figure 3.2. We expect very little contamination in the density profile, with more than 90 per cent completeness in this magnitude range (see Section 2.4.3). A Sérsicprofile plus a background parameter, i.e.

" 1 #! R n P (R) = Pe exp −bn − 1 + bg (3.2) Re where

bn = 1.9992n − 0.3271 (3.3) is ﬁtted to the combined S-Cam and ACS radial surface density of GC candidates and 28 Chapter 3. Wide-ﬁeld imaging of NGC 4365’s globular cluster system

Figure 3.2: Surface density of GC candidates brighter than the turnover magnitude plotted against galactocentric radius. The S-Cam points (solid points) were calculated using objects brighter than i0 = 23.6 and the ACS points (hollow diamonds) were calculated using objects brighter than z = 23.4. The innermost two ACS values were corrected for incompleteness using the values discussed in the text. No other normalisation correction has been made to either ACS or S-Cam radial surface density and the remarkable agreement in the region of overlap (between 1 and 3 arcmin) is an indication that both samples are almost perfectly complete to the turnover magnitude and contain very little contamination. The line plotted is a fitted Sérsicprofile with a constant background value. See text for further details. 3.3. Analysis of the GC System 29

Pe n Re bg (arcmin−2) (arcmin) (arcmin−2) 3.9 1.3 2.68 0.41 6.1 1.2 0.89 0.13

Table 3.2: Fitted values of the S´ersicﬁt to the GC surface density.

shown in Figure 3.2. Re is the effective radius of the GC system, Pe is the density at the effective radius and n is the shape parameter of the Sérsicprofile. The fitted parameters are recorded in Table 3.2. The value for contamination expected from the analysis in Section 2.4.3 (1.230.19 arcmin−2) agrees within 2σ with the value from the Sérsic profile fit (0.89 0.13 arcmin−2).

We are not able to define the edge of NGC 4365’s GC system with certainty as the profile shown in Figure 3.2 is still decreasing at ∼ 20 arcmin (134 kpc), which is the very edge of the spatial coverage available to us with S-Cam imaging. In order to confirm the edge of the GC system even larger field of view imaging would be required. We can make an estimate of the total number of GCs associated with NGC 4365 by simply doubling the number of GC candidates brighter than the turnover magnitude (i0 = 23.6 and z = 23.4). This assumes that there is no inherent asymmetry in the GC luminosity function, the GC system of NGC 4365 contains few GCs beyond 20 arcmin and that the sample is complete and free of contamination. We determine the total number of GCs in NGC 4365’s GC system to be 6450110. This is likely to be a lower limit and doubles the expected number of GCs in NGC 4365 from the value of 3246 598 calculated by Peng et al. (2008). They use the narrow field of view of one HST/ACS and three HST/WFPC2 pointings to derive an extrapolated GC surface density profile that is not publicly available. We suggest that the surface density profile that Peng et al. (2008) fit to the GC candidates underestimated the radial extent of NGC 4365’s GC system.

These total GC numbers convert to speciﬁc frequency (Harris & van den Bergh, 1981) values of SN = 3.86 0.71 (Peng et al., 2008) and SN = 7.75 0.13 determined from this work for MV = −22.31. Several other large elliptical galaxies with similar MV have speciﬁc frequency values comparable to the SN measured here. NGC 1407 has SN = 7.98 0.87,

NGC 1399 has SN = 6.72 0.81 (Spitler et al., 2008) and Peng et al. (2008) find SN = 5.2 1.4 for M84. Fitting a power law to the radial density beyond the Sérsiceffective radius we find a slope (i.e. power-law exponent) of −1.21 0.03 to allow comparisons with analysis of other galaxies in the literature.

By binning the GC candidate positions into R.A. and Dec. deﬁned squares we derive a two-dimensional density distribution for the GC candidates. This is shown in Figure 3.3 30 Chapter 3. Wide-ﬁeld imaging of NGC 4365’s globular cluster system

Figure 3.3: Surface density of GC candidates in two dimensions from S-Cam data. The colour ranges from red to blue indicating decreasing GC surface density. Ellipses showing the ellipticity and position angle of the GC system are plotted with dotted black lines and galaxy light isophotes are plotted for comparison in solid black lines. The GC system is much more elongated than the starlight. 3.3. Analysis of the GC System 31

Figure 3.4: Histogram of the azimuthal distribution of GC candidates. Poisson errors are shown on the histogram and the dotted line with best fit ellipticity and position angle is overplotted. See text for fitting and parameter details. along with selected galaxy surface brightness isophotes from the i0 filter as determined by the IRAF task ELLIPSE. McLaughlin et al. (1994) use

σ(R, θ) = kR−α[cos2(θ − PA) + (1 − e)−2 sin2(θ − PA)]−α/2 + bg (3.4) to fit for the position angle (PA) and ellipticity (e) of the GC system of M87. They fit to the density of GCs as a function of azimuthal distribution and we follow the same procedure. We use the previously determined power law exponent (α) and the fitted background value (bg), allowing the normalisation constant (k) as well as the position angle and ellipticity to vary. We find the position angle of the GC system to be 39.72.2◦ and the ellipticity to be 0.40 0.02, see Figure 3.4. The GC system of NGC 4365 has a similar position angle to the galaxy light (∼ 42◦) but is clearly much more elliptical than the starlight of NGC 4365 (which has an ellipticity of ∼ 0.25). We see qualitative agreement with this analysis in Figure 3.3.

Colour Distribution

Before performing a quantitative analysis of the surface density characteristics of GC subpopulations in NGC 4365 we can make a qualitative assessment of colour-radius substructure by examining the surface density of GCs in colour-radius space. This is shown in 32 Chapter 3. Wide-ﬁeld imaging of NGC 4365’s globular cluster system

Figure 3.5: GC density with colour and galactocentric radius for the ACS data. The colour scale indicates object number per square arcmin per colour bin ranging from black at low density to red at high density. The plot is constrained to objects with radii of 0.02 to 5.8 arcmin. (Top) Measured GC density with colour and radius is plotted. (Centre) The model of red and blue GC density calculated using the Gaussian positions and widths from the KMM results for the two subpopulations at large radii (r > 3.5 arcmin). The proportion of GCs in each Gaussian distribution is taken from the KMM results for the whole sample and the Sérsicradial density profile used is calculated in Section 3.3.1. This was scaled to have a similar object density to the data at large radii. (Bottom) The model in the central panel has been subtracted from the density distribution in the top panel to show an approximate distribution of GCs that are not well modelled by a bimodal colour distribution of GCs. This distribution is not a robust measure but it does show indications of significant numbers of intermediate colour GCs (g − z ∼ 1.1), particularly at small radii (. 2 arcmin). 3.3. Analysis of the GC System 33

Figure 3.6: GC density with colour and galactocentric radius for the S-Cam data. The colour scale indicates object number per square arcmin per colour bin ranging from black at low density to red at high density. The plot is constrained to objects with radii of 0.3 to 12 arcmin. There is indication of intermediate colour (g0 − i0 ∼ 1.05) GCs at small radii, similar to the ACS data, but it is not clear whether this intermediate colour subpopulation is evident at larger radii.

Figures 3.5 and 3.6. At galactocentric radii larger than 2 arcmin in the top panel of Figure 3.5 the blue and red GC subpopulations of NGC 4365 are visible as distinct sequences at g−z ∼ 0.9 and ∼ 1.3 respectively. At radii smaller than 2 arcmin and possibly between 2.5 and 3.5 arcmin there is evidence of a third subpopulation of GCs intermediate in colour. Note the clear overdensity of GCs candidates at intermediate to red colours and within 1 arcmin in both Figures 3.5 and 3.6. These data conﬁrm a similar observation by Larsen et al. (2005) where evidence of intermediate colour GCs were found at radii smaller than 1.25 arcmin. We are able to extend the analysis of the GC colour substructure with radius to 5.5 arcmin with HST/ACS imaging (see Figure 3.5) and to 11.5 arcmin with S-Cam imaging (see Figure 3.6). Beyond 3 arcmin in Figure 3.6 the colour distribution seems to be dominated by 2 colour modes but there is some indication of GCs at intermediate colours.

Recently, observers have found gradients in the peak colour of the GC subpopulations with galactocentric radius (see Harris, 2009a,b; Liu et al., 2011; Forbes et al., 2011). Both blue and red subpopulations’ peak colours shift to bluer colours at larger radii. Forbes et al. (2011) ﬁnd the most extreme gradient to date in the galaxy NGC 1407. If the red subpopulation of NGC 4365 were similar to that of NGC 1407 then the peak colour for the red subpopulation would change from g0 − i0 ∼ 1.19 at the galaxy centre to g0 − i0 ∼ 1.06 at 10 arcmin, and the red subpopulation would have intermediate colours at large radii. In the bottom panel of Figure 3.5 and in Figure 3.6 an overdensity of GC candidates with intermediate colours is visible at very small radii that cannot be accounted for by a 34 Chapter 3. Wide-ﬁeld imaging of NGC 4365’s globular cluster system

Figure 3.7: Rolling peak colour values found by the KMM code plotted against galactocentric radius. The bimodal case is plotted in the top panel and the trimodal case in the bottom panel. S-Cam (g0 − i0) rolling peak values are plotted for blue (dashed), green (dash-dotted) and red (dotted) modes in the multimodal ﬁts. Galaxy g0 − i0 colours are plotted as black dots with errorbars and the solid black lines show best ﬁt gradients in colour-log radius space. gradient in the peak colour of the red subpopulation. We note that if this intermediate colour overdensity at small galactocentric radii is a distinct subpopulation and there is a gradient in the red subpopulation of NGC 4365 then statistical tests for bimodality or trimodality based on colour distribution alone might not be able to distinguish between these two subpopulations conclusively.

3.3.2 Radial colour gradients

Several groups have recently found shallow but signiﬁcant gradients in the mean colour of GC subpopulations (see Harris, 2009a,b; Liu et al., 2011; Forbes et al., 2011). The radial extent of the Subaru/S-Cam imaging makes this an excellent data set to determine whether there are gradients present in the GC subpopulations of NGC 4365. The presence of an intermediate colour subpopulation will inﬂuence the gradient determined and therefore we consider the case of a bimodal and a trimodal distribution separately. The heteroscedastic KMM algorithm was run on a rolling sample of GC candidates and the peak value determined for both bimodal and trimodal instances for the S-Cam data set. The rolling sample consisted of 360 GC candidates input to KMM at each step and between each step the 120 candidates with lowest galactocentric radius were replaced with 3.3. Analysis of the GC System 35

Subpopulation Bimodal Case Trimodal Case (dex per arcmin) (dex per arcmin) Blue −0.0037 0.0002 −0.004 0.001 Green - 0.001 0.012 Red −0.0047 0.0005 −0.009 0.002

Table 3.3: GC colour gradients for the linear ﬁt to colour with radius in arcmin (Equation 3.5) for S-Cam data.

Subpopulation Bimodal Case Trimodal Case (dex per dex) (dex per dex) Blue −0.055 0.004 −0.04 0.01 Green - −0.01 0.10 Red −0.062 0.008 −0.07 0.02

Table 3.4: GC colour gradients for the logarithmic ﬁt to colour with normalised radius (Equation 3.6) for S-Cam data. the 120 candidates closest to, but at least as far as, the furthest candidate in the previous sample. A linear relation was ﬁtted to the S-Cam peak colours, to 14.5 arcmin in the bimodal case and 7 arcmin in the trimodal case, i.e.

g0 − i0 = a + b R(arcmin) (3.5)

Results are tabulated in Table 3.3. The linear blue gradient values are consistent within errors but the linear red gradient is signiﬁcantly steeper in the trimodal case. The green gradient measured in the trimodal case is consistent with zero.

We also ﬁt a line to the logarithm of the radius normalised by Re using Re = 2.1 arcmin, i.e. 0 0 g − i = c + d log(R/Re) (3.6)

Results are tabulated in Table 3.4. Here both blue and red gradients are consistent within errors for the bimodal and trimodal cases. The green subpopulation does not have a measurable gradient in this case either. Liu et al. (2011) measure colour gradients for NGC 4365 GCs using the central ACS pointing in the bimodal case, finding a blue slope of −0.056 0.024 and a red slope of −0.033 0.021 dex per dex. The radial range of their measurements do not overlap with ours and we see that the radial colour gradient of the blue subpopulation remains constant when moving to larger galactocentric radii while the radial colour gradient of the red subpopulation steepens significantly with increasing galactocentric radius. We use the Lee, Park & Hwang (2010) empirical relationship 36 Chapter 3. Wide-field imaging of NGC 4365’s globular cluster system

Subpopulation Bimodal Case Trimodal Case (dex per dex) (dex per dex) Blue −0.19 0.01 −0.13 0.03 Red −0.22 0.03 −0.26 0.06

Table 3.5: Metallicity gradients for the logarithmic ﬁt to colour with normalised radius (Equations 3.6 & 3.7) for S-Cam data. between GC colour and metallicity to give a metallicity gradient, i.e.

∆[Fe/H] = 3.48∆(g0 − i0) (3.7)

Results are shown in Table 3.5. The gradient of the mean metallicity with galactocentric radius is shallower in the trimodal case for the blue subpopulation but for the red subpopulation the gradients in the bimodal and trimodal cases are consistent. We note that the rolling peak colour found by KMM is only consistent with the galaxy colour in the trimodal case (see Figure 3.7). We conclude that the NGC 4365 GC subpopulations show a shallow but signiﬁcant colour gradient with galactocentric radius, regardless of whether the GC system is in- terpreted as being bimodal or trimodal. Forbes et al. (2011) tabulate metallicity-radius gradients in units of dex per dex for several galaxies on which this measurement has been done. The gradients measured for NGC 4365 are steeper than the mean, but not the steepest measured. Results vary from −0.10 to −0.38 for the blue subpopulation and from −0.10 to −0.43 for the red subpopulation in Forbes et al. (2011) Table 1. These gradients explain why statistical analyses of the colour distribution (collapsed in radius) of NGC 4365’s GC system give inconclusive results.

3.3.3 Colour-magnitude trends

In the determination of the colour-magnitude trends we only investigate the blue subpopulation of the GC system, we search for a ‘blue tilt’ (Strader et al., 2006). We do this in order to decide whether the blue tilt could be the cause of the intermediate colour overdensity seen in Figures 2.5, 3.5 and 3.6. The bimodal KMM mean colour values at various magnitudes were determined by dividing the GC sample into magnitude subsets with equal numbers. The mean z magnitude of GC candidates in each subset is plotted against the blue heteroscedastic Gaussian mean value found by KMM in Figure 3.8. We did this analysis for the central HST/ACS pointing as well as for the full set of eight HST/ACS pointings and found that both cases show a signiﬁcant gradient but that the 3.3. Analysis of the GC System 37

Figure 3.8: The colour-magnitude relation for blue GC candidates found in the HST/ACS photometry. The left panel shows GC candidates found in the central ACS pointing and the right panel shows candidates found in all eight ACS pointings. The KMM mean value for the blue GCs is overplotted with blue points. The one sigma range for the linear fit to those values is overplotted with a blue shaded region on both panels. On the left panel the gradient the slope found by Mieske et al. (2006) for galaxies between −21.7 < MB < −21 is overplotted with a dashed line. 38 Chapter 3. Wide-field imaging of NGC 4365’s globular cluster system slope is steeper in the central region (−0.0477 0.0018) than for all pointings combined (−0.0240.013). Due to the smaller number of GCs when restricting to the central pointing, one magnitude bin (z = 22.75) is shifted to redder colours because there are a lack of GCs bluer than g − z = 0.9 there. This does not significanlty change the fit. Mieske et al. (2006) use the central ACS pointing and combine GCs for all galaxies in the range

−21.7 < MB < −21 to determine a slope of −0.037 0.004 using the same KMM method. In Figure 3.8 we see that this lies very close to our ﬁtted line. They also ﬁnd a steepening slope when GCs closer to the galaxy centre are used. In Figure 2.5, indication of a clump of intermediate colour/green objects is seen around g − z ∼ 1.1 and 22 > z > 20 and in the bottom panel of Figure 3.5 a clear overdensity is seen inside 2 arcmin and around g − z ∼ 1.1. The colour magnitude slope for the blue GC candidates is not steep enough to cause either of these overdensities.

3.3.4 Quantifying the GC subpopulations

We compare NGC 4365’s GC colour distribution with that of other Virgo cluster galaxies, using the ACS Virgo Cluster Survey (VCS) photometry (Peng et al., 2006). They ﬁt a homoscedastic bimodal distribution using Kaye’s Mixture Model (KMM) algorithm and obtained mean g − z colour values of 0.98 and 1.36 for the blue and red subpopulations of NGC 4365 respectively, as well as a width of 0.15 for both subpopulations. They also

fit linear relations between mean g − z colour and absolute galaxy magnitude, MB, for both GC subpopulations, assuming all Virgo Cluster galaxies have the same distance. The relations predict mean g − z colour values of 0.98 0.06 and 1.40 0.08 for the blue and red subpopulations when the conventional distance for NGC 4365 (23.1 Mpc) is used. They concluded that aside from having a large number of red GCs, NGC 4365 has an unremarkable colour distribution compared to their sample (Peng et al., 2006). Fitting a bimodal homoscedastic distribution to NGC 4365’s GCs, with KMM, using all eight available ACS pointings we find similar values to the ones found by ACS VCS (i.e. g − z = 0.94 and 1.30 for the blue and red peak colours and a width of 0.12). However, on visual inspection the fit to the GC distribution is skewed to redder colours when compared to the actual colour distribution. We also fit both bimodal and trimodal heteroscedastic distributions to the GC colour distributions, obtaining mean blue and red values of g−z = 0.89 and 1.25 in the bimodal case, and blue, green and red values of 0.88, 1.12 and 1.34 in the trimodal case. In neither case does the mean blue colour lie close to the g − z vs.

MB relation (values are 1.4σ and 1.6σ away from the relation) but only in the trimodal case does the red mean colour lie close to the Peng et al. (2006) g − z v.s. MB relation, 3.3. Analysis of the GC System 39

ACS (g − z) S-Cam (g0 − i0) peak width peak width Blue 0.89 0.07 0.80 0.07 Red 1.32 0.12 1.13 0.10

Table 3.6: Gaussian values for the blue and red GC distributions as determined from an Epanechnikov smoothing kernel of the colour distribution. where the value is 0.7σ away from the relation instead of 1.8σ away. This is a motivation to describe the GC system of NGC 4365 in terms of three rather than two subpopulations (see Appendix A for further details). From this point forward we assume that the GC system of NGC 4365 contains three subpopulations, a blue, green (intermediate colour) and red subpopulation. There are radial gradients in colour and a blue tilt but in neither case are they enough to explain the strong overdensity seen at intermediate colours in both Figures 3.5 and 3.6. In addition, there is better agreement with the colour - galaxy luminosity relations of Peng et al. (2006) to the other Virgo cluster galaxies and a near perfect match between galaxy light and mean red subpopulation colours if we assume that NGC 4365 has three subpopulations. In the trimodal case, the green GCs would likely ‘contaminate’ the blue and red subpopulations to all except the bluest and reddest colours. A simple split based on colour would leave a high percentage of green GCs in both the blue and red subpopulations and visa versa. Therefore, we formulate a probability that a GC belongs to either a blue, green or red subpopulation based on its colour and galactocentric radius. Note that red (metal rich) GCs are generally found to be more centrally concentrated than blue (metal poor) GCs (Brodie & Strader, 2006) thus GC galactocentric radius is an important factor in the probability calculation. We assume that each of the three subpopulations are fairly well described by a Gaussian in colour (without skewness or kurtosis) and that the peak and width of the Gaussian is independent of galactocentric radius (i.e. no radial colour gradient) but make no further assumptions about the radial distribution of any of the three subpopulations. Given evidence of radial colour gradients we note that this assumption is strictly inaccurate. We make this assumption because colour gradients are shallow compared to the width of the GC colour distributions and the calculation of gradient values contain other signiﬁcant assumptions about the radial distribution of the subpopulations that we wish to avoid. The following paragraphs describe how we assign a probability that a GC is blue, green or red (henceforth blue-green-red probability). The peak of the blue Gaussian distribution was determined from the bluest peak of an Epanechnikov kernel smoothing of the colour distribution (Silverman, 1986) and the width 40 Chapter 3. Wide-ﬁeld imaging of NGC 4365’s globular cluster system

Figure 3.9: Colour distribution of GC candidates at all radii for S-Cam (top) and ACS (bottom) brighter than the turnover magnitude (z = 23.4 and i0 = 23.6). The Epanechnikov kernel smoothing of the colour distribution is plotted with a dotted line, the smoothing width used is the average colour error (0.04 for both ACS and S-Cam). At least three peaks can be seen. 3.3. Analysis of the GC System 41

Figure 3.10: The normalised distribution of ACS GCs with colour in the seven radial bins (shown in arcminutes on each plot) and at all radii (lower right panel). Histogram heights are shown as black circles with errorbars representing Poissonian errors. Fitted Gaussian distributions are plotted in blue and red for the respective subpopulations and the remaining GCs (once blue and red distributions are subtracted) are plotted as green squares. The peak and width parameters for the Gaussian distributions fitted in each case are constant: µb = 0.89, σb = 0.07, µr = 1.32 and σr = 0.12, with the normalisation allowed to vary. 42 Chapter 3. Wide-field imaging of NGC 4365’s globular cluster system of the blue Gaussian distribution was determined from the colour at which 68 per cent of the GCs bluewards of the peak were included in the distribution (assuming that green GCs would not be present in significant numbers at such blue colours). See Figure 3.9 for the S-Cam colour distribution and kernel smoothing. The peak position and width of the red distribution were calculated similarly and all values are shown in Table 3.6. As seen in the bottom right panels of Figures 3.10 and 3.11, where GCs at all radii are included for each sample, the Gaussian distributions for blue and red subpopulations determined in this way are a reasonable fit to the extreme blue and red ends of the GC colour distribution. To assess the number of blue, green and red GCs as a function of colour and radius, the GCs in each sample (ACS or S-Cam) were split into radial bins of roughly equal object number and for each radial bin blue and red Gaussian distributions were fitted to the normalised colour histogram. Peak and width values were held constant across radial bins, as determined from the whole sample (see Table 3.6), and the normalisations of the distributions were fitted separately for each radial bin. This was a χ2 minimization fit. The blue and red numbers were calculated by adding the GCs in the extreme colour regions (where no significant green ‘contamination’ is expected) to the percentage of the total objects expected to belong to either the blue or red subpopulations (obtained by integrating the blue and red Gaussian distributions over intermediate colours). The green GC numbers were defined by subtracting both the blue and red subpopulation numbers from the total number in the distribution. These values were used to determine a radial surface density profile for each subpopulation (discussed further in Section 3.3.5), which is used to calculate the probability that a GC at any radius or colour belongs to the blue, green or red subpopulations. The process used to determine the blue-green-red probability of a GC starts by determining the relative number of GCs in each subpopulation at the galactocentric radius of the GC. This is done by comparing the values of the radial surface density profile for each subpopulation. The relative numbers of GCs in each population are used to scale the normalisations of the three Gaussian distributions and comparing the relative values of all three Gaussian distributions at the colour of the GC we calculate the probability of an object being blue, green, or red.

3.3.5 Characterising the GC system subpopulations

Surface density

As described in Section 3.3.4 we counted the blue, green and red GCs in each radial bin assuming that the red and blue GC colour distributions are well ﬁt by Gaussians. In 3.3. Analysis of the GC System 43

Figure 3.11: The normalised distribution of S-Cam GCs with colour in seven radial bins (shown in arcminutes on each subplot) and at all radii (lower right panel). Histogram heights are shown as black circles with errorbars representing Poissonian errors. Fitted Gaussian distributions are plotted in blue and red for the respective subpopulations and the remaining GCs (once blue and red distributions are subtracted) are plotted as green squares. The peak and width parameters for the Gaussian distributions ﬁtted in each case are constant: µb = 0.80, σb = 0.07, µr = 1.13 and σr = 0.10, with the normalisation allowed to vary. 44 Chapter 3. Wide-ﬁeld imaging of NGC 4365’s globular cluster system

Figure 3.12: GC radial surface density for blue (squares), green (triangles) and red (circles) subpopulations incorporating ACS (unfilled points) and S-Cam (filled points) GCs. The surface density for both the blue and red GC subpopulations are well fit by a Sérsicprofile added to a background term (shown as blue and red solid lines). The surface density for the green subpopulation is best fit by a power law plus a background value (shown as a green solid line). The red dashed line is a scaled version of the Sérsicprofile fit to the red subpopulation that is a reasonable fit to all except the inner two green surface density points. See text for further details. 3.3. Analysis of the GC System 45

Pe n Re bg (arcmin−2) (arcmin) (arcmin−2) Blue 1.67 0.29 1.36 0.19 7.30 0.68 0.25 Green 1.71 0.17 2.02 3.17 0.21 0.03 Red 4.91 0.59 2.02 0.25 3.17 0.19 0.50

Table 3.7: Sérsicprofile fits to the radial surface density of the blue, green and red GC subpopulations. The green subpopulation n and Re values have been fixed to the red subpopulation values.

Figure 3.12 the radial surface density data for all three subpopulations are shown with the best fit surface density model in each case (data is tabulated in Appendix B). The red and blue subpopulation distributions are both well fit by a Sérsicprofile added to a background term. The values we found are recorded in Table 3.7. The background terms were estimated from the outermost radial surface density value. The effective radius, or half number radius, of the blue subpopulation (7.30 arcmin) is more than twice as large as the effective radius of the red subpopulation (3.17 arcmin) and the Sérsicn values are consistent within combined errors. The green subpopulation can be fit by the red subpopulation Sérsicprofile if the two innermost density points are omitted (points interior to 1.5 arcmin) and we fit for Pe and the bg (see Table 3.7 for the fitted values). We also fit a two parameter power law plus a background term

α ρ(R) = ρ0R + bg (3.8) to the entire radial range of the green subpopulation, finding α = −1.71 0.13, ρ0 = 12.3 1.3 arcmin−2 and bg = 0.10 0.05 arcmin−2. Comparing the slope of a power law fit to the blue (−1.13 0.06) and red (−1.60 0.04) distributions over a radial range from 0.5 to 11 arcmin (4 to 70 kpc) we can quantitatively verify that the green and red GC subpopulations are more centrally concentrated than the blue GC subpopulation and that the slope of the green and red subpopulations are consistent with each other. In Figure 3.13 the percentage of the total GC candidate number that is attributed to each subpopulation is plotted as a function of galactocentric radius. The data points have been background subtracted and the curves derived from the profile fits in Figure 3.12 are plotted without the addition of background terms. In the case of the green subpopulation the curve shown is a combination of the power law fit in the inner parts (R < 2.25 arcmin) and the Sérsic profile beyond that. We do this to adequately display the behaviour of the intermediate colour surface density both in the inner (power law behaviour) and outer (Sérsicbehaviour) parts, noting that we plot lines to guide the eye 46 Chapter 3. Wide-field imaging of NGC 4365’s globular cluster system

Figure 3.13: The percentage contribution of each subpopulation to the total number of GC candidates plotted against galactocentric radius. Curves were calculated using the fitted surface density profiles in Figure 3.12. Both the data points and plotted profiles have had the respective background values subtracted. Beyond 12 arcmin the total population numbers are low and the measurement of the percentage contribution of any one subpopulation to the total number becomes unstable. We see that the red subpopulation dominates at small galactocentric radii and the blue dominates at large galactocentric radii while the green subpopulation is only significant at very small galactocentric radii. 3.3. Analysis of the GC System 47

Figure 3.14: The half light radius (rh) of HST/ACS GC candidates against galactocentric radius for objects brighter than z = 23.4. The high probability blue, green and red GC subpopulation candidates are plotted with blue squares, green triangles and red circles respectively. There is a trend of increasing GC candidate size with increasing galactocentric radius for all GCs regardless of subpopulation colour. A power law fitted to the whole sample of GC candidates brighter than z = 23.4 has a slope of 0.49 0.04 dex per dex. Overplotted with blue dashed, green dash-dotted and red dotted lines are the power law fits for size normalisation (with slope set by whole GC sample) of each high probability subpopulation sample. and not for analysis. In agreement with previous work (see review by Brodie & Strader, 2006) the red subpopulation dominates the GC system in the inner parts and the blue subpopulation dominates in the outer parts. We also see that the green subpopulation only contributes significantly to the GC system at very small galactocentric radii (r . 2 arcmin). This is in agreement with the qualitative analysis of the colour-galactocentric radius properties of the GC system earlier in this work (see Section 3.3.1) and with Larsen et al. (2005).

GC half light sizes

Here we measure subpopulation sizes (median half light radii) as well as trends in candidate size with galactocentric radius from the ACS photometry. As described in Section 3.3.4 each GC candidate is assigned a probability of belonging to each subpopulation. For example a candidate 1.28 arcmin from the galaxy centre with g−i = 0.98 has probabilities: 48 Chapter 3. Wide-field imaging of NGC 4365’s globular cluster system pblue = 18 per cent, pgreen = 79 per cent and pred = 3 per cent. Defining the blue and red GCs with a colour probability larger than 95 per cent and green GCs with a probability larger than 80 per cent there are 315, 426 and 491 GC candidates in each of the blue, green and red high probability subsamples. There are no green GC candidates with a probability larger than 95 per cent so we compare medium probability green candidates (80-95 per cent) with high probability blue and red candidates. The aforementioned example is not defined as part of either blue, green or red high probability subsamples. We determine the median half light radius of the GCs;

0.002 rh,blue = 0.037 0.002 arcsec 0.001 rh,green = 0.027 0.001 arcsec 0.001 rh,red = 0.025 0.001 arcsec

These sizes were measured on GCs brighter than z = 23.4 (109, 223 and 269 blue, green and red high probability subsample GCs) using ISHAPE (Larsen, 1999) where the method 0.3 0.2 is described in Strader et al. (2006). The median sizes correspond to 4.10.2, 3.00.1 and 0.1 2.80.1 parsec respectively. Blue GCs have a significantly larger median size than either green or red GCs. However, the sizes of the green GCs are only slightly larger than the red GCs. This is possibly due to distinct characteristic sizes for each subpopulation but could also be explained by a continuous trend in GC size with GC colour (e.g. Jordán, 2004). As well as a possible trend of GC size with colour we also see a trend of GC size with galactocentric radius. In Figure 3.14 a trend of increasing GC candidate size with increasing galactocentric radius is visible. Fitting a two parameter power law to the GC candidates brighter than z = 23.4 we find

[0.490.04] rh(pc) = [1.00 0.04]R(kpc)

All three subpopulations show a clear size increase with distance from the galaxy centre with a power law slope of 0.49 0.04 dex per dex. This measurement extends to 4.5Re, one of only four similar measurements extending beyond ∼ 2Re. Harris (2009a) measured a slope of 0.11 on a composite sample of six massive galaxies, G´omez& Woodley (2007) measured a slope of 0.05 0.05 for metal poor GCs and 0.26 0.06 for metal rich GCs in NGC 5128 and Spitler et al. (2006b) measured a slope of 0.19 0.03 in NGC 4594. The slope we measure is steeper than any of the previous values and much closer to the value measured for our own Galaxy (0.36 0.07 for metal rich GCs, see G´omez& Woodley 3.3. Analysis of the GC System 49

Figure 3.15: The normalised mass function for GC candidates brighter than z = 23.4 from the ACS catalogue. All GCs (black) as well as the high probability subpopulation candidates (plotted in blue dashed, green dash-dotted and red dotted lines respectively). Each subpopulation has a signiﬁcantly diﬀerent mass distribution to the others.

2007). We conclude that there is more variability, between galaxies, in the relationship of GC size with galactocentric radius than previously found. Shown in Figure 3.14 are the GCs in all three high probability subpopulation samples and power law ﬁts. We set the slope of the ﬁts to 0.49 and found normalisations of 1.096 0.042, 0.994 0.059 and 0.944 0.044 for the blue, green and red samples respectively. This results agrees with analysis of GC subpopulation median sizes.

Mass function

We compare the mass function of the high probability subpopulation GC candidates (as defined in Section 3.3.5). The mass for each GC candidate was derived using the colour dependent values for the z filter mass-to-light (M/Lz) ratio in Table 5 of Jordánet al. (2007). They used PEGASE 2.0 with a modified Salpeter initial mass function from Kennicutt (1983) and an age of 13 Gyr for all GCs. We used magnitudes brighter than z = 23.4 to calculate the mass functions (plotted in Figure 3.15). We ran the Kolmogorov- 5 7 Smirnov (KS) test on each pair with a common mass range of 2 × 10 M < M < 10 M . The KS test gives a probability of 98.3 per cent that the blue and red subpopulations are drawn from different distributions and gives a probability of more than 99.9 (99.4) per cent that the green subpopulation is drawn from a different distribution than the 50 Chapter 3. Wide-field imaging of NGC 4365’s globular cluster system blue (red) subpopulation. The discrepancy between the three subpopulations is greatest 6 around 1 × 10 M in the cumulative mass distribution the KS test uses for its analysis. Analysis of the mass function of GC subpopulations indicates that the green subpopulation is distinct from both blue and red subpopulations, containing a significantly larger 6 7 percentage of objects between 10 M and 10 M than either blue or red subpopulations, as can be clearly seen in Figure 3.15. This overdensity of green subpopulation objects can also be seen in Figure 2.5 around g − z ∼ 1.1 and 22 > z > 19. It is possible that UCDs or dE nuclei contaminate the bright, mid-colour range (Strader et al., 2006) to produce this feature.

Azimuthal properties

In Figure 3.16 we show the azimuthal distribution of high probability colour subpopulation GC candidates (as defined in Section 3.3.5) for S-Cam and ACS. We restrict the samples to radii where we have complete azimuthal coverage so as not to bias the samples in any direction. The position angle data is folded at 180◦ to improve number statistics. The bottom panel of Figure 3.16 shows that there is no position angle signal within errors for any of the three subpopulations in the ACS sample. The S-Cam azimuthal distribution in the top panel does show some structure; the high probability blue GC candidates are more likely to have position angles between 0◦ and 90◦, which is consistent with the blue GC subpopulation being elongated along the major axis of NGC 4365 (the galaxy has a position angle of ∼ 42◦, see Section 3.2). The red distribution shows an almost flat azimuthal distribution (indicating an almost circular GC system) and the green distribution shows a very shallow sinusoidal distribution with a peak between 0◦ and 90◦. It is clear that the blue GC system is very elongated along the position angle of the galaxy and likely that both green and red subpopulations also have a similar position angle to the galaxy starlight. We obtain estimates of the ellipticity of the blue, green and red subpopulations by fixing the position angle (to 42◦) as well as the power law exponent and background value as described in Section 3.3.1, Equation 3.4, following the same fitting procedure. We find that the estimated ellipticity for the blue and green subpopulations are the same within errors and also consistent with the ellipticity measured for the whole GC sample in Section 3.3.1. The estimated ellipticity of the red subpopulation is much smaller than the blue and green values. It is consistent with zero and also the ellipticity measured for the galaxy light (see Table 3.8). We also split each subpopulation into an inner and outer radial bin with equal numbers in each and refit for ellipticity but did not detect radial variation in ellipticity for any of 3.3. Analysis of the GC System 51

Figure 3.16: Histograms of the azimuthal distributions of high probability blue (dashed), green (dash-dotted) and red (dotted) GC subpopulations. The top panel shows the position angles for the S-Cam GC candidates out to 11 arcmin from the galaxy centre and the bottom panel shows the ACS GC candidates out to 3.38 arcmin from the galaxy centre. The position angle of the galaxy is ∼ 42◦ and both blue and green subpopulations show GC overdensity between 0◦ and 90◦ in the S-Cam catalogue. 52 Chapter 3. Wide-ﬁeld imaging of NGC 4365’s globular cluster system

Galaxy Light Blue Green Red 0.25 0.03 0.44 0.08 0.30 0.08 0.03 0.08

Table 3.8: Estimated ellipticity for blue, green and red GC subpopulations. The position angle of each GC subpopulation is ﬁxed to be the same as that of the galaxy light (42◦).

Figure 3.17: The radial surface brightness profile of the galaxy light from the S-Cam i0 filter compared with the surface density profile of the GC subpopulations. The GC subpopulations are plotted as blue squares, green triangles and red circles. The grey points show the surface brightness measurements and the black line shows the Sérsicprofile fit to the galaxy light (see Section 3.2). All parameters are plotted against equivalent galactocentric radius, calculated as the geometric mean of the semi-major and semi-minor radii. The green and red GC surface density profiles are similar to the galaxy light surface brightness profile. the three subpopulations.

The agreement in ellipticity between the galaxy light and the red GC subpopulation is consistent with literature findings that red GC subpopulations generally follow the properties of galaxy field stars more closely than blue GC subpopulations. The blue GC subpopulation dominates in the outer regions of the galaxy and also dominates the measurement of the ellipticity of the GC system. It can be seen qualitatively in Figure 3.3 and quantitatively from the results in Table 3.8 that the blue subpopulation is significantly more elliptical than the galaxy field stars. 3.4. Discussion 53

3.3.6 Comparison with galaxy surface brightness

The galaxy surface brightness profile from the i0 filter photometry is compared with the GC surface density profiles in Figure 3.17. We compare the galaxy surface brightness profile with 2.5 times the logarithm of the GC surface density profile and arbitrarily scaled for ease of viewing, i.e. 2.5log[P (R) − bg] (3.9) where P (R) is the GC radial surface density profile and bg is the determined background value. Visually, the shape of the blue GC surface density profile is very different to that of the galaxy surface brightness whereas both red and green GC surface density profiles have shapes similar to the galaxy light. Both red and green density profiles are slightly steeper than the galaxy light profile beyond ∼ 5 arcmin.

3.4 Discussion

In previous work on NGC 4365, evidence of an additional subpopulation of GCs at intermediate (green) colours was based on small numbers and restricted to the central regions (Puzia et al., 2002; Larsen et al., 2003; Hempel & Kissler-Patig, 2004; Brodie et al., 2005; Kundu et al., 2005; Larsen et al., 2005). The large sample of GCs in this work, covering an extended radial range, allows us to revisit these claims in the optical wavelength range. We applied several independent statistical tests to the colour distribution, i.e. Chi- Squared minimization (χ2), the Kolmogorov-Smirnov (KS) test and Kaye’s Mixture Model (KMM) algorithm (see Appendix A). When the entire radial range of our imaging is used our statistical tests cannot conclusively rule out a bimodal interpretation, partly because GC subpopulations do not have strictly Gaussian colour distributions. When examining GC subpopulation colours with radius we find significant negative colour/metallicity gradients in both the bimodal and trimodal cases. Due to its radial colour gradient, the red subpopulation has somewhat greener colours at large radii. This explains why statistical tests showed inconclusive results when the full radial extent of the imaging was used. However, this cannot explain the central green overdensity that motivates splitting NGC 4365’s GCs into three subpopulations. We compared the photometrically observable properties of blue, green and red GCs (measured here to contain 43, 17 and 40 per cent of total GCs respectively). The high probability blue, green and red samples have colour ranges of g − i < 0.82, 0.89 < g − i < 1.0 and g − i > 1.1 where definitions are based on kernel smoothing of the GC colour distribution. We find evidence for different properties between the subpopulations. The 54 Chapter 3. Wide-field imaging of NGC 4365’s globular cluster system green GCs are intermediate in size to the blue and red subpopulations (clearly smaller than the blue GCs and marginally larger than the red ones) and have a significantly different mass function to both blue and red subpopulations. The size differences could be due to a continuous trend with colour rather than specific subpopulation properties (see Jordán,2004) but the differences in mass functions suggest that the three subpopulations are indeed distinct. Green GCs also have a significantly steeper surface density profile than red GCs within 2 arcmin (13.4 kpc) from the galaxy centre. However, the green and red subpopulations can be described by the same radial surface density profile, with different normalisations, beyond 2 arcmin (13.4 kpc). The ellipticity of the green GC subpopulation is similar to the blue GC subpopulation and significantly more elongated than that of the red. We do not detect any variation in ellipticity of the subpopulations with galactocentric radius to 11 arcmin (74 kpc). These results lend support to the idea that blue, green and red GCs form three distinct subpopulations.

With the same three subpopulation division we find a number of similarities between the properties of the red subpopulation and the galaxy light of NGC 4365. The g − i radial colour gradient of the galaxy light and mean colour of the red GC subpopulation are virtually identical (see also Spitler, 2010; Forbes & Forte, 2001). The ellipticity of the red subpopulation is consistent with that of the galaxy starlight and the surface density profile of the red subpopulation is closer to the starlight surface brightness profile than either blue or green subpopulations. These results confirm similar findings in the literature (Brodie & Strader, 2006) and support GC formation scenarios in which the red subpopulation is closely linked to the formation of the bulk of the galaxy field stars.

To understand the physical implications of optical colour trimodality we consider the relationship between optical colours, metallicities and ages. There is considerable debate about whether the colour distribution of GC systems is a good proxy for the metallicity distribution, even if uniform old ages for GCs are assumed (see discussions on the nonlinearity of colour-metallicity transformations by Yoon et al., 2006; Cantiello & Blakeslee, 2007; Hempel et al., 2007). It is also possible that one or more of the GC subpopulations have signiﬁcantly younger ages caused by a merger (see Hibbard & Mihos, 1995; Schweizer & Seitzer, 1998; Montuori et al., 2010) though there is little spectroscopic evidence of young ages for NGC 4365 GCs (Brodie et al., 2005).

The green subpopulation might be a consequence of a unique evolutionary history of NGC 4365. The kinematically distinct core (KDC) in the stellar light might be a signal of this. This property is only found in galaxies the SAURON team (de Zeeuw et al., 3.4. Discussion 55

2002) classify as slow rotators in the inner regions (Emsellem et al., 2007). The KDC is confined within 5 arcsec (0.56 kpc) of the galaxy centre and the stellar population inside and outside the KDC are indistinguishable in age and metallicity. Inside 5 arcsec the stars rotate around the minor axis and outside 5 arcsec they rotate around the major axis (Davies et al., 2001). Davies et al. (2001) suggest that the KDC might have been formed in the merger of “gas-rich fragments at high redshift” and state that they find no evidence of recent star formation in NGC 4365. The effective radius of the galaxy (∼ 2 arcmin or ∼ 13.4 kpc) and the radial extent of the green subpopulation (∼ 2 arcmin) are much larger than the KDC. It is possible that the formation of the green GC subpopulation is linked to the formation of the KDC but since they are not spatially correlated, and other elliptical galaxies with KDCs show no clear indication of a green subpopulation, the connection is not clear.

The core must have been formed very early on because Davies et al. (2001) find the same > 12 Gyr luminosity-weighted age for the KDC and the rest of NGC 4365, with no sharp changes in metallicity across the boundary of the KDC. If the formation of the green subpopulation is linked then we would expect the green subpopulation to also be very old. Brodie et al. (2005) found evidence that the green subpopulation was indeed old but this analysis was based on a small number of GCs. We might also expect to find rotation for the green subpopulation to be offset by ∼ 90◦ compared to the other GC subpopulations and the bulk of the starlight, if it is associated with the formation of the KDC.

Another possible signature of the unique evolutionary history could be the significant misalignment between the galaxy’s kinematic and photometric major axes. While KDCs are relatively common in elliptical galaxies, the minor axis rotation of the bulk of NGC 4365’s starlight is relatively uncommon. Van den Bosch et al. (2008) use triaxial orbit based models to explain NGC 4365’s apparent minor axis rotation and KDC. The combination of axisymmetry in the inner parts with triaxiality in the outer parts, which van den Bosch et al. (2008) describe is consistent with the age and metallicity of the KDC but bears no obvious relation to the formation of a green GC subpopulation in NGC 4365. Alternatively, Hoffman et al. (2010) show that a ∼ 2.5 Gyr old major merger remnant with 15 − 20 per cent progenitor gas fraction has a kinematic signature remarkably similar to NGC 4365. The gas rich major merger that Hoffman et al. (2010) describe is a tempting explanation for the presence of green GCs because all NGC 4365’s anomalies would have one explanation. Given that the green GC subpopulation is likely old (Brodie et al., 2005) and there is no evidence for young stars in the KDC (Davies et al., 2001) this would have to be a very early major merger. 56 Chapter 3. Wide-field imaging of NGC 4365’s globular cluster system

The current galaxy and GC system formation scenarios that describe mechanisms for producing bimodality in GC systems could also explain GC system trimodality for a subset of galaxies under certain conditions. In the multiphase collapse scenario GC systems are formed during two phases, the first phase produces metal poor GCs, is truncated by one of a variety of possible mechanisms and then later the second phase produces the metal rich GCs (Forbes et al., 1997). In this context a GC system that is trimodal in metallicity could simply be the result of three formation phases truncated twice (by the same or two different mechanisms). The truncation of the metal poor formation phase could be caused by a universal epoch of reionization but, for truncation to happen twice, at least one truncation mechanism has to be related to nearby galaxies or a process internal to the galaxy, like an active galactic nucleus. The major merger formation scenario (Zepf & Ashman, 1993), where metal poor GCs are present in galaxies before mergers and metal rich GCs are formed during major mergers, could be consistent with some galaxies containing three GC subpopulations. If a galaxy underwent a very early gaseous major merger and later another major merger it may have two subpopulations more metal rich than the metal poor GC subpopulation. Depending on the age and metallicity difference it could appear trimodal in colour. The accretion model for GC system formation (Côtéet al., 1998) found more than two modes in some of their simulated GC systems before a third GC subpopulation was considered for NGC 4365 from observations. Côtéet al. (1998) mention that more than two peaks are present in some of their simulations when a very steep luminosity function (Schechter function α = −1.8) is used. They find that the peak of the metal poor population correlates with the slope of the luminosity function and that for very steep slopes the metal poor peak is more metal poor. It is conceivable that the merger histories of some giant elliptical galaxies will therefore show the presence of a third, old, intermediate metallicity subpopulation. The questions that remain include distinguishing which of these formation scenarios best explain the GC system of NGC 4365 and determining whether the trimodality in the colour distribution of the GC system is reflected in the metallicity distribution. The method in Foster et al. (2010) for determining metallicity from the Calcium triplet indices could provide a large sample of metallicity measurements over a wide field of view to assess whether the metallicities of the three colour subpopulations we define here are distinct. The spectroscopic analysis of a large number of GCs could also be used to build 3.5. Summary and Conclusions 57 a picture of the kinematics of the GC system of NGC 4365 and its subpopulations. An understanding of NGC 4365’s GC system kinematics could distinguish whether two or three subpopulations were all formed in situ, whether the more metal rich populations were accreted or whether the subpopulations are the result of several major mergers.

3.5 Summary and Conclusions

Combining the photometric depth, size information and resolution of HST/ACS data with the spatial extent and three ﬁlter imaging of the Subaru/S-Cam data we can achieve a uniquely detailed, and unmatched spatially extensive, analysis of NGC 4365’s GC system.

The GC system extends beyond 134 kpc from the galaxy centre to 9.5Re. The blue GC subpopulation has not yet reached the background level at the very edges of our Subaru/S- Cam imaging. We place a lower limit on the total number of GCs to be 6450 110. We find further evidence to support the existence of a distinct third subpopulation at intermediate colours in the GC system of NGC 4365. We also find a trend of increasing GC size with galactocentric radius and negative gradients in the colour/metallicity of both blue and red subpopulations with galactocentric radius. The blue subpopulation shows evidence for a blue tilt. Comparing the GC system with the galaxy light we find that the red subpopulation has a similar colour, radial colour gradient, ellipticity and radial surface density slope to the galaxy light. Based on various measured properties we find it most likely that the green subpopulation is distinct from the blue and red subpopulations. Consequently, it is very likely that NGC 4365 has had a unique evolutionary history causing the existence of a third GC subpopulation in this giant elliptical, where most similar galaxies only have two. The photometric properties of the blue and red GC subpopulations are consistent with the properties found for those of other giant elliptical GC systems, therefore any formation scenario for the green subpopulation must leave the predicted properties of the other two subpopulations relatively unchanged. In Chapter 4 we will analyse the spectroscopy for NGC 4365’s GCs. Spectroscopic analysis enables the kinematics of the GC system and it’s subpopulations to be charac- terised, which should help determine the origin of the green subpopulation of NGC 4365’s GC system.

4 Globular cluster system kinematics and substructure in NGC 4365

Classifying the stars has helped materially in all studies of the structure of the universe. No greater problem is presented to the human mind. Teaching man his relatively small sphere in the cre- ation, it also encourages him by the lessons of the unity of Nature and shows him that his power of comprehension allies him with the great intelligence over-reaching all. —Annie Jump Cannon

4.1 Introduction

Globular clusters (GCs), as some of the oldest and densest stellar systems in the Universe, present intriguing questions about their own formation while also being very useful tracers of the formation of their host galaxy (Brodie & Strader, 2006). They probably formed during violent star formation episodes in their host galaxy (or host galaxy progenitors) and therefore we can use their spatial distribution and kinematic properties to unravel the formation history of galaxies. GCs are more numerous and more easily identiﬁable in elliptical galaxies (where bright star forming clumps and dust obscuration is minimal) than in spiral galaxies.

NGC 4365 is a giant elliptical galaxy (E3), MB = −21.3 mag (Ferrarese et al., 2006), with various intriguing and unusual properties as highlighted below. We use these unusual properties of NGC 4365 to constrain galaxy and GC formation scenarios by placing the extra information contained in this system in the overall framework of ‘normal’ giant

59 60 Chapter 4. Globular cluster system kinematics and substructure in NGC 4365

Figure 4.1: The positions of the six DEIMOS slitmasks plotted on the Subaru Suprime- Cam (S-Cam) i0 filter image. The scale is shown in the bottom left corner of the 35 × 27 arcmin image. It is centred on α=12:24:26.824; δ=+07:19:03.52 (J2000.0). At a distance of 23.1 0.8 Mpc (Blakeslee et al., 2009) 1 arcmin = 6.72 kpc. 4.1. Introduction 61 elliptical galaxies. The GC systems of most giant elliptical galaxies are bimodal in optical colours (Brodie & Strader, 2006; Peng et al., 2006). There is a growing body of evidence that the relative age difference between GC subpopulations is small (Strader et al., 2005) and the distinct colours of the commonly observed subpopulations are due to bimodality in GC metallicity (Strader et al., 2007; Woodley et al., 2010; Alves-Brito et al., 2011) that are likely explained by two formation mechanisms, sites or epochs. Possible formation scenarios for bimodal metallicity distributions in GC systems are the major merger (Zepf & Ashman, 1993), multiphase collapse (Forbes, Brodie & Grillmair, 1997) and accretion scenarios (Côté et al., 1998). An alternative view is that the bimodal GC colour distributions can be observed from a unimodal GC metallicity distribution, because of strong nonlinearity in the relationship between colour and metallicity (Yoon et al., 2006; Cantiello & Blakeslee, 2007; Blakeslee et al., 2010) and in this case the kinematic properties of each colour subpopulation are not expected to be significantly different. The GC system of NGC 4365 has three GC subpopulations, that includes an additional subpopulation of ‘green’ GCs between the commonly observed blue and red GC subpopulations. This additional subpopulation, thought at the time to be younger than the ‘normal’ two subpopulations, was discovered by Puzia et al. (2002) using a combination of near-IR and optical photometry. Brodie et al. (2005) suggested that the three subpopulations might be found as trimodality in optical colours and Larsen et al. (2005) found that the green subpopulation was restricted to small galactocentric radii. Specifically for NGC 4365, two independent methods have confirmed the GCs for all three subpopulations (blue, green and red) to be older than ∼ 10 Gyr. Brodie et al. (2005) observed a spectroscopic sample of 22 GCs to measure their ages and metallicities using Lick indices from the Low-Resolution Imaging Spectrograph (LRIS) on the Keck telescope. They found that NGC 4365 has three old GC subpopulations with metal poor (blue), intermediate metallicity (green) and metal rich (red) GCs. Chies-Santos et al. (2011) presented a near-IR photometry and optical analysis of 99 GCs with much higher precision than previously possible, and determined that there is no observable offset in the mean age between the GCs of NGC 4365 and GC populations of other giant ellipticals. This indicates that there is no significant population of young GCs in NGC 4365. In addition to its almost unique GC system, NGC 4365 has very unusual stellar kinematic properties. The kinematically distinct core (KDC) at its centre is relatively uncommon, seen in ∼ 10 per cent of early type (E and S0) galaxies in the ATLAS3D volume limited survey (Krajnovićet al., 2011). Also very uncommon is the starlight kinematics 62 Chapter 4. Globular cluster system kinematics and substructure in NGC 4365 outside the KDC which ‘rolls’ about the major axis rather than rotating about the minor axis (Surma & Bender, 1995). Less than 5 per cent of galaxies, in the ATLAS3D volume limited survey, show this kinematic behaviour. Davies et al. (2001) used the SAURON (Spectrographic Areal Unit for Research on Optical Nebulae) integral field spectrograph to map its kinematic and metallicity structure in two dimensions, finding stars both inside and outside the KDC to be older than 10 Gyr. NGC 4365 is one of only two early type galaxies, in the ATLAS3Dsample of 260, that have both a KDC and a near 90◦ misalignment between the photometric and kinematic major axes (i.e. rolling rather than rotating stars). Finally Bogdánet al. (2012a), see also Mihos et al. (2012, in prep), presented evidence of a stream of stellar light extending ∼ 200 kpc southwest and ∼ 100 kpc northeast from NGC 4365. This strong indicator of a very recent ∼ 10 : 1 merger (occurring with in the last few Gyr) presents a possible cause for the unique properties of NGC 4365. Chapter 3 (Blom et al., 2012a) presented an optical photometry analysis of ∼ 4000 NGC 4365 GCs, using the most spatially extended sample of GCs analysed to date (observed with Subaru/Suprime-Cam and supplemented with archival imaging from Hubble Space Telescope/Advanced Camera for Surveys) and found different spatial distributions, sizes and mass distributions for the three subpopulations of NGC 4365 GC system. They also concluded that separate formation scenarios are required to explain the existence of the three separate subpopulations of GC in NGC 4365. Kinematic analysis of the three GC subpopulations in NGC 4365 is key to disentangling the possible formation scenarios of GC systems. Several recent investigations have shown that the kinematic features of the two standard GC subpopulations (seen in other giant elliptical galaxies) are different (Lee et al., 2010; Arnold et al., 2011; Foster et al., 2011). When combined with galaxy formation models this information constrains the possible formation scenarios of GCs. For example, Foster et al. (2011) compare kinematics of the GC subpopulations in NGC 4494 with simulations such as those described in Bekki et al. (2005) and conclude that the galaxy has undergone a recent major merger of similar disk galaxies. The addition of the kinematic properties of a third GC subpopulation limits the possible formation scenarios even further. In Section 4.2 we first describe the preparation, observation and reduction of the spectroscopic sample of GCs and then investigate the colour/metallicity and line-of-sight velocity distributions of the GCs. The separation of the GCs into subpopulations and analysis of the kinematic features of the three subpopulations are presented in Section 4.3. We then discuss the results and summarise our conclusions regarding the possible 4.2. Spectroscopic sample 63

Figure 4.2: Colour distribution (reddening corrected) of GCs with radial velocity measurements compared with the total GC photometric sample. The grey histogram shows the kinematic sample and the white histogram shows the distribution of photometric candidate GCs brighter than the turnover magnitude, i0 = 23.6, and over all observable radii, ∼ 20 arcmin, (scaled down by a factor of 2.7 for comparison purposes). The kinematic sample is a subsample of the photometric candidate sample. The top x-axis shows the [Fe/H] metallicity determined from the empirical conversion published in Lee et al. (2010). We note the three GC subpopulations with peaks at g0 − i0 = 0.8, 0.97 and 1.13. We also note that the blue GCs in the photometric candidate sample are underrepresented in the kinematic sample with respect to the number of redder GCs (g0 − i0 > 0.9). formation scenarios for NGC 4365 in Sections 4.4 and 4.5. Appendix C contains the kinematic fits for all the different methods of subpopulation separation we investigated, and an alternative assumption of position angle for kinematic fitting.

4.2 Spectroscopic sample

4.2.1 Data acquisition

The GC spectroscopic sample presented in this work was obtained from six multi-object DEIMOS slitmasks observed in clear dark time on the Keck II telescope. This galaxy was observed as part of the SAGES Legacy Unifying Globulars and Galaxies Survey 64 Chapter 4. Globular cluster system kinematics and substructure in NGC 4365

(SLUGGS1) (Brodie et al. 2012 in prep). Four masks were observed on 2010, Jan- uary 11 and 12 with seeing between 0.6 and 0.9 arcsec and a further two masks were observed on 2010, February 18 with seeing of ∼ 0.8500. In Chapter 3 (Blom et al., 2012a) Subaru/Suprime-Cam (S-Cam) and Hubble Space Telescope / Advanced Camera for Sur- veys (HST/ACS) imaging was used to identify GC candidates associated with NGC 4365 and publish extinction corrected g0, r0 and i0 magnitudes for S-Cam and/or g and z magnitudes for HST/ACS candidates. A total of 443 GC candidates with i0 brighter than 23 mag were placed in slits, for which we aimed to obtain a spectrum of high enough signal- to-noise (S/N ∼ 5) to determine a line-of-sight radial velocity. For all six masks a 1200 l mm−1 grating was used, with 1 arcsec wide slits and centred on 7800 A˚ to obtain ∼ 1.5 A˚ resolution. This set up (see also Foster et al., 2011; Arnold et al., 2011; Strader et al., 2011; Romanowsky et al., 2009, 2012) allows observations from ∼ 6550 - 8900 A˚ , which covers the wavelength range in which we expect the red-shifted Calcium II triplet (CaT) absorption features (8498, 8542 and 8662 A).˚ It sometimes also includes the red-shifted Hα line (6563 A)˚ at the blue end of the observed spectrum. Footprints of the six DEIMOS masks plotted on the S-Cam i0 filter image, used to identify GCs around NGC 4365, are shown in Fig. 4.1. We obtained total exposure times of 110, 40, 105, 96, 120 and 75 minutes for each of the six masks respectively. For each mask the total exposure time was split into three or four separate exposures to minimize the effects of cosmic rays. The total exposure times for each mask were independently constrained by seeing and object visibility during the night. The median seeing was 0.8, 0.7, 0.75, 0.7, 0.9 and 0.7 arcsec for each of the six masks respectively. Masks two and six were observed at the end of their respective observing nights. The raw data were reduced to one dimensional spectra using a customised version of the deep2 (Deep Extragalactic Evolutionary Probe 2) galaxy survey data reduction pipeline (idl spec2d; Cooper et al., 2012; Newman et al., 2012). The pipeline uses dome flats, NeArKrXe arc lamp spectra and sky light visible in each slit to perform flat fielding, wavelength calibration and local sky subtraction, respectively. It outputs the object spectrum as well as the locally subtracted sky spectrum. By placing the slits at an angle of at least 5◦ offset from the position angle of each mask the deep2 pipeline is forced to determine an individual wavelength calibration for each pixel column of the spectra in each slit. This ensures that the root-mean square of the wavelength calibration is as small as possible, Newman et al. (2012) state that the deep2 pipeline achieves a root-mean-squared (rms) accuracy of 0.007A,˚ with adequate arclamp

1http://sluggs.swin.edu.au/ 4.2. Spectroscopic sample 65 exposure and a fifth order Legendre polynomial iterative fit. This rms is determined from the final mean residual about the fit. The 5◦ tilt in the slit also improves the accuracy of the sky subtraction, by oversampling each sky line to allow a more accurate fit to the sky level.

4.2.2 Obtaining line-of-sight velocities

We obtained reliable line-of-sight radial velocities for 252 GCs. This increases the previously available NGC 4365 spectroscopic GC sample (33 individual GCs in Larsen et al., 2003; Brodie et al., 2005, combined) by an order of magnitude. The line-of-sight radial velocity is determined by the peak of the cross correlation function (obtained using the IRAF task RV.FXCOR) between the spectrum and 13 stellar templates. The velocity measurement of a GC is considered reliable if the correlation function has a single, identifiable peak and at least two of the four absorption lines (three CaT and one Hα) are clearly visible in the red-shifted spectrum. In a few cases one or other of these visual checks was not convincing and the resulting GC velocity is recorded as a borderline velocity measurement (17 objects in total). The final catalogue of 269 radial velocities for NGC 4365 GCs are given in Pota et al. (2013). The colour distribution of the kinematic spectroscopic sample is compared with that of the photometric sample in Fig. 4.2. The photometric sample includes GC candidates with g0 − i0 colours between 0.5 and 1.4, showing overdensities at g0 − i0 ∼ 0.8 and between 0.95 and 1.2 as well as an underdensity at g0 − i0 ∼ 0.9. Ideally the kinematic sample will have the same colour distribution, scaled down in number, as the photometric sample. In this ideal case we would be able to extrapolate conclusions drawn from the spectroscopic samples to the entire GC population without caveat. Fig. 4.2 shows that while we do not completely achieve this ideal, the spectroscopic sample does cover the full colour range and shows a significant underdensity at intermediate colours between the blue and the green subpopulations (seen at g0 − i0 ∼ 0.9). In the spectroscopic sample blue GCs are under represented with respect to the green/red GCs compared to the photometric sample. This is because the six DEIMOS pointings have a much smaller radial extent (∼ 12 arcmin) than the S-Cam imaging (∼ 20 arcmin) and at larger galactocentric radii there are many more blue GCs than red or green GCs (see Fig. 4.5 and Section 3.3.5. We expect to see a roughly Gaussian distribution in the line-of-sight velocities of GCs around NGC 4365. In Fig. 4.3 we show a Gaussian distribution overplotted on the velocity distribution of NGC 4365’s GCs. The peak is set to the literature value for NGC 4365’s −1 recession velocity (vsys = 1243 km s , Smith et al., 2000) and the standard deviation is 66 Chapter 4. Globular cluster system kinematics and substructure in NGC 4365

Figure 4.3: Line-of-sight velocity distribution for GCs in the kinematic sample. The distribution of the secure line-of-sight velocity measurements (252 GCs) is plotted with a solid line and the borderline velocity measurements (17 GCs) are added to the distribution with a dashed line. The plotted Gaussian distribution has a peak at 1243 km s−1 and a velocity dispersion of 237 km s−1. The velocity distribution of the GCs agrees well with the Gaussian at high velocities but is skewed to low line-of-sight velocities. 4.2. Spectroscopic sample 67

Figure 4.4: Line-of-sight radial velocity (Vlos) of GCs plotted against galactocentric radius. Filled black circles show the 252 GCs with reliable radial velocity measurements, while grey circles show borderline GC detections. The solid black line is plotted at the −1 galaxy systemic velocity Vsys = 1243 km s , dotted lines show 1σ and 2σ envelopes and the dashed line shows the 3σ envelope. The σ values were calculated in bins of 29 GCs −1 with Vlos > 1243 km s , excluding borderline GCs. Note the large number of low velocity −1 (Vlos ∼ 700 km s ) > 2σ outliers. The solid grey line is plotted at the systemic velocity −1 of the nearby galaxy NGC 4342 (Vsys = 751 km s , Grogin et al., 1998). which lies ∼ 35 arcmin from NGC 4365 68 Chapter 4. Globular cluster system kinematics and substructure in NGC 4365

Figure 4.5: GC surface density with colour and galactocentric radius. The colour scale indicates object number per arcmin2 per colour bin ranging from black at low density to red at high density. The colour-radius position of spectroscopically conﬁrmed GCs are overplotted with white dots. GCs with radial velocities more than 2σ from the host galaxy are marked with black plusses (positive) and dots (negative). The distribution of GCs with observed line-of-sight radial velocities is consistent with the underlying distribution of photometrically observed GCs in the same radial range and there is no indication that the GCs with > 2σ low velocities deviate from the underlying distribution either. set to 237 km s−1. This standard deviation was calculated from one side of the velocity −1 distribution, i.e. from 68 percent of the GCs with Vlos > 1243 km s . The peak agrees well with the observed GC velocity distribution but an excess of GCs is seen above the overplotted Gaussian at low velocities in Fig. 4.3.

We can calculate the skewness (ξ3) of the velocity sample using the standardised third moment (Joanes & Gill, 1998),

pN(N − 1) PN (x − x)3 ξ = i=1 i (4.1) 3 N(N − 2) s3 where N is the total number of GCs in the sample, xi is the line-of-sight velocity of the ith GC, x is the mean line-of-sight velocity of the sample and s is the standard deviation of the sample. The calculated skewness is not strongly dependent on whether the entire GCs sample (including borderline velocity measurements) or only the secure velocity measurements are used,

ξ3,all = −0.20

ξ3,secure = −0.19

The standard error in skewness (ξ3 ) can be calculated from the number of GCs in the 4.2. Spectroscopic sample 69 sample (Joanes & Gill, 1998),

s 6N(N − 1) = = 0.15 (4.2) ξ3 (N − 2)(N + 1)(N + 3)

The measured negative skewness of the GC line-of-sight velocity distribution (−0.190.15) is only marginally signiﬁcant, given the relatively small number of GCs contributing to the skewness. We further investigate the possible causes of this negative skewness in the velocity distribution.

4.2.3 Low velocity GCs

We explore the possibility that the GCs causing marginal skewness in the measured velocity distribution are associated with the stream of stars that were found crossing NGC 4365 (North East to South West) from analysis of wide-field, very deep B filter imaging (Bogdánet al. 2012a; Mihos et al. 2012, in prep). The stream seems to be an indication of an ongoing merger with a small galaxy. The distinct velocities of these GCs may indicate that they are not associated NGC 4365’s GC system and the recently discovered stream is a likely candidate. Fig. 4.4 shows the GC line-of-sight velocities plotted against galactocentric radius. To define which GC velocities are outliers we used the GC velocities > 1243 km s−1 (where the velocity distribution is not affected by skewness) to calculate the standard deviation of the sample, σ. The calculation of σ was done in radial bins of 29 GCs each (excluding borderline velocity measurements). The calculated 1σ, 2σ and 3σ envelopes are plotted above and below the systemic velocity and show clearly that there is a larger number of GCs with low velocities outside the 2σ envelope (13) than with high velocities (3). When we plot these velocity outliers on a map of GC surface density with colour and galactocentric radius there is no strong indication that the low velocity outliers follow a separate distribution to the colour-radius distribution of NGC 4365 GCs (see Fig. 4.5). The density map shows the distribution of photometrically identified GCs overplotted with the kinematic sample of GCs (those with reliable radial velocities). The blue and red subpopulations, as well as an overdensity of green GCs in the central regions, are visible in the density plot and the GCs with radial velocity measurements follow the colour-radius distribution well. The low velocity outliers tend to have red or green colours in the inner regions and blue colours in the outer regions of the galaxy, which is consistent with the general distribution of NGC 4365 GCs in colour-radius space. In Fig. 4.6 the positions of the low velocity GCs are plotted on a map of the spatial 70 Chapter 4. Globular cluster system kinematics and substructure in NGC 4365

Figure 4.6: Spatial distribution of GCs showing individual radial velocities. The colour of individual points correspond to their radial velocity and the scale runs from 500 (dark blue) to 1800 km s−1 (dark red). GCs with radial velocities more than 2σ from the host galaxy are marked with plusses (positive) and dots (negative). 4.3. Kinematics of the GC subpopulations 71 distribution of the GCs with line-of-sight velocities. All but two of the low velocity GCs form a line running NE to SW across NGC 4365. This spatial structure of low velocity GCs is aligned with the recently discovered tidal stream (Bogdánet al. 2012a; Mihos et al. 2012, in prep) and would seem to indicate that the majority of the low velocity GCs are associated with the stream. In summary, the evidence does not conclusively show that the low velocity GCs are associated with the stellar stream. Their distribution in colour-radius space is not distin- guishable from the overall distribution of NGC 4365 GCs but their spatial distribution does suggest that they are different from the bulk of NGC 4365 GCs. We exclude all GCs with line-of-sight velocities that lie outside the 2σ envelopes and GCs with borderline velocity measurements from further kinematic analysis. This conservative sample ensures that the kinematic results are unaffected by unreliable GC velocity measurements and that the sample is not skewed to low velocities. We note here that including the low velocity outliers makes no significant difference to subsequent kinematic results, except to increase the uncertainties in our fits.

4.3 Kinematics of the GC subpopulations

4.3.1 Kinematic model description

To investigate the detailed kinematic properties of the three GC subpopulations we fit an inclined disk model to the GC system of NGC 4365, using the individual GCs as tracers. This inclined disk model is able to fit for rotation velocity and the position angle of that rotation with our sparse and non-uniform sampling of tracer particles. To fit this model we minimize (as in Foster et al., 2011);

N 2 X (Vobs,i − Vmod,i) Λ = + ln(σ2 + (∆V )2) (4.3) σ2 + (∆V )2 obs,i i=1 obs,i

Vrot Vmod,i = Vsys (4.4) r 2 1 + tan(PAi−PAkin) qkin

where Vobs,i is the observed line-of-sight velocity, ∆Vobs,i is the uncertainty on the observed velocity, σ is the fitted velocity dispersion of the sample, Vmod,i is the modelled velocity given by Equation 4.4, Vsys is the recession velocity of NGC 4365, Vrot is the th rotation velocity fitted to the sample, PAi is the position angle of the i GC, PAkin is the kinematic position angle fitted to the sample and qkin is the kinematic axis ratio (b/a) fixed 72 Chapter 4. Globular cluster system kinematics and substructure in NGC 4365 for the sample. In contrast to Foster et al. (2011) we do not use a minimisation algorithm to determine the minimum value of Λ but evaluate the function for all reasonable values of Vrot, σ and PAkin. This brute force method ensures that we have found the global minimum for the function and therefore the best fitting values.

We use the bootstrap Monte Carlo method to obtain an estimate of the uncertainty of the model fit. A random sample of GCs is chosen from the original GC sample (equal in number to the original sample but allowing the data from a given GC to be used more than once) and then Λ is minimized for that randomised sample. This is repeated 1000 times and the uncertainties are determined by calculating the central 68 per cent confidence intervals of the distribution (1σ) for each of Vrot, σ and PAkin. To use a brute force minimisation method for each iteration would be prohibitively time consuming and therefore we use a Powell minimisation algorithm (from SCIPY.OPTIMIZE in the PYTHON framework) with initial estimates set to the solution obtained from the kinematic best fitting values. The Powell minimisation algorithm is marginally sensitive to the initial conditions but we found no significant difference in the uncertainties obtained using this method and those found for a test case where the brute force method was used instead.

In general the 1σ uncertainties are not symmetric about the best fitting value because the distribution obtained from the bootstrap algorithm is not symmetric about the peak value. We calculate a central 68 per cent confidence interval rather than a symmetric one. The peak value of the distribution aligns well with the best fitting value in most cases. We do not artificially align the peak value of the distribution with the best-fit value and consequently it is possible for the best fitting value to fall outside the 1σ range of the distribution. This is a clear indication that for such a case the model is inappropriate for the data.

We use this model to determine the kinematic properties of NGC 4365’s GC system as a whole, ﬁnding: 42.3 −1 Vrot = 39.510.5 km s , 7.4 −1 σ = 212.29.2 km s and 22◦ PAkin = 147.326 .

This would seem to indicate a GC system with a deﬁned but small rotation along the photometric minor axis of the galaxy (132◦). However it is well-known that GC subpopulations within a single galaxy can show diﬀerent rotation properties (Lee et al., 2010; Arnold et al., 2011; Foster et al., 2011) and might confuse any rotation signal unless they are separated, therefore we explore further. 4.3. Kinematics of the GC subpopulations 73

Figure 4.7: Kinematics for NGC 4365 GCs as a function of g0 − i0 colour. Kinematics are fitted for a rolling sample of 30 GCs and are plotted at the mean colour of the sample. The top panel shows the fitted rotation velocity (Vrot), the centre panel shows the fitted velocity dispersion (σ) and the bottom shows the fitted kinematic position angle (PAkin). Solid lines show the kinematic fit to the data and the grey filled area shows the 1σ uncertainties (obtained from 1000 bootstrap trials). On the bottom panel the photometric major axis of NGC 4365 (see Section 3.2) is plotted with dashed lines and the minor axis is plotted with dotted lines. The bottom panel shows that the very blue, green and very red GCs all have constrained position angles of rotation and that the blue and red subpopulations rotate in a different direction to the green subpopulation. Where there is significant contamination between subpopulations in colour the rotation is unconstrained.

4.3.2 Kinematics as a function of colour

We use our kinematic model to analyse the kinematics of NGC 4365’s GCs as a function of colour. The GCs with velocities less than 2σ away from the galaxy systemic velocity and secure velocity measurements are used to form a rolling sample of 30 GCs as follows. The 30 GCs with bluest colours form the first subsample. To form the next subsample the bluest GC is discarded while the GC with the bluest colour that was not included in the first subsample (i.e. the next reddest) is added and this process is repeated until all GCs have been included in a subsample. We chose the sample size of 30 as a best compromise between samples large enough to robustly constrain the kinematics and samples small enough to give adequate colour resolution. The kinematic fitting procedure is carried out 74 Chapter 4. Globular cluster system kinematics and substructure in NGC 4365 on each subsample and the results are shown in Fig. 4.7. In this case we fix the kinematic axis ratio to one (qkin = 1) as the photometric axis ratio for the galaxy light as well as blue, green and red GC subpopulations are all different (qblue = 0.56, qgreen = 0.70, qred = 0.97 and qgal = 0.75 from Section 3.3.5). This assumption of qkin = 1 is equivalent to assuming a circular rotation disk and is chosen as the simplest case scenario. 4.3. Kinematics of the GC subpopulations 75 62 53 23 24 ) kin ◦ ( A P 174 194 ) (56) 14 22 15 26 − 1 s 1 . σ 0 . 8 (65) > > 218 202 0 (km i ed − R ) 0 P g − 1 52 11 48 34 ot s r V 39 94 (km 33 31 27 23 ) kin . ◦ ◦ ( A P 145 144 ) or 222 1 . 05 (77) 13 22 10 15 ◦ − 1 s 0 . 8 (53) < σ 0 i > 229 217 (km − 0 een g ) Gr < P − 1 57 29 39 24 ot s r ) for each subpopulation in the medium probability (probability 0 . 9 V 83 81 kin (km 38 54 28 34 ) kin ◦ ( A P 254 269 (61) ) 10 17 10 16 − 1 s 0 . 8 (53) σ 0 . 85 > < 179 178 (km 0 i lue B − ) 0 P g − 1 48 18 44 27 ot s r V 49 67 (km ) and kinematic position angle (PA ) and best colour cut splits. Numbers of GCs in each subpopulation representative sample are given in brackets cut probability Colour Split P Summary of kinematic fits to various divisions of the three GC subpopulations. We tabulate the rotation velocity Definition colour medium ), velocity dispersion ( σ rot able 4.1: ( V after the definition parameters. The major axis of NGC 4365’s stellar light is 42 denoted with T 76 Chapter 4. Globular cluster system kinematics and substructure in NGC 4365

Fig. 4.7 shows that the GCs of NGC 4365 have significantly different kinematic features, which depend on GC colour. The changes in position angle seen in Fig. 4.7 show this very clearly, e.g. blue GCs rotate with PAkin that is consistent with the photometric major axis, green GCs rotate with PAkin consistent with the photometric minor axis and red GCs with rotate with PAkin that is between the photometric major and minor axes of NGC 4365. At the colours where we expect significant overlap between subpopulations the rotation cannot be constrained. Here the values fitted for the PAkin become unstable while the values fitted for Vrot drop to zero and become inconsistent with the errors calculated. Where it is not possible to constrain the position angle of the rotation, the amplitude of rotation is also unconstrained.

Foster et al. (2011) noted that Vrot is artiﬁcially increased when the PAkin is allowed to vary. Strader et al. (2011) describe this artiﬁcial increase as a rotation bias and note that this bias is important when the amplitude of rotation is low, e.g.

V r20 rot 6 0.55 (4.5) σ N

Here N denotes the number of GCs in the sample. We apply this criterion to the kinematic

fit as a function of colour of NGC 4365’s GCs. Where the PAkin is well constrained 0 0 (g − i ∼ 0.79 − 0.84, 0.97 − 1.02 and 1.12 − 1.17) the rotation parameter (Vrot/σ) is well above the criterion value for this sample size (N = 30). We also see that the rotation parameter fails the criterion (i.e. is affected by rotation bias) in areas where the PAkin is 0 0 poorly constrained (e.g. 0.87 < g − i < 0.95) even when the fitted Vrot is measured to be significantly greater than zero.

In the blue, green and red colour ranges where we do constrain the kinematics the blue −1 subpopulation Vrot and σ ranges are 50 − 140 and 140 − 160 km s respectively while the −1 green subpopulation Vrot and σ ranges are 80 − 200 and 180 − 220 km s respectively. −1 The rotation range of the red subpopulation is Vrot ∼ 80 − 180 km s while its velocity dispersion decreases with increasingly red GC colour, σ ∼ 220 to ∼ 200 km s−1.

This observation of distinct kinematic features is a confirmation that the three GC subpopulations of NGC 4365 seen in colour and other photometric properties (Larsen et al., 2003; Brodie et al., 2005; Blom et al., 2012a) are indeed separate. They are likely to have formed via different mechanisms or at different epochs. To better constrain the position angle (PAkin) and derive clearer values for the amplitude of rotation (Vrot) and dispersion (σ), we need to divide the GCs into three uncontaminated representative samples of the three GC subpopulations. 4.3. Kinematics of the GC subpopulations 77

Figure 4.8: Observed line-of-sight velocity is plotted against position angle for the blue (left), green (centre) and red (right) GC subpopulations. We show the subpopulations as defined by the best colour cut split (see Table 4.1). The fitted rotation velocity is plotted with a solid sinusoid, and an envelope of twice the fitted velocity dispersion is shown with ◦ ◦ dotted lines for each subpopulation. The position angle (PAkin) of rotation is 269 , 144 and 194◦ for the blue, green and red subpopulations respectively. The photometric major axis of the galaxy is 42/222◦ and the overdensity of blue GCs at ∼ 210◦ is due to the strong elongation of the blue GC subpopulation along the major axis of the galaxy.

4.3.3 Dividing the sample into three subpopulations

In Chapter 3 we divided NGC 4365’s GCs into three subpopulations according to a probability of belonging to that subpopulation. Each GC was assigned a probability of belonging to the blue, green and red subpopulations (PBlue + PGreen + PRed = 1) based on its colour and galactocentric radius. To quote, “The process used to determine the blue-green-red probability of a GC starts by determining the relative number of GCs in each subpopulation at the galactocentric radius of the GC. This is done by comparing the values of the radial surface density profile for each subpopulation. The relative numbers of GCs in each population are used to scale the normalisations of the three Gaussian distributions and comparing the relative values of all three Gaussian distributions at the colour of the GC we calculate the probability of an object being blue, green, or red.” The high probability subsamples (P > 0.95) do not include any green GCs and therefore we define medium probability samples with P > 0.80 to represent the three subpopulations. Of GCs with radial velocities, 53 blue, 53 green and 65 red GCs form part of the medium probability samples. The kinematic fits to the medium probability fits are shown in Table 4.1. The blue and green sample fits agree with the kinematic properties found in Fig. 4.7 but the fit to the ◦ red sample is not well constrained (PAkin varies by 115 ).The blue medium probability sample contains GCs with colours bluer than g0 − i0 ∼ 0.85 and the green sample contains GCs with colours between g0 − i0 ∼ 0.85 and g0 − i0 ∼ 1.05. These samples contain 78 Chapter 4. Globular cluster system kinematics and substructure in NGC 4365 relatively few GCs but do not overlap and are mostly free of contamination from the other subpopulation. The red medium probability sample contains more GCs but also contains GCs with colours as ‘blue’ as g0−i0 ∼ 1.0 and is likely affected by contamination from green GCs. We expect some contamination particularly in the red sample because the medium probability red GCs extend almost as far as the peak colour of the green subpopulation (g0 −i0 = 0.98). The probability that these GCs belong to the green subpopulation is close to 20 percent and not trivial.

We also experimented with splitting the GCs into three subsamples at diﬀerent g0 − i0 colours, detailed in Appendix C. To determine the most accurate kinematics for the three colour subpopulations of NGC 4365 we need to include as many GCs as possible in each sample while minimising the number of GCs from one subpopulation contaminating the sample for another subpopulation. The best kinematic results are determined by the ﬁt which has the minimum value of Λ/ndf (see Equation 4.3) and the most constrained

PAkin. Salient results are shown in Table 4.1 and details discussed in Appendix C. The best kinematic ﬁts are found for the 61 blue GCs with g0 − i0 < 0.85, 77 green GCs with 0.9 < g0 − i0 < 1.05 and 56 red GCs with g0 − i0 > 1.1. The results for the medium probability and best colour cut samples are consistent within the uncertainties, but the best colour cut sample contains larger numbers of blue and green GCs and a relatively uncontaminated sample of red GCs. The red sample in this case ends close to the expected peak in the red subpopulation (g0−i0 = 1.13). Henceforth we use the best colour cut sample for the subpopulation representative samples.

The kinematics are ﬁtted twice for each colour split deﬁnition, once with the axis ratio

fixed to the photometric value for each GC subpopulation (qkin = qphotom) and once with the axis ratio fixed to unity (qkin = 1) as before. We do not have enough GCs in each sample to fit for qkin and it is not clear that the kinematic axis ratio would necessarily be equal to the photometric axis ratio, especially since Fig. 4.7 leads us to expect rotation in different directions to the photometric position angle. We do not find any significant differences in the fitted kinematics between the two axis ratio cases and conclude that this kinematic fitting method is not very sensitive to the kinematic axis ratio (see Appendix

C). All further kinematic ﬁts are done in the case where qkin = 1 as this is the more general case.

We compare the rotation of these subpopulation representative samples (henceforth referred to as the blue, green and red subpopulations) with each other and with the kinematics of the central starlight (to ∼ 0.5 arcmin) as measured by Davies et al. (2001) ◦ and Krajnovićet al. (2011). They found that the KDC rotates with a PAkin of 38 5 4.3. Kinematics of the GC subpopulations 79 aligned with the photometric major axis of the galaxy (42◦) and that the bulk of the ◦ starlight rotates with a PAkin of 145 6.5 , very close to the minor axis of the galaxy (132◦). Using the Stellar Kinematics from Multiple Slits (SKiMS) technique (see also Proctor et al., 2009; Foster et al., 2011), Arnold et al. (2012, in prep) find that the minor axis rotation of the starlight continues out to ∼ 4.5 arcmin. We find that the blue ◦ ◦ subpopulation rotates with a PAkin between 235 and 297 , not consistent with either the major (222◦) or the minor (312◦) axes of NGC 4365. The green subpopulation rotates 27◦ with a PAkin of 14423 , consistent with the minor axis of the galaxy and the rotation of ◦ the bulk of the starlight. The red subpopulation rotates with a PAkin between 170 and 217◦, almost consistent with the major axis but in the opposite direction to the stars of the KDC. The kinematic properties that we have found by using the best representative samples for each GC subpopulation agree with the kinematics seen in Fig. 4.7 where GCs kinematics are plotted as a function of colour. Fig. 4.8 shows the combined rotation velocity and velocity dispersion. These are plotted separately for each subpopulation and the best-fit models show good agreement with the observed radial velocities of the subpopulations. The non-uniform azimuthal distribution of GCs, seen particularly in the blue subpopulation, but also in the green subpopulation, is partly due to the elliptical spatial distribution of these subpopulations

(axis ratio qblue = 0.56 and qgreen = 0.70, see Section 3.3.5) and partly due to the asym- metric spatial coverage of the spectral observations. We do not expect this azimuthal bias to significantly influence the kinematic fits.

4.3.4 Radial kinematics for three subpopulations

We use the best colour cut samples to analyse the variation of kinematics with galactocentric radius for each subpopulation. The samples contain 61 and 56 GCs for the blue and red subpopulations respectively, but only 61 of 77 GCs for the green subpopulation as we restrict the green sample to include only GCs with galactocentric radii smaller than 5 arcmin. We impose this galactocentric radius restriction on the green subpopulation because beyond this radius it contributes less than ten per cent of the total GC population. The GC subpopulation kinematics are fitted as a function of galactocentric radius in rolling bins of 22 GCs and the results of the kinematic fits are shown in Fig. 4.9. The bin size of 22 GCs was chosen to maximize the radial range of the kinematic analysis. This is significantly smaller than the bin size used for Fig. 4.7 because inter subpopulation contamination has been reduced. We fit the kinematics in the case where the PAkin is allowed to vary with galactocentric radius. 80 Chapter 4. Globular cluster system kinematics and substructure in NGC 4365

Figure 4.9: Kinematics as a function of radius for NGC 4365’s three GC subpopulations. GC subpopulations are defined using the best colour cut in Table 4.1. Kinematics are fitted for a rolling sample of 22 GCs and are plotted at the mean galactocentric radius of the sample. The top panel shows the fitted rotation velocity (Vrot), the centre panel shows the fitted velocity dispersion (σ) and the bottom panel shows the fitted kinematics 0 0 position angle (PAkin). The blue subpopulation (all GCs bluer than g − i = 0.85) is plotted with blue dashed lines, the green subpopulation (GCs between g0 − i0 = 0.9 and 1.05 with galactocentric radius < 5 arcmin) is plotted with green dash-dotted lines and the red subpopulation (all GCs redder than g0 − i0 = 1.1) is plotted with red dotted lines. The lines show the kinematic fit to the data and the filled areas show the 1σ uncertainties (obtained from 1000 bootstrap trials). The black points show stellar kinematics extracted along the minor axis of NGC 4365 from the SAURON data (de Zeeuw et al., 2002). The effective radius of the galaxy starlight is plotted with a vertical line at 2.1 arcmin. On the bottom panel the photometric major axis of NGC 4365 is plotted with dashed lines and ◦ the minor axis is plotted with dotted lines. The variation of the PAphot is less than 3 and not visible on this scale. The kinematic fits to the three subpopulations show clearly different features. 4.3. Kinematics of the GC subpopulations 81

The blue subpopulation shows significant rotation at all radii. There is an indication ◦ ◦ that the PAkin of rotation twists from ∼ 200 to ∼ 310 between 3 and 7 arcmin (20 −1 and 47 kpc). The rotation velocity is consistent with being flat and Vrot ∼ 100 km s over the entire radial range of the blue subpopulation. The velocity dispersion of the blue subpopulation is also consistent with being flat and σ ∼ 200 km s−1. The radial kinematic profiles are consistent with the values found for the blue GCs as seen in Fig. 4.7 as well as the fitted values for the entire blue subpopulation sample (Table 4.1). The kinematics of the blue GC system varies more significantly with galactocentric radius than GC colour and by removing most of the contaminating GCs coming from the green subpopulation as well as dividing the blue GC by radius, we achieve tighter constraints on the subpopulation kinematics. The twist in the PAkin could suggest that the radial outskirts of the blue subpopulation have been affected by accretion of galaxies and associated GCs, or a tidal interaction. This kinematic twist could also be a signature of triaxiality in the blue GC subpopulation. −1 The green subpopulation rotates relatively fast at small radii (Vrot ∼ 200 km s ) but the rotation decreases quickly with increasing galactocentric radius to ∼ 75 km s−1 at 3.5 ◦ arcmin. The radially fitted PAkin is consistent with being flat (150 30 ) and consistent 27◦ with the PAkin fitted to the entire subpopulation (14423 ). The velocity dispersion of the green subpopulation also decreases with galactocentric radius from 260 to 180 km s−1. ◦ The red subpopulation rotates fastest between 2 and 4 arcmin, with PAkin ∼ 222 . The −1 rotation velocity amplitude here is Vrot ∼ 180 km s .The fitted PAkin within 2 arcmin is not as well constrained as that fitted between 2 and 4 arcmin, and within 2 arcmin the rotation parameter fails the criterion defined in Equation 4.5. The rotation of the red GC subpopulation at small galactocentric radii is consistent with zero rotation and the value −1 of ∼ 75 km s fitted here is due to rotation bias caused by the free variation of PAkin. The velocity dispersion of the red GCs decreases with radius from σ ∼ 220 to ∼ 150 km s−1. The red subpopulation is dispersion dominated in the central regions and rotation dominated in the outer regions of the galaxy. 82 Chapter 4. Globular cluster system kinematics and substructure in NGC 4365

Figure 4.10: Kinematic parameters of GCs as a function of galactocentric radius. Rota- tion parameter Vrot/σ (top) as well as the root mean squared velocity VRMS (bottom) for GC subpopulations plotted against galactocentric radius. Linestyles are the same as those in Fig. 4.9. This figure shows the case where the position angle is allowed to vary. The rotation bias is small everywhere except inside a galactocentric radius of 2 arcmin for the red subpopulation. Here the plotted value for Vrot/σ is significantly overestimated but because there is no fitting involved in obtaining the VRMS this value is not affected. 4.3.KinematicsoftheGCsubpopulations 83 5 , colour † † † • 2 i , † 68 19 11 . . 02 . . − 0 0 0 0 97 . g ± ± ± 2 (arcmin) ± fromKrajnović 1 e 30 17 . ∼ . . 12 2 . R 7 3 0 † † † † † Thisisthelargestradial 08 08 08 03 08 5 . . . . . 0 0 0 0 0 phot ± ± ± ± ± q 56 70 97 75 75 . . . . . 0 0 0 0 0 4 ) , 4 1 , • valuewasconvertedtoa • − 10 16 10 15 15 26 s 3 z 10 ± ± ± aretakenfromChapter3, ± ± − † (km 178 217 202 g 262 σ 260 The ) 3 1 − 4 , 4 • , 44 27 39 24 48 34 8 ± ± ± (kms ± 67 81 94 60.9 ot 63 r V ) ◦ • 5 ( . 28 34 27 23 23 24 5 6 ± ± ± ± ± kin 269 144 194 A 38 P 145 fromDaviesetal.(2001),whileallothervaluesarederivedinthisChapter. • ) † ◦ 1 1 , , ( † † 3 1 . ± 2 222 222 ± / / 222 9 . photom 42 42 / undefined A 40 ∼ ∼ 42 P 3 , † † † † (mag) i 0.80 0.97 1.13 1.15 1.40 ThiseffectiveradiusmeasurementisnotobtaineddirectlyfromaSérsicprofilefittothegreensubpopulation 2 g fromFerrareseetal.(2006)and GCs SummaryoftheNGC4365galaxysystemproperties.Valuesmarkedwith GCs GCs Thepositionanglesfortheblueandgreensubpopulationswerequalitativelysimilartothegalaxylightandweretherefore Stars KDC Thesearemaximumvelocitiesobservedinalimitedfield-of-viewnotmeanormedianvelocities. 1 4 Red Blue Green able4.2: value. assumedequaltoit. extentoftheKDCandnotitseffectiveradius. butbycomparingthepowerlawslopeofthegreenandredsubpopulations. Notes: etal.(2011), T 84 Chapter 4. Globular cluster system kinematics and substructure in NGC 4365

Plotting the rotation parameter (Vrot/σ) in Fig. 4.10 we see that the red subpopulation shows very diﬀerent properties to the blue and green subpopulations. This ratio is one measure of the diskyness of the system, where Vrot/σ > 1 can indicate a disky system. It is only the red GC subpopulation that shows rotationally dominated characteristics, and ◦ only at galactocentric radii beyond 2.5 arcmin, where it rotates with a PAkin of ∼ 222 (photometric major axis of NGC 4365).

q P 2 (Vlos−Vsys) The root mean squared velocity VRMS ∼ N shows variation with radius for all three subpopulations but changes most signiﬁcantly in the green subpopulation. −1 The green GC subpopulation VRMS decreases from ∼ 270 to ∼ 190 km s over 1.5 arcmin in galactocentric radius. The sharply declining VRMS seen in the case of the green subpopulation could be explained by a truncation in their density distribution or a bias towards radial orbits relative to the other GC subpopulations.

4.4 Discussion

Since most giant elliptical galaxies contain only two metallicity subpopulations of GCs it is likely that one of NGC 4365’s three GC subpopulations is the result of an additional and unusual formation/accretion process. Determining which of the GC subpopulations is additional may indicate which mechanism caused it and could be a key to uncovering the formation mechanisms of GC subpopulations in ordinary elliptical galaxies. We summarise the kinematic and photometric properties of the three GC subpopulations as well as the two main components of the galaxy starlight (the KDC and the bulk of the stars) in Table 4.2 and use these properties to discuss possible formation scenarios. We ﬁnd three distinct kinematic signatures for the three GC subpopulations. The blue subpopulation rotates at all observable radii and its kinematic position angle twists slowly with increasing galactocentric radius from 200◦ to 312◦. Its rotation is unrelated to NGC 4365’s stellar light. The green subpopulation rotates with the bulk of NGC 4365’s stellar light along the photometric minor axis of the galaxy (132◦). The red subpopulation rotates only at large radii along the photometric major axis (222◦) of NGC 4365 but in the opposite direction to the KDC of the galaxy (38◦). The distinct kinematics of the red and green GC subpopulations, seen clearly in Fig. 4.11, provide evidence that the green GCs form a separate subpopulation and are not simply the metal poor tail of the red subpopulation. The photometric analysis in Chapter 3 (Blom et al., 2012a) found that the radial density distribution and colour of the red GC subpopulation most closely matched the 4.4. Discussion 85 stellar light of NGC 4365 and, while the axis ratio of the green subpopulation more closely matched that of the starlight, they concluded that there was evidence that the red subpopulation was related to the bulk of the galaxy starlight. The analysis here shows that the rotation position angle, and possibly also the rotation velocity amplitude, of the green GC subpopulation most closely match the extrapolated stellar kinematics. Thus it is not immediately clear which of the two is the additional subpopulation.

If the red GC subpopulation is additional it could have been created by an additional, later (still > 10 Gyr ago) merger (Zepf & Ashman, 1993). This might explain why the red subpopulation rotation is decoupled from the bulk of the starlight and the green subpopulation rotates with the bulk of the stars. The problem arises when comparing this red subpopulation to those of other galaxies of similar luminosity to NGC 4365, as it is not an outlier but consistent with the colour of red subpopulations of other galaxies of similar luminosity/mass (Peng et al., 2006).

If the green GC subpopulation has been recently donated from a medium sized galaxy that accreted on to NGC 4365 then the remnant of this accreted galaxy might still be visible in the form of a stellar stream across NGC 4365 as reported by Bogdánet al. (2012a). In Chapter 3 we showed that the green subpopulation is centred at g0 − i0 ∼ 0.97 and we can estimate the absolute magnitude range for the accreted galaxy by comparing the peak colour of the green subpopulation with the relationship between peak GC subpopulation colour and galaxy luminosity empirically determined by Peng et al. (2006). In this case the green subpopulation of NGC 4365 is the original metal-rich (red) GC subpopulation of a −14 < MB < −15 dwarf galaxy and we expect to find primarily blue and green GCs in the stellar stream of NGC 4365. The total stellar light observed in the low surface brightness stream is B ∼ 13.5 mag (Bogdánet al., 2012a, this a minimum brightness for the accreted galaxy) corresponding to MB ∼ -18.3 at the distance of NGC 4365. There is a sizeable mismatch between the predicted absolute magnitude for the dwarf galaxy and the much brighter integrated stellar light from the stream, suggesting that the stellar stream is not the remnant of the galaxy associated with accreted green GCs. This explanation has several other problems. It is still to be explained why the accreted green subpopulation is so centrally concentrated while the stellar light of the accreted galaxy is still on the outskirts of NGC 4365. Also why does the green subpopulation rotate with the bulk of the stellar light of NGC 4365 (along the minor axis of the galaxy) rather than along the stream orientation, which is roughly aligned with the major axis of NGC 4365?

If the accretion event which brought the green GCs into NGC 4365 was not recent, the currently observable stellar stream may be unrelated to the green GCs. We can 86 Chapter 4. Globular cluster system kinematics and substructure in NGC 4365 still investigate this claim based on the number of green GCs (see a similar analysis by Romanowsky et al., 2012). In Chapter 3 we showed that the green subpopulation contributes 17 percent of the total GC population or 1100 20 GCs to NGC 4365’s GC system (NT ot = 6450 110). We can calculate the absolute magnitude range of the accreted galaxy using the observed relationship between GC speciﬁc frequency (SN = 0.4(M +15) NGC 10 V , Harris & van den Bergh, 1981) and galaxy luminosity (Peng et al.,

2008). To be consistent with the SN range observed for GC systems in the literature, the accreted galaxy must be brighter than MV ∼ −20.5 (MB ∼ −19.5). The number of GCs in the green subpopulation far exceeds the expected number of GCs that would be associated with a dwarf galaxy. If the green subpopulation was donated by an accreted galaxy (at any stage in the history of NGC 4365) the mean colour of the green GCs suggests a very small galaxy whereas the number of green GCs suggests a large galaxy.

If the green GCs were accreted from a group of dwarf galaxies it might account for the large number of GCs with colours that indicate a small host galaxy, but the group of dwarfs would need to contain many tens of members to account for the large number of inferred green GCs.

We also consider the possibility that both the red and green subpopulations were formed in-situ. In this case the green subpopulation would have formed unusually early in the dense core of the galaxy while the red GCs formed later when the core of the galaxy was slightly bigger and slightly more metal rich. This scenario might explain why the green subpopulation is more centrally concentrated than the red subpopulation and why the green GCs rotate with a signiﬁcant fraction of the starlight. It does not explain why we do not detect a fraction of the stars rotating with the red GCs or what would have caused two epochs of early in-situ GC formation.

−1 Bogd´anet al. (2012a) studied the lenticular galaxy NGC 4342 (Vsys = 751 km s , Grogin et al., 1998) embedded in the stellar stream across NGC 4365 (see Fig. 4.11) and inferred a massive and extended dark matter halo around NGC 4342. The central velocity dispersions of NGC 4365 (2617 km s−1) and NGC 4342 (24411 km s−1) are consistent within errors (Bernardi et al., 2002) implying similar mass galaxies. It is possible that this very massive, if relatively faint, galaxy could have passed close by NGC 4365 and dragged some of NGC 4365’s own stars out from the galaxy to form the stream. It is also possible that the stream stars have been stripped from NGC 4342 itself. These scenarios −1 both agree with our observation of low velocity GCs (at ∼ 700 km s , similar to the Vsys of NGC 4342) in NGC 4365 (Fig. 4.4) that are aligned with the stellar stream (Figs 4.6 and 4.11). The low velocity GCs are consistent with the colour distribution of NGC 4365’s 4.5. Conclusions 87 underlying GC population (Fig. 4.5) but it is still unknown how their colours compare to those of NGC 4342. It is tempting to suggest that the close passage of NGC 4342 might have stretched NGC 4365 starlight and GC system into its current shape. If NGC 4365 was round or only mildly elliptical (with a PAphot close to the current PAkin of the bulk of stellar light) before NGC 4342 made its close pass we would expect to see remnants of the original shape in the central regions of NGC 4365. Ferrarese et al. (2006) ﬁnd the galaxy ellipticity drops ◦ below 0.05 in the g and z bands while the PAphot twists to ∼ 90 at radii smaller than 0.3 arcsec. These photometric properties only appear at scales smaller than the KDC of NGC 4365 and are unlikely to relate to the bulk of NGC 4365. This scenario might explain the anomalous ‘rolling’ starlight as an artefact of very recent perturbations to NGC 4365’s photometric shape but would not explain the presence of NGC 4365’s KDC . The KDC could be a remnant of an old merger, as its age is observed to be > 10 Gyr and consistent with the galaxy starlight (Davies et al., 2001) or the consequence of our particular viewing angle into an intrinsically triaxial system (van den Bergh et al., 1991). Fig. 4.11 summarises the photometric and kinematic information of the NGC 4365 system graphically. The blue GC subpopulation is the most radially extended GC subpopulation and is more elongated than the galaxy light. Its rotation shows a slow twist from alignment with the galaxy major axis in the inner parts to the galaxy minor axis in the outer parts. The amplitude of its rotation and velocity dispersion are constant with radius. The green subpopulation is the most centrally concentrated, roughly as elongated as the galaxy light and shows ‘rolling’ kinematics like the galaxy light (PAkin ∼ galaxy minor axis). Its rotation and velocity dispersion amplitude decrease with increasing radius. The red subpopulation is almost circular and rotates only in the outer parts, along the galaxy major axis (in the opposite direction to the KDC). The low velocity GCs are aligned with the stellar stream and contain GCs of all colour subpopulations. The kinematics of the galaxy and GC system are complex but it appears that the stellar stream is not connected to the presence of any one of the GC subpopulations.

4.5 Conclusions

NGC 4365 is a giant elliptical galaxy with rare stellar kinematic properties and also, very unusually, three globular cluster (GC) subpopulations. Recently, a stellar stream running across NGC 4365 and its nearby neighbour, NGC 4342, was discovered. With high velocity resolution Keck/DEIMOS spectra we analyse the kinematics of a large number and spatially extended sample of GCs around NGC 4365. These data 88 Chapter 4. Globular cluster system kinematics and substructure in NGC 4365

Figure 4.11: Schematic representation of the NGC 4365 stellar light and GC system. The diagram is oriented with North up and East left. Positions of low velocity GC outliers are shown with black dots. Ellipses show the 2Re extent of the galaxy light (black) and three GC subpopulations in blue, green and red respectively. The shape and relative size of the stellar light stream (reported in Bogdánet al., 2012a) is shown with a grey irregular outline. The rotation direction of the galaxy light as well, as the green and red GC subpopulations, are shown in the top right of the diagram. The rotation direction of the blue GCs is not shown because there is a significant twist with galactocentric radius. A cross marks the position of NGC 4342. The low velocity GCs are generally aligned with the stellar stream and have a similar velocity to NGC 4342. 4.5. Conclusions 89 allow us to separate the GC system into its three colour subpopulations and analyse the kinematics of each subpopulation in detail. We find distinct rotation properties for each GC subpopulation. This indicates that the three GC subpopulations are distinct and that the formation history of NGC 4365 is complex. We conclude that it is unlikely that the third (green) GC subpopulation is related to the existence of the stellar stream and that it is also unlikely to have originated from an accreted galaxy. We also find a further group of low velocity GCs (covering the full range in colour), which might be related to the stellar stream extending across NGC 4365 towards NGC 4342 (∼ 35 arcmin away at a similar systemic velocity to these GCs). The stellar stream and the low velocity GCs are examined in more detail in Chapter 5.

5 Tidal interaction between the early type galaxies NGC 4365 and NGC 4342

I do not know what I may appear to the world, but to myself I seem to have been only like a boy playing on the sea-shore, and diverting myself in now and then ﬁnding a smoother pebble or a prettier shell than ordinary, whilst the great ocean of truth lay all undiscovered before me. —Sir Isaac Newton

5.1 Introduction

ΛCDM cosmology predicts that the giant galaxies we see in the local Universe have been built up to their present size by successive mergers of galaxies and accretion of small galaxies. Globular clusters (GCs) are very useful tracers of these merger and accretion events as they are dense and mostly robust to the violent interactions of galaxies. Additionally, several extragalactic spectroscopic studies have found that the overwhelming majority of GCs are > 10 Gyrs old (Strader et al., 2005). Therefore, most of the GCs that we observe locally were formed very early in the evolution of the Universe and have survived all the interactions their host galaxies were involved in. They have likely undergone several galaxy merger events, or have been stripped off smaller galaxies, and accumulated around large galaxies over time (Côtéet al., 1998; Tonini, 2013). They should maintain a chemical signature of the conditions they were formed under and a kinematic signature of the processes by which they were acquired by large galaxies (West et al., 2004). NGC 4365 is a giant elliptical galaxy with a redshift independent distance measurement

91 92 Chapter 5. Tidal interaction between NGC 4365 and NGC 4342 of 23.1 Mpc (Blakeslee et al., 2009) and a recession velocity of 1243 km s−1 (Smith et al., 2000), placing it ∼ 6 Mpc behind the Virgo Cluster. NGC 4365 is the central galaxy in the W 0 group and has been noted to have rare properties for a galaxy of its size. It has a kinematically distinct core and the bulk of its stars rotate along the minor axis (Surma & Bender, 1995; Davies et al., 2001; Krajnovićet al., 2011, Arnold et al. 2013 in prep.), while its GC system consists of three subpopulations (Puzia et al., 2002; Larsen et al., 2005; Brodie et al., 2005; Blom et al., 2012a,b) rather than the usual two (Brodie & Strader, 2006; Peng et al., 2006). NGC 4342 is a small S0 galaxy 20 arcmin away from NGC 4365 with a recession velocity of 751 km s−1 (Grogin et al., 1998) but its distance has not been measured independently of redshift. Recently, Jiang et al. (2012) highlighted it as a local example of a high redshift superdense ‘red nugget’ (Daddi et al., 2005; van Dokkum et al., 2008) and it has long been known to have an oversized central supermassive black hole for its bulge mass (Cretton & van den Bosch, 1999). In the last year, Bogdánet al. (2012a) presented a very deep B filter image of NGC 4365 showing a stellar stream extending from NE of NGC 4365, across the elliptical galaxy, bridging the gap between NGC 4365 and NGC 4342, and extending further SW. This image has prompted investigation of the nature of the interaction between NGC 4365 and NGC 4342. Bogdánet al. (2012b) found X-ray emitting hot gas around NGC 4342. This presents an interesting puzzle regarding the origin of the stellar stream. It appears as if the stellar material has been tidally stripped off NGC 4342 as it passed by NGC 4365 and what remains of NGC 4342 is now swinging back towards NGC 4365. However, the presence of X-ray emitting gas around NGC 4342 suggests a large dark matter halo that should not be present if stars have been stripped off the galaxy. Here we investigate the interaction between NGC 4365 and NGC 4342 using GCs, assessing what the spatial, colour and velocity distributions of the GCs around NGC 4342 and NGC 4365 indicate about the nature of the interaction between these two galaxies. In particular, is NGC 4365 stripping stars off NGC 4342 or is there another explanation? We describe the photometric selection of GC candidates and detail our analysis of the properties of the spatial and colour distributions of the GC system in Section 5.2. Section 5.3 contains the reduction and analysis of the spectroscopic data as well as the kinematic selection of GCs. In Section 5.4 we analyse the kinematic properties of the stream and NGC 4342 GCs and in Section 5.5 we present a counter-argument to the findings of Bogdán et al. (2012a), before discussing the results and concluding the Chapter in Sections 5.6 and 5.7 respectively. 5.2. Photometric Analysis 93

5.2 Photometric Analysis

5.2.1 Imaging Data

We use the public archival data from the MegaCam instrument on the Canada-France- Hawaii Telescope (CFHT) to identify GCs in the area around NGC 4342 and NGC 4365. The square degree u filter image stack is centred on R.A. = 185.912◦ and Decl. = 7.0562◦ (J2000.0) with a total exposure time of 4240 s. The g and i filter image stacks are centred on R.A. = 186.133◦ and Decl. = 7.2708◦ (J2000.0), and have total exposure times of 3170 and 2055 s respectively. As seen in Figure 5.1, the overlap of the three image stacks covers an area of ∼ 0.66 square degrees, encompassing NGC 4365, NGC 4342 and almost all of the stellar stream (Bogdánet al., 2012a). We use catalogues produced by the MegaPipe data reduction pipeline (Gwyn, 2008), operated by the Canadian Astronomy Data Centre (CADC), to spatially match point sources found in all three image stacks. The faintest i magnitude of included objects is 23.0 mag. We correct for foreground reddening, using

Au = 0.11, Ag = 0.08 and Ai = 0.04, determined from Galactic dust maps (Schlegel et al., 1998).

5.2.2 Globular cluster candidate selection

The selection of GC candidates by comparison with matched point sources is shown in

Figure 5.2. The figure shows (g − i)0 colour plotted against (u − g)0 colour for the point sources in the field-of-view, as well as the NGC 4365 GC candidates used to identify the locus of GCs in colour-colour space. These NGC 4365 GC candidates were matched in spatial coordinates between the MegaPipe catalogues and previous work using the Subaru/SuprimeCam (SCam) instrument (Blom et al., 2012a). We use the SCam identified GC candidates to fit a line to the GC distribution, then expand the line by 2σ to form a box and select candidates that overlap that box within individual errors. Given the i = 23.0 mag cutoff, the mean individual errors in the GC candidate sample are 0.1 mag in (u − g)0 colour and 0.04 mag in (g − i)0 colour. As described in Section 2.4.2, the radial extent of the imaging used affect the distribution of GCs within the selection box and here, because the both the reference and selection sample are restricted to large radii, there are more GCs at the blue end of the selection box than at the red end. We identify a total of 1786 GC candidates that meet these criteria. 94 Chapter 5. Tidal interaction between NGC 4365 and NGC 4342

Figure 5.1: Spatial alignment of the u ang g filter images from the MegaCam intrument on the Canada-France-Hawaii Telescope (CFHT). The u filter image is centred on NGC 4342 at R.A. = 185.912◦ and Decl. = 7.0562◦ (J2000.0) and shown on the bottom right, below the g filter image, which is centred on NGC 4365 at R.A. = 186.133◦ and Decl. = 7.2708◦ (J2000.0). The i filter image has the same position as the g filter and is not shown here. Each filter image covers one square degree in area and the region of overlap is ∼ 0.66 square degrees. 5.2. Photometric Analysis 95

Figure 5.2: Colour selection of globular cluster (GC) candidates. (g−i)0 colour is plotted against (u − g)0 colour. The black crosses show all point sources brighter than i = 21 mag and red dots show the objects matched with NGC 4365 GC candidates selected from Subaru/SuprimeCam photometry (Blom et al., 2012a). The previously photometrically selected GC candidates form a linear correlation, shown with a black line. We select additional GC candidates that are consistent, within individual errors (not shown), with the red 2σ box around the line. 96 Chapter 5. Tidal interaction between NGC 4365 and NGC 4342

Figure 5.3: Spatial distribution of globular cluster (GC) candidates around NGC 4342 and NGC 4365. (Left) Individual GC candidates are shown as grey dots and the two galaxies are shown with circles circumscribing a cross (NGC 4365 towards the top left and NGC 4342 towards the bottom right). The two-dimensional distribution of the GC candidates shows a clear overdensity around NGC 4365 and NGC 4342. There are several galaxies in the region that are as bright as NGC 4342 but none show a similar GC overdensity. The positions of these galaxies; NGC 4341, NGC 4343, IC 3267 and IC 3259 are plotted with x’s. The horizontal void running through the centre of NGC 4365 is due to CCD chip problem in the u ﬁlter. GC candidates are not recovered in the very central regions of NGC 4365 and NGC 4342 due to the high galaxy surface brightness there. (Right) The two-dimensional spatial distribution of GC candidates smoothed to a resolution of 1.8 arcmin. Here we see an overdensity of GC candidates not only around NGC 4342 and NGC 4365 but also between the two galaxies and South West of NGC 4342, that is spatially coincident with the stellar stream. The spatial coincidence of the elongated GC overdensity and the recently reported stellar stream (Bogd´anet al., 2012a) is highly unlikely to be a chance occurrence. Both galaxy centres and the horizontal void are blocked in the smoothed image. 5.2. Photometric Analysis 97

5.2.3 Spatial distribution

Given the selection of GC candidates, Figure 5.3 shows their spatial distribution. The ﬁrst panel shows the unsmoothed 2D surface density map. We see a GC overdensity corresponding to NGC 4365’s GC system and we ﬁnd an overdensity of GCs around

NGC 4342. The absolute magnitude of NGC 4342 (MB = −18.45 mag for a distance of 23.1 Mpc) is comparable to that of several other galaxies superimposed on the stellar stream (i.e. NGC 4341: MB = −17.68, NGC 4343: MB = −19.46, IC 3267: MB = −17.7,

IC 3259: MB = −17.91, assuming the same distance as NGC 4342) however we do not see a clear overdensity associated with any of them. None of the superimposed galaxies are likely to be significantly more distant than NGC 4342 (see their quoted recession velocities in Figure 5.13). We find that NGC 4342 has an order of magnitude more GCs than the nearby galaxies of similar luminosity, indicating that it has deviated significantly from a ‘normal’ evolutionary history. The 2D distribution of GC candidates is smoothed to a spatial resolution of 1.8 arcmin in the second panel of Figure 5.3. Once the data are smoothed the GC overdensity along the stellar stream is also apparent. The GC overdensity bridges the gap between NGC 4365 and NGC 4342 and extends further SW of NGC 4342, in roughly the same line as the bridge part of the GC stream. This result is robust to small variations in the magnitude cut and the spatial resolution of the smoothing. The number of GCs contributing to the visible stream overdensities is on the order of tens of GCs and it is conceivable that such a stream structure could arise by chance. We see a possible example of this chance structure extending NW from NGC 4365. This possibility is testable if the distribution of GCs expected from both NGC 4365 and NGC 4342 were simulated and compared with the actual spatial distribution. In order to simulate the spatial distribution of the GC contribution from these two galaxies, their radial and azimuthal distributions need to be determined.

Spatial properties of NGC 4342’s GC system

With ∼ 140 photometrically observed GC candidates in the spatial overdensity seen around NGC 4342 in the left panel of Figure 5.3 we can determine some of the quantitative parameters of the GC distribution. The radial surface density of GC candidates, shown in Figure 5.4, is calculated in annuli around NGC 4342. No GC candidates are detected within 0.5 arcmin from the centre of NGC 4342 because the high starlight surface brightness there inhibits object detection. Consequently the first radial bin (between 0.5 and 1.5 arcmin) is possibly also affected by this incompleteness. To ensure that there are significant GC 98 Chapter 5. Tidal interaction between NGC 4365 and NGC 4342

Figure 5.4: Radial surface density of GCs brighter than i = 23 mag around NGC 4342 plotted against galactocentric radius. The dots with error bars show the surface density points and the best fit Sérsicprofile (with an additional background term) is plotted with a solid line. The ranges of Sérsicparameters that would still fit the data points within errors are also shown. The two dotted lines forming an envelope to the best fit line show the minimum and maximum range for the fitted parameters (Pe, Re, n) respectively. The third dotted line shows a Sérsicprofile where the shape parameter n is 2.3 times the best fit value, consistent with all the data points. The data are well matched by the Sérsic profile fit but without data points closer than 1 arcmin from the galaxy centre it is not possible to constrain the fit well. See text for further details. 5.2. Photometric Analysis 99 candidate numbers in each bin we divide the data into only five annuli and fit a Sérsic profile added to a background term to these. The Sérsicprofile and background term (bg) is given by (Graham & Driver, 2005):

" 1 #! R n P (R) = Pe exp −bn − 1 + bg (5.1) Re where

bn = 1.9992n − 0.3271 (5.2)

Pe is the density at the effective radius, Re is the effective radius of the GC system, and n is the shape parameter of the Sérsicprofile. The best fit profile is given by:

−2 Pe = 1.68 0.19 arcmin

Re = 2.12 0.12 arcmin n = 0.50 0.12 bg = 0.782 0.025 arcmin−2

The value of the background term and the effective radius of the GC system are well constrained by the five surface density values obtainable from the MegaCam data set. Here the value for the background surface density will include non-GC contamination and GCs that are spatially associated with the stellar stream rather than NGC 4342 itself. We measure the effective radius of NGC 4342’s GC system to be 2.12 0.12 arcmin (∼ 14 kpc) compared with the stellar effective radius of 0.5 arcmin (∼ 3 kpc) (de Vaucouleurs et al., 1991). The best fit profile is plotted on Figure 5.4. The truncation of NGC 4342 GCs at ∼ 5 arcmin is secure and we can assert that the bound GC system of NGC 4342 does not extend beyond ∼ 34kpc. It is unusual for a GC system to have a value < 1 for the shape parameter n (Rhode & Zepf, 2004; Spitler et al., 2006a, 2008; Pota et al., 2013) and we also plot a Sérsicprofile where n = 1.15, as we suspect that the fitted error for n is underestimated due to the small dataset. The data cannot rule out this (n = 1.15) fit as it is consistent within errors with every data point. To improve the constraint on the shape parameter n, more GC data are needed in the central regions of the galaxy. To probe further into the centre of NGC 4342 better imaging and careful subtraction of the galaxy light is required. If n is indeed < 1, the GC system of NGC 4342 is sharply truncated and could imply that it has been tidally stripped.

The azimuthal distribution of GCs around NGC 4342 is consistent with a circular GC 100 Chapter 5. Tidal interaction between NGC 4365 and NGC 4342

Figure 5.5: Azimuthal distribution of GCs within 5 arcmin from the centre of NGC 4342 and brighter than i = 23 mag. This distribution indicates that the GC system of NGC 4342 is consistent with zero ellipticity (a circular GC distribution). If the GC distribution was significantly elliptical the azimuthal distribution would show a clear sinusoidal shape with the peak of the distribution at the position angle of the elliptical distribution. The photometric position angle of NGC 4342 is 166◦. system (see Figure 5.5). Even with azimuthal bins encompassing 18 degrees in radius, the distribution of GC candidates does not show a signficant peak at any one position angle. It is possible that deeper imaging, probing fainter GCs and closer to the galaxy centre would show a small but significant ellipticity. The photometric major axis position angle of NGC 4342 is 166◦ and it would be interesting to determine if the GC distribution position angle is consistent with that of the galaxy or the stellar stream (∼ 45◦).

Statistical signiﬁcance of GC overdensity on the stream

We generated 1000 simulated spatial distributions around NGC 4365 and NGC 4342 of which Figure 5.6 is an example. These distributions are the addition of a uniform background sampling added to a sampling from the spatial distributions of both NGC 4365 and NGC 4342. Each simulated spatial distribution contains 2000 objects. This number was chosen to be higher than the number of objects in the observed distribution (1786) because there are voids in the observed spatial distribution. The background was set at a 5.2. Photometric Analysis 101 density of 0.5 arcmin−2 across the entire area. This value was chosen to visually match the object density far from NGC 4365, NGC 4342 and the stream on the CFHT/MegaCam observations. The background value of 0.78 measured in Section 5.2.3 is likely to include stream GCs as well as non-GC contaminating objects. The rejection method was used to extract random positions according to the spatial distribution functions. We used the Sérsicprofile and azimuthal distribution fitted by Blom et al. (2012a) to simulate the radial and azimuthal distribution of NGC 4365 and the previously fitted Sérsicprofile for NGC 4342. The Sérsicprofile was noted to be highly uncertain in the inner regions of NGC 4342, but the fitted profile is sufficient for these purposes as we are primarily concerned with the outer regions of both galaxies’ spatial distributions. Figure 5.6 also shows the specific areas which were compared to the observed spatial distribution. The overdensity in the two stream areas was calculated by dividing the density in the stream defined areas by the density in the off-stream areas alongside (see Figure 5.6). For the observed spatial distribution we calculated a stream overdensity of 1.66 for the stream area between NGC 4365 and NGC 4342 and an overdensity of 1.90 for the stream area SW of NGC 4342. These observed stream overdensities are compared with the distribution of stream overdensities in Figure 5.7. An overdensity equal to or greater than the observed value on the stream between NGC 4365 and NGC 4342 occurs by chance 6.9 percent of the time but only 2.3 percent of the time on the stream beyond NGC 4342. An overdensity of GCs between NGC 4365 and NGC 4342 could be due to the effects of smoothing large and overlapping GC systems from both galaxies, but we also see a clear overdensity of GCs along the stellar stream South West of NGC 4342. The extension of the GC distribution is highly unlikely to occur by chance and is best explained by tidal stripping of GCs from one galaxy onto another.

5.2.4 Colour distribution

0 0 We have calculated the transformation from (g − i )0 in the SDSS photometric system

(used for the SCam photometry) to (g − i)0 in the MegaCam photometric system. The data used for the calculation and the resulting transformation are shown in Figure 5.8. We find a shift of −0.102 mag from the SDSS to MegaCam colours, that is not strongly dependent on (g − i)0 colour over the colour range of GC candidates. Comparing the colour distributions of GC candidates associated with NGC 4365, NGC 4342 and the stellar stream (see Figure 5.9) we see a significant population of red (metal-rich) GCs present in NGC 4365’s GC system that is absent from the colour distribution of either the stream or NGC 4342. There is no significant difference between the 102 Chapter 5. Tidal interaction between NGC 4365 and NGC 4342

Figure 5.6: An example of the 1000 simulations of the spatial distribution of the GCs around NGC 4365 and NGC 4342. Each simulated spatial distribution contains 2000 GC candidates, created by the addition of a uniform background, the Sérsicprofile of NGC 4342 GCs and the spatial distribution of NGC 4365 GCs. NGC 4365’s GC spatial distribution includes the Sérsicprofile and azimuthal distribution fitted in Blom et al. (2012a). The solid orange and magenta regions mark the areas of stream overdensity between the two galaxies and SW of NGC 4342 respectively. These stream overdensities can be seen in Figure 5.3. The dotted orange and magenta off-stream regions on either side of the solid boxes are used to compare off-stream densities. 5.2. Photometric Analysis 103

Figure 5.7: Cumulative distribution of the calculated stream overdensity for 1000 simulated GC spatial distributions. The dashed orange and solid magenta lines show the on-stream density divided by the off-stream density for the stream region between the two galaxies and SW of NGC 4342 respectively. The dash-dotted orange and dotted magenta lines show the overdensity measured in the same stream regions for the observed GC candidate spatial distribution. The horizontal black dashed line shows the 95th percentile. The orange distribution shows that the region between NGC 4365 and NGC 4342 is more likely to show an overdensity due to the significant ellipticity and stream aligned position angle of NGC 4365. The intersection of the orange dash-dotted and dotted lines falls below the 95th percentile but the intersection of the magenta dotted and solid lines falls well above the 95th percentile. The GC stream overdensity SW of NGC 4342 is a statistically significant deviation from the GC distributions of NGC 4342 and NGC 4365. 104 Chapter 5. Tidal interaction between NGC 4365 and NGC 4342

0 0 Figure 5.8: Scaling between Subaru/SuprimeCam (g −i )0 and CFHT/MegaCam (g−i)0 colours. Points with error bars are the objects with photometry from both SCam and MegaCam and MegaCam i magnitude brighter than 23 mag. The black line shows a one- to-one correlation between the (g − i)0 colours. The SCam colours have been shifted by −0.102 mag and the comparison between shifted SCam and MegaCam colours is consistent with a one-to-one correlation over the GC colour range. 5.2. Photometric Analysis 105

Figure 5.9: Histogram of (g − i)0 GC colours for NGC 4365, NGC 4342 and the stellar stream. The colour distribution of GCs within 5 arcmin of NGC 4342 is plotted with a dashed line and the colour distribution of GCs that overlap with the stream regions as deﬁned in Figures 5.3 and 5.6 is plotted with a dotted line. The colour distribution of NGC 4365 is plotted with a solid line. It has been shifted from Subaru/SuprimeCam 0 0 (g − i )0 (Blom et al., 2012a) to CFHT/MegaCam (g − i)0 colours and scaled down for comparison. The colour distribution of NGC 4342’s GC system is indistinguishable from the colour distribution of the GC stream and both lack the red GCs seen in the colour distribution of NGC 4365’s GC system. 106 Chapter 5. Tidal interaction between NGC 4365 and NGC 4342

GC colour distribution of the stream population and that of NGC 4342. This supports the theory that the stream GCs are actually GCs from the outer parts of NGC 4342’s GC system that have been stripped oﬀ the galaxy during a tidal interaction with NGC 4365.

5.3 Spectroscopic Analysis

5.3.1 Spectroscopic Data

Three slitmasks were observed on 2012, April 17 with the DEep Imaging Multi-Object Spectrograph (DEIMOS) on the Keck II telescope. The slitmasks were positioned along the stream South West of NGC 4365 and around NGC 4342. GC candidates within each slitmask were chosen for observation based on solely their brightness. The slitmasks were each observed for a total of 2 hours, split over 4 observations of 30 min each. The median seeing was 1.2 arcseconds. For all 3 masks, the slits were 1 arcsecond wide and a 1200 l/mm grating, centred on 7800 A,˚ was used in conjunction with the OG550 filter. This setup (Romanowsky et al., 2009; Arnold et al., 2011; Foster et al., 2011; Strader et al., 2011; Romanowsky et al., 2012) enables observations from ∼ 6550 − 8900 A˚ with a wavelength resolution of ∼ 1.5 A.˚ The Calcium II triplet (CaT) absorption features (8498, 8542 and 8662 A)˚ are within the observed wavelength range up to recession velocities of ∼ 8 200 km s−1. This includes NGC 4342 (V = 751 km s−1, Grogin, Geller & Huchra, 1998) and NGC 4365 (V= 1243 km s−1 Smith et al., 2000). We use these absorption features to determine the line-of-sight velocity for each candidate GC. The data were reduced using a modified version of the deep2 (Deep Extragalactic Evolutionary Probe 2) galaxy survey data reduction pipeline (idl spec2d Cooper et al., 2012; Newman et al., 2012). The pipeline uses dome flats, NeArKrXe arc lamp spectra and sky light visible in each slit to perform flat fielding, wavelength calibration and local sky subtraction, respectively. Newman et al. (2012) state that the deep2 pipeline achieves a root-mean-squared (rms) accuracy of 0.007A,˚ with adequate arclamp exposure and a fifth order Legendre polynomial iterative fit. This rms is determined from the final mean residual about the fit. The pipeline outputs sky subtracted object spectra as well as the sky spectra themselves. Each object spectrum was cross correlated (using the iraf task rv.fxcor) with the CaT features in 13 stellar template spectra observed with the same setup. The velocity of the object is calculated by taking the mean of the 13 cross correlation values and the error on the velocity is calculated by adding in quadrature the standard deviation of the velocities and the mean of the errors on those velocities (as determined by rv.fxcor). 5.3. Spectroscopic Analysis 107

Figure 5.10: Comparison of independent velocity measurements from DEIMOS spectra. Velocities are plotted with error bars on a one-to-one relationship line. All the velocities are consistent with each other except the lowest velocity, at −200km s−1 as identiﬁed by the second measurement. We exclude this object from further analysis and measure an rms dispersion about the one-to-one line of 8.8 km s−1. 108 Chapter 5. Tidal interaction between NGC 4365 and NGC 4342

The process of velocity determination was independently repeated by a collaborator (C. Foster) and results are shown in Figure 5.10. We only include a velocity measurement in further analysis, if it has a reliable velocity (as determined by both investigators) and the measured velocities are consistent within the measured errors (i.e. the 1σ error bar of one measurement overlaps the other velocity measurement). The root-mean-squared dispersion between the two independent measurements is 8.8 km s−1. The rms dispersion between the independent measurements of recession velocity result from the inclusion of a few reliable, but lower signal-to-noise spectra.

5.3.2 Velocity selection

If the GC candidates on the stream and around NGC 4342 originate from the same system, we expect their recession velocities to be consistent. A superposition of NGC 4342 GCs on a population of stream GCs that have been stripped from another, now completely disrupted galaxy, might show an oﬀset in recession velocity.

We ﬁnd 33 objects with measurable recession velocities between -200 and 1600 km s−1 but assume the 11 objects with velocities smaller than 400 km s−1 to be stars in our own Galaxy. The remaining 22 GCs are plotted in Figure 5.11 showing GC recession velocity against galactocentric radius from NGC 4365. The new velocities presented in this work scatter about the recession velocity of NGC 4342 (V = 751 km s−1). The velocities of GCs around NGC 4365 are also shown along with the 2σ dispersion envelopes as a function of galactocentric radius. As highlighted in Blom et al. (2012b), there is a grouping of low velocity GCs that are also consistent with NGC 4342’s recession velocity.

Four of the 22 GCs with measured velocities are more likely to be associated with NGC 4365 than NGC 4342 and the stream. Two have signiﬁcantly higher velocities than NGC 4342 (> 1400 km s−1) and another two have reasonably high velocities (> 1050 km s−1); all are radially consistent with NGC 4365’s GC system. We associate 18 new GC recession velocities with the sample of 13 stream GCs identiﬁed in Blom et al. (2012b) giving a total of 31 GCs associated with NGC 4342 and the stream. The positions, recession velocities and photometric colours of these GCs are listed in Table 5.1. We conclude that GCs associated with NGC 4342 and the stream are indistinguishable in velocity. 5.3. Spectroscopic Analysis 109

Figure 5.11: Phase-space diagram of globular cluster line-of-sight velocity as a function of distance from NGC 4365. New GC velocities presented here are shown as triangles with black outlines and error bars while previously published GC velocities (Blom et al., 2012b) are shown without. The colour of each symbol indicates its (g − i)0 colour. NGC 4365 and NGC 4342 are marked with crosses at 0 arcmin, V = 1243 km s−1 and ∼20 arcmin, V = 751 km s−1 respectively. The solid line shows the 2σ envelopes of the GC velocity distribution and the dashed line indicates the adopted boundary, below which we classify objects as low velocity stream GCs. The 31 stream and NGC 4342 GCs have a similar mean velocity to NGC 4342. The (g − i)0 colours of low velocity GCs are mostly blue with some green GCs but no red GCs. Red GCs are only found in the central regions of NGC 4365. The stream GCs are likely tidally stripped from the original GC system of NGC 4342. 110 Chapter 5. Tidal interaction between NGC 4365 and NGC 4342

I.D. R.A. Decl. V Verr (g − i)0 (degrees) (km s−1) (mag) 1 185.9401 7.031459 1060.96 32.67 0.739 2 185.9213 7.035125 863.96 24.33 0.666 3 185.9598 7.045965 706.69 16.77 0.846 4 185.9558 7.045967 831.14 14.73 0.708 5 185.9771 7.048600 758.41 22.65 0.911 6 185.9753 7.082502 795.31 17.96 0.861 7 185.9512 7.099172 897.06 17.68 0.786 8 186.0573 7.182025 627.83 35.51 0.854 9 185.7894 6.944254 870.57 20.42 0. 674 10 185.8825 6.988903 811.86 34.09 0.953 11 185.8480 7.014668 769.10 41.91 0.741 12 185.9185 7.019106 770.51 19.25 0.773 13 185.8866 7.040393 661.51 32.29 0.727 14 185.9483 7.063918 778.85 25.04 0.814 15 185.7881 7.030769 724.31 29.35 0.696 16 185.8933 7.083244 1008.32 16.11 0.748 17 185.8355 7.081957 460.64 33.42 0.775 18 185.8873 7.147331 664.27 39.24 0.758 19 186.0902 7.243450 796.64 11.37 0.925 20 186.1140 7.328690 538.85 7.39 0.984 21 186.0164 7.152190 519.93 9.97 0.870 22 186.0573 7.182080 624.60 6.26 0.659 23 186.0640 7.224360 734.53 5.15 0.760 24 186.1108 7.315610 698.58 15.57 0.923 25 186.1011 7.325980 712.22 12.39 0.845 26 186.1203 7.300140 720.09 8.61 0.712 27 186.1341 7.368620 682.49 14.93 1.077 28 186.0563 7.251730 758.64 12.42 0.642 29 186.1403 7.175740 668.30 4.22 0.808 30 186.1637 7.145680 797.62 5.46 0.680 31 186.0988 7.293170 673.24 4.55 0.877

Table 5.1: Positions, recession velocities and photometric colours (CFHT/MegaCam) for the 31 GCs associated with NGC 4342 and the stream. GCs numbered 1 to 18 are presented for the ﬁrst time in this work and GCs numbered 19 to 31 are low velocity outliers previously presented in Blom et al. (2012b). 5.4. Kinematic properties of NGC 4342’s GC system 111

5.4 Kinematic properties of NGC 4342’s GC system

Taking the stream GCs as part of the GC system of NGC 4342 we calculate a mean velocity of 742 23 km s−1 and a velocity dispersion of 127 16 km s−1 (31 GCs). If GCs are being stripped off NGC 4342 to form the GC stream we expect the GCs closest to NGC 4365 may not be bound to NGC 4342. Therefore in Figure 5.12 the GCs are divided into three radial bins: the outer bin for GCs that are less than 15 arcmin from the centre of NGC 4365 and the inner two bins for GCs within 3 arcmin and 10 arcmin of NGC 4342 respectively. The 8 GCs within 3 arcmin from NGC 4342 have a mean projected distance of 2.14 arcmin, a mean velocity of 835 49 km s−1 and a velocity dispersion of 139 37 km s−1. The 9 GCs between 3 and 10 arcmin from NGC 4342 have a mean projected distance of 4.2 arcmin, a mean velocity of 75043 km s−1 and a velocity dispersion of 13032 km s−1. The 14 GCs within 15 arcmin from NGC 4365 have a mean projected distance from NGC 4343 of 14.9 arcmin, a mean velocity of 68222 km s−1 and a velocity dispersion of 83 16 km s−1. There is a significant decrease in mean recession velocity (from 835 to 750 to 682 km s−1) with increasing distance from NGC 4342, but it is not possible to disentangle a possible velocity shear along the GC stream from the effect of one-sided selection in velocity space as NGC 4365’s GC system overlaps the GCs along the stream. The outermost velocity dispersion point in Figure 5.12 is also probably artificially decreased because the velocities overlap with NGC 4365’s GCs and the GCs there are unlikely to be bound to NGC 4342. We do not see a significant decrease in the measured velocity dispersion at 4.2 arcmin (3 to 10 arcmin from NGC 4342) compared to that at 2.14 arcmin (within 3 arcmin from NGC 4342). If the mean GC velocity is fixed to the recession velocity of NGC 4342 (751 km s−1) the calculated velocity dispersion increases to 16544 and 11022 km s−1 for the innermost (mean distance at 4.2 arcmin) and outermost bins respectively (mean distance at 14.9 arcmin). If three GCs with 2σ velocities are excluded from the sample, the calculated velocity dispersion for the inner bins decrease to ∼ 75 23 km s−1. In Figure 5.13 the 31 velocity selected NGC 4342/stream GCs are plotted on an image of the stellar stream. The recession velocities of the GCs and nearby galaxies are marked by the colour of the points (i.e. NGC 4341: V = 922 km s−1, NGC 4343: V = 1014 km s−1, IC 3267: V = 1231 km s−1, IC 3259: V = 1406 km s−1). We note that NGC 4342 is the only galaxy in the area that has a low enough recession velocity to be consistent with the recession velocities of the GCs around it and extending along the stream. Given that the GCs in the area are moving with NGC 4342 and are strongly spatially aligned with the stream we conclude that the GC system of NGC 4342 is being stripped through tidal 112 Chapter 5. Tidal interaction between NGC 4365 and NGC 4342

Figure 5.12: Velocity dispersion of globular clusters associated with NGC 4342 and the stream are plotted with ﬁlled points as a function of the distance from the centre of NGC 4342. Errorbars in the x-direction show the radial extent of the bins. The grey region shows the velocity dispersion range as calculated for the entire sample of 31 GCs, i.e. 111 to 143 km s−1. The outermost ﬁlled point shows a velocity dispersion for GCs that are unlikely to be bound to NGC 4342. The open symbols with errorbars show the velocity dispersion calculated when three GC velocities are excluded from the sample. These GCs fall between 2.1 and 2.5 σ from the mean velocity of NGC 4342’s GC system. 5.5. Dark matter constraints 113

Figure 5.13: Imaging of the giant elliptical NGC 4365 and other nearby galaxies in the W 0 group. A very deep image which reveals a ∼ 300 kpc long stream of stars (Bogd´anet al., 2012a). Globular clusters (GCs) in the stream are represented by triangles and galaxies by circles, with the colour of the points denoting their recession velocity (velocities > 1900 km s−1 are assigned the reddest colour). The GCs have a mean velocity consistent with the recession velocity of NGC 4342 (located at R.A. ∼ 185.9◦ and Dec. ∼ 7.05◦) indicating that they were tidally stripped along with some stars from NGC 4342. The other galaxies in the W 0 group have recession velocities intermediate between those of NGC 4342 and NGC 4365. interaction with NGC 4365. This implies that dark matter is being stripped form NGC 4342 as well.

5.5 Dark matter constraints

This Section, including Figure 5.14, was prepared by Dr. Aaron J. Romanowsky, a coauthor on Blom et al. (2013), and is included in this Thesis for continuity of argument. We now consider the main challenge to the tidal stripping scenario raised by Bogd´an et al. (2012a), which is the dark matter (DM) halo of NGC 4342. Because the DM would have much lower initial binding energies on average than the stars, it should be stripped 114 Chapter 5. Tidal interaction between NGC 4365 and NGC 4342

ﬁrst, and severe stripping of the stars should not begin until a large fraction of the DM is stripped away. Thus at ﬁrst glance, the inference from the X-ray gas of a massive DM halo would seem to rule out the tidal scenario. However, there are still two key points to consider: (1) quantifying the degree of possible stripping; (2) more critically examining the X-ray mass inference itself, given the strong disturbances observed in the gas.

The ﬁrst point is especially relevant in the light of a new determination that the tidal stripping may have been at only the ∼ 50 percent level (Forbes et al. 2013) rather than the & 90 percent level adopted by Bogd´anet al. (2012a). There would still be a non- zero amount of DM remaining, and it is not immediately obvious how much to expect. Indeed, simulations of the tidal stripping of satellite galaxies have predicted surprisingly resilient DM halos. In one case, Goerdt et al. (2008) examined small disk galaxies with ΛCDM-like halos within a Virgo cluster-like potential, and concluded that “Even when the entire [baryonic] disk is tidally stripped away, the nucleus stays intact and can remain dark matter dominated even after severe stripping.” Libeskind et al. (2011) simulated the mass stripped from small galaxies (subhalos) located within the larger DM halo of a massive galaxy. They found after a billion years, tidal stripping had indeed removed most of the DM, but 30 percent remained along with 60 percent of the original stellar mass (i.e. 40 percent of the stellar mass was stripped away).

These works suggest that it is plausible for a significant fraction of DM to remain around NGC 4342 even after strong stripping. Investigating the question further would require a simulation tailored to the NGC 4342 and NGC 4365 system, which is beyond the scope of this work. However, we can at least compare the DM profile inferred from the X-rays to a standard ΛCDM halo and see if there is any evidence of deviation from the norm as may be expected after stripping. We therefore construct a simplified mass model for NGC 4342 that consists of a stellar component plus a DM halo with a profile predicted by ΛCDM simulations (Navarro, Frenk & White, 1997). The stellar mass-to- light ratio is estimated based on the B − V = 0.96 colour to be M/LB ' 4.2 (Bell et al. 2003, for a Chabrier 2003 initial mass function), which translates to a stellar mass of 10 M? ' 1.6 × 10 M . From Dutton et al. (2010) we expect a corresponding virial mass 12 of Mvir ' 1.2 × 10 M , and from Prada et al. (2012) we predict a halo concentration of cvir ' 10.7. The predicted circular velocity profile is plotted in Figure 5.14 (“mini halo”); one expects that outside the very central regions, the profile will be fairly constant with a −1 circular velocity of vc ∼ 140–150 km s . We also plot the X-ray result, showing the full range of reported uncertainties from different azimuthal sectors in order to allow 5.5. Dark matter constraints 115

Figure 5.14: Circular velocity profile of NGC 4342. Predictions are shown based on the stellar mass only (dashed curve), a standard dark matter halo for a small galaxy (solid curve), or for a large galaxy (dotted curve). The results based on X-rays are shown as a solid curve with shaded uncertainty region, and are in strong conflict with any reasonable ΛCDM-based halo. This Figure was prepared by Dr. Aaron J. Romanowsky, a co-author on Blom et al. (2013), and is included in this Thesis for continuity of argument. conservatively for systematic uncertainties from the ram pressure effects. The X-ray profile −1 is dramatically different from the ΛCDM-based prediction, implying vc ∼ 260–300 km s , which brings us back to our second question above, about the reliability of the X-ray constraints.

We next consider a much more massive DM halo, corresponding to an ∼ L∗ elliptical 12 galaxy, with Mvir = 6.0 × 10 M and cvir = 9.5. As shown in Figure 5.14 (“big halo”), in −1 this case the vc profile rises to ∼ 200 km s at ∼ 20 kpc, but still conflicts with the X-ray result. A halo that matches the X-ray profile requires an extremely high concentration, 12 e.g., cvir ∼ 40 for Mvir ∼ 3 × 10 M , which would make it an outlier at a ∼ 5σ level from the expected mean ΛCDM relation.

One point to note here is that Bogd´anet al. (2012a) considered the assumption of hy- drostatic equilibrium to be questionable beyond ∼ 5 kpc, and relied on information within only ∼ 10 kpc. By adding some combination of adiabatic halo contraction (Blumenthal 116 Chapter 5. Tidal interaction between NGC 4365 and NGC 4342 et al., 1986) and a higher stellar mass-to-light ratio, the model predictions (for both mini and big halos) would be in better agreement with the X-rays inside ∼ 5 kpc, while also perhaps being more consistent with the central stellar kinematics from (Cretton & van den Bosch, 1999). However, this also implies that in these central regions, the X-rays cannot really discriminate between massive and stripped halos, and that at larger radii, they may be completely unreliable (note that a similar problem was also seen in another galaxy by Romanowsky et al. 2009). Additional information about the halo mass could in principle come from the dynamics of the GCs around NGC 4342. We have explored this possibility using spherical, anisotropic Jeans models, and comparing to the velocity dispersion data. Unfortunately, we found that the results are very sensitive to the inclusion or exclusion of a few velocities.

5.6 Discussion

The properties of the GC system around NGC 4342 indicate that the galaxy is being tidally stripped. We find a distribution of GCs peaked at the position of NGC 4342. Detailed analysis of the GC spatial distribution reveals a statistically significant GC stream overlapping with NGC 4365’s GC system (NE of NGC 4342) and extending a roughly equal distance SW of NGC 4342. The (g − i)0 colour distribution of the GCs close to NGC 4342 is indistinguishable from that of GCs in the stream and significantly different from that of NGC 4365’s GC colour distribution. This suggests that the stream GCs have been drawn from the outskirts of NGC 4342’s GC system. In addition to the similarity of their spatial and colour properties, the recession velocities of stream and NGC 4342 GCs (V = 74223 km s−1) match the recession velocity measured from NGC 4342 starlight (V = 751 km s−1) very well. At first glance the evidence from stellar light also supports this interpretation. Bogdán et al. (2012a) report the presence of a stellar stream extending from the NE corner of NGC 4365 across the galaxy, through NGC 4342 and further SW, which aligns well with the GC stream. As NGC 4342 is a much less luminous galaxy (MB = −18.45 mag) than

NGC 4365 (MB = −21.3 mag) it seems reasonable to assume that NGC 4365 is tidally stripping NGC 4342 as it moves through the W 0 group. Closer investigation of NGC 4342’s properties brings this into question as Bogdánet al. (2012b) find X-ray evidence of hot gas surrounding NGC 4342 and infer a large dark matter halo. If NGC 4342 still has a large dark matter halo it becomes very difficult to explain how a large percentage of stars and GCs have been stripped off the galaxy during a tidal interaction without stripping an even larger fraction of the dark matter (Libeskind 5.6. Discussion 117 et al., 2011). If the majority of the dark matter and GCs as well as 50 percent of the stars have been stripped off NGC 4342 (see Forbes et al. 2013), the original dark matter halo of NGC 4342 must have been very large and that dark matter is unlikely to have escaped the potential well of NGC 4365 at the centre of the W 0 group. If the stars have instead been stripped off NGC 4365 and are in the process of accreting onto NGC 4342 we expect GCs from the outskirts of NGC 4365 to be accreting onto NGC 4342 as well. Given the measured recession velocities of NGC 4365, NGC 4342 and the stream GCs, this interpretation of the stream GCs originally coming from NGC 4365 is highly unlikely.

It is possible that measurements of the X-ray halo do not imply as large a dark matter halo as first thought. Bogdánet al. (2012b) noted that the X-ray emitting hot gas is sharply truncated on the NW edge of NGC 4342 due to ram-pressure as it moves towards NGC 4365 at 1300 km s−1 at an angle of 22◦ with respect to the viewing plane. This hydrodynamic disturbance is likely to affect the dark matter inference beyond 5 kpc from NGC 4342. We showed in Figure 5.14 that inside 5 kpc the X-ray measurement cannot distinguish between a galaxy with a very large dark matter halo or one with a moderate amount or even no dark matter. With a moderate, or very small, dark matter halo there are no reasons why NGC 4342 could not have been partially stripped of its GCs and stars. Further support for the tidal stripping scenario comes from investigation of NGC 4342 in the context of galaxy scaling relations and comparison to well know tidally stripped galaxies (Forbes et al. 2013).

We also consider a third galaxy with the same recession velocity and GC system properties as NGC 4342. If interaction with NGC 4342 has completely disrupted this galaxy and both its stars and GCs are now being accreted onto NGC 4342, there is no need for stars to be stripped from NGC 4342 before its dark matter halo. It is not far- fetched to assume a galaxy with an almost identical GC colour distribution to NGC 4342 as many galaxies have GC systems with similar colour distributions (Peng et al., 2006), but it might be somewhat more problematic to also assume that such a galaxy had exactly the same recession velocity as NGC 4342. However, many of the galaxies around NGC 4342 have similar recession velocities and recent observations have shown that small galaxies accreting along streams or ﬁlaments might not be as uncommon as previously thought (Ibata et al., 2013). Nevertheless, we consider this possibility as less likely than NGC 4342 itself simply being stripped. 118 Chapter 5. Tidal interaction between NGC 4365 and NGC 4342

5.7 Conclusions

We find an overdensity of globular clusters (GCs) centred on the S0 galaxy NGC 4342 and extending into a statistically significant GC stream (at the 97.7 percent level) that is spatially coincident with the stellar stream crossing NGC 4365, as reported by Bogdán et al. (2012a). The photometric colours of the stream GCs match the colours of the GCs around NGC 4342 and their measured recession velocities (V = 742 23 km s−1) match the measured recession velocity of NGC 4342 itself (V = 751 km s−1). The evidence from these observations indicate that stars and GCs from NGC 4342 are being tidally stripped by the larger galaxy NGC 4365, situated at the centre of the W 0 group. This finding apparently contradicts the conclusion that NGC 4342’s stars and GCs have not have been stripped because it hosts a large dark matter halo that has not been stripped (Bogdánet al., 2012b). We found that where the dark matter halo inference is secure (within 5 kpc from NGC 4342 where it is not hydrodynamically disturbed) it cannot differentiate between a large dark matter halo or none at all. Therefore, it is possible that NGC 4342 stars and GCs are being accreted by NGC 4365. After considering a third galaxy, with the same recession velocity and GC colour as NGC 4342, that has been disrupted to form the stream we conclude that NGC 4342 is most likely being tidally stripped to form the stellar and GC stream. 6 Conclusions and future work

Science always doesn’t go forwards. Its a bit like doing a Rubik’s Cube, you sometimes have to make more of a mess with a Rubik’s Cube before you can get it to go right. —Dame Jocelyn Bell Burnell

Each individual Chapter contains a discussion of the results presented therein and highlights the implications for theoretical scenarios of galaxy evolution. Therefore, we will present a summary of the results and conclusions from the individual chapters here, before continuing to describe potential avenues for further research. We conclude this thesis with commentary on how the research presented here contributes to current studies of galaxy evolution.

6.1 Chapter Summary

In Chapter Two we presented a procedure for finding globular clusters (GCs) around the giant elliptical galaxy NGC 4365. We contrasted the techniques of finding GCs with space based imaging (where GCs are spatially resolved but the field of view of each image is small) with ground based imaging (where GCs are point sources, indistinguishable in angular size from stars but the instrument field of view can be much larger). The photometric properties of NGC 4365’s GC system were analysed in Chapter Three of this thesis. One of the main aims of this Chapter was to determine if the previously identified third GC colour subpopulation (Puzia et al., 2002; Brodie et al., 2005; Larsen et al., 2005) is definitively separate from the usual two GC subpopulations. The analysis was done on a combination of one wide-field, three filter Subaru/Suprime-Cam image and eight, two filter Hubble Space Telescope/Advanced Camera for Surveys pointings.

119 120 Chapter 6. Conclusions and future work

We found a signiﬁcant gradient of mean colour with GC magnitude for the blue GC subpopulation and a signiﬁcant gradient in mean colour with galactocentric radius for both blue and red GC subpopulations. Neither the colour-magnitude (blue-tilt) or colour- radius relations are strong enough to explain the overdensity of intermediate colour GCs at small galactocentric radii. Therefore GC system of NGC 4365 was divided into three GC subpopulations: the blue, green (intermediate colour) and red subpopulations. During our investigation of the properties of each subpopulation we found:

• The radial surface density of the blue GC subpopulation is clearly distinct from that of the red subpopulation and the green subpopulation radial surface density is only signiﬁcantly diﬀerent to that of the red subpopulation in the inner 1.2 arcmin (2 data points).

• The median half light radius (mean size) of the blue GC subpopulation is more than 3σ different from the green and read GC subpopulation sizes. The difference in median size between the green and red subpopulation sizes is significant, but only at the 1σ level.

• Each of the three GC subpopulations have signiﬁcantly diﬀerent mass functions (computed from absolute magnitude using a standard mass-to-light ratio) and photometric ellipticity.

• All three GC subpopulations follow a power law galactocentric radius - GC size trend with a slope of 0.49. This slope is most likely due to the galaxy potential well that the GCs reside in. In general, the GCs at small galactocentric radii undergo more tidal forces and consequently lose more of their of outer stars than GCs at large galactocentric radii.

These results lead us to conclude that the third (green) GC subpopulation is most likely physically distinct from both the blue and red GC subpopulations. In light of this conclusion we discussed the possible connection of the green GC subpopulation to the Kine- matically Distinct Core (KDC) (Surma & Bender, 1995; Davies et al., 2001) at the centre of NGC 4365 and found that connection unlikely given the disparity in physical extent, 5 arcsec for the KDC and 2 arcmin for the green subpopulation. We also discussed the impact that the existence of three GC subpopulations would have on the current GC formation scenarios. While the major merger (Zepf & Ashman, 1993), multiphase collapse (Forbes et al., 1997) and accretion (Côtéet al., 1998) scenarios primarily aim to explain the existence of ubiquitous GC bimodality, we find that each of the formation scenarios 6.1. Chapter Summary 121 is able to explain the presence of an intermediate colour (green) subpopulation in a small percentage of large galaxies. To further investigate the properties of each GC subpopulation and distinguish between possible formation scenarios for NGC 4365’s GC system we analysed the velocity of a subsample of the GCs. The data and analysis are presented in Chapter 4. Spectral observations were performed with the Keck II/DEep Imaging Multi-Object Spectrograph (DEIMOS). We found that each of the blue, green and red GC subpopulations rotate in different directions and discovered a distinct population of low velocity GCs around NGC 4365. In summary:

• The green GC subpopulation rotates in the same direction as the stellar component of NGC 4365, along the minor photometric axis (i.e. rolling).

• The red GC subpopulation rotation is only signiﬁcant beyond 2 arcmin for the galaxy centre and there is rotates perpendicular to the stellar component and in the opposite direction to the rotation of the KDC.

• The rotation of the blue GC subpopulation twists in direction by ∼ 90◦ from 3 to 7 arcmin. This is consistent with members of the blue subpopulation having been accreted from several diﬀerent dwarf galaxies.

• The 13 low velocity GCs around NGC 4365 form a spatially linear feature that is consistent with the stellar stream presented in Bogd´anet al. (2012a).

The green subpopulation does not originate from the galaxy that is causing the stellar stream feature, as the low velocity GC are not preferentially green in colour and the rotation of the green subpopulation is perpendicular to the stream. We also discussed the possibility that NGC 4365’s green GCs were accreted from a single, intermediate sized galaxy and found this to be unlikely. The number of green GCs suggests a fairly luminous galaxy (Harris & van den Bergh, 1981; Peng et al., 2008) and the mean colour of the green GCs suggests a much less luminous galaxy (Peng et al., 2006). We conclude that it is most likely that the red GCs were formed in situ, the green GCs are also likely to have formed in situ but earlier in a much more eﬃcient phase of GC formation, and the blue GCs have been accreted from smaller galaxies over the lifetime of NGC 4342. Given the evidence of a stellar stream from Bogd´anet al. (2012a) we investigated the GC system of NGC 4342 and the stream between NGC 4365 and NGC 4342 in Chapter 5. We used publicly available imaging from the MegaCam instrument on the Canada-France- Hawaii Telescope (CFHT) to identify GCs in the NGC 4365, NGC 4342 and stream region. 122 Chapter 6. Conclusions and future work

We found a statistically signiﬁcant overdensity of GCs forming a stream from NGC 4365 to NGC 4342 and beyond. We also obtained Keck/DEIMOS spectroscopy of a subsample of GCs on NGC 4342 and the stream. We added the low velocity GCs around NGC 4365 to the stream sample as the recession velocities are similar and they coincide spatially with the stellar stream. Analysing the photometric and spectroscopic properties of these GCs we found:

• The spatial distribution of GCs shows a density peaks at the positions of NGC 4365 and NGC 4342 as well as an elongated feature extending SW of NGC 4365 and crossing NGC 4342.

• The colour distribution of the GCs around NGC 4342 is indistinguishable from the colour distribution of GCs spatially coincident with the stellar stream. Their colour distributions are signiﬁcantly diﬀerent from that of NGC 4365.

• Measured recession velocities for combined NGC 4342 and stream GCs match the recession velocity of NGC 4342.

From these results we conclude that the GCs in the stream are being tidally stripped from NGC 4342 onto NGC 4365. This result suggests that the stellar stream has been formed through stripping of stars from NGC 4342 and conflicts with the Bogdánet al. (2012a) conclusion that NGC 4342 cannot have undergone significant stellar stripping because it still hosts a large dark matter halo as evidenced by its hot gas halo (Bogdánet al., 2012b). The question of whether NGC 4342 is being tidally stripped could be resolved with very deep spectral analysis of the stellar stream but such observations are prohibitively time consuming.

6.2 Avenues for Further Research

6.2.1 Substructure within galaxy groups and clusters

Recent studies have shown clear evidence of tidal streams around our own Milky Way Galaxy and around the Andromeda galaxy (M31, McConnachie et al., 2009). These studies have used very large stellar surveys, across enormous fields-of-view to identify low surface brightness tidal features using areas of individual star count overdensity. These are important confirmations of the frequent minor mergers and harassment events predicted by ΛCDM cosmology. However such observations are limited to the Local Group. It is possible to find evidence of very low surface brightness tidal features, on the scale 6.2. Avenues for Further Research 123 of galaxy interaction, much further. Mihos et al. (2005) and Bogdánet al. (2012a) have presented images of tidal streams, using low surface brightness integrated stellar light as far as the Virgo Cluster and Tal et al. (2009) found tidal features in galaxies more than 30 Mpc away. However, this work requires a prohibitive amount of observing time (100+ nights on the UK Schmidt telescope on Kitt Peak, and summation of the entire Sloan Dig- ital Sky Survey, respectively) and extremely careful monitoring of observing conditions to avoid introduction of spatially coherent noise into the imaging. In Chapter 5 of this thesis we uncovered a GC overdensity (Blom et al., 2013) overlapping the low surface brightness stellar feature presented in Bogdánet al. (2012a). Figure 6.1 shows the smoothed spatial distribution of GCs compared with the low surface brightness stream feature. Lee et al. (2010) presented a detection of large scale structure in the GC distribution of the Virgo Cluster but did not have the spatial resolution to identify particular tidal overdensities. The GC data was obtained from publicly available, relatively shallow imaging obtained with the MegaCam imager on the Canada-France-Hawaii Telescope (CFHT). Given the relative binding energies for GCs and stars to the galaxy potential we expect GCs to be stripped at least as easily as stars (and possibly more easily than stars) during a merger or accretion event, and therefore expect GC overdensities to follow low surface brightness stellar indicators of ongoing accretion or merging. This expectation seems to be supported by this discovery in the W 0 group and the fact that similar GC overdensities have been detected by Mackey et al. (2010) on the stellar streams observed around M31. The benefit of observing GC overdensities is that the bulk of GCs, at the Virgo Cluster distance, are brighter than i = 24 mag arcsec−2 (i = 24 mag over a seeing disc 1 arcsec2 in area), more than 4 magnitudes brighter in surface brightness than the stellar stream features. Even with shallower imaging, significant numbers of GCs can be identified and overdensities, if present, detected. Many theories of galaxy formation suggest that galaxy groups accelerate the evolution of galaxies, due to the increased frequency of galaxy interactions. This search for intra- group structure is a direct method to evaluate that claim. It is especially important to uncover the interaction processes for early type galaxies, thought to undergo multiple wet and dry mergers during their lifetime.

6.2.2 Searching for the causes of signiﬁcant kinematic misalignment

Most elliptical galaxies rotate with a kinematic position angle very similar to their photometric position angle. As shown in Figure 6.2, the ATLAS3D team (Krajnovi´cet al., 2011) found very few galaxies with extreme (∼ 90◦) kinematic misalignment, and only 2 of those 124 Chapter 6. Conclusions and future work

Figure 6.1: Reproduction of Figures 5.13 and 5.3 for direct comparison. (Top) A very deep image which reveals a ∼ 300 kpc long stream of stars (Bogd´anet al., 2012a). (Bottom) Smoothed GC surface density from CFHT data within the dashed region of the deep image. The GCs reveal a similar spatial distribution to the stellar stream, with concentrations around NGC 4365 and NGC 4342. Refer to Chapter 5 for more details. 6.2. Avenues for Further Research 125

Figure 6.2: Reproduction of Figure 8a from Krajnovi´cet al. (2011). It shows the kinematic misalignment angle (diﬀerence between galaxy photometric and kinematic axes) as a function of the galaxy ellipticity. NGC 4365 (hosting 3 globular cluster subpopulations) and NGC 4406 (M86) are circled, both host a Kinematically Distinct Core and are the dominant galaxy in their respective groups. 126 Chapter 6. Conclusions and future work

(NGC 4365 and NGC 4406) also have a Kinematically Distinct Core feature. Both NGC 4365 and NGC 4406 are the most massive galaxy in their groups, just beyond the Virgo Cluster. The GC system in NGC 4406 has not yet been studied kinematically, neither has a detailed and careful study of its photometric properties been done with deep, wide-field imaging. Blom et al. (2012a,b) found that NGC 4365 hosts a very unusual colour trimodal GC system with distinct kinematic features associated with each GC subpopulation. The question, Does stellar kinematic misalignment correlate with n > 2 GC subpopulations?, is one first step to finding the cause of both of these unusual galaxy properties. Photo- metric analysis of the GC subpopulations in the systems shown to have extreme kinematic misalignment would provide an answer and insight into the assembly history of these mis- aligned galaxies. The answer to another question, Are unusual GC systems or extreme kinematic misalignment present preferentially in galaxy groups?, could provide significant insight into the environments that influence the formation of unusual galaxies. The same group galaxies targeted for wide-field studies, in search of GC overdensities indicating tidal interactions, would provide some answers to this question. The sample of group GC systems could also be compared to the GC systems analysed as part of the SAGES Legacy Unifying Globulars and Galaxies Survey (SLUGGS)1 .

6.3 Broad impact of this work

There are many unanswered questions regarding the formation and assembly history of the giant elliptical galaxy NGC 4365. However, this thesis has definitively shown that NGC 4365 has three distinct GC subpopulations and presented a possible formation scenario. We propose that green GCs formed within the central halo that would become NGC 4365, but unusually early during the GC formation era and that the red GCs also formed in situ as NGC 4365 was forming but slightly later. We determine it most likely that NGC 4365 accreted blue GCs from smaller galaxy halos over its lifetime. This interpretation is in best agreement with the accretion scenario of GC formation (Côtéet al., 1998), borrowing ideas from the multiphase collapse scenario (Forbes et al., 1997) and agrees well with the results of recent cosmological simulations of galaxy and GC system formation (Tonini, 2013). The spatial and mass resolution as well as simulation size required to simulate the formation of GCs and GC systems at the same time as the formation and evolution of their parent galaxies, has evaded galaxy modellers thus far. With continuing improvements in computing facilities it is likely that these technical constraints will soon no longer hamper

1http://sluggs.swin.edu.au 6.3. Broad impact of this work 127 the progress of GC galaxy evolution studies. It is important for the observations of GC systems to also continue to probe deeper into the properties of GC systems. GC systems contain a wealth of information about the chemical environment in the early universe when they were formed, as well as dynamical tracers of the interactions their host galaxies have undergone. With more wide-ﬁeld data the properties of large galaxies and galaxy groups in the nearby universe can be uncovered in greater detail. With deeper data we can probe GC systems further away to improve statistical analysis of GC system properties as well as investigations of nearby GC systems with more accurate photometry and spectroscopy. For example, probing most of the way down the GC luminosity function might allow the intrinsic skewness and kurtosis of GC subpopulations to be determined. This thesis is an early example of the detailed analysis of galaxy evolutionary history that can be done with analysis of its GC system. Future studies can build on the techniques presented here to study a wider range of galaxies and come closer to answering our questions about how galaxies formed and became the beautiful object we see in the sky today.

Bibliography

Alves-Brito A., Hau G. K. T., Forbes D. A., Spitler L. R., Strader J., Brodie J. P., Rhode K. L., 2011, MNRAS, 417, 1823

Arimoto N., Yoshii Y., 1987, A&A, 173, 23

Arnold J. A., Romanowsky A. J., Brodie J. P., Chomiuk L., Spitler L. R., Strader J., Benson A. J., Forbes D. A., 2011, ApJ, 736, L26+

Ashman K. M., Bird C. M., Zepf S. E., 1994, AJ, 108, 2348

Ashman K. M., Conti A., Zepf S. E., 1995, AJ, 110, 1164

Ashman K. M., Zepf S. E., 1992, ApJ, 384, 50

Bailin J., Harris W. E., 2009, ApJ, 695, 1082

Bassino L. P., Faifer F. R., Forte J. C., Dirsch B., Richtler T., Geisler D., Schuberth Y., 2006, A&A, 451, 789

Baugh C. M., Cole S., Frenk C. S., 1996, MNRAS, 283, 1361

Bedin L. R., Piotto G., Anderson J., Cassisi S., King I. R., Momany Y., Carraro G., 2004, ApJ, 605, L125

Bekki K., Beasley M. A., Brodie J. P., Forbes D. A., 2005, MNRAS, 363, 1211

Bell E. F., McIntosh D. H., Katz N., Weinberg M. D., 2003, ApJS, 149, 289

Bender R., Burstein D., Faber S. M., 1992, ApJ, 399, 462

Bernardi M., Alonso M. V., da Costa L. N., Willmer C. N. A., Wegner G., Pellegrini P. S., Rit´eC., Maia M. A. G., 2002, AJ, 123, 2990

Blakeslee J. P., Cantiello M., Peng E. W., 2010, ApJ, 710, 51

Blakeslee J. P., JordánA., Mei S., CôtéP., Ferrarese L., Infante L., Peng E. W., Tonry J. L., West M. J., 2009, ApJ, 694, 556

Blom C., Forbes D. A., Brodie J. P., Foster C., Romanowsky A. J., Spitler L. R., Strader J., 2012b, MNRAS, 426, 1959

Blom C., Spitler L. R., Forbes D. A., 2012a, MNRAS, 420, 37

Blumenthal G. R., Faber S. M., Flores R., Primack J. R., 1986, ApJ, 301, 27

129 130 Bibliography

Blumenthal G. R., Faber S. M., Primack J. R., Rees M. J., 1984, Nature, 311, 517

Bogd´an A.,´ Forman W. R., Kraft R. P., Jones C., Blom C., Randall S. W., Zhang Z., Zhuravleva I., Churazov E., Li Z., Nulsen P. E. J., Vikhlinin A., Schindler S., 2012b, ApJ, 755, 25

Bogd´an A.,´ Forman W. R., Zhuravleva I., Mihos J. C., Kraft R. P., Harding P., Guo Q., Li Z., Churazov E., Vikhlinin A., Nulsen P. E. J., Schindler S., Jones C., 2012a, ApJ, 753, 140

Brodie J. P., Strader J., 2006, ARA&A, 44, 193

Brodie J. P., Strader J., Denicol´oG., Beasley M. A., Cenarro A. J., Larsen S. S., Kuntschner H., Forbes D. A., 2005, AJ, 129, 2643

Burstein D., Davies R. L., Dressler A., Faber S. M., Stone R. P. S., Lynden-Bell D., Terlevich R. J., Wegner G., 1987, ApJS, 64, 601

Cantiello M., Blakeslee J. P., 2007, ApJ, 669, 982

Carlberg R. G., 1984, ApJ, 286, 403

Chabrier G., 2003, PASP, 115, 763

Chen C., Cˆot´eP., West A. A., Peng E. W., Ferrarese L., 2010, ApJS, 191, 1

Chies-Santos A. L., Larsen S. S., Kuntschner H., Anders P., Wehner E. M., Strader J., Brodie J. P., Santos J. F. C., 2011, A&A, 525, A20+

Cole S., Aragon-Salamanca A., Frenk C. S., Navarro J. F., Zepf S. E., 1994, MNRAS, 271, 781

Conroy C., Spergel D. N., 2011, ApJ, 726, 36

Cooper M. C., Newman J. A., Davis M., Finkbeiner D. P., Gerke B. F., 2012, in Astro- physics Source Code Library, record ascl:1203.003 spec2d: DEEP2 DEIMOS Spectral Pipeline. p. 3003

Cˆot´eP., Marzke R. O., West M. J., 1998, ApJ, 501, 554

Cˆot´eP., McLaughlin D. E., Cohen J. G., Blakeslee J. P., 2003, ApJ, 591, 850

Cˆot´eP., McLaughlin D. E., Hanes D. A., Bridges T. J., Geisler D., Merritt D., Hesser J. E., Harris G. L. H., Lee M. G., 2001, ApJ, 559, 828 Bibliography 131

Cretton N., van den Bosch F. C., 1999, ApJ, 514, 704

Daddi E., Renzini A., Pirzkal N., Cimatti A., Malhotra S., Stiavelli M., Xu C., Pasquali A., Rhoads J. E., Brusa M., di Serego Alighieri S., Ferguson H. C., Koekemoer A. M., Moustakas L. A., Panagia N., Windhorst R. A., 2005, ApJ, 626, 680

Davies R. L., Kuntschner H., Emsellem E., Bacon R., Bureau M., Carollo C. M., Copin Y., Miller B. W., Monnet G., Peletier R. F., Verolme E. K., de Zeeuw P. T., 2001, ApJ, 548, L33

Davis M., Efstathiou G., Frenk C. S., White S. D. M., 1985, ApJ, 292, 371 de Vaucouleurs G., de Vaucouleurs A., Corwin Jr. H. G., Buta R. J., Paturel G., Fouqu´e P., 1991, Third Reference Catalogue of Bright Galaxies. Volume I: Explanations and references. Volume II: Data for galaxies between 0h and 12h. Volume III: Data for galaxies between 12h and 24h. de Zeeuw P. T., Bureau M., Emsellem E., Bacon R., Carollo C. M., Copin Y., Davies R. L., Kuntschner H., Miller B. W., Monnet G., Peletier R. F., Verolme E. K., 2002, MNRAS, 329, 513

Decressin T., Charbonnel C., Meynet G., 2007, A&A, 475, 859

Dutton A. A., Conroy C., van den Bosch F. C., Prada F., More S., 2010, MNRAS, 407, 2

Eggen O. J., Lynden-Bell D., Sandage A. R., 1962, ApJ, 136, 748

Emsellem E., Cappellari M., Krajnovi´cD., van de Ven G., Bacon R., Bureau M., Davies R. L., de Zeeuw P. T., Falc´on-BarrosoJ., Kuntschner H., McDermid R., Peletier R. F., Sarzi M., 2007, MNRAS, 379, 401

Ferrarese L., CôtéP., JordánA., Peng E. W., Blakeslee J. P., Piatek S., Mei S., Merritt D., MilosavljevićM., Tonry J. L., West M. J., 2006, ApJS, 164, 334

Forbes D. A., 1996, AJ, 112, 954

Forbes D. A., 2005, ApJ, 635, L137

Forbes D. A., Brodie J. P., Grillmair C. J., 1997, AJ, 113, 1652

Forbes D. A., Forte J. C., 2001, MNRAS, 322, 257

Forbes D. A., Franx M., Illingworth G. D., Carollo C. M., 1996, ApJ, 467, 126 132 Bibliography

Forbes D. A., Ponman T., O’Sullivan E., 2012, MNRAS, 425, 66

Forbes D. A., Spitler L. R., Strader J., Romanowsky A. J., Brodie J. P., Foster C., 2011, MNRAS, 413, 2943

Foster C., Forbes D. A., Proctor R. N., Strader J., Brodie J. P., Spitler L. R., 2010, AJ, 139, 1566

Foster C., Spitler L. R., Romanowsky A. J., Forbes D. A., Pota V., Bekki K., Strader J., Proctor R. N., Arnold J. A., Brodie J. P., 2011, MNRAS, 415, 3393

Geisler D., Lee M. G., Kim E., 1996, AJ, 111, 1529

Goerdt T., Moore B., Kazantzidis S., Kaufmann T., Macci`oA. V., Stadel J., 2008, MN- RAS, 385, 2136

G´omezM., Woodley K. A., 2007, ApJ, 670, L105

Goudfrooij P., Hansen L., Jorgensen H. E., Norgaard-Nielsen H. U., de Jong T., van den Hoek L. B., 1994, A&AS, 104, 179

Graham A. W., Driver S. P., 2005, PASA, 22, 118

Gratton R. G., 1985, A&A, 147, 169

Gratton R. G., Carretta E., Bragaglia A., 2012, A&A Rev., 20, 50

Grogin N. A., Geller M. J., Huchra J. P., 1998, ApJS, 119, 277

Gwyn S. D. J., 2008, PASP, 120, 212

Hanes D. A., 1977, MNRAS, 179, 331

Hanes D. A., Whittaker D. G., 1987, AJ, 94, 906

Harris W. E., 1991, ARA&A, 29, 543

Harris W. E., 2009a, ApJ, 699, 254

Harris W. E., 2009b, ApJ, 703, 939

Harris W. E., 2010, ArXiv e-prints

Harris W. E., Harris G. L. H., Barmby P., McLaughlin D. E., Forbes D. A., 2006, AJ, 132, 2187 Bibliography 133

Harris W. E., van den Bergh S., 1981, AJ, 86, 1627

Hempel M., Kissler-Patig M., 2004, A&A, 428, 459

Hempel M., Kissler-Patig M., Puzia T. H., Hilker M., 2007, A&A, 463, 493

Herschel W., 1786, Royal Society of London Philosophical Transactions Series I, 76, 457

Herschel W., 1789, Royal Society of London Philosophical Transactions Series I, 79, 212

Hesser J. E., Harris W. E., Vandenberg D. A., Allwright J. W. B., Shott P., Stetson P. B., 1987, PASP, 99, 739

Hibbard J. E., Mihos J. C., 1995, AJ, 110, 140

Hoﬀman L., Cox T. J., Dutta S., Hernquist L., 2010, ApJ, 723, 818

Hubble E., 1932, ApJ, 76, 44

Hubble E. P., 1926, ApJ, 64, 321

Ibata R. A., Lewis G. F., Conn A. R., Irwin M. J., McConnachie A. W., Chapman S. C., Collins M. L., Fardal M., Ferguson A. M. N., Ibata N. G., Mackey A. D., Martin N. F., Navarro J., Rich R. M., Valls-Gabaud D., Widrow L. M., 2013, Nature, 493, 62

Jacoby G. H., Branch D., Ciardullo R., Davies R. L., Harris W. E., Pierce M. J., Pritchet C. J., Tonry J. L., Welch D. L., 1992, PASP, 104, 599

Jiang F., van Dokkum P., Bezanson R., Franx M., 2012, ApJ, 749, L10

Joanes D. N., Gill C. A., 1998, Journal of the Royal Statistical Society. Series D (The Statistician), 47, pp. 183

Jord´anA., 2004, ApJ, 613, L117

JordánA., CôtéP., Blakeslee J. P., Ferrarese L., McLaughlin D. E., Mei S., Peng E. W., Tonry J. L., Merritt D., MilosavljevićM., Sarazin C. L., Sivakoff G. R., West M. J., 2005, ApJ, 634, 1002

JordánA., McLaughlin D. E., CôtéP., Ferrarese L., Peng E. W., Mei S., Villegas D., Merritt D., Tonry J. L., West M. J., 2007, ApJS, 171, 101

JordánA., Peng E. W., Blakeslee J. P., CôtéP., Eyheramendy S., Ferrarese L., Mei S., Tonry J. L., West M. J., 2009, ApJS, 180, 54 134 Bibliography

Kauﬀmann G., White S. D. M., Guiderdoni B., 1993, MNRAS, 264, 201

Kennicutt Jr. R. C., 1983, ApJ, 272, 54

Kormendy J., Fisher D. B., Cornell M. E., Bender R., 2009, ApJS, 182, 216

Krajnovi´cD., Emsellem E., Cappellari M., Alatalo K., Blitz L., Bois M., Bournaud F., Bureau M., Davies R. L., Davis T. A., de Zeeuw P. T., Khochfar S., Kuntschner H., Lablanche P.-Y., McDermid R. M., Morganti R., Naab T., Oosterloo T., Sarzi M., Scott N., Serra P., Weijmans A.-M., Young L. M., 2011, MNRAS, 414, 2923

Kundu A., Whitmore B. C., 2001, AJ, 121, 2950

Kundu A., Zepf S. E., Hempel M., Morton D., Ashman K. M., Maccarone T. J., Kissler- Patig M., Puzia T. H., Vesperini E., 2005, ApJ, 634, L41

Larsen S. S., 1999, A&AS, 139, 393

Larsen S. S., Brodie J. P., 2003, ApJ, 593, 340

Larsen S. S., Brodie J. P., Beasley M. A., Forbes D. A., Kissler-Patig M., Kuntschner H., Puzia T. H., 2003, ApJ, 585, 767

Larsen S. S., Brodie J. P., Huchra J. P., Forbes D. A., Grillmair C. J., 2001, AJ, 121, 2974

Larsen S. S., Brodie J. P., Strader J., 2005, A&A, 443, 413

Larson R. B., 1974, MNRAS, 166, 585

Larson R. B., 1990, in Capuzzo-Dolcetta R., Chiosi C., di Fazio A., eds, Physical Pro- cesses in Fragmentation and Star Formation Vol. 162 of Astrophysics and Space Science Library, Formation of star clusters. pp 389–399

Lee M. G., Park H. S., Hwang H. S., 2010, Science, 328, 334

Lee M. G., Park H. S., Hwang H. S., Arimoto N., Tamura N., Onodera M., 2010, ApJ, 709, 1083

Libeskind N. I., Knebe A., Hoﬀman Y., Gottl¨ober S., Yepes G., 2011, MNRAS, 418, 336

Liu C., Peng E. W., JordánA., Ferrarese L., Blakeslee J. P., CôtéP., Mei S., 2011, ApJ, 728, 116

Mackey A. D., Huxor A. P., Ferguson A. M. N., Irwin M. J., Tanvir N. R., McConnachie A. W., Ibata R. A., Chapman S. C., Lewis G. F., 2010, ApJ, 717, L11 Bibliography 135

Mar´ın-Franch A., Aparicio A., Piotto G., Rosenberg A., Chaboyer B., Sarajedini A., Siegel M., Anderson J., Bedin L. R., Dotter A., Hempel M., King I., Majewski S., Milone A. P., Paust N., Reid I. N., 2009, ApJ, 694, 1498

Masters K. L., JordánA., CôtéP., Ferrarese L., Blakeslee J. P., Infante L., Peng E. W., Mei S., West M. J., 2010, ApJ, 715, 1419

McConnachie A. W., Irwin M. J., Ibata R. A., Dubinski J., Widrow L. M., Martin N. F., CôtéP., Dotter A. L., Navarro J. F., Ferguson A. M. N., Puzia T. H., Lewis G. F., Babul A., Barmby P., BienayméO., Chapman S. C., Cockcroft R., Collins M. L. M., Fardal M. A., Harris W. E., Huxor A., Mackey A. D., PeñarrubiaJ., Rich R. M., Richer H. B., Siebert A., Tanvir N., Valls-Gabaud D., Venn K. A., 2009, Nature, 461, 66

McLaughlin D. E., Harris W. E., Hanes D. A., 1994, ApJ, 422, 486

Mieske S., JordánA., CôtéP., Kissler-Patig M., Peng E. W., Ferrarese L., Blakeslee J. P., Mei S., Merritt D., Tonry J. L., West M. J., 2006, ApJ, 653, 193

Mihos J. C., Harding P., Feldmeier J., Morrison H., 2005, ApJ, 631, L41

Milone A. P., Bedin L. R., Piotto G., Anderson J., King I. R., Sarajedini A., Dotter A., Chaboyer B., Mar´ın-Franch A., Majewski S., Aparicio A., Hempel M., Paust N. E. Q., Reid I. N., Rosenberg A., Siegel M., 2008, ApJ, 673, 241

Montuori M., Di Matteo P., Lehnert M. D., Combes F., Semelin B., 2010, A&A, 518, A56+

Muratov A. L., Gnedin O. Y., 2010, ApJ, 718, 1266

Naab T., Johansson P. H., Ostriker J. P., 2009, ApJ, 699, L178

Navarro J. F., Frenk C. S., White S. D. M., 1997, ApJ, 490, 493

Newman J. A., Cooper M. C., Davis M., Faber S. M., Coil A. L., Guhathakurta P., Koo D. C., Phillips A. C., Conroy C., Dutton A. A., Finkbeiner D. P., Gerke B. F., Rosario D. J., Weiner B. J., Willmer C. N. A., Yan R., Harker J. J., Kassin S. A., Konidaris N. P., Lai K., Madgwick D. S., Noeske K. G., Wirth G. D., Connolly A. J., Kaiser N., Kirby E. N., Lemaux B. C., Lin L., Lotz J. M., Luppino G. A., Marinoni C., Matthews D. J., Metevier A., Schiavon R. P., 2012, ArXiv e-prints 136 Bibliography

Ouchi M., Shimasaku K., Okamura S., Furusawa H., Kashikawa N., Ota K., Doi M., Hamabe M., Kimura M., Komiyama Y., Miyazaki M., Miyazaki S., Nakata F., Sekiguchi M., Yagi M., Yasuda N., 2004, ApJ, 611, 660

Owers M. S., Randall S. W., Nulsen P. E. J., Couch W. J., David L. P., Kempner J. C., 2011, ApJ, 728, 27

Peng E. W., JordánA., CôtéP., Blakeslee J. P., Ferrarese L., Mei S., West M. J., Merritt D., MilosavljevićM., Tonry J. L., 2006, ApJ, 639, 95

Peng E. W., JordánA., CôtéP., Takamiya M., West M. J., Blakeslee J. P., Chen C., Ferrarese L., Mei S., Tonry J. L., West A. A., 2008, ApJ, 681, 197

Piotto G., Bedin L. R., Anderson J., King I. R., Cassisi S., Milone A. P., Villanova S., Pietrinferni A., Renzini A., 2007, ApJ, 661, L53

Pota V., Forbes D. A., Romanowsky A. J., Brodie J. P., Spitler L. R., Strader J., Foster C., Arnold J. A., Benson A., Blom C., Hargis J. R., Rhode K. L., Usher C., 2013, MNRAS, 428, 389

Prada F., Klypin A. A., Cuesta A. J., Betancort-Rijo J. E., Primack J., 2012, MNRAS, 423, 3018

Proctor R. N., Forbes D. A., Romanowsky A. J., Brodie J. P., Strader J., Spolaor M., Mendel J. T., Spitler L., 2009, MNRAS, 398, 91

Puzia T. H., Zepf S. E., Kissler-Patig M., Hilker M., Minniti D., Goudfrooij P., 2002, A&A, 391, 453

Rhode K. L., Zepf S. E., 2004, AJ, 127, 302

Romanowsky A. J., Strader J., Brodie J. P., Mihos J. C., Spitler L. R., Forbes D. A., Foster C., Arnold J. A., 2012, ApJ, 748, 29

Romanowsky A. J., Strader J., Spitler L. R., Johnson R., Brodie J. P., Forbes D. A., Ponman T., 2009, AJ, 137, 4956

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

Schweizer F., Seitzer P., 1998, AJ, 116, 2206

Shapley H., 1918, PASP, 30, 42 Bibliography 137

Silverman B. W., 1986, Density estimation for statistics and data analysis

Smith R. J., Lucey J. R., Hudson M. J., Schlegel D. J., Davies R. L., 2000, MNRAS, 313, 469

Spitler L. R., 2010, MNRAS, 406, 1125

Spitler L. R., Forbes D. A., Strader J., Brodie J. P., Gallagher J. S., 2008, MNRAS, 385, 361

Spitler L. R., Larsen S. S., Strader J., Brodie J. P., Forbes D. A., Beasley M. A., 2006a, AJ, 132, 1593

Spitler L. R., Larsen S. S., Strader J., Brodie J. P., Forbes D. A., Beasley M. A., 2006b, AJ, 132, 1593

Strader J., Beasley M. A., Brodie J. P., 2007, AJ, 133, 2015

Strader J., Brodie J. P., Cenarro A. J., Beasley M. A., Forbes D. A., 2005, AJ, 130, 1315

Strader J., Brodie J. P., Spitler L., Beasley M. A., 2006, AJ, 132, 2333

Strader J., Romanowsky A. J., Brodie J. P., Spitler L. R., Beasley M. A., Arnold J. A., Tamura N., Sharples R. M., Arimoto N., 2011, ApJS, 197, 33

Strader J., Smith G. H., 2008, AJ, 136, 1828

Surma P., Bender R., 1995, A&A, 298, 405

Tal T., van Dokkum P. G., Nelan J., Bezanson R., 2009, AJ, 138, 1417

Tonini C., 2013, ApJ, 762, 39

Usher C., Forbes D. A., Brodie J. P., Foster C., Spitler L. R., Arnold J. A., Romanowsky A. J., Strader J., Pota V., 2012, MNRAS, 426, 1475

Valcarce A. A. R., Catelan M., 2011, A&A, 533, A120 van den Bergh S., Morbey C., Pazder J., 1991, ApJ, 375, 594 van den Bosch R. C. E., van de Ven G., Verolme E. K., Cappellari M., de Zeeuw P. T., 2008, MNRAS, 385, 647 van Dokkum P. G., Franx M., Kriek M., Holden B., Illingworth G. D., Magee D., Bouwens R., Marchesini D., Quadri R., Rudnick G., Taylor E. N., Toft S., 2008, ApJ, 677, L5 138 Bibliography

West M. J., CôtéP., Marzke R. O., JordánA., 2004, Nature, 427, 31

Woodley K. A., Harris W. E., Puzia T. H., G´omez M., Harris G. L. H., Geisler D., 2010, ApJ, 708, 1335

Yagi M., Kashikawa N., Sekiguchi M., Doi M., Yasuda N., Shimasaku K., Okamura S., 2002, AJ, 123, 66

Yoon S., Yi S. K., Lee Y., 2006, Science, 311, 1129

Zepf S. E., Ashman K. M., 1993, MNRAS, 264, 611 A Statistical tests for trimodality

A.1 Kaye’s Mixture Model algorithm

The most common tool to determine the statistical significance of GC bimodality is the Kaye’s Mixture Model (KMM) code (Ashman et al., 1994), designed to calculate the ‘p’ statistic for the significance of a homoscedastic bimodal Gaussian distribution being preferred over a unimodal Gaussian distribution. KMM is also capable of determining the significance of heteroscedastic multimodal distributions being preferred over a unimodal Gaussian distribution but the ‘p’ statistic in these cases is an approximation. Applied to our GC colour distribution for NGC 4365 both bimodal and trimodal distributions result in a ‘p’ statistic less than 10−4 (a probability larger than 99.99 per cent that the multimodal distributions are preferred over the unimodal distribution). The traditional KMM code is however not capable of determining the probability that a trimodal distribution is preferred over a bimodal distribution. Using an extension to the traditional KMM code, written by Dr Matthew Owers (Ow- ers et al., 2011), we were able to assess a probability for three Gaussian modes over two Gaussian modes bearing in mind that the code was untested for application to GC colour distributions. We attempted to test the code at the same time as using it for our analysis by including checks on how the code performed with varying GC input numbers, analysis of other galaxies’ GC distributions and redundant calculations for the probability of four modes over three modes and bimodality over unimodality. See results in Table A.1. Comparing results for NGC 4365 with NGC 1407 (data from Spitler et al. 2011 in prep. also used in Romanowsky et al. 2009; Foster et al. 2010; Forbes et al. 2011) the extended KMM test finds that both are trimodal (NGC 4365 with a probability of 99.7 per cent and NGC 1407 with a probability larger than 99.9 per cent) but there is little evidence in the literature for NGC 1407 being trimodal. When the number of GCs is halved for

139 140 Appendix A. Statistical tests for trimodality

NGC 4365 (data from Peng et al., 2006) the probability of trimodality decreases below 95 per cent (to 91.8 per cent), even though the smaller sample is more central and we expect trimodality to be more prominent there. It is probable that in the case of NGC 1407 the preference for trimodality is due to signiﬁcant skewness in the subpopulation distributions (caused by strong radial gradients in subpopulation mean colour, see Forbes et al. 2011). A.1. Kaye’s Mixture Model algorithm 141 n 76 45 61 43 642 298 302 437 696 415 417 584 203 348 523 204 595 272 147 333 dal modes m σ 0.1 0.1 0.05 0.09 0.08 0.04 0.12 0.06 0.08 0.12 0.14 0.11 0.15 0.07 0.19 0.17 0.10 0.09 0.08 0.13 µ Quadrumo 0.78 0.96 1.01 1.15 0.77 0.91 0.98 1.14 0.88 1.11 1.19 1.36 0.91 0.92 1.08 1.34 0.83 1.00 1.29 1.49 23 6 0 i - - p over 3) 0.252 0.596 0.915 (4 n colours 51 600 567 512 616 562 426 550 488 620 120 192 593 255 489 z + 1 modes are preferred over − colours where m n ) are shown and between each modality g σ 0 i 0.04 0.09 0.09 0.04 0.11 0.06 0.08 0.12 0.12 0.07 0.17 0.17 0.11 0.09 0.16 − rimodal 0 T g µ 0.78 0.96 1.13 0.76 0.93 1.13 0.88 1.12 1.13 0.92 0.98 1.33 0.85 1.00 1.41 p over 2) 0.003 0.000 0.003 0.082 0.328 (3 n 642 784 820 605 318 587 320 475 1037 1053 dal σ 0.05 0.12 0.06 0.09 0.08 0.15 0.13 0.17 0.13 0.15 ) and number of objects assigned ( σ Bimo µ 0.79 1.05 0.79 1.06 0.89 1.25 0.94 1.33 0.98 1.43 µ ), width ( galaxies with HST/ACS photometry and p over 1) 0.000 0.000 0.000 0.000 0.000 (2 n galaxies with Subaru/S-Cam photometry and 905 795 1679 1604 1658 Comparing dal σ 0.16 0.16 0.22 0.24 0.26 Unimo µ Comparing 0.95 0.92 1.12 1.19 1.24 Results from the extended KMM test for testing the signiﬁcance with which 24 24 24 4365 1407 4365 4365 point. 6 6 6 t. point. t. point. M60 z z z Galaxy all cen cen NGC NGC NGC NGC able A.1: the ‘ p ’ statistic is shown. in colour. For each mode the mean value ( T 142 Appendix A. Statistical tests for trimodality

VCC number 1226 1316 731 1535 Other name M49 M87 NGC 4365 NGC 4526 M49 0.79 0.86 1.6 × 10−7 1.4 × 10−3 M87 - 0.45 4.6 × 10−9 6.0 × 10−4 NGC 4365 - - - 9.2 × 10−3

Table A.2: KS test results comparing galaxies in the ACS Virgo Cluster Survey. The ‘p’ statistic shown is an inverse measure of the probability that GCs in the different galaxies are drawn from different underlying colour distributions. Only M49 and M87 are inconsistent with being drawn from different colour distributions and hence could be said to have the same underlying colour distribution. Other galaxy comparisons show a probability greater than 99 per cent of being drawn from different underlying colour distributions.

A.2 The Kolmogorov-Smirnov test

The Kolmogorov-Smirnov (KS) test shows clearly that most giant elliptical galaxies have GC colour distributions that are significantly different from each other (subpopulation peak colours and widths vary between galaxies, see Table A.2) (using GC catalogues from Peng et al., 2006). A sample of galaxies with reasonably large GC numbers were chosen to compare to each other and most galaxies show a probability of larger than 99 per cent of being drawn from a colour distribution different to any other galaxy in the sample. Two very large galaxies (M87 and M49) are consistent with being drawn from the same underlying colour distribution. It is clear that each galaxy’s GC system is formed independently of any other GC system but that similar galaxy properties can result in similar GC systems. We also used this test to compare colour distributions simulated from bimodal and trimodal descriptions of the data (as output by KMM) with the actual GC colour distribution and found that the observed GC distribution is significantly different from both simulated distributions. Table A.3 shows that for all comparisons the KS test shows probabilities of

Simulated Actual colour distributions distributions ACS S-Cam Bimodal 10−95 10−256 Trimodal 10−99 10−262

Table A.3: KS test results comparing simulated bimodal and trimodal distributions with NGC 4365’s colour distribution. The ‘p’ statistic shown is an inverse measure of the probability that the simulated distributions are drawn from different colour distributions to the actual ones. A.3. Chi-Squared minimization 143 almost 100 per cent that the simulated and actual distributions are significantly different. Neither of the simulated distributions is a good enough description of the observed distribution, likely because they cannot take into account the skewness and peakiness (kurtosis) of the observed distributions. The colour modes in GC systems are not perfectly Gaussian and therefore this method of analysis cannot distinguish between bimodal and trimodal Gaussian descriptions of the colour distribution to determine which is more likely.

A.3 Chi-Squared minimization

Comparing the bimodal and trimodal results for a reduced Chi-Squared (χ2) minimization is dependent on binning the data in colour. We have chosen to do this binning via an Epanechnikov kernel smoothing of the data rather than an histogram. With different colour bin sizes the statistical results will also be different. We have chosen bin sizes for the Epanechnikov kernel that are equal to the colour errors in the data. Kernel smoothed data is overpotted with bimodal and trimodal distributions in Figures A.1 and A.2. Using the entire radial range of our data we can achieve the best number statistics but the results are not conclusive. For the ACS GCs, where we have better colour resolution, the best-fitting trimodal Gaussian distribution has a slightly smaller reduced χ2 value than the best-fitting bimodal distribution but the best fit third mode is very close to blue modes and contains many fewer objects. This ‘third mode’ is not best described as a third GC subpopulation but as an indication that the two modes of the underlying GC distribution are not purely Gaussian but probably Gaussians with both skewness and kurtosis. We do see a significantly smaller reduced χ2 value and a central third mode with a large proportion of the GCs in the S-Cam catalogue. 144 Appendix A. Statistical tests for trimodality

Figure A.1: Results of bimodal and trimodal ﬁts to the Epanechnikov kernel smoothing of the HST/ACS GC colour distribution. The distribution is plotted in grey, showing poissonian errors, and the individual Gaussians as well as the summation of all three are overplotted for both the bimodal (dashed lines) and trimodal (dotted lines) ﬁts. The reduced χ2 value is 1.40 for the bimodal case and 1.26 for the trimodal case.

Figure A.2: Results of bimodal and trimodal ﬁts to the Epanechnikov kernel smoothing of the Subaru/S-Cam GC colour distribution. The distribution is plotted in grey, showing poissonian errors, and the individual Gaussians as well as the summation of all three are overplotted for both the bimodal (dashed lines) and trimodal (dotted lines) ﬁts. The reduced χ2 value is 2.57 for the bimodal case and 1.07 for the trimodal case. B Surface density data

B.1 Surface density data

The GC surface density data are recorded in Table B.1.

145 146 Appendix B. Surface density data

rgal Total err rgal Blue errB Red errR rgal Green errG (arcmin−2) (arcmin−2) (arcmin−2) (arcmin−2) GC candidate radial surface density obtained from HST/ACS 0.20 119 30 0.62 12.5 2.3 38.0 4.1 0.63 24.6 3.3 0.32 125 20 1.2 10.2 1.5 20.0 2.1 1.2 9.9 1.5 0.50 81 11 1.8 8.5 1.1 14.5 1.5 2.1 3.51 0.46 0.75 66.5 6.8 2.4 5.48 0.74 7.57 0.86 3.4 1.92 0.27 1.10 43.4 4.0 3.0 5.27 0.69 5.06 0.67 1.50 31.6 2.9 3.8 4.03 0.52 6.22 0.64 1.90 26.6 2.4 5.0 3.47 0.40 2.44 0.33 2.30 14.9 1.6 2.75 15.9 1.4 3.15 13.6 1.5 GC candidate radial surface density obtained from Subaru/S-Cam 1.10 43.4 4.0 1.2 11.4 1.6 21.8 2.2 1.5 5.86 0.73 1.50 32.0 2.9 1.8 9.0 1.2 12.3 1.4 2.7 2.25 0.32 1.90 24.9 2.3 2.4 6.39 0.79 8.21 0.90 4.2 1.05 0.16 2.30 16.4 1.7 3.0 5.23 0.69 5.85 0.73 5.8 0.74 0.11 2.75 14.9 1.3 3.8 4.06 0.45 4.79 0.49 8.0 0.399 0.056 3.25 11.8 1.1 4.5 3.88 0.40 2.94 0.35 10.5 0.318 0.044 3.75 10.70 0.95 5.4 3.34 0.36 1.93 0.28 13 0.328 0.038 4.25 8.54 0.80 6.2 2.52 0.25 1.85 0.22 16 0.174 0.024 4.75 7.57 0.71 7.4 1.77 0.17 1.18 0.14 5.25 5.88 0.60 8.6 1.48 0.15 0.90 0.12 5.75 5.70 0.56 9.9 1.32 0.13 0.760 0.099 6.30 5.05 0.46 11.1 1.08 0.11 0.594 0.082 6.90 3.77 0.38 12.4 0.645 0.081 0.717 0.086 7.50 3.36 0.34 13.8 0.653 0.071 0.530 0.064 8.20 2.47 0.24 15.2 0.582 0.067 0.431 0.058 8.90 3.10 0.30 17.0 0.438 0.051 0.434 0.051 9.60 2.71 0.24 20.5 0.362 0.046 0.502 0.054 10.5 1.83 0.17 11.5 2.10 0.17 12.5 1.74 0.15 13.5 1.67 0.14 15.0 1.49 0.10 17.2 1.274 0.087 20.2 1.033 0.093

Table B.1: GC radial surface density data for the total GC system and each subpopulation. We show surface density with diﬀerent radial binning (columns typeset in bold) for the total GC system, the blue and red subpopulations and the green subpopulation. C Additional supopulation kinematic ﬁts

C.1 Total subpopulation kinematic ﬁts

The colour peaks and widths of the blue, green and red subpopulations cause significant overlap in the colour ranges of the subpopulations. We explore the effect of splitting the populations in different ways to determine the best way to minimize the contamination between samples of the subpopulations and maximize the number of GCs in each subpopulation sample. The first subpopulation split definition is based on the probability that each GC is assigned of being in either the blue, green or red subpopulations by Blom et al. (2012a). They assigned each GC a probability of belonging to the blue, green and red subpopulations (PBlue + PGreen + Pred = 1) based on its colour and galactocentric radius. We define medium probability samples with P > 0.80 to represent the three subpopulations and find 53 blue, 53 green and 65 red GCs with line-of-sight radial velocity measurements. These samples are mostly free of contamination from the other subpopulations but have a relatively small number of GCs to be analysed for each subpopulation. The next subpopulation split definitions cut the GCs into three subsamples at different g0 − i0 colours. For the first colour cut split definition we split the sample at g0 − i0 = 0.9 and g0 −i0 = 1.1 where we see dips in number of the photometric sample colour distribution (see Fig. 3.9). This subpopulation split definition divides the kinematic sample roughly equally between the three subpopulations. The next two colour cut split definitions widen the green sample definition first to redder colours (colour cut 2) and secondly to bluer colours (colour cut 3). The repeated fits for the blue (in colour cut 1 and colour cut 2) and red samples (in colour cut 1 and colour cut 3) show the level to which results are affected by the randomisation in the bootstrap technique of determining uncertainties. We also show the kinematic fits for both cases of the kinematic axis ratio. In the first

147 148 Appendix C. Additional supopulation kinematic ﬁts case qkin is ﬁxed to the photometric value for each GC subpopulation (qkin = qphotom from

Blom et al., 2012a) and in the second the axis ratio is fixed to unity (qkin = 1). We do not find any significant differences in the fitted kinematics between the two axis ratio cases and conclude that this kinematic fitting method is not very sensitive to the kinematic axis ratio. In all split definitions for all subpopulations, the kinematic fits for the two cases are consistent within 1σ uncertainty and the computed Λ/ndf values are equal. All further kinematic fits are done in the case where qkin = 1 as this is the more general case. C.1. Total subpopulation kinematic fits 149 ), rot = 0 . 97 23 24 66 51 62 53 51 44 51 40 23 21 49 48 57 44 . Values ) kin ◦ red ( q A P 175 174 167 167 195 167 167 194 Colour P (79) (79) (56) ) 15 26 14 23 14 22 12 18 13 19 15 25 12 19 12 20 = 0 . 70, − 1 s 1 . 1 . 05 1 . 05 0 . 8 (65) σ > > > > 218 218 216 216 202 216 216 0 202 green 0 0 i (km i i q ed − R − − 0 ) P 0 0 g . 56, g g − 1 48 34 56 13 52 11 45 16 45 15 50 32 44 19 45 15 ot s r = 0 V 39 39 43 43 95 43 43 94 (km blue 27 23 21 15 33 31 25 16 17 14 28 24 16 14 31 25 ) kin ◦ ( A . Probability is denoted with P 141 145 142 126 124 134 134 ◦ 144 ) 1 . 05 (96) 1 . 05 (77) 1 . (100) 10 15 12 20 13 22 9 17 9 13 9 14 10 14 10 15 − 1 or 222 < 0 . 8 (53) s < < 0 ◦ σ 0 0 i i i > 228 229 217 221 222 216 217 − 217 − − (km 0 0 0 een g g g Gr ) < < < P 57 41 − 1 39 24 57 29 43 33 43 29 40 26 41 30 36 23 ot s r 0 . 9 0 . 9 0 . 85 V 83 96 81 67 83 67 81 103 (km ) for each subpopulation in all the split definitions. Results are shown kin 28 34 24 18 38 54 12 11 40 50 17 12 43 55 27 13 ) kin ◦ ( A P 243 254 251 267 251 267 248 269 (61) (80) (80) ) 10 16 10 17 10 17 10 15 9 15 9 14 8 15 9 16 − 1 s 0 . 9 0 . 9 0 . 8 (53) 0 . 85 σ < < > < 178 179 188 189 188 189 178 0 0 178 0 i i (km i lue − − B − 0 0 ) 0 P g g g − 1 44 27 52 23 48 18 41 22 41 18 43 21 46 20 47 25 ot s ) is fixed to the photometric values for each subpopulation ( q r V 58 49 56 40 56 40 70 67 phot (km q = in 1 1 1 1 kin k phot phot phot phot ) and kinematic position angle (PA q q q q q Defn Results for kinematic fits to various divisions of the three GC subpopulations. We tabulate the rotation velocity ( V . 75) and where it is fixed to 1. Numbers of GCs in each subpopulation representative sample are given in brackets after prob. cut 1 cut 2 cut 3 Split = 0 col. col. col. Definition gal med. q able C.1: velocity dispersion ( σ where the axis ratioand ( q the definition parameters. Thefor major the axis best of colour the cuts NGC are 4365 highlighted stellar in light bold. is 42 T 150 Appendix C. Additional supopulation kinematic fits

Split Blue Green Red Deﬁnition medium probability 12.1 12.6 12.3 colour cut 1 11.9 11.8 12.2 colour cut 2 11.9 15.3 12.2 colour cut 3 12.0 19.5 12.2

Table C.2: Minimisation parameter divided by degrees of freedom (Λ/ndf) for kinematic ﬁts to various divisions of the three GC subpopulations. Values for the best colour cuts are highlighted in bold.

The best kinematic fits are found for the blue GCs in colour cut 3, green GCs in colour cut 1 and red GCs in colour cut 2. We determine this best kinematic fit to be the subpopulation sample has the most constrained position angle (for q=1), see Table C.1, and lowest value of Λ/ndf, see Table C.2. The blue subpopulation representative sample in colour cut 3 Λ/ndf is marginally larger than in colour cut 1 or 2 but the position angle range is significantly smaller (62◦) than in either colour cut 1 or 2 (90◦ and 98◦ respectively). The green subpopulation representative sample position angle range in colour cut 1 (50◦) is only slightly smaller than that of colour cut 2 (52◦) but the Λ/ndf is significantly smaller. The red subpopulation representative sample Λ/ndf is the same for each colour cut but the position angle range is significantly smaller in colour cut 2 (47◦) ◦ ◦ than either of the other colour cuts (91 and 101 ) where the PAkin is not constrained to < 90◦.

C.2 Radial kinematic ﬁts with ﬁxed position angle

Here we compare the radial kinematic fits for the three subpopulations in the case where the PA is fixed to the value fitted for the whole subpopulation (Fig. C.1) to the case where the PA is allowed to vary with radius (Fig. 4.9). The fitted values for σ do not change at all when the position angle of rotation is fixed. The uncertainties in Vrot, are larger when

PAkin is fixed than when the PAkin is allowed to vary. If the reduction in number of free parameters, in the case where PAkin is fixed, does allow a more stable fit to the Vrot and σ we would expect the uncertainties (68 per cent confidence intervals) to be smaller there.

Since the opposite is true it is more likely that the fit is better where the PAkin is allowed to vary, and since the PAkin is well constrained at almost all radii in that case it is also likely that the Vrot is also more reliable than the case where PAkin is fixed. The one exception is for red GCs less than 2 arcmin from the galactic centre. The varying PAkin is not well constrained there and rotation parameter (Vrot/σ) fails the C.2. Radial kinematic fits with fixed position angle 151

Figure C.1: Kinematics as a function of radius for NGC 4365’s three GC subpopulations. GC subpopulations are deﬁned using the best colour cuts highlighted in Table C.1 and the PAkin is ﬁxed to the values found there. Symbols are as in Fig. 4.9.

criterion in Equation 4.5 (Strader et al., 2011). In the case where PAkin is fixed Vrot is correctly fitted to be consistent with zero. The decrease in Vrot at certain other points in Fig. C.1 (e.g. ∼ 3 arcmin for the blue subpopulation) is due to the offset between the value the PAkin was set at and the PAkin at which the system is actually rotating at that radius.