LONG TERM OPTICAL AND X-RAY MONITORING OF BLACK HOLE

CANDIDATES IN EXTRAGALACTIC GLOBULAR CLUSTERS

By

Kristen C. Dage

A DISSERTATION

Submitted to

Michigan State University

in partial fulfillment of the requirements

for the degree of

Astrophysics and Astronomy−Doctor of Philosophy

2020

ABSTRACT

LONG TERM OPTICAL AND X-RAY MONITORING OF

BLACK HOLE CANDIDATES IN EXTRAGALACTIC

GLOBULAR CLUSTERS

By

Kristen C. Dage

The question of whether globular clusters (GCs) host black holes (BHs) has been a long-

standing question in astronomy. Because of the dynamics of the crowded globular cluster

environment, they should be extremely efficient at forming black hole-black hole binaries

(BBHs), such as those detected merging by LIGO. LIGO is detecting many more (and more

massive!) BBHs than was ever expected, and a first step to understanding the formation

of these binaries is to understand BHs in GCs. However, very few ways exist to study BHs

in GCs observationally, and there is only one effective method to study BH candidates in

extragalactic GCs: by studying ultraluminous X-ray sources (ULXs) that are associated with

GCs. Because their X-ray luminosities far exceed the Eddington limit for a stellar mass BH,

ULXs in GCs are good BH candidates. However, they are very rare, with only a few per

galaxy (with only 10 currently known sources among 3 galaxies).

In this thesis, we study a total of 9 ULX sources that are associated with globular clusters

in both X-ray and optical. One of these, RZ2109, shows extreme variability across more than

15 years of monitoring in X-ray, with changes of almost an order of magnitude on the order

of days. Typically any variability in flux is expected to be correlated with a change in the

spectral properties of the source (i.e. some physical change in the system), however, the

shape of the X-ray spectrum of RZ2109 remains the same across all observations. X-ray and

optical studies of all 9 GC ULXs indicate a potential correlation between the X-ray spectrum

and presence of optical emission beyond the cluster continuum–one of the few X-ray-optical

correlations known for astronomical objects. Lastly, we present long-term monitoring in

optical of RZ2109 shows a broad [OIII] emission line which declines long-term over almost

ten years, and then re-brightens.

“Life is a space voyage, and not a short one, either, but one that’ll last for billions of

years.” (Kurt Vonnegut)

In loving memory of John Howkins and Gary Dage

iv

ACKNOWLEDGMENTS

I want to thank my advisor, Steve Zepf, who is the bee’s knees, and my wonderful collab-

orators: Arash Bahramian, Arunav Kundu, Mark Peacock, and Tom Maccarone. Thank

you for sharing this journey with me, and for being wonderful researchers and friends. I’ve

enjoyed all of our discussions and the different perspectives you bring.

I want to thank the people who inspired me to become a better scientist and researcher,

especially (Dr.!) Lara Backer, for being an inspiration, to my grandfather John Howkins for

always asking really annoying questions and making me want to ask questions too, to my

dad, Gary, for being the embodiment of an innovator, inventor, and engineer and encouraging

my curiosity in the world. To my mom, Enid, for making math fun (even though we swore

up and down we didn’t really like it that much).

I want to thank the people who taught me: especially Dave Stoddard and Art Wiggins

for being the first people to introduce me to the joy of physics. Thanks to Frances Smith,

Ruth Favro, Mike LaChance, Jacqueline Vansant, Thaddeus Radzilowski, Dorothy McLear,

and Rick Simek, as well as Jim Hetrick, Nick Licata, Don Bord, Jeff Prentis, Ed Brown and

Leslie Reynolds. And special thanks to Will Clarkson for giving me my first chance to do

astronomy, and to Jay Strader and Laura Chomiuk for welcoming me into their herd.

Lastly, I want to thank everyone who has been there for me, Rufus, Nadia Khalyleh,

Jamilah Alhashidi, Kit, Gerald, Beth and Lara Backer, Dylan Heberle, Joshua Torres, Amal,

Yasmeen and Amna Khalyleh, and all the new friends I’ve made along the way: Sara Ayoub,

Jessica Maldonado, Philipp Grete & Petja Ivanova, Laura Shishkovsky, Teresa Panurach,

Megan Davis, Shannon Schmoll, Chelsea Harris, Huei Sears, Chris Britt, Benoit Cote, Ryan

Urquhart, Erica Thygesen, Carl Fields, Brandon Barker, and especially Kim Crosslan.

v

TABLE OF CONTENTS

LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii

LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii

KEY TO ABBREVIATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv

Chapter 1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1 Overview of Accretion Processes in X-Ray Binaries
. . . . . . . . . . . . . .
1.2 Ultraluminous X-Ray Sources . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.1 The Eddington Limit . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.2 The Ultraluminous State . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.3 Neutron Star ULXs . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Globular Clusters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3.1 Black Holes in Globular Clusters? . . . . . . . . . . . . . . . . . . . .
1.3.2 ULXs in Globular Clusters . . . . . . . . . . . . . . . . . . . . . . . .
1.3.3 RZ2109 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4
Implications of This Work . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.5 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Chapter 2 X-ray variability from the ultraluminous black hole candidate
X-ray binary in the globular cluster RZ 2109 . . . . . . . . . . .
2.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Data and analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2.2.1 Observations

3.1
3.2 Data and Analysis

Chapter 3 X-Ray Spectral Variability of Ultraluminous X-Ray Sources in
Extragalactic Globular Clusters . . . . . . . . . . . . . . . . . . .
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.1 Globular Cluster ULX Sample . . . . . . . . . . . . . . . . . . . . . .
Spectral Fitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.2
3.2.2.1 NGC 4472 GC ULXs . . . . . . . . . . . . . . . . . . . . . .
3.2.2.2 NGC 4649 GC ULXs . . . . . . . . . . . . . . . . . . . . . .
3.2.2.3 NGC 1399 GC ULXs . . . . . . . . . . . . . . . . . . . . . .
3.2.3 Long and Short-term X-ray Variability . . . . . . . . . . . . . . . . .
Long-Term Variability . . . . . . . . . . . . . . . . . . . . .
Short-term (inter-observational) variability . . . . . . . . . .
3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.2.3.1
3.2.3.2

1
2
4
4
5
6
8
9
10
11
12
13

15
15
19
20
25
29

33
33
37
37
38
40
43
49
54
54
57
60

vi

3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

64

Chapter 4

4.3 X-Ray Data and Analysis

Slow Decline and Rise of the Broad [OIII] Emission Line in
Globular Cluster Black Hole Candidate RZ2109 . . . . . . . . .
4.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Optical Data and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.1 Gemini . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.2
SOAR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.3 Equivalent Width Measurement . . . . . . . . . . . . . . . . . . . . .
4.2.4 Normalisation of the Spectra . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.1 Chandra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.2
Swift/X-Ray Telescope . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4 Results: Time Behaviour of [OIII] Emission . . . . . . . . . . . . . . . . . .
4.4.1 Time Evolution of the [OIII] emission . . . . . . . . . . . . . . . . . .
4.4.2 Time Evolution of the Core and Red and Blue Wings of the Emission
Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4.3 X-ray Behaviour Versus [OIII] . . . . . . . . . . . . . . . . . . . . . .
4.4.4 Rise in [OIII] Emission . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5 Comparison to Milky Way GC Sources . . . . . . . . . . . . . . . . . . . . .
4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Chapter 5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1 Summary and Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

70
70
73
73
74
74
76
76
77
78
79
79

81
85
86
87
88

91
91
93

BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

96

vii

LIST OF TABLES

Chandra and background flare filtered XMM-Newton observations,
background subtracted average source count rates and raw source
counts in 0.3-10 keV.
. . . . . . . . . . . . . . . . . . . . . . . . . .

19

Chandra Fit Parameters for xspec best fit model tbabs*(diskbb+pegpwrlw).
Hydrogen column density (NH ) frozen to 1.6 ×1020 cm−2.
Chandra Fluxes (0.5-8 keV) for xspec best fit model tbabs*(diskbb+pegpwrlw).

. . . . .

23

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

23

XMM-Newton Fit Parameters for xspec model tbabs*(diskbb+pegpwrlw).
Hydrogen column density (NH ) frozen to 1.6 ×1020 cm−2.
24
Chandra Fluxes (0.5-8 keV) for xspec best fit model tbabs*(diskbb+pegpwrlw).

. . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

24

Table 2.1:

Table 2.2:

Table 2.3:

Table 2.4:

Table 2.5:

Table 2.6:

All luminosities in 0.2-10 keV.

. . . . . . . . . . . . . . . . . . . . .

31

Table 3.1:

Coordinates and optical properties of the GC ULX sample. Optical
properties from (Zepf et al., 2007, RZ2109) , (Maccarone et al., 2003,
GCU1), Peacock et al. (2014, GCU2, GCU3, GCU4), (Strader et al.,
2012b, GCU5, GCU6) and (Paolillo et al., 2011, GCU7, GCU8) All
magnitudes are in z-band unless otherwise noted. Distances from
Macri et al. 1999; Blakeslee et al. 2009, 2001. RZ2109 and GCU1’s
V magnitudes were converted to z by using the relation V = g -
0.39(g − z) + 0.07 (Peacock et al., 2010), and metallicities converted
to g− z from B-R with the following relationship: g− z = 1.305(B-R)
. . . . . . . . . . . . . . . . . . . . .
- 0.543 (Peacock et al., 2010)..

Table 3.2:

Observations of NGC 4472, with raw source counts (0.5-8.0 keV) for
GCU1, GCU2, GCU3 and GCU4. Observations marked with * were
too off axis to measure counts from. Observations marked with - had
the source off the chip.
. . . . . . . . . . . . . . . . . . . . . . . . .

39

40

viii

Table 3.3:

Table 3.4:

Table 3.5:

Table 3.6:

Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best fit
single-component model tbabs*diskbb for GCU1 and GCU2 in NGC
4472. Hydrogen column density (NH ) frozen to 1.6 ×1020 cm−2). Fit
parameters marked with ∗∗ either encountered an xspec error when
computing a lower bound, or had a lower bound consistent with zero,
and are presented as an upper limit. Lower count observations fit
with C-stat have their statistics presented in parentheses. All fluxes
shown are unabsorbed.
. . . . . . . . . . . . . . . . . . . . . . . . .

Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best fit
single-component model tbabs*diskbb for GCU3 and GCU4 in NGC
4472. Hydrogen column density (NH ) frozen to 1.6 ×1020 cm−2). Fit
parameters marked with ∗∗ either encountered an xspec error when
computing a lower bound, or had a lower bound consistent with zero,
and are presented as an upper limit. Lower count observations fit
with C-stat have their statistics presented in parentheses. All fluxes
shown are unabsorbed.
. . . . . . . . . . . . . . . . . . . . . . . . .

Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best
fit single-component model tbabs*pegpwrlw for GCU1 and GCU2
in NGC 4472. Hydrogen column density (NH ) frozen to 1.6 ×1020
cm−2). Fit parameters marked with ∗∗ either encountered an xspec
error when computing a lower bound, or had a lower bound consistent
with zero, and are presented as an upper limit. Lower count obser-
vations fit with C-stat have their statistics presented in parentheses.
. . . . . . . . . . . . . . . . . . .
All fluxes shown are unabsorbed.

Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best
fit single-component model tbabs*pegpwrlw for GCU3 and GCU4
in NGC 4472. Hydrogen column density (NH ) frozen to 1.6 ×1020
cm−2). Fit parameters marked with ∗∗ either encountered an xspec
error when computing a lower bound, or had a lower bound consistent
with zero, and are presented as an upper limit. Lower count obser-
vations fit with C-stat have their statistics presented in parentheses.
All fluxes shown are unabsorbed. . . . . . . . . . . . . . . . . . . . .

44

45

46

47

Table 3.7:

F-test probability values for single component versus two component
models for GC ULXs in NGC 4472 with over 100 source counts. We
compare statistics between tbabs*(diskbb+pegpwrlw) and tbabs*diskbb
only in columns titled “Disk”, and tbabs*(diskbb+pegpwrlw) with
tbabs*pegpwrlw only in columns titled “PL”. Blank table entries
are where the source had fewer than 100 counts in a given observation. 47

Table 3.8:

Chandra Observations of NGC 4649, with raw source counts (0.5-8.0
keV) for GCU5 and GCU6. . . . . . . . . . . . . . . . . . . . . . . .

48

ix

Table 3.9:

Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best fit
single-component model tbabs*diskbb of of GC ULXs in NGC 4649.
(Hydrogen column density (NH ) frozen to 2.0 ×1020 cm−2). Fit
parameters marked with ∗∗ either encountered an xspec error when
computing a lower bound, or had a lower bound consistent with zero,
and are presented as an upper limit. Lower count observations fit
with C-stat have their statistics presented in parentheses. All fluxes
shown are unabsorbed.
. . . . . . . . . . . . . . . . . . . . . . . . .

Table 3.10: Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best fit
single-component model tbabs*pegpwrlw of of GC ULXs in NGC
4649. (Hydrogen column density (NH ) frozen to 2.0 ×1020 cm−2).
Fit parameters marked with ∗∗ either encountered an xspec error
when computing a lower bound, or had a lower bound consistent with
zero, and are presented as an upper limit. Lower count observations
fit with C-stat have their statistics presented in parentheses. All
fluxes shown are unabsorbed.
. . . . . . . . . . . . . . . . . . . . .

50

51

Table 3.11: F-test probability values for single component versus two component
models for GC ULXs in NGC 4649 with over 100 source counts. We
compare statistics between tbabs*(diskbb+pegpwrlw) and tbabs*diskbb
only in columns titled ”Disk”, and tbabs*(diskbb+pegpwrlw) with
. . . . . . . . . . . .
tbabs*pegpwrlw only in columns titled ”PL”.

51

Table 3.12: Chandra Fit Parameters and Fluxes (0.5-8.0 keV) for spectral best
fit values of GCU6 in NGC 4649 for tbabs*tbabs*pegpwrlw where
NH was not consistent with zero. All fluxes shown are unabsorbed.

Table 3.13: χ2/d.o.f. for tbabs*tbabs*diskbb,where the best fit NH was consis-
tent with zero. (Hydrogen column density (NH ) frozen to 2.2 ×1020
cm−2.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Table 3.14: Chandra Observations of NGC 1399, with raw source counts (0.5-8.0
keV) for GCU7 and GCU8. . . . . . . . . . . . . . . . . . . . . . . .

Table 3.15: Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best fit
single-component model tbabs*diskbb of GC ULXs in NGC 1399.
Hydrogen column density (NH ) frozen to 1.34 ×1020 cm−2). Fit
parameters marked with ∗∗ either encountered an xspec error when
computing a lower bound, or had a lower bound consistent with zero,
and are presented as an upper limit. Lower count observations fit
with C-stat have their statistics presented in parentheses. All fluxes
. . . . . . . . . . . . . . . . . . . . . . . . .
shown are unabsorbed.

52

52

52

55

x

Table 3.16: Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best fit
single-component model tbabs*pegpwrlw of GC ULXs in NGC 1399.
Hydrogen column density (NH ) frozen to 1.34 ×1020 cm−2). Fit
parameters marked with ∗∗ either encountered an xspec error when
computing a lower bound, or had a lower bound consistent with zero,
and are presented as an upper limit. Lower count observations fit
with C-stat have their statistics presented in parentheses. All fluxes
shown are unabsorbed.
. . . . . . . . . . . . . . . . . . . . . . . . .

56

Table 3.17: F-test probability values for single component versus two component
models for GC ULXs in NGC 1399 with over 100 source counts. We
compare statistics between tbabs*(diskbb+pegpwrlw) and tbabs*diskbb
only in columns titled “Disk”, and tbabs*(diskbb+pegpwrlw) with
tbabs*pegpwrlw only in columns titled “PL”.
. . . . . . . . . . . .

57

Table 3.18: χ2 values comparing a model of constant luminosity to the luminosity
. . . . . . . . . . . . . .

of the GC ULXs in this sample over time.

Table 3.19: Best fit slopes and uncertainties of sources that varied in kT or LX .

57

64

Table 4.1:

[OIII] Equivalent Width (Angstrom).

. . . . . . . . . . . . . . . . .

76

Table 4.2:

X-ray luminosities from 0.2-10 keV measured by Swift/XRT.

. . . .

78

xi

LIST OF FIGURES

Figure 1.1: Artist’s representation of an X-ray binary. Because the compact ob-
ject (in this case a black hole) is so much more massive than the
companion star, it accretes material from the star, which forms an
accretion disk around the central object. Image from Rob Hynes. . .

Figure 1.2: Credit: Optical image of M82, with the X-ray point sources overlaid
on the galaxy, and a cut out of the observed X-ray pulses connected
to M82-X2. ESA/XMM-Newton; NASA/Chandra and SDSS . . . .

Figure 1.3: Optical image of globular cluster 47 Tucanae. Image credit: NASA,
ESA, and the Hubble Heritage (STScI/AURA)-ESA/Hubble Collab-
oration Acknowledgment: J. Mack (STScI) and G. Piotto (University
of Padova, Italy).
. . . . . . . . . . . . . . . . . . . . . . . . . . . .

Figure 2.1: Top panel shows best fit powerlaw index for Chandra and XMM ob-
servations, middle panel shows disk temperature, lowest panel shows
calculated unabsorbed model flux. Chandra data is represented by
blue squares, and the XMM-Newton data by orange circles. All pa-
. . . . . . . . . . . .
rameters and fluxes are in the 0.5-8 keV band.

Figure 2.2:

Luminosity variability of RZ2109 in energy band 0.2-10.0 keV. Chan-
dra data is represented by blue squares, and the XMM-Newton data
by orange circles.
. . . . . . . . . . . . . . . . . . . . . . . . . . . .

Figure 2.3:

Left: Fitted spectrum with residuals of Chandra ObsID 321 (2000-
06-12). Right: Fitted spectrum with residuals of XMM-Newton ob-
servation 0112550601 (2002-06-05, pn only).
. . . . . . . . . . . . .

3

7

9

29

30

30

Figure 2.4: Fitted spectrum with residuals of XMM-Newton observation 0200130101

(2004-01-01, pn only).

. . . . . . . . . . . . . . . . . . . . . . . . .

32

Figure 2.5:

Left: Fitted spectrum with residuals of Chandra ObsID 12889 (2011-
02-14). Right: Fitted spectrum with residuals of Chandra ObsID
12888 (2011-02-21).
. . . . . . . . . . . . . . . . . . . . . . . . . . .

Figure 2.6:

Left: Fitted spectrum with residuals of XMM-Newton observation
0761630101 (2016-01-05, pn only). Right: Fitted spectrum with resid-
uals of XMM-Newton observation 0761630301 (2016-01-09, pn only).
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

32

32

xii

Figure 3.1: X-ray image of NGC 4472 (ObsID 12888, filtered to 0.5-8.0 keV) with
. . . . . . . . . . . . . . . . . . .

regions for GCU1-GCU4 overlaid.

Figure 3.2: X-ray image of NGC 4649 (ObsID 12976, filtered to 0.5-8.0 keV) with
. . . . . . . . . . . . . . . . .

regions for GCU5 & GCU6 overlaid.

Figure 3.3: X-ray image of NGC 1399 (ObsID 319, filtered to 0.5-8.0 keV) with
. . . . . . . . . . . . . . . . .

regions for GCU7 & GCU8 overlaid.

Figure 3.4: Upper: LX vs. time for NGC 4472 GC ULXs with mean luminosity
(data from Tables 3.3, 3.4,3.5, 3.6). Middle: LX vs. time for NGC
4649 GC ULXs with mean luminosity (data from Tables 3.9,3.10).
Lower: LX vs. time for NGC 1399 GC ULXs with mean luminosity
(data from Tables 3.15,3.16 ).
. . . . . . . . . . . . . . . . . . . . .

Figure 3.5: Upper: Light curve for GCU1 from ObsID 12889. Middle: Light
curve for GCU1 from ObsID 12888. Both are binned by 5 ksec and 2
ksec. Lower: Lomb-Scargle periodogram of ObsIDs 12888 &12889 of
GCU1 binned by 5 ksec compared to the Lomb-Scargle periodograms
of the background, as well as red noise and white noise.
. . . . . . .

Figure 3.6:

Spectra and fitted models (fit residuals on lower panel) of GCU7 and
GCU8. GCU7 is best fit by a disk model, while GCU8 is much harder
and better fit by a power law model.
. . . . . . . . . . . . . . . . .

Figure 3.7: Upper: Plot of optical cluster colour (g − z) versus best fit spectral
parameter for GC ULXs. Lower: Plot of absolute z-band magnitude
versus best fit spectral parameter for GC ULXs. Neither absolute
magnitude nor colour appear to be correlated with best fit spectral
parameter in this sample. However, it is worth noting how red in
colour GCU8 is. (Optical cluster values from Table 3.1) . . . . . . .

Figure 3.8:

linmix best fits of LX vs kT for RZ2109, GCU7 (Slopes consistent
with zero) and GCU1 (slope inconsistent with zero).
. . . . . . . .

41

49

53

58

59

61

62

65

Figure 3.9:

Γ vs. Luminosity (0.5-8.0 keV) for GC ULXs best fit by tbabs*pegpwrlw.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

66

Figure 3.10: Upper:Tin vs. Luminosity (0.5-8.0 keV)for GC ULXs best fit by
tbabs*diskbb. Data for RZ2109 is taken from Dage et al. (2018).
Lower: Histogram of best fit kT for all sources. . . . . . . . . . . . .

67

xiii

Figure 4.1: Un-normalised Gemini spectrum from 2009. The purple shaded re-
gions show one example of the 25˚A wide regions sampled out of larger
regions (blue) that are averaged to approximate the continuum val-
ues at the start and end of the emission region. The orange shaded
area shows the emission region we are measuring over from 4964˚A to
5058˚A . The blue line is the continuum fit in the region of interest.
.

Figure 4.2: Normalised spectra from observations spanning 2007-2018. Observa-
tions are from the W.M. Keck Observatory (December 2007, [dark
blue]) Gemini South (March 2009, [yellow], May 2011, [light blue]),
and SOAR (March 2016, [light pink], March 2018, [light purple], April
. . . . . . . . . . . . . . . . . . . . . . . . . . .
2019 [dark purple]).

Figure 4.3: Upper panel: Change in X-ray luminosity over time. Purple star is
ROSAT, Blue squares are Chandra observations, orange circles are
taken from XMM-Newton observations, yellow triangles are upper
limits from Swift/XRT. See Maccarone et al. (2010b), Dage et al.
(2018), and Section 4.3 of this paper for X-ray analysis. Lower panel:
Change in L(5007) over time, L(5007) equivalent widths are presented
in Table 4.1.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Figure 4.4:

Intervals of the [OIII]λ5007 emission line components, 5006-5026˚A
for the [OIII] blue wing, 5026-5037˚A for the [OIII] core, and 5037-
5057˚A for the [OIII] red wing.
. . . . . . . . . . . . . . . . . . . . .

Figure 4.5: Flux in each regime of the emission line over the course of the obser-
vations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

75

77

80

83

84

Figure 4.6: Flux in each regime of the λ4959, λ5007 emission cores.

. . . . . . .

84

Figure 4.7: Measurements from early April 2019, and late April 2019 verifying
. . . .

the increase in the [OIII] emission on two different gratings.

86

Figure 4.8:

Spectra from 2016-2019 showing decline and subsequent increase.

.

87

xiv

KEY TO ABBREVIATIONS

• BBH - Black Hole Black Hole Binary
• BH - Black Hole
• GC - Globular Cluster
• IMBH - Intermediate Mass Black Hole
• LIGO - Laser Interferometer Gravitational-Wave Observatory
• LMXB - Low Mass X-Ray Binary
• NS - Neutron Star
• TDE - Tidal Disruption Event
• UCXB - Ultracompact X-ray Binary
• ULX - Ultraluminous X-Ray Source

xv

Chapter 1

Introduction

One of the long-standing questions in astronomy is “Do globular clusters host black holes?”

This question has been under debate since 1969 (Spitzer, 1969). As some of the densest

and oldest places in the universe, globular clusters (GCs) can probe unique condition for

black hole (BH) formation and the creation of exotic binaries. Currently, the only method

to probe the number of black holes in extragalactic globular clusters (see Section 1.3.1 for

more detail) is by searching for ultraluminous X-ray sources (ULXs) which are coincident

with the globular clusters of other galaxies.

Ultraluminous X-ray sources are an exotic class of objects which are much brighter in the

X-ray than what is physically expected for a stellar mass compact object, as they typically

exceed the Eddington limit, which predicts luminosity limits based on the object’s mass (See

Section 1.2.1 for more detail). The observed luminosities are produced by accretion onto a

compact object, although the exact physical processes behind the high luminosities are still

not well understood.

Younger populations of ULXs have been well studied in the star-forming regions of spiral

galaxies, which may be due to accretion onto highly magnetized neutron stars (NSs) or BHs.

However, ULXs also exist in older populations: globular clusters associated with elliptical

galaxies. Studying this class of ULXs in GCs has implications both for properties of ULXs

as a whole, as well as for understanding the question regarding the number of black holes in

globular clusters. Because these sources are so rare, understanding their accretion processes

1

and signatures can provide clues to their formation histories, which in turn may one day

explain why these sources are so rare. Therefore, a brief overview of accretion processes in

X-ray binaries is required before delving deeper into the subject.

1.1 Overview of Accretion Processes in X-Ray Binaries

X-ray binaries are a class of binary systems which emit X-ray radiation. The binary contains

two objects: a compact object (i.e. a neutron star or black hole), which feeds off of the

companion star (the “donor”) and accretes material off the star. The angular momentum

in the accreted plasma is conserved and thus forms an accretion disk around the compact

object1.

Figure 1.1 shows a classic representation of such a system. If the donor star gets close

enough to the compact object, it will fill its Roche lobe2 and become pear shaped, with the

material exiting through the L1 Lagrangian point and forming the accretion disk around the

compact object.

Viscous stress that is thought to originate in magentorotational instability (Balbus &

Hawley, 1998) causes matter to lose angular momentum and fall in towards the surface of

the compact object. This releases gravitational potential energy, ( GM ˙M

R ) 3. The compact

object implies that the radius will be very small, which implies the disk temperature is high,

thereby producing X-rays. Because the compact object is so massive, the emission is highly

energetic and luminous.

1Some high-mass X-ray binaries do not always have accretion disks, they can be fed solely by winds

emanating from a massive donor star, which do not produce a stable accretion disk.

2Roche lobe: a teardrop shaped region that represents the smallest equipotential surface that can exist

around a binary system. See http://astronomy.swin.edu.au/cosmos/R/Roche-lobe for more.

3Where G is the gravitational constant, R is the distance between the two objects, M is the mass of the

accretor, and ˙M is the mass transfer rate.

2

Figure 1.1: Artist’s representation of an X-ray binary. Because the compact object (in this
case a black hole) is so much more massive than the companion star, it accretes material
from the star, which forms an accretion disk around the central object. Image from Rob
Hynes.

The observed X-ray emission from X-ray binaries can have several separate sources within

the accretion disk. The accretion disk itself emits in X-rays at lower photon energies (Shakura

& Sunyaev, 1973; Hanawa, 1989). This emission is thermal, multi-color blackbody emission

(i.e. blackbody models with multiple temperatures superimposed) and influenced by the

geometry of the disk. Non-thermal X-ray emission can also be observed in higher X-ray

energies, and is produced by inverse Comptonization (Titarchuk, 1994).

X-ray reprocessing can also occur in the accretion disk. The emitted X-rays can interact

with clumpy material in the disk, or hit the outer edges (which for some geometries will puff

up with respect to the inner disk). The X-rays will be absorbed, and then re-emitted at a

lower energy (UV or optical).

3

1.2 Ultraluminous X-Ray Sources

ULXs are defined as non-nuclear point sources (i.e. not the active galactic nucleus in the

given galaxy) which have X-ray luminosities exceeding the Eddington limit for a 10 M(cid:12)
black hole, with values of 1039 erg/s or above. Because most X-ray binaries do not typically

reach these high luminosities, ULX sources are likely to be indicators of either higher mass

black holes, or a lower mass compact object (either neutron star or lower mass black hole)

accreting in a non-standard physical regime.

1.2.1 The Eddington Limit

The Eddington Limit was derived by Sir Arthur Eddington as a way to determine the max-

imum luminosity produced by a radiating body under certain conditions4. This maximum

luminosity occurs when the gravitational force is equal to the radiation pressure:

Prad < Fgrav.

(1.1)

We have implicitly assumed that the radiation is spherically symmetric, and that the central

mass is exerting force on a gas comprised of free protons and electrons. The radiation

pressure is ΦE/c, where ΦE is the energy flux transferred to a free electron:

σT L
4πR2 . Here, L
is the source luminosity, R is the radius, and σT represents the cross-section for Thomson

scattering. Thus, we find:

σT L
4cπR2 ≤

GM mp

R2

.

(1.2)

4See http://www-ppl.s.chiba-u.jp/lecture/radiation/node2.html for more details on the deriva-

tion.

4

Upon simplifying the equation and evaluating the constants, we find the Eddington limit:

LEdd = 1.2 × 1038 M
M(cid:12)

erg/s.

(1.3)

Since the Eddington limit depends on the mass of the central object, a luminosity of 1039

erg/s must be generated in a very physically different way depending on the mass of the ac-

cretor. For a neutron star, which are typically less massive than 2 M(cid:12), this luminosity is well
above its Eddington limit. In this case, either the source is accreting in a physically different

regime than other NS XRBs, or the geometry of the system does not fit the assumptions

used to derive the Eddington limit (see Section 1.2.3).

For black holes around 10M(cid:12), luminosities on this order of magnitudes are around or
only slightly above their Eddington limits. By contrast, intermediate mass black holes (a

few hundred to thousands of stellar masses, i.e. smaller than the super massive BHs that

reside at the center of galaxies but larger than anything produced by the deaths of single

stars) would be accreting in the sub-Eddington regime, while producing luminosities near or

above 1039 erg/s because of their very high mass. By studying the observational signatures

of ULXs, we can begin to disentangle the nature of the compact object.

1.2.2 The Ultraluminous State

For X-ray binaries accreting at normal rates (i.e. sub-Eddington), they are typically classified

into two states, the so-called high/soft state, or the low/hard state. The first state happens

when the thermal component (disk) dominates; the source is brighter, but has many more

lower (softer) photons. In the low/hard state, the X-ray spectrum is mainly a steep power-

law component, while the luminosity is relatively lower with a smaller contribution from

5

the disk. For quiescent black holes, and ones in the low/hard state, the power output is

dominated by the associated radio jet (Gallo et al., 2003).

However, work by Gladstone et al. 2009 showed that ULXs belonged to an entirely

different accretion regime. Gladstone systematically fit a number of the highest quality

spectra of ULXs in star-forming regions, and found that the spectra could be well modeled

by a disk plus a Comptonized corona. One physical interpretation of this, assuming the

accretor is a lower mass BH or NS is that a super-Eddington accretion flow is present along

with the accretion disk.

Follow-up work by Sutton et al. 2013 determined that ULXs fell into three classes: soft

ultraluminous and hard ultraluminous regimes, and a broadened disk model. These classes

are based on which model the sources are best fit by, i.e. a disk model, a power-law model,

or a combination of both. The subsequent discovery of pulsating ULXs by Bachetti et al.

2013 makes the physical interpretation of these classes of models difficult, as the compact

object could have a wide range of masses, and the spectral shape does not differentiate a

neutron star from an intermediate mass black hole.

1.2.3 Neutron Star ULXs

Given their low mass, neutron stars were thought to be unlikely accretors in ULX binaries.

However, all of that changed when pulsations were detected in the ULX source M82 X-2

(Bachetti et al., 2014). Because coherent pulsations require a surface, a black hole accretor

is out of the question, as they are by nature without any type of surface and unable to

produce pulsations. Since then, pulsations have been observed in other systems: NGC 5907

ULX, NGC 7793 P13, NGC 300 ULX-1 (Greene et al., 2019, and references therein). Another

NS ULX has been identified via a cyclotron resonance feature (Brightman et al., 2018).

6

Figure 1.2: Credit: Optical image of M82, with the X-ray point sources overlaid on the
galaxy, and a cut out of the observed X-ray pulses connected to M82-X2. ESA/XMM-
Newton; NASA/Chandra and SDSS

The physical mechanisms to explain why the lower mass neutron star accretors can reach

luminosities on the order of 1040 erg/s are currently under debate. Geometric beaming is one

possibility (King et al., 2017), but is not a satisfactory explanation for the high luminosity

origins of all ULX sources; for example, sources with nebular optical emission lines have a

relatively low inferred beaming factor (Binder et al., 2018). However, the presence of a high

magnetic field could also provide another source of pressure to draw extra material off the

donor star and overcome radiation pressure (Brightman et al., 2018).

In either case, it is worth noting that all of the detected NS ULXs, and indeed the

well studied population of ULXs, are found in star forming regions of spiral galaxies and

seem to have massive and hydrogen rich companions. NS ULXs, then, not only seem to be

associated with a high magnetic field to drive their high accretion rates, but also a massive

donor star. Both of these phenomena go hand-in-hand with a young environment, as massive

7

donor stars have short lifespans, and high magnetic fields decay over timescales of millions

of years. This motivates the question as to whether these ULXs in star-forming regions of

galaxies are powered by the same physical mechanisms as the ULXs with lower mass, metal

poor donors which reside in globular clusters associated in elliptical galaxies.

1.3 Globular Clusters

Globular clusters are made up of ten thousands to millions of stars packed in a dense environ-

ment of tens to hundreds of parsecs across (e.g. Fig 1.3). Because of this, the environment is

very dynamic−the cross section for interaction is high and binaries can form easily; in fact,
X-ray binaries tend to be over-represented in globular clusters (Clark, 1975; Pooley et al.,

2003). Clusters represent an older population of stars, and can produce many black holes

as byproducts of the deaths of the initially present shorter-lived stars. This makes globular

clusters an attractive environment for forming binaries of a black hole with a second black

hole (BBHs). In fact, the more massive objects in a binary will preferentially exchange their

companion for a more massive object, leading to the formation of up to two generations of

BBHs, where one of the black holes from the first BBH is swapped out for a second black

hole passing by (Rodriguez et al., 2019). The formation of BBHs is currently of significant

interest, as a total of ten BBHs have now been observed by LIGO 5, and these mergers are

more common than was expected (Abadie et al., 2010). However, we must first consider the

nature of black holes in globular clusters.

5https://www.ligo.caltech.edu/page/four-new-detections-o1-o2-catalog

8

Figure 1.3: Optical image of globular cluster 47 Tucanae. Image credit: NASA, ESA, and
the Hubble Heritage (STScI/AURA)-ESA/Hubble Collaboration Acknowledgment: J. Mack
(STScI) and G. Piotto (University of Padova, Italy).

1.3.1 Black Holes in Globular Clusters?

The exact number of black holes in globular clusters is currently an open question, and

it has been under debate for quite some time if globular cluster should even retain black

holes. The more massive objects in the clusters are expected to fall towards the cluster

center, and then be ejected by the dynamical influences of the other binaries (Spitzer, 1969;

Kulkarni et al., 1993). More recent numerical and observational work has shown that this

may not be the case (Morscher et al., 2013; Shishkovsky et al., 2018; Giesers et al., 2018,

2019). Actually, black holes may possibly drive the cluster formation and evolution, and

the nature and number of black holes certainly should influence the cluster structure (Giersz

et al., 2019). Observational studies of BHs in GCs has a large impact both on understanding

cluster evolution, as well as giving clues to the conditions that BBHs form in.

Observing black holes is difficult, and it is even more difficult to search for them in

9

globular clusters. Globular cluster black hole candidates in our own Galaxy are few and

far between, with only a handful of candidates (Strader et al., 2012a; Chomiuk et al., 2013;

Bahramian et al., 2017; Shishkovsky et al., 2018), and two confirmed (Giesers et al., 2018,

2019). There are two methods to search for black holes in galactic globular clusters. The first

method studies black holes in quiescence, as there is a correlation between radio luminosity

from the jet, and X-ray luminosity (Plotkin et al., 2012), which can allow BHs to be separated

from other X-ray emitting binaries such as NSs. The second method relies on having an

extensive spectroscopic survey of the clusters, and using the radial velocities to dynamically

measure the black hole mass (Giesers et al., 2018). Both of these techniques will only work

with Galactic globular clusters, as they require the clusters to be resolved, and the ability

to detect X-ray faint sources. However, the Milky Way is home to less than 200 globular

clusters, whereas the number of known globular clusters outside our own Galaxy is over

hundreds of thousands. The best way to probe the much larger extragalactic population of

globular clusters for black hole candidates is by searching for ULXs.

1.3.2 ULXs in Globular Clusters

Because the X-ray luminosity of ULX sources exceeds the Eddington limit for a 10 solar

mass black hole, these sources are good black hole candidates. However, not all ULXs are

black holes! As previously discussed in Section 1.2.3, some ULXs in star-forming regions of

galaxies were found to have pulsations, implying that the compact object is a neutron star

(Bachetti et al., 2013).

For ULXs in globular clusters, a neutron star accretor is highly unlikely. While NSs are

common in GCs, the conditions observed for the entire known samples of NS ULXs cannot

be met in the cluster. For example,there are a number of major differences between the

10

globular cluster environments and the star-forming regions: both due to physical differences

in the binaries, and the formation and evolution history of the systems. Physically, the best

explanation for high accretion rates onto neutron stars involves extremely strong magnetic

fields. However, the highest magnetic field measured in any globular cluster X-ray binary

is several orders of magnitude lower than what is necessary to drive super Eddington ac-

cretion onto a neutron star. Furthermore, the donor stars of ULXs in star-forming regions

tend to be massive and hydrogen rich (Pintore et al., 2018). However, globular clusters are

older populations which no longer host massive stars, and the donor stars of these binary

systems must be lower mass and metal poor. The vast difference between these potential

donor stars must therefore also be reflected in accretion mechanisms at high transfer rates for

low mass X-ray binaries and high mass X-ray binaries. Lastly, the formation and evolution

mechanisms are entirely different. The binary components of ULXs in star-forming regions

of galaxies evolve together. The dynamic processes and interactions inside of the globular

cluster environment form binaries by a completely different method. The crowded cluster

environment has a much larger interaction rate, and the higher mass bodies preferentially

exchange their binary companions for more massive objects (Rabolli, 1990; Hut et al., 1992).

Therefore, many binary systems in a globular cluster are unlikely to remain with its origi-

nal binary companion for long, and can exchange out multiple companions throughout its

evolution (Heggie, 1975; Hut & Bahcall, 1983).

1.3.3 RZ2109

RZ2109 was the first discovered black hole candidate in a globular cluster (Maccarone et al.,

2007). It is a highly variable and ultra luminous X-ray source that is physically associated

with a spectroscopically confirmed globular cluster in NGC 4472. Because extragalactic

11

globular clusters are unresolved in X-ray, variability is key in determining whether the high

X-ray luminosity is due to a single source, or confusion from multiple X-ray binaries hosted

in the cluster, as the light from multiple sources would not vary. RZ2109s extreme variability

points to the X-ray emission being produced by a single, super Eddington binary, and the

compact object being a black hole.

Optical spectroscopy of RZ2109 revealed a broad and luminous [OIII] emission line, with

no significant hydrogen emission beyond the cluster continuum (Zepf et al., 2007; Steele

et al., 2014). To date, this is the only observed [OIII] line in a GC which is broadened by

thousands of kilometers per second. The high oxygen to hydrogen ratio implies that the

donor star in the system is a white dwarf (a so-called ”ultracompact X-ray binary”), making

RZ2109 the only candidate ultracompact X-ray binary in an extragalactic globular cluster.

HST/STIS measurements show that the [OIII] emission line is spatially resolved (Peacock

et al., 2012b). This implies that the emission is due to a large oxygen nebula flooding the

cluster (on the scale of parsecs). RZ2109’s uniqueness in multiple wavelengths has exciting

implications for the nature of black hole sources in globular clusters.

1.4

Implications of This Work

The question of whether globular clusters host black holes, and if so, how many and what

masses, has been a long-standing question in astronomy. Currently, two BHs in a GC have

been dynamically measured, and ∼ 15 candidates have been identified. Because of the
dynamics of the crowded GC environment, they should be extremely efficient at forming

BBH binaries, such as those detected merging by LIGO. LIGO is detecting many more

BBH binaries than was ever expected, and a first step to understanding the formation of

12

these binaries is to understand BHs in GCs. However, very few ways exist to study BHs

in GCs observationally, and there is only one effective method to study BH candidates in

extragalactic GCs: ULXs in GCs, which is the focus of this thesis.

Therefore, this work not only has multiple implications for both ULXs as a whole and

globular clusters, but also the history behind the progenitors of the LIGO BBHs. Studying

globular cluster ULX sources can not only shed light on the nature of black holes in globular

clusters, but can also define a new class of ULXs in a very old and dynamic environment.

Their nature and numbers has implications for understanding the progenitors of the LIGO

sources, the question of black holes in globular clusters, even the formation and evolution of

the clusters themselves, in addition to extending the knowledge of ULXs as a whole.

Major results of publications in this work include a potential correlation between X-ray

behavior and presence of optical emission for globular cluster ULXs, and new discoveries

about the variable nature of RZ2109.

1.5 Outline

Chapter two consists of the peer reviewed article written by Dage et al (2018) which is

published in the Astrophysical Journal. It covers long-term monitoring (almost 20 years)

in X-ray of RZ2109. The data shows significant variability, with the source dropping in

luminosity by almost an order of magnitude over the course of only two days. Curiously,

the shape of the X-ray spectrum remains the same. This is an unusual find, as typically any

variability in flux is expected to be correlated with a change in the spectral properties of the

source (i.e. some physical change in the system). The persistence of spectral parameters,

along with the rapid variability, make RZ2109 stand out from the rest of the ULX population.

13

Chapter three consists of the peer reviewed article written by Dage et al (2019) which is

published in the Monthly Notices of The Royal Astronomical Society. We compare RZ2109

to the larger sample of GC ULXs, by expanding our sample and analyzing X-ray data of

eight more globular cluster ULX sources in NGC 4472, NGC 4649 and NGC 1399. The

sources are either best fit by power-law models, or by softer multi-component blackbody

disk models.

In the case of the softer (disk) models, the fits were bimodal: either the sources are

better fit by a higher disk temperature or (like RZ2109) had a lower disk temperature with

the spectral parameter constant while varying in luminosity. There is a likely correlation

between the disk temperature and whether or not the source has optical emission, as the

optical spectra of the low kT sources show optical emission, and a number of the higher kT

sources have no emission beyond the cluster continuum

Chapter four consists of the peer reviewd article written by Dage et al (2019) which is

published in the Monthly Notices of The Royal Astronomical Society. We find that the optical

emission of RZ2109 spends almost ten years in a long-term decline, before re-brightening.

Because the oxygen is likely being ionized by the X-ray emission, one expects a correlation,

albeit over longer timescales (due to the nebula size). We compare the optical emission to

the highly variable X-ray emission, and discuss what implication these observations have for

the size scale of the nebula.

14

Chapter 2

X-ray variability from the

ultraluminous black hole candidate

X-ray binary in the globular cluster

RZ 2109

2.1

Introduction

Determining whether or not globular clusters host black holes, and if so how many and

what masses has been the subject of both longstanding interest and current theoretical work

(e.g. Spitzer 1969 to Chatterjee et al. 2017). This interest has grown dramatically with the

detection of merging black holes by LIGO (Abbott et al., 2016a), as black hole mergers in

globular clusters are one of the leading possibilities for the origin of the LIGO sources (e.g,

Abbott et al., 2016b; Rodriguez et al., 2016). There is little question black holes are present

early in the life of a globular clusters - hundreds to thousands of stellar mass black holes

are expected to be produced in typical globular clusters as the result of standard stellar

evolution (e.g., Ivanova et al., 2010). Early work suggested that dynamical interactions

among the black holes may eject many or almost all of them (e.g. Kulkarni et al. 1993,

15

Sigurdsson & Hernquist 1993). However, recent simulations have found that this process is

not as efficient as earlier expectations and current models generally predict the retention of

a significant number of black holes in globular clusters (e.g. Morscher et al. 2015, Heggie &

Giersz 2014, Sippel & Hurley 2013).

From an observational perspective, some of the first and strongest evidence for the pres-

ence of black holes in globular clusters has come from X-ray and optical studies of extra-

galactic globular clusters. One way to find black hole candidates in globular clusters is to

identify globular cluster X-ray sources with luminosities in excess of the Eddington limit

for an accreting neutron star, which may be indicative of a more massive black hole pri-

mary. No such source is found in Galactic globular clusters. However, over the very large

sample of extragalactic globular clusters a number of such sources have now been identified

in Chandra observations of early-type galaxies in the local universe. Because extragalactic

globular clusters are unresolved in the X-rays, short term variability in these ultraluminous

X-ray sources (ULXs) is critical for eliminating the possibility that the high X-ray luminos-

ity arises from the superposition of multiple accreting neutron stars (e.g., Kalogera et al.,

2004). A handful of such high LX , variable sources are now known, starting with RZ2109

(Maccarone et al., 2007), making these among the best candidates for accreting black holes

in globular clusters (e.g. Roberts et al. (2012), Maccarone et al. (2011), Irwin et al. (2010)

and references therein).

Additional information about these ultraluminous sources can help constrain the nature

of the accreting system. For example the source in RZ2109 has strong [OIII]4959,5007

emission with a velocity width of several thousand km/s and absolutely no hydrogen observed

in emission (Zepf et al., 2008). The strong presence of [OIII]5007 and absence of Hβ is

indicative of accretion from a very O-rich and H-poor donor, for which a CO white dwarf

16

seems most likely (Steele et al., 2014). The broad velocity and high luminosity of the outflow

is also consistent with outflows from sources accreting around their Eddington limit, based

on both empirical and theoretical work (Zepf et al., 2008, and references therein) Moreover,

the [OIII]5007 can not be strongly beamed. Therefore, the lack of similar [OIII] emission in

objects without high LX places valuable limits on any beaming of the X-rays in the RZ2109

source.

It is also interesting to compare the ULXs discovered in globular clusters with the more

widely known ULXs in star forming galaxies (see Kaaret et al. (2017) for a review of the

latter). While sharing the property of high X-ray luminosity, there are many differences

between these populations. The donor stars in the star forming ULX systems are typically

high mass stars, while such stars are long dead in globular clusters, and the donor stars in

globular cluster systems are either white dwarfs (as in RZ2109, Steele et al., 2014) or other

lower mass stars. Moreover, for a ULX in a star forming galaxy the accreting compact object

will have been recently formed, and in the case of a neutron star may have an extremely

high magnetic field. In contrast, globular clusters have populations of compact objects that

formed long ago which can make close binary systems through dynamical interactions within

the globular cluster (Ivanova et al., 2010).

These underlying physical differences may be matched to observational differences be-

tween the star forming and globular cluster ULXs. One of the most striking results in the

study of ULXs in star forming galaxies is that at least some of the ULXs have accretors that

are neutron stars rather than black holes and are thus accreting at many times their formal

Eddington limit (e.g. Bachetti et al. 2014, F¨urst et al. 2016, Israel et al. 2017b, Israel et al.

2017a). Models to account for these generally involve some combination of extremely large

magnetic fields and beaming (review by Kaaret et al. (2017) and references therein). Because

17

most of these models predict that pulses will not be observed at all times and depend on

various geometries, it is possible that many ULXs in star forming galaxies are neutron star

accretors (King et al. 2017, Middleton & King 2017).

The likely absence of both extreme magnetic fields and substantial beaming differenti-

ates globular cluster ULX sources like RZ2109 from some star forming ULXs and supports

identifying the RZ2109 accretor as a black hole (Peacock et al., 2012b). However, studies of

ULXs in star forming galaxies provide an extensive set of phenomenological and theoretical

work on super-Eddington accretion onto compact objects and its observational manifesta-

tions to which observations of RZ2109 can be compared (e.g. Poutanen et al. 2007, Gladstone

et al. 2009, Sutton et al. 2013, Middleton et al. 2015). Broadly speaking these papers re-

late accretion rate relative to Eddington and viewing angle to observed properties such as

X-ray spectrum, luminosity, and possible variability based on various assumptions about the

underlying astrophysics (see review by Kaaret et al., 2017).

The goal of this paper is to analyze the now large number of X-ray observations of

RZ2109 over 16 years, with the aim of constraining the nature of this likely accreting black

hole system. RZ2109 is a very well-studied system, with extensive optical spectroscopy

(Steele et al. (2014), ?, Zepf et al. (2008) and references therein), and several extant X-ray

studies utilizing Chandra and XMM-Newton. Here we report multiple new Chandra and

XMM-Newton observations, and combine these with archival data to study the variability

of the X-ray emission from RZ2109 over a broad range of time scales. The paper is arranged

so that the observations are presented in Section 3.2, the results from the analysis of these

observation in Section 3.3, and the conclusions in Section 3.4.

18

2.2 Data and analysis

RZ2109 has been observed numerous times by XMM-Newton and Chandra observatories

over the last 16 years. We reduced and analyzed all of these observations as tabulated in

Table 2.1. The background flare filtered (see further in text for details) XMM-Newton net

count rates were obtained by filtering the energy in the range 0.3-10 keV in the spectral

extraction. We then loaded the spectra into XSPEC, and obtained the net count rates

(with no model) by using the show rate command. The Chandra source count rates were

calculated using funtools 1 to list the counts in the source and background areas based on an

image filtered in the 0.3-10 keV range. We then background subtracted the average source

counts and divided by the observation length.

Table 2.1: Chandra and background flare filtered XMM-Newton observations, background
subtracted average source count rates and raw source counts in 0.3-10 keV.

ObsID

Date

Exposure Avg. Rate Src. Counts

322 2
321
8095
11274
12978
12889
12888
16260
16261
16262
01125506013
0200130101
0761630101
0761630201
0761630301

2000-03-19
2000-06-12
2008-02-23
2010-02-27
2010-11-20
2011-02-14
2011-02-21
2014-08-04
2015-02-24
2016-04-30
2002-06-05
2004-01-01
2016-01-05
2016-01-07
2016-01-09

1https://github.com/ericmandel/funtools

count/s
4.3 ×10−3
8.7×10−3
1.1 ×10−2
2.1×10−4
9.6×10−4
2.4×10−3
1.0×10−3
3.1×10−3
9.0×10−5
5.6×10−4
2.3 ×10−2
4.5 ×10−3
2.1×10−2
4.3 ×10−4
3.9×10−4

48
398
60
19
19
425
230
79
3
16
282
465
1147
29
217

(ks)
10
40
5
40
20
140
160
25
25
25
11
72
44
35
65

19

2.2.1 Observations

The Chandra observations include both three new datasets we obtained in 2014, 2015, and

2016, and archival data going back to 2000. We used ciao version 4.9 4 (Fruscione et al.,

2006) for analysis of all Chandra data. For most on-axis observations, we manually extracted

the spectrum from the source and background regions using specextract. For off-axis

observations we used acis-extract 5 (Broos et al., 2012) to extract the source regions

(Broos et al., 2012). Specifically, in observations 12888 and 12889, the source is located

on ACIS chip 8 and thus far off-axis. Additionally in observation 12888, it is located on

the edge of this chip and is affected by dithering and edge effects. Given these issues and

low signal-to-noise ratio of the detection in these observations, we used acis-extract to

extract the spectra. Source extraction regions constructed by acis-extract are polygons

approximating Chandra-ACIS point spread function based on MARX (Davis et al., 2012)

simulations6. acis-extract also applies PSF corrections to ancillary response files (ARF)

and exposure and background scaling corrections to the spectrum to take into account edge

effects.

RZ2109 was also observed with XMM-Newton in 2002, 2004, 20087, and three times in

2016 (see Table 2.1). We used sas 16.1.08 to extract the spectra from the MOS1, MOS2

and pn detections.

We set FLAG== 0 to screen conservatively 9, and originally extracted single and double

4http://cxc.harvard.edu/ciao/
5http://www2.astro.psu.edu/xray/docs/TARA/ae users guide.html
6We note that MARX 5.0, 5.1, and 5.2 simulate PSF of off-axis

inaccurately (see
https://github.com/Chandra-MARX/marx/pull/21). We have used MARX 5.3.2 for this work, which has
addressed this issue.

sources

7Observation 0510011501 did not have enough information left post-background flare filtering and thus

is not used in this analysis.

8https://www.cosmos.esa.int/web/xmm-newton/sas
9https://heasarc.gsfc.nasa.gov/docs/xmm/hera guide/node33.html

20

events (pattern <= 4) as recommended for XMM pn, however, in observations heavily

impacted with flares (0200130101, 0761630201, and 0761630301), we only extracted single

events (pattern == 0). For MOS1 and MOS2, we select (pattern<=12). The data was

filtered for high background flares by only selecting times at which the background was

constant. Those times were determined by examining the background light curve from the

PPS. We ignored any counts below 0.2 keV. To account for differences among the three

detectors when fitting, a constant factor was added to the best fit models; the value for pn

was frozen at 1.0, while the values for MOS1 and MOS2 were free.

We used xspec version 12.9.110 (Arnaud, 1996) to analyze the X-ray spectra of both

new and archival Chandra and XMM-Newton observations. All xspec analysis used the

abundance of elements from Wilms et al. (2000).

We used the F-test function to compare the χ2 statistics of a single disk component model

with an absorption term to a two component model (disk component added to a powerlaw

model pegged from 0.5-8 keV, also with absorption) for the three Chandra observations

with the highest counts (321, 12888, and 12889). The probability that the improved fit

statistics of the two component model is due to chance is respectively: 0.002, 0.006, 1.4e-

05. Similarly, the F-test probabilities of a single powerlaw model with absorption compared

to the absorbed two component model are: 0.047, 0.047 and 0.001. Therefore, we fit a

multicolor disk (MCD) model (diskbb) added to a power law pegpwrlw, and multiplied by

the absorption component to all observations. We note for completeness that if a power law

is fit as a single-component model to the Chandra data, its index is in the range from 3.2 to

4.3, while the XMM-Newton data typically have a single-component powerlaw index around

3.0. In all our fits, we include an absorption term, tbabs, fixed to a foreground hydrogen

10https://heasarc.gsfc.nasa.gov/xanadu/xspec/

21

column density of NH = 1.6 × 1020 cm−2 11. We found no evidence for a second absorption
column; we fit the highest count Chandra data with a second absorption parameter and

found that in each case, the best fit value was consistent with zero.

For the bulk of the observations χ2 was used as the fitting statistic. Spectra with more

than 100 source counts were binned in groups of 20; spectra with fewer counts than that

were binned with 1 count per bin and fit with c-stats (Cash, 1979).

To estimate the unabsorbed fluxes in the 0.5-8keV range we used xspec’s multiplicative

model cflux 12 with the best fit spectral model. After adding in the cflux component and

refitting, we then used the error command in xspec on the flux parameter and obtained

upper and lower bounds on the fluxes of each observation to the 90% confidence interval.

All parameter errors were also obtained in this manner.

While all of our fitting is carried out in the 0.5-8keV range appropriate for Chandra data,

many X-ray results are given in the 0.2-10 keV range. Therefore to compare to other work,

we calculate the 0.2-10 keV fluxes and luminosities based on our spectral fits to the data

from 0.5-8 keV, also using cflux.

Tables 2.2, 2.3, 2.4, and 2.5 show best fit parameters, fit statistics and the fitted flux

for Chandra and XMM-Newton observations respectively. The fluxes and fit parameters are

also plotted in Figure 2.1.

Two Chandra observations have extremely low average source counts. Maccarone et al.

(2010b) have previously found 19 source count rates with a background of 10.8 counts for

obsID 11274, which, despite being highly off-axis, is significant at the 95% confidence level

(Gehrels, 1986). In obsID 16261, we detect three counts in the source region, which - consid-

11http://cxc.harvard.edu/toolkit/colden.jsp
12https://heasarc.gsfc.nasa.gov/xanadu/xspec/manual/XSmodelCflux.html

22

Table 2.2: Chandra Fit Parameters for xspec best fit model tbabs*(diskbb+pegpwrlw).
Hydrogen column density (NH ) frozen to 1.6 ×1020 cm−2.
Date

Disk Norm13

Γ

χ2
ν/d.o.f. 14

Powerlaw Flux
(erg cm2 s −1)
3.0(+2.5
−1.4)×10−14
3.8(±0.8)×10−14
2.2(+2.7
−1.8)×10−14
9.2 (+10.0
−6.2 ) ×10−16
6.7 (+10.0
−4.5 ) ×10−15
1.9 (±0.4 )×10−14
1.0 (±0.2 )×10−14
2.6(±0.8)×10−14
9.3(+27.7
−7.9 )×10−16
6.9(+5.0
−3.5)×10−15

N/A 15
3.14/16
N/A 16

—

N/A 19
1.19/16
1.69/7
N/A 20

—

N/A 22

Tin
(keV)

2000-03-19
2000-06-12
2008-02-23
2010-02-27 17
2010-11-20
2011-02-14
2011-02-21
2014-08-04
2015-02-24 21
2016-04-30

0.11 +0.04
−0.03
0.12 ±0.02
0.13 +0.03
−0.02
—
0.13 +0.08
−0.04
0.13 ±0.02
0.11 ±0.03
0.12 +0.04
−0.03
—
0.09 ±0.05

107 +670
−90
68+85
−34
60+164
−42
—
3.1+38.3
−2.9
21 +36
−14
30 +135
−23
39 +233
−34
—
242 2169
−239

1.3 +1.2
−1.1
1.7 ±0.5
0.5 +1.4
−2.5
(3.5) 18
0.2+1.6
−2.1
1.3 ±0.6
1.8 ±0.6
2.0 ±0.7
(3.5)
1.8+1.5
−1.6

Table 2.3: Chandra Fluxes (0.5-8 keV) for xspec best fit model tbabs*(diskbb+pegpwrlw).

Date

2000-03-19
2000-06-12
2008-02-23
2010-02-27
2010-11-20
2011-02-14
2011-02-21
2014-08-04
2015-02-24
2016-04-30

Unabsorbed Flux

(erg cm2 s−1)
5.7(+2.4
−1.6) ×10−14
7.0 (±0.9) ×10−14
4.5(+1.2
−1.0)×10−14
9.2(+10.0
−6.2 ) ×10−16
9.6(+8.8
−4.6)× 10−15
03.6(±0.4)×10−14
1.6(±0.3)×10−14
2.8(+0.6
−0.5)×10−14
9.3(+27.7
−7.9 )×10−16
2.0(+1.1
−0.8)×10−14

23

Table 2.4: XMM-Newton Fit Parameters for xspec model tbabs*(diskbb+pegpwrlw). Hy-
drogen column density (NH ) frozen to 1.6 ×1020 cm−2.
Γ
Date

Disk Norm24

Powerlaw Flux

χ2
ν/d.o.f. 25

Tin
(keV)

2002-06-05
2004-01-01
2016-01-05
2016-01-07 Upper Limit Only —-
2016-01-09

0.14 +0.06
−0.02
0.17 ±0.02
0.16±0.01
0.15+0.15
−0.02

18+25
−16
1.8 +0.8
−0.5
7.2 +2.6
−1.8
0.2 +1.4
−0.1

2.6 +0.7
−3.9
0.6 +0.7
−0.9
1.0 ±0.4
1.4 +1.4

—

3.9 (± 1.5)×10−14
9.2(+3.7
−3.4)×10−15
2.7 (± 0.5)×10−14
—

−0.6 ≤7.0×10−15

1.73/16
2.31/45
1.38/56

—

0.59/6

Table 2.5: Chandra Fluxes (0.5-8 keV) for xspec best fit model tbabs*(diskbb+pegpwrlw).

Date

2002-06-05
2004-01-0126
2016-01-05
2016-01-07
2016-01-09

CCF Constant

pn/MOS1/MOS2
(1.0)/1.20/0.83
(1.0)/1.57/1.59
(1.0)/0.85/1.08

—

N/A 27

Unabsorbed Flux
(erg cm2 s−1)
6.3 (+3.8
−1.0)×10−14
1.7 (±0.4)×10−14
5.0 (±0.6) ×10−14
8.9(+13.0
−8.2 )×10−16
2.3(+5.2
−2.2)×10−15

ering the expected scaled background of 1 count - is also significant at the 95% confidence

level. To estimate a flux for both of these observations, we took the background subtracted

count rates and used pimms 23 to fit with a powerlaw index of 3.5, which was the common

best fit to the single powerlaw model of the other Chandra data. XMM-Newton observation

0761630201 had very few counts left post background flare filtering. This, in conjunction

with a relatively high background ((cid:39) 50%) meant that detailed spectral analysis was not
possible. However, we were able to fit a single component disk model to the data and obtain

a flux using cflux. This lends significant uncertainty to this flux estimate.

23http://cxc.harvard.edu/toolkit/pimms.jsp

24

2.3 Results

The overall goal of this paper is to monitor variations in the X-ray luminosity of RZ2109

over the time covered by all of the available data, ranging from 2000 to 2016. Figure 2.2

shows the luminosities in the 0.2-10 keV range, which were calculated using the fluxes from

Section 3.2 and a distance of 16.1 Mpc (Macri et al., 1999). The luminosities are also listed

in Table 2.6.

One of the main results apparent from Figure 2.2 and Table 2.6 is that RZ2109 varies sig-

nificantly over all of the time scales observed, from days to years. During some observations

RZ2109 is observed to have LX ∼ 4 × 1039 erg s−1, while at other times it is observed to
have LX ∼ 2− 3× 1038 erg s−1 or even fainter, along with various times at which RZ2109 is
found to be between these luminosities. This variability is surprising because there is strong

evidence in other ways that the source is a stellar mass black hole accreting material from a

carbon-oxygen white dwarf at a very high rate, at or somewhat above its formal Eddington

limit (e.g., Peacock et al., 2012a; Roberts et al., 2012). In such a case the source is expected

to be persistent because accretion disks in high luminosity, short period ultracompacts like

this are not expected to have ionization instabilities (e.g., Maccarone et al., 2010a). While

there are beginning to be counter-examples to this argument (see Maccarone et al., 2010a),

understanding such systems may give important clues to the formation and accretion pro-

cesses in these globular cluster black hole sources. It is interesting to compare the variability

of RZ2109 to that of other ultraluminous X-ray sources in extragalactic globular clusters.

Of the six such sources published - Maccarone et al. (2007), Brassington et al. (2010); Shih

et al. (2010); Irwin et al. (2010); Maccarone et al. (2011); Roberts et al. (2012), all vary,

with at least three of them varying by more than an order of magnitude. There is a need

25

to be careful about variability in this list, because variability is also one of the criteria to

ensure that most of the X-ray flux comes from a single source and not multiple sources in

the globular cluster, and variability is one of the criteria used in these papers. However, in

these cases, the ultraluminous sources are typically the brightest X-ray sources among the

globular cluster sources in each galaxy. So substantial and large variability appears to be

the norm for ultraluminous X-ray sources in extragalactic globular clusters.

Given the luminosity variability observed in these sources, it is natural to test whether

there is any corresponding spectral variability. A key feature of the variability RZ2109

found here is that there is often no evidence for corresponding changes in the X-ray spectra

of RZ2109. The similarity of the X-ray spectra at different observed fluxes can be seen in

Tables 2.2, 2.3, 2.4 and 2.5 and Figure 2.1. This result is different than that found in the

original variability discovery for RZ2109 which showed that the decrease in flux seen within

the 2002 XMM-Newton observation was driven by a decrease in the soft component that can

be interpreted as a change in the column density of absorbing material along the line of sight

(Maccarone et al. 2007, Shih et al. 2008). While this is still true for the flux change within the

2002 XMM-Newton observation, such a model can not explain most of the variability among

the many observations shown here. Other ULX sources in extragalactic globular clusters

show a range of behavior in the relationship between luminosity and spectral variability.

Shih et al. (2010) find a ULX in an extragalactic globular cluster in NGC 1399 with more

than a factor of ten decrease in luminosity and no evident change in spectral shape. On the

other hand, a different extragalactic globular cluster ULX in NGC 4649 studied by Roberts

et al. (2012) does exhibit spectral changes in some observations. The overall picture is that

some spectral variability happens, but there are clear observations in multiple sources of

little or no spectral variability even with order of magnitude luminosity changes.

26

It is also natural to ask whether there is any overall longterm trend of LX with time

for RZ2109. Unfortunately the data are not quite adequate for addressing this question

specifically. It is intriguing that many of the fainter fluxes appear to be found at more recent

times. However, there are observations in 2014 and 2016 in which LX (cid:39) 3×1039 erg s−1, and
thus well within the ULX regime and not much different than observations in 2000 and 2002.

Given the short term variability clearly evident in RZ2109, one way to get at the long-term

changes in RZ2109 may be to study its [OIII]5007 emission which appears to be emitted over

a region of light months to light years, and therefore samples the overall photoionizing flux

averaged over those timescales (Peacock et al., 2012b). The physical origin of the variability

RZ2109 has not yet been established. As noted above, the accretion disk is not expected to

be subject to ionization instabilities, so a different mechanism must be operating.

One natural mechanism to produce changes in the accretion rate over time is to have small

changes in the eccentricity of the orbit (e.g., Hut & Paczynski, 1984), perhaps due to the

Kozai mechanism (Kozai, 1962), in which a lower mass third star in an outer orbit introduces

eccentricity into the main two body system. For typical parameters for an RZ2109-like

system and a mass of the third star of 0.5M(cid:12), the Kozai timescale is roughly a decade. This
therefore may account for any long term trend we may see in RZ2109, but not very short

term variability. This is why it was attractive originally to try to explain RZ2109’s variability

with varying absorption, but such an explanation fails to account for the absence of spectral

changes seen in many datasets since. Other possibilities for short-term variability are listed

in the Maccarone et al. (2010a) study of similar variability of the Galactic ultracompact

binary 4U 0513−40. These possibilities include tidal disc instabilities (e.g., Whitehurst,
1988; Osaki, 1995) and irradiation of the donor star leading to modulations of the accretion

rate (e.g., Hameury et al., 1986). It is not yet clear whether these can account for variability

27

observed in RZ2109.

The observed X-ray spectrum and variability of RZ2109 can also be compared to the

well-known ULXs observed in star forming galaxies (Kaaret et al., 2017). Compared to most

of these ULXs not in globular clusters, RZ2109 is significantly softer and much more variable.

Within the field ULX population, there is a class of objects known as ultrasoft ULXs (ULSs)

that are both softer and more variable than most ULXs (e.g., Earnshaw & Roberts, 2017;

Urquhart & Soria, 2016). These papers also find that the variability in ULSs is primarily

at higher X-ray energies, although there is an exception to this general characteristic (Feng

et al., 2016). Thus RZ2109 differs from most field ULSs in that its variability is either mostly

at low energies (e.g., Shih et al., 2008), or the luminosity varies equally at all energies, as

shown above.

That the globular cluster ULXs are different than most ULXs in star forming galaxies

is not surprising. Globular cluster ULXs are essentially low-mass X-ray binaries (LMXBs)

and nearly all field ULXs are high mass X-ray binaries (HMXBs). The binaries that make

the globular cluster ULXs are also likely to be formed by dynamical interactions within

the globular cluster (Ivanova et al., 2010), unlike the binary stellar evolution that makes

field ULXs. As a result, the accretor in globular cluster ULXs is likely to be much older

and thus not have the large magnetic fields often proposed for field ULXs. Differences in

the donor stars will also be significant in globular cluster ULXs compared to field ULXs.

Globular cluster sources will have old, low-mass donors while star forming ULXs typically

have young, higher mass donors. This leads to differences in the orbital parameters and the

composition of the accreted material, among others, which then may have implications for

the resulting observables in the ULXs.

28

Figure 2.1: Top panel shows best fit powerlaw index for Chandra and XMM observations,
middle panel shows disk temperature, lowest panel shows calculated unabsorbed model flux.
Chandra data is represented by blue squares, and the XMM-Newton data by orange circles.
All parameters and fluxes are in the 0.5-8 keV band.

2.4 Conclusions

We confirm long-term variability in the X-ray emission from the globular cluster black-hole

candidate source XMMUJ122939.9+075333 in the extragalactic globular cluster RZ2109.

The system shows strong luminosity variability over long and short time scales, dropping

anywhere from as low as LX (cid:39) 7 × 1037 erg s−1 to as high as LX (cid:39) 5 × 1039 erg s−1.
The system also underwent significant changes in luminosity over very short time scales:

observations four days apart from each other showed a drop in luminosity by almost a factor

of five. Over 16 years of X-ray monitoring, the spectral shape remains extremely soft. The

fitted Chandra and XMM-Newton spectra for the high count observations can be seen in

Figures 2.3-3.5. While the fit quality is too low to make any statements in regards to

29

0.00.51.01.52.02.53.03.54.0Power-lawPhotonindex5200053000540005500056000570000.050.100.150.200.25DiskbbTin(keV)520005300054000550005600057000MJD10−1610−1510−1410−13Flux0.5-8.0keV(ergcm2/s)Figure 2.2: Luminosity variability of RZ2109 in energy band 0.2-10.0 keV. Chandra data is
represented by blue squares, and the XMM-Newton data by orange circles.

spectral variability, it is remarkable how consistently soft the spectra are.

Figure 2.3: Left: Fitted spectrum with residuals of Chandra ObsID 321 (2000-06-12). Right:
Fitted spectrum with residuals of XMM-Newton observation 0112550601 (2002-06-05, pn
only).

Acknowledgements

KCD, SEZ, and MBP acknowledge support from Chandra grant GO4-15089A. SEZ and

MBP also acknowledge support from the NASA ADAP grant NNX15AI71G. This research

30

5100052000530005400055000560005700058000MJD1037103810391040Luminosity0.2-10.0keV(erg/s)10−510−410−310−210−1NormalizedCountss−1keV−1100101Energy(keV)−3−2−10123χ10−510−410−310−210−1NormalizedCountss−1keV−1100101Energy(keV)−3−2−10123χTable 2.6: All luminosities in 0.2-10 keV.

Date

2000-03-19
2000-06-12
2002-06-05
2004-01-01
2008-02-23
2010-02-27
2010-11-20
2011-02-14
2011-02-21
2014-08-04
2015-02-24
2016-01-05
2016-01-07
2016-01-09
2016-04-30

(erg s−1)
2.9×1039
3.9×1039
4.4×1039
8.4×1038
3.3×1039
1.1×1038
2.7×1038
1.8×1039
8.7×1038
2.0×1039
9.8×1036
2.8×1039
4.9×1038
6.1×1036
2.0×1037

Luminosity Lower Bound Upper Bound Obs
(erg s−1)
4.0×1039
4.4×1039
4.9×1039
1.0×1039
4.7×1039
7.4 ×1037
5.5×1038
2.1×1039
1.1×1039
2.7×1039
6.3×1037
3.1×1039
5.2×1038
3.5×1037
1.2×1038

(erg s−1)
5.6×1039
6.0×1039
5.5×1039
1.3×1039
7.0 ×1039
1.3 ×1038
1.2×1039
2.5×1039
1.3×1039
3.3×1039
2.74×1038
3.4×1039
5.6×1038
7.4×1037
3.7×1038

Chandra
Chandra
XMM
XMM
Chandra
Chandra
Chandra
Chandra
Chandra
Chandra
Chandra
XMM
XMM
XMM
Chandra

has made use of data obtained from the Chandra Data Archive and the Chandra Source

Catalog, and is based on observations obtained with XMM-Newton, an ESA science mission

with instruments and contributions directly funded by ESA Member States and NASA. We

also acknowledge use of NASA’s Astrophysics Data System and Arxiv.

Software The following softwares and packages were used for analysis: ciao, software

provided by the Chandra X-ray Center (CXC), heasoft obtained from the High Energy

Astrophysics Science Archive Research Center (HEASARC), a service of the Astrophysics

Science Division at NASA/GSFC and of the Smithsonian Astrophysical Observatory’s High

Energy Astrophysics Division, sas, the XMM-Newton Science Analysis System, SAOImage

DS9, developed by Smithsonian Astrophysical Observatory, numpy (Van Der Walt et al.,

2011), matplotlib (Hunter, 2007) and astropy (Astropy Collaboration et al., 2013).

31

Figure 2.4: Fitted spectrum with residuals of XMM-Newton observation 0200130101 (2004-
01-01, pn only).

Figure 2.5: Left: Fitted spectrum with residuals of Chandra ObsID 12889 (2011-02-14).
Right: Fitted spectrum with residuals of Chandra ObsID 12888 (2011-02-21).

Figure 2.6: Left: Fitted spectrum with residuals of XMM-Newton observation 0761630101
(2016-01-05, pn only). Right: Fitted spectrum with residuals of XMM-Newton observation
0761630301 (2016-01-09, pn only).

32

10−510−410−310−210−1NormalizedCountss−1keV−1100101Energy(keV)−3−2−10123χ10−510−410−310−210−1NormalizedCountss−1keV−1100101Energy(keV)−3−2−10123χ10−510−410−310−210−1NormalizedCountss−1keV−1100101Energy(keV)−3−2−10123χ10−510−410−310−210−1NormalizedCountss−1keV−1100101Energy(keV)−3−2−10123χ10−510−410−310−210−1NormalizedCountss−1keV−1100101Energy(keV)−3−2−10123χChapter 3

X-Ray Spectral Variability of

Ultraluminous X-Ray Sources in

Extragalactic Globular Clusters

3.1

Introduction

Ultraluminous X-ray sources (ULXs) are non-nuclear X-ray sources with luminosities signif-

icantly greater than the Eddington limit or a 10 solar mass black hole (BH). LX 1039 erg s−1

is often adopted as a general guideline for being well above this limit and therefore a ULX.

This implies the accreting object in a ULX is either a black hole, or a neutron star (NS)

with super-Eddington accretion and/or emission beamed along our line of sight. The most

well studied ULX population is that in star forming galaxies (e.g. review by Kaaret et al.

2017). The ULXs with luminosities lower than 3 ×1039 erg s−1 are typically fit best by a
singly peaked, broadened disk. The higher luminosity ULXs are fit by a two component disk

and power law model, and are either “soft” or “hard”, depending on the slope of the power

law. The slope differentiates between ∼ Eddington and super-Eddington models (Gladstone
et al., 2009; Sutton et al., 2013). These ULXs show strong luminosity variability that are

often accompanied by significant changes in states/spectral shapes (Sutton et al., 2013; Ba-

33

chetti et al., 2013; Walton et al., 2013; Feng et al., 2016). A third group of ULXs have

also been identified, with luminosities below 3 ×1039 erg s−1 and soft blackbody dominated
emission and are known as of “supersoft” ULXs (Urquhart & Soria, 2016).

Some optical counterparts have been identified for ULXs in star forming regions. These

are primarily OB type giants or supergiants (Pintore et al., 2018, and references therein).

High mass companion stars could provide an explanation for the high mass transfer rates

which drive the high luminosities and account for their connection to star-forming regions

(see Gladstone et al. 2009 and references therein).

Of the many sources confirmed as ULXs, a number of those associated with star form-

ing galaxies have been found to have pulsations, implying that the compact object is a

neutron star (e.g., Bachetti et al. 2014 to Pintore et al. 2018). Although ULXs in star

forming galaxies have been studied extensively, ULXs also exist in a completely different

environment: globular clusters (GCs). Since 2007, five globular cluster ULX sources have

been studied in some depth: XMMUJ122939.7+075333 (RZ2109; Maccarone et al., 2007),

CXOJ0338318-352604 (Irwin et al., 2010), CXOKMZJ033831.7−353058 (Shih et al., 2010),
CXOU 1229410+0757442 (Maccarone et al., 2011), and CXOUJ1243469+113234 (Roberts

et al., 2012). Some of these sources are proposed to have black hole primaries due to their

highly variable ULX emission and other properties (Maccarone et al., 2007; Shih et al., 2010).

Interestingly, nebular emission is observed from some of these GC ULXs. Such emission

is likely associated with the ULX, since it is extremely rare in globular clusters, which

lack young stars and are known to have few planetary nebulae (Peacock et al., 2012b, and

references therein). Their forbidden optical emission lines limit beaming to a factor of a

few or less for some of these GC sources. While geometric beaming has been proposed as a

mechanism for producing the super-Eddington X-ray emission from accreting neutron stars

34

(King et al., 2001), Peacock et al. (2012a) find that the optical emission from the GC ULX

sources are too luminous to accommodate a large beaming factor and should be isotropically

emitting. (See also Pakull & Mirioni (2002), and Binder et al. (2018) for a discussion of

beaming in ULX sources.)

The presence of black holes in globular clusters, and if so, the properties the globular

cluster black holes has been debated for many decades (see Spitzer 1969, Chatterjee et al.

2017, Park et al. 2017). While black holes are initially expected to form in globular clusters

(Ivanova et al., 2010), early numerical simulations indicated that dynamical interactions

within the globular cluster would cause the black holes to be ejected (Kulkarni et al., 1993;

Sigurdsson & Hernquist, 1993). However, more recent work, (e.g., Morscher et al. 2013,

2015; Heggie & Giersz 2014; Sippel & Hurley 2013) show that the ejection time scales are

much longer, and that stellar mass black holes remain well mixed in the cluster. In addition

to this, globular clusters are prime environments for black hole mergers, and if black holes

are retained in globular clusters, then globular clusters could be the progenitors of recent

LIGO detections of merging black holes (see Abbott et al., 2016b; Rodriguez et al., 2016).

Studying GC ULX sources could shed some light on the nature of black holes in globular

clusters.

It is interesting to compare the ULX population in star forming regions to those in

globular clusters, which are two very different environments. Firstly, the dense globular

cluster environment is conducive to dynamical formation of binary systems, thus, GC ULXs

are likely dynamically formed (Ivanova et al., 2010). The large interaction cross-section in

this environment means that it is unlikely that the current binary partner of any compact

object in a GC was a binary with the object when the stars initially formed and evolved. In

contrast, ULX sources in star forming regions likely evolved from primordial binaries, and

35

that the two objects in the binary evolved at the same time.

The donor stars in the two types of ULXs are also likely to be very different; due to the

ages of globular clusters, 13 Gyr or so, the BHs and NSs were born from the massive stars

in these GCs many Gyr ago. This very old age for these primordial BH and NS populations

differs greatly from the young ages of the NSs and BHs formed from massive stars in currently

star forming regions. For example, one such GC ULX source likely has a white dwarf as its

donor star (Steele et al., 2014), but the ULXs in star-forming regions should have high mass

donor stars, as mentioned previously.

One other difference between these ULX populations is that those in star forming regions

have hydrogen emission present in their optical spectra (e.g., Fabrika et al., 2015), while at

least three of the globular cluster ULXs have no hydrogen emission (Zepf et al., 2008; Irwin

et al., 2010; Roberts et al., 2012). The most well studied GC ULX, RZ2109, also has very

different X-ray behaviour from ULXs in star forming regions; it varies by more than an order

of magnitude in X-ray luminosity between many different observations, but exhibits little or

no variation in kT over those same observations (Shih et al., 2008; Dage et al., 2018).

Several studies have considered the nature of the GC ULX sources and their optical

emission, although most of them tend to focus on the source RZ2109 (e.g., Zepf et al., 2008;

Steele et al., 2011; Peacock et al., 2012a,b; Steele et al., 2014). There are also studies of

the optical spectrum of the GC ULX CXOJ0338318-352604 in the galaxy NGC 1399 (Irwin

et al., 2010). The emission lines of RZ2109 have been variously modelled as a ∼ 50 − 100
M(cid:12) mass black hole tidally disrupting a horizontal branch star (Clausen et al., 2012a) and
as the ejecta from a R Corona Borealis star being photoionised by an unrelated X-ray source

elsewhere in the cluster Maccarone & Warner (2011). The goal of this paper is to consider

all known GC ULX sources to better understand their nature and constrain these models.

36

In this paper, we undertake an analysis of the eight known globular cluster ULXs with

LX > 1039 erg s−1 to broaden our understanding of these sources. As noted above, three

of these sources have been studied previously, while five are new to this paper. Section 3.2

discusses the Chandra observations and analysis, the results are presented in Section 3.3.

The major results of the paper are discussed further in Section 3.4.

3.2 Data and Analysis

3.2.1 Globular Cluster ULX Sample

We consider low mass X-ray binaries (LMXBs) in the sample of seven local early-type galaxies

as presented in Peacock et al. (2014). The galaxies studied are all within 20 Mpc and have

deep Chandra observations of > 100 ksec. We select the targets that are located in GCs in

these galaxies and that have X-ray luminosities greater than 1039 erg s−1. The X-ray fluxes

of the point sources in these galaxies are published by Paolillo et al. 2011 (NGC 1399),

Brassington et al. 2008 (NGC 3379), Brassington et al. 2009 (NGC 4278), Joseph 2013

(NGC 4472), Li et al. 2010 (NGC 4594), Luo et al. 2013 (NGC 4649), and Sivakoff et al.

2008 (NGC 4697). Globular cluster X-ray sources were then identified from these catalogues

by matching to optical counterparts in aligned HST optical images (see Peacock et al.,

2014; Luo et al., 2013, for details). The resulting catalogue of high probability GC LMXB

candidates is used to select ULX sources.

Sivakoff et al. (2007) predict that the X-ray luminosity of bright GC sources is dominated

by a single high luminosity object and not from multiple fainter sources with a combined

high luminosity. Many of the sources in our sample are also highly variable in X-ray, which

also implies that most or all of the luminosity comes from a single source (see Maccarone

37

et al. 2007 and Section 3.2.3 for further discussion of source variability). See Table 3.1 for

source coordinates and optical cluster properties. We present new data and analysis of three

of the previously studied sources (Irwin et al. 2010; Maccarone et al. 2011; Roberts et al.

2012), and new analysis of five GC ULXs which have not yet been previously studied in

depth.

3.2.2 Spectral Fitting

We use archival Chandra data of NGC 1399, NGC 4472 and NGC 4649 (See Tables 3.2.

For these observations, we use ciao-4.91’s (Fruscione et al., 2006) specextract function

to extract the spectra, with approximately 2.5(cid:48)(cid:48) circular regions on the sources, and set

a series of 5-10 similarly shaped regions in source-less areas around the sources to select

the background regions. We follow Dage et al. (2018) for all fitting of Chandra spectra.

Observations with counts greater than 100 were binned by 20, those with less were binned

by 1. We fit the spectra with with xspec2 (Arnaud, 1996), using χ2 statistics for the more

detailed spectra and C-statistics3 (Cash, 1979) for the spectra which were binned in counts

of 1. We set the abundance of elements to Wilms (Wilms et al., 2000), and freeze the value

of the equivalent hydrogen column absorption (nH) to the value for that galaxy4. We use

the “ignore bad” command to remove bad channels.

All of the data were fit with two separate single component models. The first is a multi-

temperature blackbody disk (tbabs*diskbb) (Mitsuda et al., 1984). The second is a pegged

power law model (tbabs*pegpwrlw) with the normalisation pegged from 0.5-8.0 keV. We also

1http://cxc.harvard.edu/ciao/
2https://heasarc.gsfc.nasa.gov/xanadu/xspec/
3https://heasarc.gsfc.nasa.gov/xanadu/xspec/manual/XSappendixStatistics.html
4http://cxc.harvard.edu/toolkit/colden.jsp

38

Table 3.1: Coordinates and optical properties of the GC ULX sample. Optical properties
from (Zepf et al., 2007, RZ2109) , (Maccarone et al., 2003, GCU1), Peacock et al. (2014,
GCU2, GCU3, GCU4), (Strader et al., 2012b, GCU5, GCU6) and (Paolillo et al., 2011,
GCU7, GCU8) All magnitudes are in z-band unless otherwise noted. Distances from Macri
et al. 1999; Blakeslee et al. 2009, 2001. RZ2109 and GCU1’s V magnitudes were converted
to z by using the relation V = g - 0.39(g − z) + 0.07 (Peacock et al., 2010), and metallicities
converted to g − z from B-R with the following relationship: g − z = 1.305(B-R) - 0.543
(Peacock et al., 2010)..

Dec

Object RA
RZ2109
GCU1
GCU2
GCU3
GCU4
GCU5
GCU6
GCU7
GCU8

12:29:39.9 +07:53:33.3
12:29:41.0 +07:57:44.2
12:29:34.5 +08:00:32.1
12:29:42.3 +08:00:08.1
12:29:34.5 +07:58:51.6
12:43:46.9 +11:32:34
12:43:44.5 +11:31:50
03:38:31.8
-35:26:04
-35:27:05.7
03:38:32.6

g − z
0.84

z
20.4
20.8 NGC 4472 (16.8 Mpc)
22.1
19.5
20.1
20.3
22.2
20.7
19.9

0.92
1.42
1.11
1.55
1.60
1.98
2.24

Host Galaxy (Distance)
NGC 4472 (16.8 Mpc)

NGC 4472 (16.8 Mpc)
NGC 4472 (16.8 Mpc)
NGC 4472 (16.8 Mpc)
NGC 4649 (16.5 Mpc)
NGC 4649 (16.5 Mpc)
NGC 1399 (20.0 Mpc)
NGC 1399 (20.0 Mpc)

fit the high count (100 counts) data with a two component model tbabs*(diskbb+pegpwrlw),

and used F-test to determine if any improvement was statistically significant. Lastly, we de-

termine to what extent (if any) there is intrinsic absorption in these systems by fitting a sec-

ond absorbing column to the high count data (tbabs*tbabs*pegpwrlw and tbabs*tbabs*diskbb.)

The pegpwrlw model is normalised such that the best fit to the power law norm is the

unabsorbed flux from 0.5-8.0 keV. To determine the unabsorbed flux in the models fit by

diskbb,we multiply the models by cflux, with the energy range between 0.5 and 8 keV and

fit. To calculate the luminosities, we use the distances of 20.0 Mpc for NGC 1399 (Blakeslee

et al., 2001), 16.8 Mpc for NGC 4472 (Macri et al., 1999) and 16.5 Mpc for NGC 4649

(Blakeslee et al., 2009).

39

Table 3.2: Observations of NGC 4472, with raw source counts (0.5-8.0 keV) for GCU1,
GCU2, GCU3 and GCU4. Observations marked with * were too off axis to measure counts
from. Observations marked with - had the source off the chip.

ObsID

Date

322
321
8095
11274
12978
12889
12888
16260
16261
16262

2000-03-19
2000-06-12
2008-02-23
2010-02-27
2010-11-20
2011-02-14
2011-02-21
2014-08-04
2015-02-24
2016-04-30

ObsLen GCU1 GCU2 GCU3 GCU4
(ks)
10.36
39.59
5.09
39.67
19.78
135.59
159.31
24.74
22.76
24.73

Cts
41
196
21
180
*
488
641
50
*
52

Cts
34
142
11
115
-
492
506
*
42
50

Cts
42
134
31
458
12
1067
1559

6
58
136

Cts
15
178
11
196
2
491
644
*
*
89

3.2.2.1 NGC 4472 GC ULXs

There are five globular cluster ULX sources in NGC 4472 with LX > 1039 erg s−1. The

brightest of these, RZ2109 has been previously well studied in both X-ray and optical

(Maccarone et al. 2007, Zepf et al. 2008, Shih et al. 2008, Steele et al. 2011, Dage et al.

2018). A second globular cluster ULX, CXOU 1229410+0757442, hereafter GCU1, has also

been studied in some detail by Maccarone et al. (2011). There are three other GC ULXs,

CXOU 1229345+08003209, (hereafter GCU2), CXOU 1229423+08000808 (hereafter GCU3),

CXOU1229345+07585155 (hereafter GCU4), which have not yet been studied in depth (see

Figure 3.1 for a Chandra image of the source locations.). At least 10 different Chandra

observations exist for these sources spanning from the year 2000 to 2016 (see Table 3.2),

enabling a study of their spectral properties and behaviour on the scale of years to decades.

Below, we discuss the spectral fitting results for all sources.

CXOU 1229410+0757442 (GCU1)

40

Figure 3.1: X-ray image of NGC 4472 (ObsID 12888, filtered to 0.5-8.0 keV) with regions
for GCU1-GCU4 overlaid.

41

For GCU1, Maccarone et al. (2011) have previously examined earlier data through ObsID

11274, and could not find a statistical difference between an absorbed power law model versus

an absorbed disk model, preferring the disk model for physical reasons. While the lower count

spectra were ambiguous as to whether a single disk model is a better fit than a single power

law model, χ2 statistics for the deep observations (ObsIDs 12888 & 12889) indicate that the

disk model alone is a much better fit than a single power law. Values for both models are

presented in Tables 3.3, 3.5.

We fit a power law with a second absorbing column (tbabs*tbabs*pegpwrlw) to obser-

vations with greater than 100 source counts. We find that the nH is not inconsistent with

zero in three of these observations. By comparison, fitting tbabs*tbabs*diskbb resulted in

the second absorbing column consistent with zero that had significantly better χ2 values in

all cases.

We also find that adding a power law component to the single component disk is not

a statistically significant improvement to the fit of GCU1. Specifically, we used xspec’s

F-test tool 5 which found no statistically significant improvement in the fit from adding the

second component. We compare the statistics of the two component model to both single

component models; these values are presented in Table 3.7. We note that while ObsID 11274

seems to favor a two component model, the best fit two-component power law photon index

is 6.8, which is not typical for X-ray binaries. We do not report the best fit values for the

two-component models as they are not physically realistic or statistically significant.

CXOU 1229345+08003209 (GCU2)

GCU2 had consistently better fit statistics for a single component disk model (Tables 3.3,

3.5). The F-test values (Table 3.7) again indicate that a single component disk model is the

5https://heasarc.gsfc.nasa.gov/xanadu/xspec/manual/node83.html

42

best fit for this source.

We rule out the necessity of a second absorbing column for GCU2 by again fitting

tbabs*tbabs*pegpwrlw and tbabs*tbabs*diskbb to the source. Only ObsIDs 11274 and

12888 have a second absorption component for the power law model that is non-zero, how-

ever, the disk model,which has the best fit second absorbing column consistent with zero,

has significantly better fit statistics than the intrinsically absorbed power law.

CXOU 1229423+08000808 (GCU3)

We fit a power law model to longer observations of GCU3 with a second absorbing column,

but found the best fit absorption to be consistent with zero in all cases. We fit GCU3 with

both single component models (as reported in Tables 3.3, 3.5), as well as a two component

model. The statistics generally favour a single component disk fit, or are inconclusive. The

two component model only showed a significant statistical improvement in one observation

(Table 3.7), however, the best fit two-component power law for that had a very high index

of 5.7, which again is unphysical and much steeper than the typical value for X-ray binaries.

CXOU1229345+07585155 (GCU4)

We found that the best fit value for the second absorbing column of this source was con-

sistent with zero across all of the longer observations. GCU4 was statistically better fit by

tbabs*pegpwrlw (Tables 3.3, 3.5). The F-test values generally do not favour a two compo-

nent model (Table 3.7), except for ObsID 12889. However, it again has an unphysical power

law of 5.0.

3.2.2.2 NGC 4649 GC ULXs

NGC 4649 hosts two GC ULXs, CXOUJ1243469+113234 (hereafter GCU5) (Roberts et al.,

2012) and CXOU1243445+113150 (hereafter GCU6) (see Figure 3.2 for a Chandra image of

43

Table 3.3: Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best fit single-
component model tbabs*diskbb for GCU1 and GCU2 in NGC 4472. Hydrogen column
density (NH ) frozen to 1.6 ×1020 cm−2). Fit parameters marked with ∗∗ either encountered
an xspec error when computing a lower bound, or had a lower bound consistent with zero,
and are presented as an upper limit. Lower count observations fit with C-stat have their
statistics presented in parentheses. All fluxes shown are unabsorbed.

ObsID/Date

322 (2000-03-19)
321 (2000-06-12)
11274 (2010-02-27)
12889 (2011-02-14)
12888 (2011-02-21)
16261 (2015-02-24)
16262 (2016-04-30)

ObsID/Date

322 (2000-03-19)
321 (2000-06-12)
11274 (2010-02-27)
12889 (2011-02-14)
12888 (2011-02-21)
16261 (2015-02-24)
16262 (2016-04-30)

GCU1

Disk Norm
(10−4)

Tin
keV

0.75 (+0.33
−0.25) ≤ 136.13∗∗
0.98 (+0.24
12.9 (+15.9
−0.20)
−7.2 )
1.7 (+0.33
7.7 (+5.5
−0.25)
−3.4)
18.5 (+6.5
1.13 (±0.10)
−4.8)
14.1 (+4.1
1.29 (±0.1)
−3.2)
0.87 (+0.29
−0.19) ≤ 64.01∗∗
26.9(+34.0
0.98 (+0.29
−0.22)
−15.2)
GCU2

Tin
keV

Disk Norm
(10−4)

1.29 (+1.01
−0.42) ≤ 24.2 ∗∗
3.8 (+4.4
1.36 (+0.39
−2.1)
−0.29)
1.45 (+0.56
−0.37) ≤ 7.89∗∗
4.5(+2.7
1.37 (±0.20)
−1.7)
7.3 (+4.7
1.16 (±0.17)
−2.9)
1.25 (+0.78
−0.36) ≤ 5.67∗∗
1.10 (+0.63
−0.32) ≤ 32.0∗∗

χ2
ν/d.o.f.
or (C-stat)
(26.83/36)

1.51/5
1.74/22
1.25/49
1.10/69

(41.04/51)

0.55/5

χ2
ν/d.o.f.
or (C-stat)
(31.83/32)

0.70/6
1.35/5
1.11/24
1.36/24

(36.00/37)
(37.53/49)

Disk Flux
(10−14 erg cm−2 s −1)
2.9 (+1.1
−0.8)
2.3 (±0.5)
11.1 (±1.5)
5.8(±0.5)
7.6(±0.5)
2.7(+0.7
−0.6)
4.9(±0.9)

Disk Flux
(10−14 erg cm−2 s −1)
3.2 (+1.6
−1.1)
2.5 (±0.6)
2.5 (±0.7)
3.0(±0.4 )
2.6(±0.3)
2.1(+0.8
−0.6)
2.40 (+0.87
−0.63)

44

Table 3.4: Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best fit single-
component model tbabs*diskbb for GCU3 and GCU4 in NGC 4472. Hydrogen column
density (NH ) frozen to 1.6 ×1020 cm−2). Fit parameters marked with ∗∗ either encountered
an xspec error when computing a lower bound, or had a lower bound consistent with zero,
and are presented as an upper limit. Lower count observations fit with C-stat have their
statistics presented in parentheses. All fluxes shown are unabsorbed.

GCU3

Disk Norm χ2

ν/d.o.f.

Disk Norm χ2

ν/d.o.f.

or (C-stat)
0.43/8
1.36/8
0.95/23
0.92/30
(78.08/77)

or (C-stat)
(30.64/38)
1.32/7
(24.54/22)
1.69/7
1.48/25
1.82/33
(43.73/48)
(44.97/58)
(48.94/50)

Disk Flux
(10−14 erg cm−2 s −1)
3.2 (±0.7)
3.6(±0.6)
2.9 (±0.4)
3.1(±0.3)
3.6(+1.1
−0.8)

Disk Flux
(10−14 erg cm−2 s −1)
4.2 (+2.0
−1.3)
3.7 (+1.0
−0.8)
3.0 (+2.4
−1.3)
3.8 (± 0.9)
3.0 (± 0.4)
3.3 (± 0.4)
2.2 (+1.0
−0.6)
3.6 (+1.2
−0.9)
3.0 (+1.3
−0.9)

ObsID/Date

321 (2000-06-12)
11274(2010-02-27)
12889 (2011-02-14)
12888 (2011-02-21)
16262 (2016-04-30)

ObsID/Date

322 (2000-03-19)
321 (2000-06-12)
8095 (2008-02-23)
11274 (2010-02-27)
12889 (2011-02-14)
12888 (2011-02-21)
16260 (2014-08-04)
16261 (2015-02-24)
16262 (2016-04-30)

Tin
(keV)

1.41 (+0.40
−0.30)
1.19 (+0.26
−0.20)
1.39 (+0.25
−0.20)
1.31 (±0.15)
1.27 (+0.62
−0.32)

Tin
(keV)

1.20 (+0.90
−0.42)
0.98 (+0.38
−0.24)
1.2 (+1.9
−0.5)
1.52 (+0.63
−0.39)
1.28 (+0.22
−0.18)
1.39 (+0.25
−0.19)
1.4 (+1.2
−0.5)
1.46(+0.88
−0.42)
1.5 (+1.2
−0.5)

(10−4)
4.1 (+4.7
−2.3)
9.5(+8.3
−4.7)
3.9 (+2.8
−1.7)
5.4(+2.6
−1.8)
≤ 20.9∗∗
GCU4

(10−4)
10.6 (+21.8
−10.6)
20.9 (+35.9
−14.0)
≤ 10.6∗∗
3.7 (+0.6
−0.3)
5.9 (+4.0
−2.5)
4.7 (+3.3
−2.0)
2.7 (+7.9
−2.7)
≤ 13.8∗∗
≤ 15.5∗∗

45

Table 3.5: Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best fit single-
component model tbabs*pegpwrlw for GCU1 and GCU2 in NGC 4472. Hydrogen column
density (NH ) frozen to 1.6 ×1020 cm−2). Fit parameters marked with ∗∗ either encountered
an xspec error when computing a lower bound, or had a lower bound consistent with zero,
and are presented as an upper limit. Lower count observations fit with C-stat have their
statistics presented in parentheses. All fluxes shown are unabsorbed.

ObsID/Date

Γ

322 (2000-03-19)
321 (2000-06-12)
11274 (2010-02-27)
12889 (2011-02-14)
12888 (2011-02-21)
16261 (2015-02-24)
16262 (2016-04-30)

1.7 (± 0.5)
1.7 (± 0.2)
1.3 (± 0.1)
1.6 (± 0.10)
1.5 (± 0.1)
2.0 (± 0.4)
1.8 (±0.3)

ObsID/Date

Γ

322 (2000-03-19)
321 (2000-06-12)
11274 (2010-02-27)

12889 (2011-02-14)
12888 (2011-02-21)
16261 (2015-02-24)
16262 (2016-04-30)

1.4 (±0.5)
1.3 (± 0.2)
1.4 (± 0.3)
1.4 (± 0.1)
1.6 (± 0.1)
1.6 (± 0.5)
1.9 (± 0.5)

GCU1
χ2
ν/d.o.f.
or (C-stat)
(22.74/36)
3.41/5
1.77/22
2.25/49
1.71/69
(47.61/51)
0.73/5

GCU2
χ2
ν/d.o.f.
or (C-stat)
(35.31/32)
3.11/6
3.27/5

9.29/24
965459.3/24
(38.69/37)
(36.02/49)

PL Flux
(10−14 erg cm−2 s −1)
4.5 (+2.5
−1.5)
2.6 (± 0.6)
12.8 (± 1.4)
6.8 (± 0.5)
8.8 (± 0.5)
3.3 (+0.9
−0.7)
5.9 (± 1.0)

PL Flux
(10−14 erg cm−2 s −1)
3.9 (+1.9
−1.3)
3.0 (± 0.7)
2.9 (± 0.7)
3.5 (± 0.4)
3.2 (± 0.3)
2.5 (+0.9
−0.7)
2.91 (+0.91
−0.72)

46

Table 3.6: Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best fit single-
component model tbabs*pegpwrlw for GCU3 and GCU4 in NGC 4472. Hydrogen column
density (NH ) frozen to 1.6 ×1020 cm−2). Fit parameters marked with ∗∗ either encountered
an xspec error when computing a lower bound, or had a lower bound consistent with zero,
and are presented as an upper limit. Lower count observations fit with C-stat have their
statistics presented in parentheses. All fluxes shown are unabsorbed.

ObsID/Date

Γ

321 (2000-06-12)
11274(2010-02-27)
12889 (2011-02-14)
12888 (2011-02-21)
16262 (2016-04-30)

1.3 (±0.2)
1.4 (± 0.2)
1.4 (±0.1)
1.5 (±1.0)
1.5 (± 0.4)

ObsID/Date

Γ

322 (2000-03-19)
321 (2000-06-12)
8095 (2008-02-23)
11274 (2010-02-27)
12889 (2011-02-14)
12888 (2011-02-21)
16260 (2014-08-04)
16261 (2015-02-24)
16262 (2016-04-30)

1.9(±0.5)
1.6 (±0.2)
1.4 (±0.6)
1.5 (±0.2)
1.5 (±0.1)
1.5 (±0.1)
1.5 (±0.4 )
1.5 (±0.5)
1.5 (±0.4)

GCU3
χ2
ν/d.o.f.
or (C-stat)
0.75/8
2.56/8
0.90/23
1.64/30
(81.63/77)

GCU4
χ2
ν/d.o.f.
or (C-stat)
(220.21/38)
0.64/7
(25.54/22)
0.82/7
1.42/25
1.20/33
(41.84/48)
(43.75/58)
(48.88/50)

PL Flux
(10−14 erg cm−2 s −1)
3.7 (± 0.7)
4.5 (+0.8
−0.7)
3.3 (± 0.4)
3.6 (± 0.3)
4.4 (+1.3
−1.0)

PL Flux
(10−14 erg cm−2 s −1)
4.3 (+1.7
−1.2)
5.3 (±0.9)
3.8 (+2.6
−1.6)
4.4(±0.9)
3.63±0.4 )
4.0 (±0.4)
2.5(+0.9
−0.69)
4.1(+1.2
−0.9)
3.5(+1.2
−0.9)

Table 3.7: F-test probability values for single component versus two component models
for GC ULXs in NGC 4472 with over 100 source counts. We compare statistics be-
tween tbabs*(diskbb+pegpwrlw) and tbabs*diskbb only in columns titled “Disk”, and
tbabs*(diskbb+pegpwrlw) with tbabs*pegpwrlw only in columns titled “PL”. Blank table
entries are where the source had fewer than 100 counts in a given observation.

GCU1

ObsID Disk
321
11274
12889
12888

1
0.07
0.18
0.14

PL
0.31
0.06
5.65e-8
1.37e-7

GCU2
Disk
1

PL
0.06

GCU3
Disk
0.01

PL
0.99

GCU4
Disk
0.15

PL
0.99

0.10
0.36

0.17
5.83e-4

2.31e-3
0.10

4.07e-3
6.01e-6

0.06
1.35e-3

0.01
1

47

Table 3.8: Chandra Observations of NGC 4649, with raw source counts (0.5-8.0 keV) for
GCU5 and GCU6.

ObsID

Date

785
8182
8507
12976
12975
14328

2000-04-20
2007-01-30
2007-02-01
2011-02-24
2011-08-08
2011-08-12

ObsLen GCU5 GCU6
(ks)
38.11
52.37
17.52
101.04
84.93
13.97

Cts
135
196
45
265
260
39

Cts
65
468
111
773
505
84

the source locations.). CXOUJ1243469+113234 (GCU5)

As noted in Roberts et al. (2012), GCU5 shows better fit statistics for the single power

law component. ObsID 12975 is an exception here; the single disk component had much

better fit statistics than the single power law. However, Roberts et al. (2012) found that this

observation had a second intrinsic absorbing column that was significant. We do not account

for such a component here, but it could explain why in this particular instance, a disk was a

better fit than a single power law, when a power law has typically a better fit than a disk in

previous observations. The F-test values for comparison between a single component model

and a two component model are listed in Tables 3.9, 3.10 and the best fit values for either

single component models are listed in Tables 3.12,3.13.

CXOU1243445+113150 (GCU6)

GCU6 is ambiguous as to whether a single disk model or a single power law model is a better

fit. See Tables 3.9, 3.10 for a comparison of the single disk versus power law model.The F-

test values for GCU6 (Table 3.11) indicate that the two component model is not a better fit

to the data. We find that this source has a significant (non-zero) second absorbing column

for ObsIDs 8182 and 12975. When comparing these observations to a disk model with a

48

Figure 3.2: X-ray image of NGC 4649 (ObsID 12976, filtered to 0.5-8.0 keV) with regions
for GCU5 & GCU6 overlaid.

best fit second absorbing column that is consistent with zero, we find that it is statistically

ambiguous as to which model is a better fit. We present these fits in Table 3.12.

3.2.2.3 NGC 1399 GC ULXs

NGC 1399 hosts three GC ULXs. One, CXOKMZJ033831.7−353058 (Shih et al., 2010)
has faded beyond detection by 2005, and we do not study it here. CXOJ0338318-352604,

(hereafter GCU7) has been previously studied by Irwin et al. (2010), while the third,

CXOU0338326-35270567 (hereafter GCU8) is bright but has not been previously exten-

sively studied in X-ray. We present new analysis on both old and new data (Table 3.14). see

Figure 3.3 for a Chandra image of the source locations and any nearby sources.

CXOJ0338318-352604 (GCU7)

GCU7 shows better fit statistics for the single component disk model (Table 3.16). The F-test

values indicate that the two component model is not necessary, except perhaps for ObsID

49

Table 3.9: Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best fit single-
component model tbabs*diskbb of of GC ULXs in NGC 4649. (Hydrogen column density
(NH ) frozen to 2.0 ×1020 cm−2). Fit parameters marked with ∗∗ either encountered an
xspec error when computing a lower bound, or had a lower bound consistent with zero,
and are presented as an upper limit. Lower count observations fit with C-stat have their
statistics presented in parentheses. All fluxes shown are unabsorbed.

GCU5

ObsID/Date

785 (2000-04-20)
8182 (2007-01-30)
8507 (2007-02-01)
12976 (2011-02-24)
12975 (2011-08-08)
14328 (2011-08-12)

ObsID/Date

785 (2000-04-20)
8182 (2007-01-30)
8507 (2007-02-01)
2.0 (+0.9
−0.6)
12976 (2011-02-24)
12975 (2011-08-08)
14328 (2011-08-12)

Tin
(keV)

1.08 (+0.60
−0.30)
2.02(+0.50
−0.36)
1.27 (+0.45
−0.31)
1.09 (±0.13)
2.18 (+0.54
−0.37)
2.1 (+2.1
−0.7)

Tin
(keV)

1.32 (+0.94
−0.43)
1.28(+0.34
−0.25)
1.5

1.12 (+0.31
−0.23)
2.15 (+0.90
−0.51)
0.65 (+0.32
−0.15)

Disk Flux
(10−14 erg cm−2 s −1)
1.1 (+0.5
−0.3)
8.6 (±1.2)
4.6(±1.1)
5.3 (±0.5 )
6.2(±0.8)
6.4(+2.3
−1.7)

Disk Flux
(10−14 erg cm−2 s −1)
2.3 (+1.1
−0.7)
2.9 (±0.6)
(43.23/44)

5.3(±0.3)
3.2(±0.6)
1.6 (+0.6
−0.4)

Disk Norm χ2

ν/d.o.f.

(10−4)
≤ 12.5∗∗
2.9(2.5
−1.5)
9.2(+14.4
−5.9 )
19.4(+9.9
−6.6)
1.6 (+1.3
−0.8)
≤ 7.20∗∗
GCU6

or (C-stat)
(65.99/59)
1.68/21
2.00/4
1.93/36
0.74/23
(49.51/76)

Disk Norm χ2

ν/d.o.f.

or (C-stat)
1.71/5
0.95/8

(10−4)
≤ 10.3∗∗
5.5 (6.2
−3.1)
1.27 (+0.92
−0.41) ≤ 13.6∗∗
5.7 (+7.2
−3.3)
0.9 (+1.2
−0.6)
≤ 64.5∗∗

2.04/12
1.67/12
(31.03/37)

50

Table 3.10: Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best fit single-
component model tbabs*pegpwrlw of of GC ULXs in NGC 4649. (Hydrogen column density
(NH ) frozen to 2.0 ×1020 cm−2). Fit parameters marked with ∗∗ either encountered an
xspec error when computing a lower bound, or had a lower bound consistent with zero,
and are presented as an upper limit. Lower count observations fit with C-stat have their
statistics presented in parentheses. All fluxes shown are unabsorbed.

ObsID/Date

Γ

785 (2000-04-20)
8182 (2007-01-30)
8507 (2007-02-01)
12976 (2011-02-24)
12975 (2011-08-08)
14328 (2011-08-12)

1.6 (±0.3)
1.2 (± 0.1)
1.5 (± 0.3)
1.6 (± 0.1)
1.1 (± 0.1)
1.2 (± 0.3)

ObsID/Date

Γ

785 (2000-04-20)
8182 (2007-01-30)
8507 (2007-02-01)
12976 (2011-02-24)
12975 (2011-08-08)
14328 (2011-08-12)

1.3 (±0.2)
1.3 (± 0.2)
1.5 (± 0.4)
1.6 (± 0.2)
1.1 (± 0.2)
2.0 (± 0.5)

GCU5
χ2
ν/d.o.f.
or (C-stat)
(61.54/59)
1.47/21
2.52/4
1.50/36
1.20/23
(48.53/77)

GCU6
χ2
ν/d.o.f.
or (C-stat)
0.95/5
1.41/8
(35.81/43)
1.34/12
1.96/12
(27.63/37)

PL Flux
(10−14 erg cm−2 s −1)
1.3 (+0.5
−0.4)
9.3 (± 1.1)
5.3 (+1.3
−1.2)
6.2 (± 0.6)
7.0 (± 0.7)
7.02(+2.05
−1.60)

PL Flux
(10−14 erg cm−2 s −1)
3.2 (+0.9
−0.8)
3.7 (± 0.6)
2.4(+1.1
−0.7)
2.2 (± 0.3)
3.6 (±0.5)
2.1 (± 0.2)

Table 3.11: F-test probability values for single component versus two component mod-
els for GC ULXs in NGC 4649 with over 100 source counts. We compare statistics be-
tween tbabs*(diskbb+pegpwrlw) and tbabs*diskbb only in columns titled ”Disk”, and
tbabs*(diskbb+pegpwrlw) with tbabs*pegpwrlw only in columns titled ”PL”.

GCU5

ObsID Disk
785
8182
12976
12975

-
0.09
2.72 ×10−3
0.99

GCU6
Disk
0.33
0.65
0.12
1

PL
0.80
0.20
1
0.47

PL
-
0.31
0.19
6.41 ×10−3

51

Table 3.12: Chandra Fit Parameters and Fluxes (0.5-8.0 keV) for spectral best fit values of
GCU6 in NGC 4649 for tbabs*tbabs*pegpwrlw where NH was not consistent with zero.
All fluxes shown are unabsorbed.

ObsID/Date

NH

(1020 cm−2)

Γ

PL χ2

ν/d.o.f. PL Flux

(10−14 erg cm−2 s −1)

8182 (2007-01-30)
12975 (2011-08-08)

0.17 +0.17
−0.13
0.25+0.25
−0.23

1.8 (± 0.4)
1.5 (± 0.4)

0.84/7
1.85/11

3.7 (± 0.6)
3.7 (±0.5)

Table 3.13: χ2/d.o.f. for tbabs*tbabs*diskbb,where the best fit NH was consistent with
zero. (Hydrogen column density (NH ) frozen to 2.2 ×1020 cm−2.

ObsID/Date

Disk χ2

ν/d.o.f.

8182 (2007-01-30)
12975 (2011-08-08)

1.1/7
1.82/11

Table 3.14: Chandra Observations of NGC 1399, with raw source counts (0.5-8.0 keV) for
GCU7 and GCU8.

ObsID

Date

320
319
239
240
2389
4172
9530
14527
16639
14529

1999-10-18
2000-01-18
2000-01-19
2000-06-16
2001-05-08
2003-05-26
2008-06-08
2013-07-01
2014-10-12
2015-11-06

ObsLen GCU7 GCU8
(ks)
3.38
56.04
3.60
43.53
14.67
44.50
59.35
27.79
29.67
31.62

Cts
23
629
22
44
10
294
366
230
162
226

Cts
12
448
14
12
13
114
248
136
171
132

52

Figure 3.3: X-ray image of NGC 1399 (ObsID 319, filtered to 0.5-8.0 keV) with regions for
GCU7 & GCU8 overlaid.

319 (Table 3.17). However, the two component model has a best fit power law index of 5.7.

Fitting this source with a second absorption component plus power law gives an absorption

that is consistent with zero in almost all cases, and for all fits of tbabs*tbabs*pegpwrlw,

the power law index is between 3-4, which implies that the spectrum is very soft and unlikely

to be fit by a power law model of any sort.

Clausen et al. (2012a) model this system as the tidal disruption of a horizontal giant

branch star by an intermediate mass black hole, while Maccarone & Warner (2011) model

this as a R Corona Borealis star illuminated by a bright X-ray source. Neither scenario is

ruled out by the observed shallow decline of the LX . Other binary system models might

also be consistent with these data.

CXOU0338326-35270567 (GCU8)

We find that there is no evidence for a second absorbing column of any sort for this source.The

F-test values for GCU8 rule out a two component model (Table 3.17), and the statistics

indicate that a single power law component is the best fit model for this source. (Tables

3.15, 3.16). Note that the best fit values of the disk norms for GCU8 are upper limits, even

53

in the higher count observations binned by 20 and fit with χ2 statistics, most likely because

the single disk fit for this source was not a good fit to the point where xspec had difficulty

fitting the normalisations to these spectra.

GCU8 was marginally detected in ObsID 2389. We used the WebPIMMS tool to estimate

an upper limit on the unabsorbed flux using a count rate of 6.8×10−4 ct/s, and a fixed
powerlaw index of 1.7, with the NH fixed to frozen to 1.34 ×1020 cm−2 .

3.2.3 Long and Short-term X-ray Variability

RZ2109 has been shown to vary both long and short term (see Maccarone et al. 2007; Shih

et al. 2010; Dage et al. 2018, for example), which is leading evidence for its compact object

being a black hole accretor. Other GC ULXs show some significant long-term variability,

and quantifying the variability of these sources on either long or short timescales can shed

light on the nature of the objects that make up these ULX systems.

3.2.3.1 Long-Term Variability

We found the average luminosity comparing the data to a range of 5000 luminosities drawn

from the lowest LX to the highest for the source and computing the reduced χ2 in each case.

We plot the luminosity with the χ2 closest to 1.0 as the mean luminosity. This is plotted for

all the sources in Figure 3.4.

To quantify the variability of these sources, we do a χ2 minimisation fit of the data using

scipy8 with a model of constant luminosity, using the best fit luminosity from above as the

mean value. We find that GCU2, GCU3, GCU4 and GCU6 have χ2 values less than 1 (e.g.,

no evidence for long-term variability), while GCU1, GCU5, GCU6 and GCU7 have very

8https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.chisquare.html

54

Table 3.15: Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best fit single-
component model tbabs*diskbb of GC ULXs in NGC 1399. Hydrogen column density
(NH ) frozen to 1.34 ×1020 cm−2). Fit parameters marked with ∗∗ either encountered an
xspec error when computing a lower bound, or had a lower bound consistent with zero,
and are presented as an upper limit. Lower count observations fit with C-stat have their
statistics presented in parentheses. All fluxes shown are unabsorbed.

GCU7

Disk Norm χ2

ν/d.o.f.

ObsID/Date

320 (1999-10-18)
319 (2000-01-18)
239 (2000-01-19)
240 (2000-06-16)
2389 (2001-05-08)
4172 (2003-05-26)
9530 (2008-06-08)
14527 (2013-07-01)
16639 (2014-10-12)
14529 (2015-11-06)

ObsID/Date

320 (1999-10-18)
319 (2000-01-18)
240 (2000-06-16)
2389 (2001-05-08)
4172 (2003-05-26)
9530 (2008-06-08)
14527 (2013-07-01)
16639 (2014-10-12)
14529 (2015-11-06)

Tin
(keV)

0.28 (+0.19
−0.09)
0.39 (±0.04)
0.27 (+0.18
−0.10)
0.52 (+0.29
−0.18)
0.21 (+0.07
−0.11)
0.36 (+0.08
−0.06)
0.34 (±0.04)
0.39 (+0.08
−0.06)
0.43 (±0.07)
0.35 (±0.07)

Tin
(keV)
≤ 1.8
1.27(+0.21
−0.17)
1.8 (+1.3
−0.5)
-
2.6 (+2.0
−0.8)
1.28 (± 0.30)
1.36(+0.47
−0.29)
2.0(+2.9
−0.7)
2.0(+1.2
−0.5)

(10−4)
≤ 46.8∗∗
7.3 (+3.5
−2.4)
≤ 288.4∗∗
1.3 (+1.6
−0.7)
≤ 1272∗∗
7.7(+11.9
−4.7 )
11.9(+8.2
−5.0)
7.9(+9.2
−4.4)
6.6(+7.1
−3.3)
14.2(+21.3
−8.5 )

GCU8

(10−4)
≤ 4.09∗∗
13.5 (+5.2
−8.2)
≤ 3.14∗∗
-
≤ 3.78∗∗
8.7 (+8.3
−4.4)
11.0 (+14.1
−6.9 )
≤ 9.09∗∗
≤ 9.46∗∗

55

or (C-stat)
(8.50/12)
1.70/24
(12.36/13)
(22.28/17)
(8.94/12)
0.96/5
1.25/12
0.98/6
0.94/7
1.65/5

or (C-stat)
(19.54/18)
1.84/31
(53.01/48)
-
1.74/12
1.25/16
0.28/9
0.96/6
0.67/9

Disk Flux
(10−14 erg cm−2 s −1)
4.8 (+4.0
−3.2)
2.5(+0.2
−0.3)
2.3 (+1.3
−1.0)
1.5(+0.8
−6.2)
4.3(+2.5
−1.8)
1.7(±0.3)
2.0(±0.2)
2.7(±0.4)
3.4(±0.5)
2.7(±0.4)

Disk Flux
(10−14 erg cm−2 s −1)
7.8 (+9.2
−3.9)
6.7 (± 0.9)
5.5(+1.7
−1.3)
≤ 0.5 6
8.7 (+1.8
−1.5)
4.5 (± 0.8)
7.2 (+1.6
−1.3)
6.3 (+2.7
−1.7)
8.4 (+2.2
−1.8)

Disk Norm χ2

ν/d.o.f.

Table 3.16: Chandra Fit Parameters and Fluxes (0.5-8 keV) for spectral best fit single-
component model tbabs*pegpwrlw of GC ULXs in NGC 1399. Hydrogen column density
(NH ) frozen to 1.34 ×1020 cm−2). Fit parameters marked with ∗∗ either encountered an
xspec error when computing a lower bound, or had a lower bound consistent with zero,
and are presented as an upper limit. Lower count observations fit with C-stat have their
statistics presented in parentheses. All fluxes shown are unabsorbed.

ObsID/Date

Γ

320 (1999-10-18)
319 (2000-01-18)
239 (2000-01-19)
240 (2000-06-16)
2389 (2001-05-08)
4172 (2003-05-26)
9530 (2008-06-08)
14527 (2013-07-01)
16639 (2014-10-12)
14529 (2015-11-06)

3.1 (+1.0
−0.9)
2.4 (± 0.1)
3.1 (+1.1
−0.9)
2.7 (+0.8
−0.7)
3.8 (+1.1
−1.0)
2.9 (±0.3)
2.6 (± 0.2)
2.7 (± 0.3)
2.7 (±0.2)
3.0 (±0.3)

ObsID/Date

Γ

320 (1999-10-18)
319 (2000-01-18)
240 (2000-06-16)
2389 (2001-05-08)
4172 (2003-05-26)
9530 (2008-06-08)
14527 (2013-07-01)
16639 (2014-10-12)
14529 (2015-11-06)

-

1.3 (±0.7)
1.4 (±0.1)
1.5 (±0.5)
1.1 (±0.2)
1.5 (±0.2)
1.3 (±0.2)
1.1 (±0.3)
1.2 (±0.2)

PL Flux
(10−14 erg cm−2 s −1)
2.1 (+1.5
−1.0)
2.9 (± 0.3)
2.5 (+1.7
−1.1)
1.6(+0.9
−0.7)
4.1 (+2.4
−1.8)
2.1 (± 0.3)
2.4 (± 0.3)
3.2 (±0.5)
4.0 (± 0.6)
3.2 (±0.5)

PL Flux
(10−14 erg cm−2 s −1)
8.8 (+1.7
−3.9)
8.3 (±0.9)
5.9 (± 1.7)
≤ 0.5 7
9.5(±1.4)
5.6(±0.7)
9.7(±1.7)
8.0+1.9
−1.7)
10.2(+1.9
−1.7)

GCU7
χ2
ν/d.o.f.
or (C-stat)
(10.91/12)
8.32/24
(14.53/13)
(24.99/17)
(10.30/12)
0.76/5
2.64/12
3.71/6
1.99/7
3.23/5

GCU8
χ2
ν/d.o.f.
or (C-stat)
(18.87/20)
1.37/31
(51.46/48)
-
1.24/12
0.66/16
0.42/9
0.78/6
0.52/9

56

Table 3.17: F-test probability values for single component versus two component mod-
els for GC ULXs in NGC 1399 with over 100 source counts. We compare statistics be-
tween tbabs*(diskbb+pegpwrlw) and tbabs*diskbb only in columns titled “Disk”, and
tbabs*(diskbb+pegpwrlw) with tbabs*pegpwrlw only in columns titled “PL”.

GCU7
ObsID Disk
319
0.08
0.22
4172
0.56
9530
0.54
14527
16639
0.11
0.54
14529

PL

GCU8
Disk

7.63 ×10−4

0.46
0.04
0.11
0.03
0.37

6.14 ×10−3

0.19
0.01
0.51
0.65
0.42

PL
0.44

1
1

0.12

1

0.99

Table 3.18: χ2 values comparing a model of constant luminosity to the luminosity of the GC
ULXs in this sample over time.

Source GCU1 GCU2 GCU3 GCU4 GCU5 GCU6 GCU7 GCU8
χ2

26.33

3.00

0.30

5.22

4.12

0.22

0.17

0.55

large χ2 values, indicating that they are much more variable. These values are presented in

Table 3.18.

3.2.3.2 Short-term (inter-observational) variability

Two GC ULXs (Maccarone et al., 2007; Shih et al., 2010) show variability on short time

scales. We extracted light curves of GC ULXs from our sample in any observations that had

500 source counts or greater (see Tables 3.2, 3.8, and 3.14). The source GCU1 shows interest-

ing behaviour in ObsIDs 12888 and 12889. These observations were taken a week apart, and

were each near 150ks in length. The fluxes in each observation were significantly different

(see Tables 3.3, 3.5), and yet within the observation, no clear variability was observed.

We also search for inter-observational variability or periodicities in any Chandra light

curves with greater than 500 source counts (See Tables 3.2, 3.8, 3.14). We implement a

generalized Lomb-Scargle periodogram algorithm 9 (Lomb, 1976; Scargle, 1982) on combined

9http://www.astroml.org/modules/generated/astroML.time_series.lomb_scargle.html

57

Figure 3.4: Upper: LX vs.
time for NGC 4472 GC ULXs with mean luminosity (data
from Tables 3.3, 3.4,3.5, 3.6). Middle: LX vs. time for NGC 4649 GC ULXs with mean
luminosity (data from Tables 3.9,3.10). Lower: LX vs. time for NGC 1399 GC ULXs with
mean luminosity (data from Tables 3.15,3.16 ).

58

520005300054000550005600057000Time(MJD)012345678Luminosity(1039ergs−1cm2)BestFitLine-GCU3BestFitLine-GCU4BestFitLine-GCU2BestFitLine-GCU1LX-GCU3LX-GCU4LX-GCU2LX-GCU1530005350054000545005500055500Time(MJD)012345678Luminosity(1039ergs−1cm2)BestFitLine-GCU6BestFitLine-GCU5LX-GCU6LX-GCU551000520005300054000550005600057000Time(MJD)012345678Luminosity(1039ergs−1cm2)BestFitLine-GCU8BestFitLine-GCU7LX-GCU8LX-GCU7Figure 3.5: Upper: Light curve for GCU1 from ObsID 12889. Middle: Light curve for
GCU1 from ObsID 12888. Both are binned by 5 ksec and 2 ksec. Lower: Lomb-Scargle
periodogram of ObsIDs 12888 &12889 of GCU1 binned by 5 ksec compared to the Lomb-
Scargle periodograms of the background, as well as red noise and white noise.

59

020406080100120140Time (ks)0.00000.00250.00500.00750.01000.01250.0150Rate (ct/s)2ks5ks020406080100120140160Time (ks)0.0050.0100.0150.020Rate (ct/s)2ks5ks1000015000200002500030000350004000045000Period (s)0.000.050.100.150.200.250.30PowerDataRed NoiseWhite NoiseBackgroundlong observations of the four sources in NGC 4472 and two in NGC 4649 to search for

any trends within the observation. The background subtracted light curves as well as the

background were extracted from event files filtered to 0.5-8.0 keV using CIAO’s dmextract

tool, and binned by 2 ksec and 5 ksec.

To determine how significant (if at all) any periods identified in the periodogram are, we

computed the periodograms of red noise, white noise and the background and compare them

to the periodogram of the data for bins of 2 ksec and 5 ksec. We use the DELCgen package

(Connolly, 2016) to generate red noise simulations of the Chandra light curves (Timmer

& Koenig, 1995). We also extract background light curves and compute the Lomb-Scargle

periodogram in the same manner. For white noise, we shuffle10 our original light curves and

re-compute the periodogram.

We find that there are no clear significant periods in this data and that white noise is

the main cause of spurious signals in the Lomb-Scargle periodogram (see Figure 3.5, (lower

panel) for the various contributions of noise to the detected signals in the data).

3.3 Results

We fit a sample of eight GC ULXs located in NGC 1399, NGC 4472 and NGC 4649 (see Table

3.1) with two different single component models, an absorbed disk or an absorbed power law,

and take data from the literature for a ninth (Dage et al., 2018). We also consider a power law

with intrinsic absorption, but find that for almost all the sources, the intrinsic absorption

component was either consistent with zero, or not a statistical improvement over a disk

model.

10https://docs.scipy.org/doc/numpy-1.15.0/reference/generated/numpy.random.shuffle.html

60

Figure 3.6: Spectra and fitted models (fit residuals on lower panel) of GCU7 and GCU8.
GCU7 is best fit by a disk model, while GCU8 is much harder and better fit by a power law
model.

We also consider a two component model, however, it was not enough of a statistical

improvement over the single component model. In the few cases where the two component

model was a statistical improvement, the power law index was always unphysical. The χ2

statistics for the single component models either indicated that one model was a better fit

than the other, or the statistics were comparable, except in the case of GCU6, which was

ambiguous.

Figure 3.6 shows the best fit models and residuals for spectra from GCU7 and GCU8.

GCU8 was consistently brighter than GCU7 (see Table 3.14), and the best fit model was a

power law with no intrinsic absorption. The lower luminosity source, GCU7, was best fit

by a diskbb model. In this case specifically, the system producing the emission in GCU8 is

physically different than that of GCU7.

The data never require that the sources change between a disk to a power law between

observations, although it is difficult to [strongly] rule out the possibility. The ninth GC ULX

source, RZ2109 is the only source that is clearly a two component model (Shih et al., 2008);

however its power law indices were not well constrained (Dage et al., 2018). Like these other

61

10−510−410−310−210−1NormalizedCountss−1keV−1GCU7GCU8100101Energy(keV)−2.50.02.5χFigure 3.7: Upper: Plot of optical cluster colour (g − z) versus best fit spectral parameter
for GC ULXs. Lower: Plot of absolute z-band magnitude versus best fit spectral parameter
for GC ULXs. Neither absolute magnitude nor colour appear to be correlated with best fit
spectral parameter in this sample. However, it is worth noting how red in colour GCU8 is.
(Optical cluster values from Table 3.1)

GC ULX sources, RZ2109 also shows no strong evidence for changes in spectral state.

We also compare optical cluster colour (g-z) and magnitude (z) to the best fit spectral

index in Figure 3.7. There does not appear to be clear correlation between optical colour

and X-ray behaviour of the sources.

The ultimate aim of this paper is to study the behaviour of the brightest (LX > 1039 erg

s−1) GC ULXs. To quantify how the spectral parameters of these sources change with their

luminosity, we plot the values of the spectral component versus luminosity in Figures 3.9

and 3.10. Any GC ULX with largely better statistics for a power law was plotted in Figure

3.9; any source with consistently better statistics for a disk is plotted in Figure 3.10. GCU6

was ambiguous as to whether a disk or a power law was a better fit, so we therefore present

it on both plots.

The power law sources show little variability in either Γ or luminosity. Within uncer-

tainties, both GCU8 and GCU4 have similar luminosity and power law index values across

62

DiskbbTin(keV)12Color(g-z)RZ2109(NGC4472)GCU7(NGC1399)GCU6(NGC4649)GCU2(NGC4472)GCU3(NGC4472)GCU1(NGC4472)GCU5(NGC4649)GCU8(NGC1399)GCU4(NGC4472)GCU6(NGC4649)0123DiskbbTin(keV)−12−10−8AbsoluteMagnitude(z)123Power-lawIndexΓall of their own observations, with GCU4 at a lower luminosity than GCU8. GCU6 also

does not vary significantly in either parameter. GCU5 seems to follow the same behaviour,

except for ObsID 785 (2000-04-20), the first observation taken of this source. It seems as

though it begins with a low luminosity, then brightens and stays fairly consistent at that

luminosity/Γ.

The disk sources appear to be bimodal: they are either sources with Tin greater than 0.5

keV, or much less than that temperature. Of the sources with Tin > 0.5 keV, GCU1 shows

the most variability. GCU6 shows some variability in Tin and LX . GCU2 and GCU3 do not

show significant variability in either luminosity or disk temperature.

Finally, the two sources with nearly steady disk temperatures below 0.5 keV, RZ2109

at (cid:39) 0.15 keV and GCU7 at (cid:39) 0.4 keV do show significant variability in luminosity while
having no large change in Tin. It is of note that the only sources that vary significantly with

LX but not visibly with the spectral parameter are RZ2109 and GCU7. Interestingly, both

RZ2109 and GCU7 show optical emission lines (Zepf et al., 2008; Irwin et al., 2010).

To determine the extent to which variations in luminosity and variations in spectral fit

parameter (either kT or Γ) are correlated and estimate the correlation slope, we determine

the best fit line to the data for each individual source. To carry out this fit, we use linmix11

(Kelly, 2007) which uses Bayesian inferences, and develops MCMC sampling to allow linear

fits while accounting for uncertainties in both variables. However, this implementation does

not allow asymmetric parameter uncertainties which typically rise in X-ray spectral fits.

Thus, we conservatively chose the larger uncertainty value on each parameter for both lower

and upper values. We fit the correlation with Tin as a function of LX , which allows a simple

test for lack of correlation for systems which luminosity seems to vary independent of disk

11Python port by J. Meyers: https://github.com/jmeyers314/linmix

63

Table 3.19: Best fit slopes and uncertainties of sources that varied in kT or LX .

Object
RZ2109
GCU1
GCU7

Slope
0.0 ± 0.02
0.29 ± 0.16
0.08 ± 0.2

temperature (RZ2109 and GCU7).

We used this method to fit the slopes of RZ2109, GCU1, and GCU7. The best fit slopes

and uncertainties are reported in Table 3.19. We did not fit slopes for GCU2, GCU3, or

GCU6 as they do not appear to vary significantly in either kT or LX . The difference in

slopes between the sources below 0.5 keV (RZ2109 and GCU7) and the sources above 0.5

keV (GCU1) is suggestive of a dichotomy between low kT and high kT sources, with the

low kT sources having a slope that is likely consistent with zero, and some high kT sources

having a non-zero slope. See Figure 3.8 for the best fit slopes and errors of RZ2109, GCU1

and GCU7.

3.4 Conclusions

We consider a total sample of nine ultraluminous X-ray sources (LX ≥ 1039 erg s−1) physi-
cally associated with globular clusters for which we do new data analysis for eight and rely

on our previous study of RZ2109 for the ninth. We find that the sources are best fit by

a single component - either an absorbed disk or an absorbed power law. Two component

absorbed disk plus power law model is not statistically required for any of these sources.

The two component fits also give unphysical power law indices.

When we compare the luminosity to the spectral parameters of the sources, we find that

sources best fit as power laws have either no clear variability in either power law index or

64

Figure 3.8: linmix best fits of LX vs kT for RZ2109, GCU7 (Slopes consistent with zero)
and GCU1 (slope inconsistent with zero).

65

Figure 3.9: Γ vs. Luminosity (0.5-8.0 keV) for GC ULXs best fit by tbabs*pegpwrlw.

66

0.81.01.21.51.82.02.22.5Power-lawIndexΓ10381039Luminosity0.5-8.0keV(erg/s)GCU8(NGC1399)GCU6(NGC4649)GCU5(NGC4649)GCU4(NGC4472)Figure 3.10: Upper:Tin vs. Luminosity (0.5-8.0 keV)for GC ULXs best fit by tbabs*diskbb.
Data for RZ2109 is taken from Dage et al. (2018). Lower: Histogram of best fit kT for all
sources.

67

10381039Luminosity0.5-8.0keV(erg/s)RZ2109(NGC4472)GCU7(NGC1399)GCU6(NGC4649)GCU3(NGC4472)GCU2(NGC4472)GCU1(NGC4472)0.51.01.52.02.5DiskbbTin(keV)02468Numberluminosity, or only show mild variability in each parameter. When comparing luminosity

to inner disk temperature, the sources split into two temperature ranges, one group with

temperatures above 0.5 keV, and the other with low Tin. The part of the sample best fit

by disks with temperatures below 0.5 keV and show strong variability in luminosity with

no clear variability in disk temperature. The disk sources with temperatures below 0.5 keV

are the only ones to show variability only in luminosity, without significant corresponding

spectral changes.

Prior work (Angelini et al., 2001; Kundu et al., 2002; Sarazin et al., 2003; Jord´an et al.,

2004) has shown that more luminous (and presumably more massive) globular clusters are

more likely to contain low mass X-ray binaries. Beyond that, our GC ULX sample does not

show any clear correspondence of X-ray behaviour or luminosity to the optical photometric

properties of the cluster. However, some aspects of the X-ray behaviour appear to correlate

with the presence of optical emission lines. RZ2109, GCU7 and GCU5 all have published

optical spectra. RZ2109 shows bright and broad [OIII] emission lines beyond the globular

cluster continuum, with no hydrogen emission detected (Zepf et al., 2008). GCU7 has nar-

row [OIII] and [NII] emission lines beyond the cluster continuum, also with no hydrogen

emission (Irwin et al., 2010). RZ2109 and GCU7 both have similar behaviours in X-ray,

with consistent, low disk temperatures and luminosity variability.

GCU5 shows no optical emission lines at all, including no hydrogen emission lines. GCU5

has very different X-ray behaviour, as it has a better fit as a single power law model that does

not show the same kind of luminosity variability seen in the other two sources. Additionally,

when it is fit as a single disk component, it has temperatures greater than 1 keV (Roberts

et al., 2012). We note that none of the three sources has hydrogen emission present.

We postulate that X-ray behaviour may be linked with optical emission, as the sources

68

with low disk temperatures both have optical emission, but the source with no optical emis-

sion has a vastly different behaviour in X-ray. However, optical follow-up on other sources

in our sample is necessary to confirm such a claim.

Acknowledgments

KCD, SEZ, and MBP acknowledge support from Chandra grant GO4-15089A and NASA

grant number HST-AR-13923.001-A from the Space Telescope Science Institute, which is

operated by AURA, Inc., under NASA contract NAS 5-26555. SEZ and MBP also acknowl-

edge support from the NASA ADAP grant NNX15AI71G. This research has made use of

data obtained from the Chandra Data Archive and the Chandra Source Catalog. We also

acknowledge use of NASA’s Astrophysics Data System and Arxiv. The authors thank Jay

Strader and Ryan Urquhart for helpful discussion. We also thank the anonymous referee for

their suggestions to improve the paper.

The following softwares and packages were used for analysis: ciao, software provided by

the Chandra X-ray Center (CXC), heasoft obtained from the High Energy Astrophysics

Science Archive Research Center (HEASARC), a service of the Astrophysics Science Di-

vision at NASA/GSFC and of the Smithsonian Astrophysical Observatory’s High Energy

Astrophysics Division, SAOImage DS9, developed by Smithsonian Astrophysical Observa-

tory, astroML, linmix (Kelly, 2007), numpy (Van Der Walt et al., 2011), DELCgen

(Connolly, 2016), matplotlib (Hunter, 2007), Palettable12, scipy, and astropy (Astropy

Collaboration et al., 2013).

12https://jiffyclub.github.io/palettable/

69

Chapter 4

Slow Decline and Rise of the Broad

[OIII] Emission Line in Globular

Cluster Black Hole Candidate RZ2109

4.1

Introduction

RZ2109 is a globular cluster associated with the elliptical galaxy NGC 4472, and is host to a

stellar mass black hole candidate. The X-ray source, XMMUJ122939.7+075333, associated

with RZ2109 is ultraluminous (∼ 1039 erg s−1) and highly variable, with the count rate drop-
ping by a factor of seven in only a few hours (Maccarone et al., 2007). Optical spectroscopy

obtained by Zepf et al. (2007) revealed a very broad and luminous [OIII]λλ4959,5007 emis-

sion lines, with a velocity width of around 2000 km s−1 (Zepf et al., 2008). The oxygen

emission is also variable (Steele et al., 2011). RZ2109 is the only globular cluster associated

with NGC 4472 to show [OIII]5007 emission, which places constraints on geometric beaming

of the system. If the large X-ray luminosity were produced by strong beaming effects, it

would be likely that there would be several sources with strong [O III] whose X-ray emission

was beamed away from our line of sight (see Peacock et al., 2012a, for more detail).

Not only is the [OIII] emission in RZ2109 broad, but Peacock et al. (2012b) found that

70

the oxygen emission line is spatially resolved by HST, which implies that there is an oxygen

nebula within the cluster with a size scale of ∼5±2 pc. Such a large size for the nebula
negates a previous hypothesis that this oxygen emission is due to ionised nova ejecta that

is serendipitous in a globular cluster with a bright X-ray source as suggested by Ripamonti

& Mapelli (2012), as the nova ejecta shell required to produce only [OIII] emission must

be at least an order of magnitude smaller. All evidence points to the [OIII] emission being

associated with an outflow powered by the ultraluminous X-ray source.

Steele et al. (2014) modelled the emission line nebula and found that the material is

hydrogen-depleted, and that the emission from the material is consistent with accretion-

powered outflow driven by a white dwarf donor being accreted onto a black hole, making

XMMUJ122939.7+075333 (hereafter RZ2109) one of the very few candidate ultracompact

X-ray binary (UCXB) systems in a globular cluster. UCXBs are hydrogen deficient systems

with a white dwarf donor and either neutron star or black hole accretor, and typically have

very short orbital periods of 80 minutes or less (see Van Haaften et al., 2012; Heinke et al.,

2013).

In addition to being a strong black hole UCXB candidate, the RZ2109 source is also one

of the few known stellar mass black hole candidates in a globular cluster (Peacock et al.,

2012b; Dage et al., 2018), and possibly the brightest one. Recent theoretical work has

shown that GCs are likely to host BHs, and even suggests that black holes drive the whole

dynamical evolution of clusters, (e.g. Kremer et al., 2019), so it is not surprising studies are

starting to find them, even if their signatures are challenging to observe. Currently there are

a handful of accreting black hole candidates in Milky Way globular clusters (Strader et al.,

2012a; Chomiuk et al., 2013; Miller-Jones et al., 2015; Shishkovsky et al., 2018), and there is

one dynamically-confirmed black hole candidate in which there is no evidence for accretion

71

(Giesers et al., 2018). There are a handful of other black hole candidates in extragalactic

globular clusters, including globular clusters in NGC 1399, NGC 4472 and NGC 4649 (Shih

et al., 2010; Irwin et al., 2010; Maccarone et al., 2011; Roberts et al., 2012; Dage et al.,

2019). Unlike the Milky Way sources, this extragalactic population of globular cluster black

hole candidates are all ultraluminous X-ray sources (ULXs), and are likely exhibiting a very

different accretion regime than their very much closer analogues. Some ULXs were found

to exhibit coherent pulsations (e.g. Brightman et al., 2018), implying that the compact

objects in the binary are neutron stars. Many models for neutron star ULXs also involve

extremely high magnetic fields (Brightman et al., 2018). These may be plausible for recently

formed neutron stars in star forming regions in the field, but implausible for the old stellar

populations of globular clusters. The best physical explanations for these systems requires

significant beaming (King et al., 2017), which has been ruled out for RZ2109 (Peacock et al.,

2012a), see also Dage et al. (2019).

Of the extragalactic systems, RZ2109 is the best studied, having been monitored long-

term in both X-ray and optical. Studying this system can address questions about the

evolution of black holes in globular clusters, and shed light on the nature of black holes

in globular clusters, as simulations predict their presence (e.g. Morscher et al., 2013, 2015;

Rodriguez et al., 2016). The nature of black holes in globular clusters has become increasingly

relevant in light of the LIGO discoveries, as globular clusters are one possible source for the

progenitors of the merging black hole binaries detected by LIGO (Abbott et al., 2016b). In

fact, recent simulations suggest that multiple generations of black hole binaries can form in

globular clusters (Rodriguez et al., 2019).

In this work, we present new optical spectroscopy of RZ2109 from 2011 to 2018 and add

to the Steele et al. (2011) study of the optical variability beginning in 2007. RZ2109 is

72

also highly variable in X-ray (Maccarone et al., 2007; Shih et al., 2008). The X-ray source,

which is likely to drive the ionisation of the oxygen nebula (Peacock et al., 2012b) has

been monitored long-term in X-ray, from 2000-2016 (Dage et al., 2018). We compare the

variability of the flux of the oxygen line to the X-ray variability from Dage et al. (2018)

to search for a link between the variability in both wavelengths. Section 4.2 describes the

data and analysis. The implications of these measurements on the size scale of the system

in Section 4.4, and Section 4.6 discusses the impact of these results.

4.2 Optical Data and Analysis

Optical spectra have been obtained for RZ2109 since 2007. Previous observations have been

reduced and presented in Steele et al. (2011): it was observed with Keck in 2007 (Zepf et al.,

2008), on the William Herschel Telescope (WHT) in 2008, on the Southern Astrophysical

Research Telescope (SOAR) early in 2009, and on the Gemini South Telescope later in

2009.

In this paper we analyse new data starting from 2011 until 2019 to compare to

previous observations and measurements. The newest observations were taken with Gemini

South/Gemini Multi-Object Spectrograph (GMOS) and SOAR/Goodman High Throughput

Spectrograph(GHTS), and are described in more detail below. The data were reduced using

IRAF (Tody, 1986, 1993).

4.2.1 Gemini

Data were taken using GMOS on Gemini South (Hook et al., 2004) under observing program

GS-2011A-Q-41. Data from both programs span a wavelength range of 4134˚A to 5765˚A,

73

with a B1200 grating, resolution R=37441, and a 1.0” slit. Data for GS-2011A-Q-41 were

taken on UT 2011-05-01, 2011-05-02, 2011-05-04, and 2011-05-05. RZ2109 was observed on

Gemini South in 2015 (GS-2015A-Q-1), however, the 2015 data suffered significantly from

defects on the detector which severely impacted the utility of these observations.

4.2.2 SOAR

RZ2109 was observed on the 4.1 m Southern Astrophysical Research Telescope (SOAR)

using the GHTS (Clemens et al., 2004) on UT 2012-03-14, 2016-03-14, 2018-03-14, 2018-03-

15, 2019-04-06 and 2019-04-07. The observations were taken using the 0.95” longslit (0.95”

in 2018 and 2019), with the SYZY 930 grating. The resolution is R ∼ 1500. It was also
observed on 28-04-2019 and 02-05-2019 with a higher resolution (1200 l/mm) grating and

0.95” longslit.

4.2.3 Equivalent Width Measurement

We develop a method in python to measure the equivalent width of the [OIII] emission

of the un-normalised spectra in the following manner: we select spectral bandpasses with

wavelengths longer and shorter than that of the [OIII] emission and which are relatively

featureless and hence good indicators of the cluster continuum. An average value is drawn

from two regions, a lower region of 4800-4875˚A , and an upper region of 5100-5175˚A 2. Out

of these two regions, we randomly select 25 ˚A wide regions over which to average. Similar to

Steele et al. (2011), we measure the [OIII] emission between 4964˚A and 5058˚A (due to the

breadth of the emission lines - ∼1475 km s−1, Zepf et al. 2008).

1https://www.gemini.edu/sciops/instruments/gmos/spectroscopy-overview/gratings
2https://github.com/kcdage/equivalent_width

74

Figure 4.1: Un-normalised Gemini spectrum from 2009. The purple shaded regions show one
example of the 25˚A wide regions sampled out of larger regions (blue) that are averaged to
approximate the continuum values at the start and end of the emission region. The orange
shaded area shows the emission region we are measuring over from 4964˚A to 5058˚A . The
blue line is the continuum fit in the region of interest.

We adopt the average value in the bluer range as the continuum value at 4964˚A , and

the average value in the redder range as the continuum value at 5058˚A , and fit a straight

line across the broad emission region. The very much lower S/N 2009 and 2012 SOAR

spectra were smoothed with a 1D box filter3. Figure 4.1 shows one example for the 2009

Gemini spectrum. We calculate the equivalent width, and to estimate uncertainties, repeat

this process 100 times, randomly drawing different regions to calculate the continuum values

from. The final reported equivalent width is the average of all these trials, and the error

comes from the standard deviation.

The equivalent width of RZ2109 has been measured in Steele et al. (2011) from obser-

vations spanning from 2007-2012, taken with Keck, WHT, SOAR, and Gemini. We also

confirm our measurements by remeasuring a subsample of the spectra from Steele et al.

(2011). The data are presented in Table 4.1.

3http://docs.astropy.org/en/stable/api/astropy.convolution.Box1DKernel.html

75

460047004800490050005100520053005400Wavelength(Angstrom)150200250300350400450500550Flux(ArbitraryUnits)Gemini-2009BestfitlineRandomsampleEmissionlineTotalsamplingTable 4.1: [OIII] Equivalent Width (Angstrom).

Instrument Equivalent Width Error
Date
Keck
2007-12
WHT
2008-01
SOAR
2009-02
Gemini
2009-03
Gemini
2011-05
SOAR
2012-03
2016-03-14
SOAR
2018-03-14, 2018-03-15 SOAR
2019-04-06,2019-04-07
SOAR

31.6
29.3
25.3
32.7
26.0
24.4
18.9
12.9
16.5

1.6
3.2
7.3
0.7
0.6
3.7
0.9
0.9
1.1

4.2.4 Normalisation of the Spectra

All further analysis in this paper is based off of the higher resolution Keck, SOAR and

Gemini observations. The spectral resolution of the 2009, 2011 Gemini spectra are within

about 20% of the 2016, 2018 and 2019 SOAR spectra. We normalise these spectra with

respect to the Keck spectrum (which has been flux calibrated) using the following procedure

(in python)4: we calculate a moving average across 4500˚A to 5500˚A , but ignores 4940˚A

-5100˚A in order to mask the [OIII] emission5. Then we fit a 5th order polynomial to the

smoothed continuum. We divide the fitted Keck slope by the SOAR or Gemini slope to find

the relative normalisation as a function of wavelength and apply it to each spectrum. The

normalised spectra are presented in Figure 4.2.

4.3 X-Ray Data and Analysis

RZ2109 was observed by Chandra for 30 ks on 2019-04-17 (ObsID 21647), and was followed

up by a series of shorter Swift-XRT observations triggered after the observed rise of the

4https://github.com/kcdage/spec_slope_fit/
5For the lower wavelength range,we use a lower cutoff of 4646˚A instead of 4500˚A in the case of the

SOAR 2016 data, which had a narrower wavelength range.

76

Figure 4.2: Normalised spectra from observations spanning 2007-2018. Observations are
from the W.M. Keck Observatory (December 2007, [dark blue]) Gemini South (March 2009,
[yellow], May 2011, [light blue]), and SOAR (March 2016, [light pink], March 2018, [light
purple], April 2019 [dark purple]).

optical luminosity.

4.3.1 Chandra

RZ2109 was detected in the latest Chandra observation, with 7 source counts against a

background of 2 counts detected in a 3.5” radius region in the 0.3-10keV energy range,

making this a 3 σ detection (Gehrels, 1986). Similar to Dage et al. (2018), the source was

extracted with CIAO version 4.10 using the dmextract tool (Fruscione et al., 2006) and

binned by counts of 1. We fit the resultant spectrum with xspec (Arnaud, 1996), using

Cash statistics (Cash, 1979) to fit, and Pearson Chi-Squared as the test statistic, with solar

abundances from Wilms et al. (2000).

While RZ2109 is typically best-fit by an absorbed two component disk plus power-law

model (Dage et al., 2018), given the poor statistics and low number of counts, an absorbed

77

46004800500052005400Wavelength(Angstrom)0.00.51.01.52.02.53.03.54.0Flux(10−17erg/s/cm2/Ang)Gemini2011Gemini2009SOAR-2016SOAR-2018SOAR-2019Table 4.2: X-ray luminosities from 0.2-10 keV measured by Swift/XRT.

Date

2007: 11-13, 12-(25, 27)
2010: 3-(22, 26, 30)
2019: 4-15, 5-(13, 17, 27, 31)

Exposure Length LX (0.2-10 keV)
seconds
8327.0
6061.0
5800.0

×1039 erg s−1
1.79 +1.16
−0.98
≤ 2.45
≤ 1.26

power-law model (tbabs*pegpwrlw, with the hydrogen absorption column fixed to 1.606

×1020cm−1) was the more appropriate choice. We caution that it is not justifiable to inter-
pret any best fit spectral parameters physically, due to low count numbers, and we fit solely

to estimate the X-ray luminosity. The best fit flux was 2.2 +6

0.5-8 keV energy range. The best fit power-law index is 3.0+5

−0.8 ×10−15erg cm−2 s−1 in the
−1. The X-ray luminosity7 of this
observation in 0.2-10 keV extended energy range calculated using PIMMS8 and a power-law

index of 3.0 is 2.0 +5.3

−0.7 ×1038 erg s−1.

4.3.2 Swift/X-Ray Telescope

Swift/XRT data on RZ2109 were taken in three epochs in November and December 2007,

March 2010, April and May 2019. We reprocessed all Swift/XRT data using xrtpipeline

(HEASoft 6.25). The source was not detected in any individual observations, thus we merged

observations in each epoch (using Xselect) and investigated source brightness. The source

was detected at ∼2-sigma significance in 2007, but was not detected in 2010 or 2019. The
X-ray luminosities from 0.2-10 keV are presented in Table 4.2.

6http://cxc.harvard.edu/toolkit/colden.jsp
7Assuming a distance of 16.8 Mpc (Macri et al., 1999).
8http://cxc.harvard.edu/toolkit/pimms.jsp

78

4.4 Results: Time Behaviour of [OIII] Emission

Based on the observed changes in the [OIII] emission, we can consider possible implications

these changes over time would have for the size scales of the oxygen nebula. In this section, we

specifically address what could be causing the changes to the relative strengths of components

of the [OIII]λ5007 emission line, and if we see any link between the change over time in X-ray

flux and [OIII]λ5007 emission flux.

4.4.1 Time Evolution of the [OIII] emission

We have been monitoring the [OIII]λλ4959,5007 emission of RZ2109 since 2007 to examine

how the emission has changed over time, and how changes in the optical emission compare

to any changes in the X-ray luminosity. Figure 4.3 plots the full [OIII]λ5007 equivalent

widths from Steele et al. (2011) and Table 4.1 to the X-ray luminosities of RZ2109 taken

from Maccarone et al. (2010b), Dage et al. (2018), and Table 4.2.

One of the goals of this work is to use these data to constrain the size of the [OIII]

emitting nebula. From Figure 4.3 it is clear that the [OIII]λ5007 luminosity decreases by

about a factor of two over a timescale of about 3,000 days. Because the light crossing time

of the emitting region provides a rough lower limit to the timescale on which the emission

can be seen to vary, the data shown in Figure 4.3 place an upper limit on the size of the

emitting region to be about four light years across.

This approach gives an upper limit to the size of the emitting region, but does not provide

further information about its size. An alternative approach is to assume the observed decline

in the emission line is due to the oxygen nebula expanding and lowering the overall density of

material by a factor of two while the central source remains constant. Given this assumption

79

Figure 4.3: Upper panel: Change in X-ray luminosity over time. Purple star is ROSAT, Blue
squares are Chandra observations, orange circles are taken from XMM-Newton observations,
yellow triangles are upper limits from Swift/XRT. See Maccarone et al. (2010b), Dage et al.
(2018), and Section 4.3 of this paper for X-ray analysis. Lower panel: Change in L(5007)
over time, L(5007) equivalent widths are presented in Table 4.1.

80

5000052000540005600058000MJD1037103810391040Luminosity0.2-10.0keV(erg/s)ROSATSwift/XRTChandraXMM-NewtonSwift/XRT5000052000540005600058000Time(MJD)15202530L([OIII]5007)EW(˚A)GeminiWHT(LaPalma)SOARKeckthe size scale can be estimated based on the following equations:

rfinal = rinitial + (Vexpansion) × tdecline

r3
final = 2 × r3

initial.

(4.1)

(4.2)

Adopting eight years as the time over which the emission is seen to decline and an

expansion rate for the emitting region of 1000 km/s (approximately one half the line width),

we find that under these assumptions, the current size scale would be roughly 0.04 pc. We

can also compare this to the absolute lower limit for the size of the emitting region set by the

volume required at the critical density of the [OIII]λ λ4959,5007 emission line required to

produce the observed line luminosity. As also shown in (Steele et al., 2011), this minimum

size is a few 10−3pc. The half-light radius of the emitting region could be limited to greater
than ∼ 10−3 pc at the low end and ∼2 pc at the high end.

4.4.2 Time Evolution of the Core and Red and Blue Wings of the

Emission Line

Differential behaviour between the red and blue wings of the observed [OIII] emission can also

provide a constraint on the size scale of the nebula. We will observe changes in the red wing

which should lag changes in the blue wing due to the difference in light travel time between

the material moving toward us and material moving away. Any such change is convolved

with possible changes in the structure of the emitting region, but such a comparison may

still provide a useful test of the consistency of models of the spatial scale of the nebulae.

81

To estimate this possible lag, we utilise the work of Steele et al. (2014) who modelled

the structure of the λ5007 emission region and found that it is best described by a Gaussian

core with red and blue shifted wings. We compute the equivalent widths in the wings and

the core of the emission line in the normalised spectra by defining three regions: 5006-5026˚A

as the blue wing, 5026-5037˚A as the core, and 5037-5057˚A as the red wing (see Figure 4.4).

We calculate the errors by extending the regions by 2˚A on either side, recomputing the

equivalent width and taking the difference. The λ4959 core emission was measured across

4978-4989˚A , the same width of region as the λ5007 emission core, but centered around the

λ4959 core. The uncertainty was propagated by recomputing the equivalent width for a

range of continuum values.

As can be seen in Figure 4.5 while all three structures appear to decay with time, they

do not appear to do so in concert. At most epochs, the red emission wing is brighter than

the blue wing, but fainter than the core. However, in the past two years, the core continued

dropping, then rose, while the red wing seems to have possibly flattened and the blue wing

appears to continue to fade. The changes in the core are highly significant compared to the

measurement uncertainties, but this is not the case for the wings. Both the λ4959 and λ5007

core emission follow the same trend (Figure 4.6).

As noted above, if the decline of the observed [OIII] flux is due to the central ionising

source declining, and the timescale over which it declines is indicative of the size scale of the

emitting region, then the red wing of the emission line should lag the behaviour of the blue

wing of the emission line, and the time lag should be similar to the overall spatial scale of

the emitting region.

Based on the data presented in Figure 4.4, the red wing may lag the blue wing by around

2500 days (∼ 7 years). This is similar to the size scale derived from the time it takes the

82

Figure 4.4: Intervals of the [OIII]λ5007 emission line components, 5006-5026˚A for the [OIII]
blue wing, 5026-5037˚A for the [OIII] core, and 5037-5057˚A for the [OIII] red wing.

83

4980500050205040506050805100Wavelength(Angstrom)0.00.51.01.52.02.53.03.54.0Flux(10−17erg/s/cm2/Ang)Gemini2009Gemini2011SOAR2016SOAR2018SOAR2019Bluewing,5006-5026˚ACore,5026-5037˚ARedwing,5037-5057˚AFigure 4.5: Flux in each regime of the emission line over the course of the observations.

Figure 4.6: Flux in each regime of the λ4959, λ5007 emission cores.

overall luminosity to decline by a factor of two. We note that these two measurements are

independent - the overall luminosity could decline without any difference in the red and blue

wings, or without the red wing being consistent with lagging the blue wing. However, it

remains the case that one can construct smaller size scale models where the changes are due

to the structure of emitting regions and not due to light travel times.

These implications are a contrast to the spatially resolved HST-STIS spectroscopy anal-

ysed by Peacock et al. 2012b.

In this work, the HST-STIS spectroscopy appeared to be

84

5500055500560005650057000575005800058500Time(MJD)−18−16−14−12−10−8−6−4−20EquivalentWidthCorefluxRedwingfluxBluewingflux5500055500560005650057000575005800058500Time(MJD)0.02.55.07.510.012.515.017.5EquivalentWidthλ5007CoreFluxλ4959Fluxspatially resolved, with the emitting region having an extent of 5 ± 2 pc. The variability
results appear to suggest a spatial scale on the low end of this very difficult measurement.

4.4.3 X-ray Behaviour Versus [OIII]

Any relation between changes in the X-ray and [OIII] λ5007 luminosities can be tested

by monitoring RZ2109 in both X-rays and [OIII]λλ4959,5007 emission. RZ2109 has been

observed in both of these wavelengths for more than a decade. In particular, if the optical

emitting region is of order a parsec in size, any change observed in the ten or so years we

have been observing it must be due to changes in the central X-ray source illuminating the

region, as such a large region can not change its structure so quickly.

In this case there

should be a correlation between the X-ray flux averaged over the light travel time of the

system and the [OIII] λ5007 flux.

Figure 4.3 compares the full equivalent widths from Steele et al. (2011) and Table 4.1 to

the X-ray luminosities of RZ2109 taken from Maccarone et al. (2010b), Dage et al. (2018),

and new data presented in this paper (Table 4.2).

Ideally, we would compare the time changes in the X-ray luminosity to those of the

[OIII] λ5007 luminosity to help distinguish between these possibilities for the size of the

[OIII] λ5007 emitting nebula. Fundamentally, because the X-ray emitting region is orders

of magnitudes smaller than the [OIII] λ 5007 nebular emitting region for any physically

plausible model, the [OIII] λ 5007 emission should lag changes in the X-ray emission by a

time comparable to the light crossing time of the [OIII] λ 5007 emitting nebula. In order to

carry out this test, we need to establish the effective LX seen by the [OIII] λ 5007 emitting

region on the relevant timescales of months to years. However, determining the effective

LX over longer time periods is made difficult by the widespread short-term variability of

85

Figure 4.7: Measurements from early April 2019, and late April 2019 verifying the increase
in the [OIII] emission on two different gratings.

RZ2109, so that any X-ray observation may not give a representative value for the LX at

that time. Given this caution it is still notable that the observed LX in the top panel of

Figure 4.3 suggest an overall decline of the LX of RZ2109.

4.4.4 Rise in [OIII] Emission

While the [OIII] emission has been observed in a long-term decline since 2010, in April 2019

the flux in the core was observed to be increasing. Subsequent measurements approximately

a month later with a higher resolution grating verify the observed increase (See Figure 4.7).

A close up of the spectrum from 2016-2019 is shown in Figure 4.8 to highlight the decline

and rise of the emission line.

86

4800490050005100520053000.751.001.251.501.752.002.252.502.75SOAR2019930l/mmSOAR20191200l/mmFigure 4.8: Spectra from 2016-2019 showing decline and subsequent increase.

4.5 Comparison to Milky Way GC Sources

There is one candidate globular cluster BH UCXB candidate among Galactic globular clus-

ters, 47 Tuc X9. This source has LX ∼ a few ×1033 erg s−1, but unlike typical compact
binaries in of this luminosity, also shows bright radio continuum emission (Miller-Jones et al.,

2015). X-ray timing suggests an orbital period about 28 min, and the X-ray spectrum shows

emission lines consistent with Oxygen VII and VIII and indicates overabundance of oxygen

(and possibly carbon) in the system. (Bahramian et al., 2017), but no evidence for optical

hydrogen or helium emission/absorption (Tudor et al., 2018), suggesting a CO white dwarf

donor, as inferred for RZ2109. Carbon emission lines were also detected in the FUV spec-

trum (Knigge et al., 2008). Hence, 47 Tuc X9 appears as reasonable analogue for RZ2109

excepting its lower X-ray luminosity. This can be understood in the context of the evolu-

tionary timescales of UCXBs: RZ2109 is expected to be in a relatively brief period (∼ 105
yr) of high mass transfer but will evolve to a fainter, more extended (∼ 107–108 yr), longer
period system akin to 47 Tuc X9 (Van Haaften et al., 2012; Church et al., 2017).

87

4980500050205040506050805100Wavelength(Angstrom)1.01.52.02.53.03.54.0Flux(10−17erg/s/cm2/Ang)SOAR2016SOAR2018SOAR2019We can also consider whether the size scale of the RZ2109 nebula and its implications for

the age of the system are consistent with such a scenario. If the size of the RZ2109 nebula

is on the order of a parsec, then the system has to be old enough so that the outflowing

material can reach this radius. At the observed outflow velocity of about 103 km/s, this

corresponds to an estimated age of about 103 years. Alternatively, if the system is much

smaller with a size of 10−1−10−2pc, then the age can be correspondingly younger. These are
formally lower limits to the age of the system, as the system could have been expanding for

longer, and the size scale of the emitting nebula could not be the fullest extent of the ejected

material but rather where the density and ionisation parameter are such that it where most

of the [OIII] is produced. Therefore, age constraints from the size estimates for the nebula

are readily consistent with an age much less than the 105 years duration of this evolutionary

stage, although they do not absolutely require it.

There is also good reason to believe that the early evolution of the system leads to strong

mass ejection. A CO WD donor, as implied by the optical and X-ray data, must have had

an initial mass of at least 0.4 M(cid:12) (Prada Moroni & Straniero, 2009), and white dwarfs at
or about this mass will require non-conservative mass transfer with strong wind emission to

remain in stable mass transfer(Van Haaften et al., 2012). Thus there is both a mechanism

to drive oxygen into a nebula and sufficient time for it to reach the size scales we estimate.

4.6 Conclusions

RZ2109 shows very broad [OIII]λλ4959,5007 emission, which is also variable. We have been

monitoring it with multiple telescopes since 2007 and find that the emission has declined

over the last nine years of monitoring.

88

Clausen & Eracleous (2011); Clausen et al. (2012b) model the X-ray and optical behaviour

of an intermediate mass black hole tidally disrupting a white dwarf. The decline of the [OIII]

luminosity predicted in the model in Clausen & Eracleous (2011) happens much more quickly

than what we observe here. Similarly, RZ2109 is declining on a much slower timescale than

predicted by the model from Ripamonti & Mapelli (2012) of a serendipitous bright X-ray

source and nova shell ejection.

It is also unlikely that these two unique phenomena are

serendipitous in the same cluster.

The oxygen emission is likely caused by X-ray ionisation of the oxygen nebula in the clus-

ter, therefore we expect a link between the X-ray variability and oxygen emission. However,

due to the large size of the nebula, it is unclear over what timescales a potential link between

the X-ray and optical variability would be observed. The similarity of the declines between

both the [OIII] core and the red and blue high velocity wings are an interesting challenge to

models of the physical origin.

While there may be a hint of a correlation between X-ray luminosity and optical flux,

future monitoring in X-ray and optical of RZ2109 could help determine if this really is the

case. Future optical monitoring of RZ2109’s [OIII] emission line can help address whether

the oxygen emission will continue to decline, and how the broad and narrow components of

the [OIII] emission vary.

Acknowledgements

KCD, SEZ, and MBP acknowledge support from Chandra grant GO4-15089A. SEZ and

MBP also acknowledge support from the NASA ADAP grant NNX15AI71G. This research is

based on observations obtained at the Southern Astrophysical Research (SOAR) telescope,

89

which is a joint project of the Minist´erio da Ciˆencia, Tecnologia, e Inova¸c˜ao (MCTI) da

Rep´ublica Federativa do Brasil, the U.S. National Optical Astronomy Observatory (NOAO),

the University of North Carolina at Chapel Hill (UNC), and Michigan State University

(MSU). This research has made use of the XRT Data Analysis Software (XRTDAS) developed

under the responsibility of the ASI Science Data Center (ASDC), Italy. We also acknowledge

use of NASA’s Astrophysics Data System and Arxiv. KCD acknowledges Jamie Kennea and

the Swift ToO Team, and thanks James Miller-Jones and ICRAR/Curtin University. JS

acknowledges support from the Packard Foundation. The authors thank the referee for

thoughtful and insightful comments.

The following software and packages were used for analysis: IRAF, distributed by the

National Optical Astronomy Observatories, which are operated by the Association of Uni-

versities for Research in Astronomy, Inc., under cooperative agreement with the National

Science Foundation, SAOImage DS9, developed by Smithsonian Astrophysical Observatory,

numpy (Van Der Walt et al., 2011), Palettable9, and matplotlib (Hunter, 2007).

9https://jiffyclub.github.io/palettable/

90

Chapter 5

Conclusions

5.1 Summary and Perspectives

This work is aimed at probing the multiwavelength nature of black hole candidates in extra-

galactic globular clusters by studying ULXs physically associated with globular clusters. Not

only have we identified a number of black hole candidates in extragalactic globular clusters,

but we have observationally probed their fascinating and strange accretion signatures.

Long-Term X-ray Monitoring of RZ2109: Chapter Two details the results of data

spanning 16 years of X-ray monitoring of the source RZ2109. The combination of the high

X-ray luminosity, short term X-ray variability, X-ray spectrum, and optical emission suggest

that this system is likely an accreting black hole in a globular cluster. A total of 15 spectra

from Chandra and XMM-Newton, were analysed and fit with xspec models. The spectra

are all dominated by a soft component, which is very soft with typical fit temperatures

of T (cid:39) 0.15 keV for a multicolor blackbody disk. The resulting X-ray fluxes show strong
variability on short and long timescales, while the X-ray spectrum often shows no significant

change even with luminosity changes as large as a factor of five. The persistence of spectral

parameters, along with the rapid variability, make RZ2109 stand out from the rest of the

ULX population. While the underlying physical nature cannot be completely discerned

without a thorough study of the lower energy X-ray emission of the source, the steadfastness

of the best-fit spectral parameters indicates that the shape of the emission, and therefore

91

the configuration of the central engine and the overall accretion process do not vary over the

wavelengths probed.

X-ray Analysis of the Larger Globular Cluster ULX Sample: Chapter Three

contains the analysis for a further eight globular cluster ULX sources in NGC 4472, NGC

4649 and NGC 1399. Their behavior in both X-ray and optical were compared to RZ2109.

The eight GC ULXs were fit with either a single power law and single disk models, where

the data never requires that any of the sources change between a disk and a power law across

successive observations.

In the case of the softer (disk) models, the fits were bimodal: either the sources were better

fit by a higher disk temperature (T > 0.5 keV) or (like RZ2109) had a lower disk temperature

(T < 0.5 keV) with the spectral parameter constant while varying in luminosity. The GC

ULXs best fit by a single disk show a bimodal distribution: they either have temperatures

well below 0.5 keV, or variable temperatures ranging above 0.5 keV up to 2 keV. The GC

ULXs with low kT have significant changes in luminosity but show little or no change in

kT. By contrast, the sources with higher kT either change in both kT and LX together, or

show no significant change in either parameter. There is a likely correlation between the

disk temperature and whether or not the source has optical emission, as the optical spectra

of the low kT sources show optical emission, and a number of the higher kT sources have

no emission beyond the cluster continuum. This study indicates that the sources fall into

three different “classes” (i.e. three different types of physical accretion processes), and most

importantly no evidence of state changes between the different accretion processes. The

low kT sources clearly have a unique and steady accretion mechanism, while the higher kT

sources appear to act like very luminous Galactic BH sources, and the harder power-law

sources have a strong non-thermal component.

92

Slow Decline and Rise of [OIII] Emission in RZ2109: Chapter Four describes

10 years of long-term optical spectroscopic monitoring of RZ2109. RZ2109 is particularly

notable because optical spectroscopy shows it has broad, luminous [OIII]λλ4959,5007 emis-

sion, while also having no detectable hydrogen emission. The flux of the emission line is

significantly lower in recent observations from 2016-2018 than it was in earlier observations

in 2007-2011, and the emission line changes shape over time. Both the core and the wings of

the emission line decline over time, with some evidence that the core declines more rapidly

than the wings. However, the most recent observations (in 2019) unexpectedly show the

emission line core re-brightening. This has implications that the size scale of the emission

nebula may be smaller than previously thought.

RZ2109 and other globular cluster ULXs comprise a very unique population sitting at the

cornerstone of two major fields: globular cluster dynamics, and extreme accretion physics.

Their number and characteristics can inform on the nature of black holes in globular clusters,

and therefore the progenitors to the sources detected by LIGO. The field of ULXs as a whole

can also benefit from multiwavelength studies of these sources, as it is quite exciting that

the old and dynamic globular cluster environment can host such high luminosity sources.

5.2 Future Directions

This work is at the crossroads of two major fields of study: extreme accretion physics and

globular cluster dynamics. Globular cluster ULXs represent a new class of objects accreting

at super-Eddington rates, as well as a probe of the brightest globular cluster black hole

candidate population. The existence of these sources present a number of questions, such

as:

93

How do ULXs in globular clusters compare to ULXs in star-forming regions of galaxies?

The components of star-forming ULXs are likely to be neutron stars with massive, hydrogen

rich donor stars. By contrast, ULXs in globular clusters are likely to be stellar mass BHs

with low mass, metal poor donors. However, little work to date has been done to extensively

compare the X-ray observations of the two types of sources, which is something that will be

able to be accomplished once more GC ULXs have been identified and studied.

How do ULXs in globular clusters associated with spiral galaxies compare to those asso-

ciated with elliptical galaxies? Currently, all known globular cluster ULX sources are found

near elliptical galaxies. Searching for ULXs in globular clusters associated with spiral galax-

ies could have exciting implications for the link between ULXs associated with star-forming

regions of galaxies and ULXs in globular clusters associated with elliptical galaxies.

Do GCs that host ULXs have different structures than those which do not? This ques-

tion is fueled by the physical effects that hosting a black hole is theoretically expected to

have on the structure of the host cluster. Identifying more GC ULXs will enable a study

and comparison of the optical properties of the clusters which host ULXs compared to those

which do not.

All of these questions have long-lasting implications for understanding a new class of

ULXs (and understanding ULXs as a whole), for understanding the dynamics and forma-

tion of BHs in GCs, and for understanding the sources which lead to the BH-BH merger

events detected by LIGO. The key to beginning to answer these questions is to discover new

94

populations of ULXs in globular clusters in both spiral and elliptical galaxies and to carry

out detailed multiwavelength analysis of the sources to probe their high-energy nature and

search for outflows.

95

BIBLIOGRAPHY

96

BIBLIOGRAPHY

Abadie J., et al., 2010, Classical and Quantum Gravity, 27, 173001

Abbott B. P., et al., 2016a, Physical Review Letters, 116, 061102

Abbott B. P., et al., 2016b, , 818, L22

Angelini L., Loewenstein M., Mushotzky R. F., 2001, , 557, L35

Arnaud K. A., 1996, in Jacoby G. H., Barnes J., eds, Astronomical Society of the Pacific

Conference Series Vol. 101, Astronomical Data Analysis Software and Systems V. p. 17

Astropy Collaboration et al., 2013, , 558, A33

Bachetti M., et al., 2013, , 778, 163

Bachetti M., et al., 2014, , 514, 202

Bahramian A., et al., 2017, , 467, 2199

Balbus S. A., Hawley J. F., 1998, Reviews of Modern Physics, 70, 1

Binder B. A., Levesque E. M., Dorn-Wallenstein T., 2018, preprint, p. arXiv:1807.05309

(arXiv:1807.05309)

Blakeslee J. P., Lucey J. R., Barris B. J., Hudson M. J., Tonry J. L., 2001, , 327, 1004

Blakeslee J. P., et al., 2009, , 694, 556

Brassington N. J., et al., 2008, The Astrophysical Journal Supplement Series, 179, 142

Brassington N. J., et al., 2009, The Astrophysical Journal Supplement Series, 181, 605

Brassington N. J., et al., 2010, , 725, 1805

Brightman M., et al., 2018, Nature Astronomy, 2, 312

Broos P., Townsley L., Getman K., Bauer F., 2012, AE: ACIS Extract, Astrophysics Source

Code Library (ascl:1203.001)

Cash W., 1979, , 228, 939

Chatterjee S., Rodriguez C. L., Kalogera V., Rasio F. A., 2017, , 836, L26

Chomiuk L., Strader J., Maccarone T. J., Miller-Jones J. C. A., Heinke C., Noyola E., Seth

A. C., Ransom S., 2013, , 777, 69

Church R. P., Strader J., Davies M. B., Bobrick A., 2017, , 851, L4

97

Clark G. W., 1975, , 199, L143

Clausen D., Eracleous M., 2011, , 726, 34

Clausen D., Eracleous M., Sigurdsson S., Irwin J. A., 2012a, in European Physical Journal

Web of Conferences. p. 01005, doi:10.1051/epjconf/20123901005

Clausen D., Sigurdsson S., Eracleous M., Irwin J. A., 2012b, , 424, 1268

Clemens J. C., Crain J. A., Anderson R., 2004, in Moorwood A. F. M., Iye M., eds, Vol.
5492, Ground-based Instrumentation for Astronomy. pp 331–340, doi:10.1117/12.550069

Connolly S. D., 2016, DELightcurveSimulation: Light curve simulation code (ascl:1602.012)

Dage K. C., Zepf S. E., Bahramian A., Kundu A., Maccarone T. J., Peacock M. B., 2018,

preprint, p. arXiv:1806.01848 (arXiv:1806.01848)

Dage K. C., Zepf S. E., Peacock M. B., Bahramian A., Noroozi O., Kundu A., Maccarone

T. J., 2019, , 485, 1694

Davis J. E., et al., 2012, in Space Telescopes and Instrumentation 2012: Ultraviolet to

Gamma Ray. p. 84431A, doi:10.1117/12.926937

Earnshaw H. M., Roberts T. P., 2017, , 467, 2690

Fabrika S., Ueda Y., Vinokurov A., Sholukhova O., Shidatsu M., 2015, Nature Physics, 11,

551

Feng H., Tao L., Kaaret P., Gris´e F., 2016, , 831, 117

Fruscione A., et al., 2006, in Society of Photo-Optical Instrumentation Engineers (SPIE)

Conference Series. p. 62701V, doi:10.1117/12.671760

F¨urst F., et al., 2016, , 831, L14

Gallo E., Fender R. P., Pooley G. G., 2003, , 344, 60

Gehrels N., 1986, , 303, 336

Giersz M., Askar A., Wang L., Hypki A., Leveque A., Spurzem R., 2019, , 487, 2412

Giesers B., et al., 2018, , 475, L15

Giesers B., et al., 2019, , 632, A3

Gladstone J. C., Roberts T. P., Done C., 2009, , 397, 1836

Greene J. E., Strader J., Ho L. C., 2019, arXiv e-prints, p. arXiv:1911.09678

Hameury J. M., King A. R., Lasota J. P., 1986, , 162, 71

Hanawa T., 1989, , 341, 948

98

Heggie D. C., 1975, , 173, 729

Heggie D. C., Giersz M., 2014, , 439, 2459

Heinke C. O., Ivanova N., Engel M. C., Pavlovskii K., Sivakoff G. R., Cartwright T. F.,

Gladstone J. C., 2013, , 768, 184

Hook I. M., Jørgensen I., Allington-Smith J. R., Davies R. L., Metcalfe N., Murowinski
R. G., Crampton D., 2004, Publications of the Astronomical Society of the Pacific, 116,
425

Hunter J. D., 2007, Computing in Science and Engineering, 9, 90

Hut P., Bahcall J. N., 1983, , 268, 319

Hut P., Paczynski B., 1984, , 284, 675

Hut P., McMillan S., Romani R. W., 1992, , 389, 527

Irwin J. A., Brink T. G., Bregman J. N., Roberts T. P., 2010, , 712, L1

Israel G. L., et al., 2017a, Science, 355, 817

Israel G. L., et al., 2017b, , 466, L48

Ivanova N., Chaichenets S., Fregeau J., Heinke C. O., Lombardi J. C. J., Woods T. E., 2010,

, 717, 948

Jord´an A., et al., 2004, , 613, 279

Joseph T., 2013, PhD thesis, University of Southampton, https://eprints.soton.ac.uk/

354111/

Kaaret P., Feng H., Roberts T. P., 2017, Annual Review of Astronomy and Astrophysics,

55, 303

Kalogera V., King A. R., Rasio F. A., 2004, , 601, L171

Kelly B. C., 2007, , 665, 1489

King A. R., Davies M. B., Ward M. J., Fabbiano G., Elvis M., 2001, , 552, L109

King A., Lasota J.-P., Klu´zniak W., 2017, , 468, L59

Knigge C., Dieball A., Ma´ız Apell´aniz J., Long K. S., Zurek D. R., Shara M. M., 2008, , 683,

1006

Kozai Y., 1962, , 67, 591

Kremer K., Chatterjee S., Ye C. S., Rodriguez C. L., Rasio F. A., 2019, , 871, 38

Kulkarni S. R., Hut P., McMillan S., 1993, , 364, 421

99

Kundu A., Maccarone T. J., Zepf S. E., 2002, , 574, L5

Li Z., et al., 2010, , 721, 1368

Lomb N. R., 1976, , 39, 447

Luo B., et al., 2013, The Astrophysical Journal Supplement Series, 204, 14

Maccarone T. J., Warner B., 2011, , 410, L32

Maccarone T. J., Kundu A., Zepf S. E., 2003, , 586, 814

Maccarone T. J., Kundu A., Zepf S. E., Rhode K. L., 2007, , 445, 183

Maccarone T. J., Long K. S., Knigge C., Dieball A., Zurek D. R., 2010a, , 406, 2087

Maccarone T. J., Kundu A., Zepf S. E., Rhode K. L., 2010b, , 409, L84

Maccarone T. J., Kundu A., Zepf S. E., Rhode K. L., 2011, , 410, 1655

Macri L. M., et al., 1999, , 521, 155

Middleton M. J., King A., 2017, , 470, L69

Middleton M. J., Walton D. J., Fabian A., Roberts T. P., Heil L., Pinto C., Anderson G.,

Sutton A., 2015, , 454, 3134

Miller-Jones J. C. A., et al., 2015, , 453, 3918

Mitsuda K., et al., 1984, , 36, 741

Morscher M., Umbreit S., Farr W. M., Rasio F. A., 2013, , 763, L15

Morscher M., Pattabiraman B., Rodriguez C., Rasio F. A., Umbreit S., 2015, , 800, 9

Osaki Y., 1995, , 47, L11

Pakull M. W., Mirioni L., 2002, arXiv e-prints, pp astro–ph/0202488

Paolillo M., Puzia T. H., Goudfrooij P., Zepf S. E., Maccarone T. J., Kundu A., Fabbiano

G., Angelini L., 2011, , 736, 90

Park D., Kim C., Lee H. M., Bae Y.-B., Belczynski K., 2017, , 469, 4665

Peacock M. B., Maccarone T. J., Kundu A., Zepf S. E., 2010, , 407, 2611

Peacock M. B., Zepf S. E., Maccarone T. J., 2012a, , 752, 90

Peacock M. B., et al., 2012b, , 759, 126

Peacock M. B., Zepf S. E., Maccarone T. J., Kundu A., Gonzalez A. H., Lehmer B. D.,

Maraston C., 2014, , 784, 162

100

Pintore F., et al., 2018, , 477, L90

Plotkin R. M., Markoff S., Kelly B. C., K¨ording E., Anderson S. F., 2012, , 419, 267

Pooley D., et al., 2003, , 591, L131

Poutanen J., Lipunova G., Fabrika S., Butkevich A. G., Abolmasov P., 2007, , 377, 1187

Prada Moroni P. G., Straniero O., 2009, , 507, 1575

Rabolli M., 1990, , 244, 322

Ripamonti E., Mapelli M., 2012, , 423, 1144

Roberts T. P., et al., 2012, , 760, 135

Rodriguez C. L., Chatterjee S., Rasio F. A., 2016, , 93, 084029

Rodriguez C. L., Zevin M., Amaro-Seoane P., Chatterjee S., Kremer K., Rasio F. A., Ye

C. S., 2019, , 100, 043027

Sarazin C. L., Kundu A., Irwin J. A., Sivakoff G. R., Blanton E. L., Randall S. W., 2003, ,

595, 743

Scargle J. D., 1982, , 263, 835

Shakura N. I., Sunyaev R. A., 1973, , 500, 33

Shih I. C., Maccarone T. J., Kundu A., Zepf S. E., 2008, , 386, 2075

Shih I. C., Kundu A., Maccarone T. J., Zepf S. E., Joseph T. D., 2010, , 721, 323

Shishkovsky L., et al., 2018, , 855, 55

Sigurdsson S., Hernquist L., 1993, , 364, 423

Sippel A. C., Hurley J. R., 2013, , 430, L30

Sivakoff G. R., et al., 2007, , 660, 1246

Sivakoff G. R., Jord´an A., Juett A. M., Sarazin C. L., Irwin J. A., 2008, preprint, p.

arXiv:0806.0626 (arXiv:0806.0626)

Spitzer Jr. L., 1969, , 158, L139

Steele M. M., Zepf S. E., Kundu A., Maccarone T. J., Rhode K. L., Salzer J. J., 2011, , 739,

95

Steele M. M., Zepf S. E., Maccarone T. J., Kundu A., Rhode K. L., Salzer J. J., 2014, , 785,

147

Strader J., Chomiuk L., Maccarone T. J., Miller-Jones J. C. A., Seth A. C., 2012a, , 490, 71

101

Strader J., et al., 2012b, , 760, 87

Sutton A. D., Roberts T. P., Middleton M. J., 2013, , 435, 1758

Timmer J., Koenig M., 1995, , 300, 707

Titarchuk L., 1994, , 434, 570

Tody D., 1986, in Crawford D. L., ed., Vol. 627, Instrumentation in astronomy VI. p. 733,

doi:10.1117/12.968154

Tody D., 1993, in Hanisch R. J., Brissenden R. J. V., Barnes J., eds, Vol. 52, Astronomical

Data Analysis Software and Systems II. p. 173

Tudor V., et al., 2018, , 476, 1889

Urquhart R., Soria R., 2016, , 456, 1859

Van Der Walt S., Colbert S. C., Varoquaux G., 2011, preprint, (arXiv:1102.1523)

Van Haaften L. M., Nelemans G., Voss R., Wood M. A., Kuijpers J., 2012, , 537, A104

Walton D. J., et al., 2013, , 779, 148

Whitehurst R., 1988, , 232, 35

Wilms J., Allen A., McCray R., 2000, , 542, 914

Zepf S. E., Maccarone T. J., Bergond G., Kundu A., Rhode K. L., Salzer J. J., 2007, , 669,

L69

Zepf S. E., et al., 2008, , 683, L139

102