MNRAS 521, 4257–4276 (2023) https://doi.org/10.

1093/mnras/stad836
Advance Access publication 2023 March 20

Periodic X-ray sources in the massive globular cluster 47 Tucanae:

Evidence for dynamically formed cataclysmic variables
Tong Bao ,1,2 ‹ Zhiyuan Li 1,2 ‹
and Zhongqun Cheng 3,4 ‹
1 School of Astronomy and Space Science, Nanjing University, Nanjing 210046, China
2 Key Laboratory of Modern Astronomy and Astrophysics (Nanjing University), Ministry of Education, Nanjing 210046, China
3 School of Physics and Technology, Wuhan University, Wuhan 430072, China
4 2WHU-NAOC Joint Center for Astronomy, Wuhan University, Wuhan 430072, China

Accepted 2023 March 16. Received 2023 March 16; in original form 2022 July 27

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

ABSTRACT
We present a systematic study of periodic X-ray sources in the massive globular cluster 47 Tuc, utilizing deep archival Chandra
observations that resolve the cluster core and recently available eROSITA observations that cover the cluster outskirt. By applying
the Gregory-Loredo algorithm, we detect 20 periodic signals among 18 X-ray sources, ranging between 205–95731 s. Fourteen
periods are newly discovered in the X-ray band. We classify these periodic sources into four quiescent low-mass X-ray binaries,
1 ms pulsar, two coronally-active binaries, and eleven cataclysmic variables (CVs) based on their X-ray temporal and spectral
properties, as well as multiband information. Despite a small sample subject to potential selection bias against faint and non-
magnetic CVs, the 11 CVs together define an orbital period distribution significantly different from that of the CVs previously
found in the solar neighbourhood and the Galactic bulge. In particular, there exists in 47 Tuc an apparent paucity of short-period
CVs below the period gap, which might be attributed to a high occupation fraction of non-magnetic CVs. Also characteristic
of the 47 Tuc CVs are an overabundance of long-period CVs with a subgiant donor, a substantial fraction of CVs within the
period gap, and a steep radial surface density profile. These are best understood as a group of CVs having recently formed via
dynamical interactions in the dense cluster core. Despite sufficient sensitivity of the X-ray data, only one periodic source is
found between one-third of the half-light radius and the tidal radius, the nature of which is unclear.
Key words: stars: kinematics and dynamics – globular clusters: individual: (NGC 104) – X-rays: binaries.

Compared to normal stars, close binaries are significantly more

1 I N T RO D U C T I O N
luminous at certain wavelengths, and thus are more easily discerned
As self-gravitationally bound systems evolving with a high stellar even in the dense cluster core, making themselves unique probes for
density, globular clusters (GCs) have long been regarded as the mecca the fundamental dynamical processes in which they are frequently
for close binaries, in particular, low-mass X-ray binaries (LMXBs) involved (Grindlay et al. 2001; Heinke et al. 2003; Edmonds et al.
with an accreting black hole (BH) or neutron star (NS), cataclysmic 2003a; Heinke et al. 2006). In the X-ray band, while the presence and
variables (CVs) with an accreting white dwarf (WD), coronally overabundance of luminous LMXBs (with X-ray luminosities LX 
active binaries (ABs), as well as their potential descendants, e.g. 1035 erg s−1 ) in GCs were already recognized since the Uhuru era
millisecond pulsars (MSPs) and blue straggler stars. Theoretically, (Clark 1975; Katz, Horwitz & Klapisch 1975), it is the Chandra X-ray
the formation and evolution of these exotic objects are strongly Observatory that finally opened up the possibility of resolving weak
affected by or directly resulted from stellar dynamical encounters X-ray sources (with 1030 erg s−1  LX  1034 erg s−1 ) in essentially
(including flybys, tidal captures, and collisions) in the dense GC every Galactic GC (Bassa et al. 2008; Servillat et al. 2008; Haggard,
environment. Moreover, close binaries are more likely to participate Cool & Davies 2009; Lu et al. 2009; Heinke 2010; Lan et al. 2010).
in dynamical encounters than single stars, during which they can This effectively makes available a sizable sample of weak X-ray
release (or sometimes absorb) a substantial amount of kinetic energy sources, ranging from a few to a few hundred sources per cluster,
(Heggie 1975; Hills 1975), thereby playing a crucial role in the for statistical analyses and population studies (Pooley et al. 2002;
gravothermal evolution of the host cluster (Heggie & Hut 2003). The Heinke et al. 2005b; Pooley & Hut 2006; Maxwell et al. 2012).
demography of close binaries in GCs is key to our understanding of Indeed, statistical studies of the weak X-ray sources have been
not only the profound physics of binary formation and evolution, but fruitful. In particular, Pooley et al. (2003) and Pooley & Hut (2006)
also the long-term evolution of GCs and their potential role as the found a positive correlation betwen the number of detected CVs
factory of gravitational waves and seeds of massive BHs. and the stellar encounter rate of individual GCs, which led them to
suggest that the majority of CVs in GCs have a dynamical origin,
similar to the well-established case of LMXBs. On the other hand,
E-mail: [email protected] (TB); [email protected] (ZL); Cheng et al. (2018), who used the cumulative X-ray emissivity (with
[email protected] (ZC) luminous LMXBs subtracted) as a proxy of the abundance of weak

© 2023 The Author(s)

Published by Oxford University Press on behalf of Royal Astronomical Society
4258 T. Bao, Z. Li and Z. Cheng
X-ray sources, found an underabundance rather than overabundance owing to dynamical interactions unique to the GC environment, as
of CVs and ABs in most GCs relative to the Galactic field (see also predicted by theoretical work (e.g. Ivanova et al. 2006; Shara &
Heinke et al. 2020). This was understood as stellar encounters being Hurley 2006). However, a firm conclusion on the role of dynamical
efficient in disrupting a large fraction of primordial, wide binaries effects must await for a statistically representative sample of GC
before they can otherwise evolve into CVs and ABs (Cheng et al. CVs.
2018). This finding is supported by the MOCCA simulations (Belloni This motivates us to build a larger and less biased sample of GC
et al. 2016, 2017), which predicted that detectable CVs in GCs are CVs, and more generally of X-ray-emitting close binaries, which
predominantly composed of CVs formed via the common envelope have a well-determined orbital period. In this work, we focus on 47
phase rather than dynamical interactions. This holds in the latest Tucanae (=NGC 104), which is one of the most massive and best
version of MOCCA simulation (Belloni et al. 2019), which, on the studied GCs in the Galaxy. Utilizing deep Chandra observations with
other hand, also predicts that strong dynamical interactions are able an unparalleled resolving power, as well as the recently available
to trigger CV formation in binaries that otherwise would not have eROSITA observations that cover the cluster outskirt, we conduct for
become CVs. Direct observational evidence for dynamically-formed the first time a systematic search for X-ray periodic sources in 47
CVs in GCs, however, remains scarce. Tuc. It is noteworthy that the resultant sample, like other flux-limited

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

The spatial (radial) distribution of the weak X-ray sources also samples, may suffer from selection bias against short-period CVs,
offers important insight to dynamical processes unique to the GC en- which tend to be fainter than their long-period counterparts in both
vironment, although a meaningful analysis of this kind is necessarily the X-ray and optical bands.
restricted to those GCs with a sufficiently large number of detected This paper is organized as follows. In Section 2, we describe
X-ray sources. In a series of studies, Cheng and collaborators (Cheng the Chandra and eROSITA observations and our data reduction
et al. 2019a, b, 2020a, b) examined the radial surface density profile procedure. In Section 3, we outline the basic principle of the period
of weak X-ray sources in 47 Tuc, Terzan 5, M28, and ω Cen. They searching algorithm and the detailed applications to the Chandra
found a significant dip in the radial profile, which can be understood and eROSITA data. The period searching results are presented in
as the effect of mass segregation, i.e. the X-ray sources as close Section 4, followed by a tentative classification of the periodic X-ray
binaries are on average heavier than the single stars and are more sources based on their temporal, spectral, and multiwavelength prop-
likely to sink into the cluster core through two-body relaxation. erties in Section 5. Section 6 presents the orbital period distribution
Despite the aforementioned observational and theoretical ad- of the classified CVs, compares it with that of the field CVs, and
vances in our understanding of the GC X-ray populations, key addresses important implications on the dynamical formation of GC
questions regarding their origin, evolutionary route, and present- CVs. A summary of our study is given in Section 7.
day abundances remain largely unanswered. In this and subsequent
work, we attempt to shed light on the demography of the GC X-ray
sources, in particular CVs, by examining their temporal properties, 2 DA TA P R E PA R A T I O N
taking advantage of the extensive Chandra data accumulated over the
past two decades. Specifically, close binaries often exhibit periodic 2.1 Chandra observations
modulations in their X-ray emission, which typically arise from The core region of 47 Tuc has been the target of Chandra with its
binary orbital motion and/or spin of the accretor. Knowledge on Advanced CCD Imaging Spectrometer (ACIS) for 19 times. The first
the periodic signal, when complemented by spectral and multiwave- five observations, taken in 2000, were performed with the ACIS-I3
length information, hold promise for revealing the nature of the X-ray CCD on-axis. The following 14 observations, taken between 2002
source. and 2015, were performed with the ACIS-S3 CCD on-axis, among
In the case of CVs, the orbital period, which is controlled by the which 13 were carried out with the subarray mode to minimize
evolution of the donor in response to the mass transfer, is arguably photon pile-up from the bright sources. Table 1 summarizes the basic
the most powerful observational tool for studying CV formation information of the 19 Chandra/ACIS observations (see also table 1 in
and evolution. The standard model of CV evolution predicts the Cheng et al. 2019a for more details). These observations were utilized
existence of a period gap between ∼2–3 h and a period minimum at by Cheng et al. (2019a) to perform the most sensitive detection of X-
∼80 min. These periodic features are related to the dominant angular ray sources in 47 Tuc to date, resulting in a catalog of 537 independent
momentum loss (AML) mechanism, which is responsible for driving point sources, from which we search for periodic X-ray signals.
the CV evolution and shrinking their orbits. The AML is magnetic While these sources are found at a maximum projected distance of
braking for orbital periods Porb  3 h, whereas in short-period CVs 7. 5 from the cluster centre, the majority of them are located within
(Porb ࣠ 2 h), the AML is dictated by gravitational radiation. Systems the cluster’s half-light radius of 3. 17, where the combined ACIS data
within the period gap have their mass transfer highly suppressed and have the highest sensitivity (see fig. 1 of Cheng et al. 2019a).
thus few are observable. These periodic features are well supported We downloaded and uniformly reprocessed the archival data,
by CVs found in the solar neighbourhood (Ritter & Kolb 2003). using CIAO v.4.13 and calibration data files v.4.9.4, and following
Until recently, an orbital period has been found for only 15 CVs the standard procedure1 to obtain the level-2 event file for each
distributed in six GCs (Knigge 2012). Nine of these were found in 47 observation. Astrometry alignments among individual observations
Tuc, thanks to dedicated Hubble Space Telescope (HST) observations were fulfilled by matching the centroids of commonly detected points
(Edmonds et al. 2003a, b, hereafter E03a, E03b). However, three of sources, assisted with the CIAO tool reproject aspect. We used the
the eight periods were later rejected (Rivera Sandoval et al. 2018, observation with the longest exposure (69 ks), ObsID 2738, as the
hereafter RS18). This sample, albeit small and likely biased, exhibits reference frame. We further corrected the photon arrival time of
an orbital period distribution in stark contrast with that of the field each registered event to the Solar system barycenter (i.e. Temps
CVs in the sense that most of these GC CVs have an orbital period Dynamique Barycentrique time) by using the CIAO tool axbary. We
above the period gap, while ∼70 per cent of the field CVs are found
below the gap. This is considered tempting evidence that the GC CV
population is substantially different from the field population, likely 1 http://cxc.harvard.edu/ciao

MNRAS 521, 4257–4276 (2023)

 Periodic X-ray sources in 47 Tuc 4259
Table 1. Chandra and eROSITA observations of 47 Tuc. observations were taken on 2019-11-19, each with a shorter exposure
of 8–9 ks.
ObsID Date R.A. Decl. Live time We downloaded and reprocessed the data using the eROSITA
(1) (2) (3) (4) (5) Science Analysis Software System (eSASS; Predehl et al. 2021).
Chandra We applied the barycentric correction for photon arrival times, based
78 2000-03-16 5.97704 −72.07297 3.87 on the publicly released orbit file for individual observations2 , and
953 2000-03-16 5.97695 −72.07304 31.67 utilizing the barycen task that is part of the HEASoft (v6.29) package.
954 2000-03-16 5.97716 −72.07304 0.85 Unfortunately, orbit file is not available for the very first observation,
955 2000-03-16 5.97696 −72.07294 31.67 obsID = 700012, which may introduce substantial uncertainty in the
956 2000-03-17 5.97700 −72.07282 4.69 photon arrival time. Thus, we discarded this observation and used
2735 2002-09-29 6.07523 −72.08251 65.24 the remaining seven observations for subsequent timing analysis.
2736 2002-09-30 6.07516 −72.08274 65.24 The left-hand panel of Fig. 1 displays the combined field of view of
3384 2002-09-30 6.07515 −72.08263 5.31
the eROSITA observations.
3385 2002-10-01 6.07498 −72.08296 5.31
2737 2002-10-02 6.07491 −72.08330 65.24
Similarly, we generated 0.2–5 keV PSF maps for each observation

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

3386 2002-10-03 6.07480 −72.08349 5.54 using the eSASS task apetool, according to which the source light
2738 2002-10-11 6.07322 −72.08552 68.77 curve and spectrum were then extracted. We adopt the source list
3387 2002-10-11 6.07299 −72.08544 5.73 of Saeedi et al. (2022), which includes 888 sources detected over
16527 2014-09-05 6.01654 −72.07804 40.88 0.2–5 keV and located within a projected distance of 42 arcmin from
15747 2014-09-09 6.01653 −72.07805 50.04 the cluster centre, whose centroid positions have been registered
16529 2014-09-21 6.01989 −72.07845 24.7 to the more accurate Chandra positions provided by Cheng et al.
17420 2014-09-30 6.01989 −72.07840 9.13 (2019a). Due to the moderate angular resolution of eROSITA and
15748 2014-10-02 6.01984 −72.07846 16.24 source crowding in the cluster core, Saeedi et al. (2022) excluded the
16528 2015-02-02 6.01851 −72.08395 40.28
central 1. 7 region, which is fortunately well resolved by the Chandra
eROSITA observations (right-hand panel of Fig. 1). We note that Saeedi et al.
700012† 2019-09-28 6.5339 −72.1763 19.5 (2022) classified 92 background active galactic nuclui (AGNs) and
700011 2019-11-01 5.5180 −71.9862 25.8 26 Galactic foreground stars, chiefly on the basis of tentative optical
700163 2019-11-02 6.5339 −72.1763 25.3 counterparts. These sources are included in our timing analysis for
700013 2019-11-02 6.3338 −71.9249 25.2
completeness.
700014 2019-11-02 5.7108 −72.2375 25.2
700173 2019-11-19 5.5180 −71.9862 8.8
700174 2019-11-19 6.5339 −72.1763 8.4
3 P E R I O D S E A R C H I N G P RO C E D U R E
700175 2019-11-19 6.3338 −71.9249 8.3
Note. (1)–(2) Observation ID and date. (3)–(4) J2000 sky coordinates of the 3.1 The Gregory-Loredo algorithm
telescope aimpoint, in degrees. (5): Effective exposure time, in units of ks.
† This observation is not used due to the lack of satellite orbital information. The Gregory-Loredo (GL) algorithm (Gregory & Loredo 1992),
which belongs to the category of phase-folding methods, applies
have examined the light curve of each observation and confirmed the Bayesian probability theory to evaluate the likelihood of a
that the instrumental background was quiescent for the vast majority periodic model against a model of constant flux and subsequently
of time intervals. Thus, all the science exposures are preserved for determines the most probable period, P. The key of this algorithm is
the subsequent timing analysis, which ensures an uninterrupted light the multiplicity of the phase distribution of events,
curve within each observation.
N!
Point-spread function (PSF) maps corresponding to a predefined Wm (ω, φ) = , (1)
enclosed count radius (ECR), later used for extraction of source light n1 ! n2 ! n3 ! · · · nm !
curve and spectrum, were generated for each observation, over the where N represents the total number of counts, ni (ω, φ) is the number
photon energy range of 0.5–8 keV. All PSF maps were weighted of counts falling into the i-th of m phase bins, given the frequency
by a fiducial spectrum, which is an absorbed bremsstrahlung with m
ω = 2π/P and phase φ, satisfying ni (ω, φ) = N . The multiplicity
a plasma temperature of 10 keV and a column density of NH = i=1
2.3 × 1020 cm−2 , representative of the X-ray sources in the cluster. is the number of ways that the binned distribution could have arisen
We note that a small fraction of the 537 sources catalogued by Cheng by chance and the more ni differ from each other, the smaller the
et al. (2019a) could be more significant in a sub-band (e.g. 0.5–2 keV) multiplicity. In other words, the more the stepwise model defined by
than in the full 0.5–8 keV band. We have examined sub-band light the m phase bins deviates from constant, the more likely there exists
curves but found no extra significant periodic signals. Therefore, we a periodic signal, the probability of which is inversely proportional
will focus on the 0.5–8 keV band for the Chandra-detected sources to the multiplicity. A brief summary in mathematical form of the
in the following. Bayes’s theorem and the GL algorithm can be found in Appendix A
of Bao & Li (2020).
The GL algorithm has recently been applied to 1-Ms Chandra ob-
2.2 eROSITA observations
servations of a Galactic bulge field (the so-called Limiting Window;
As one of the calibration targets, 47 Tuc was observed by eROSITA Bao & Li 2020) and the 7-Ms Chandra deep field south (Bao & Li
with the pointing mode for eight times shortly after its launch. The 2022) to detect periodic X-ray sources. These studies demonstrated
basic information of these observations are listed in Table 1. Saeedi the reliability and efficiency of the GL algorithm for probing periodic
et al. (2022) carried out source detection over the combined field of
view of the first five observations, taken on 2019-09-28, 11–01, and
11–02 and having individual exposures of 20–26 ks. Three additional 2 https://erosita.mpe.mpg.de/edr/eROSITAObservations/OrbitFiles

MNRAS 521, 4257–4276 (2023)

4260 T. Bao, Z. Li and Z. Cheng

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

Figure 1. Left: 0.2–5 keV counts image of the 47 Tuc field, combining 7 eROSITA observations. The dashed red box outlines a 2 × 2 arcmin region containing
the cluster core, which is unresolved due to source crowding and the limited angular resolution of eROSITA. The yellow circle represents the half-light radius of
3. 17. The only eROSITA source with a significant periodic signal is marked by a cyan circle and labelled following the source ID of Saeedi et al. (2022). Right:
0.5–8 keV Chandra/ACIS image of the 2 × 2 arcmin region outlined in the left-hand panel. The orange circle represents the core radius of 0. 36, and the cluster
centre is marked by the ‘+’ sign. Seventeen Chandra sources with a significant periodic signal are marked by magenta circles and labelled following the IDs of
Cheng et al. (2019a).

signals in X-ray sources with a moderate number of counts and identified, a second search is performed excluding a narrow interval
sampled by an irregular observing cadence, which is typical of X- around the identified period, which ensures that a possible second
ray observations. Here, we follow the recipe outlined in Bao & Li period within the same period searching range will not be missed.
(2020) to perform a systematic search for periodic signals from the X- The chosen period ranges are optimal for detecting the orbital
ray sources in the Chandra and eROSITA fields of 47 Tuc. We adopt and spin periods of CVs. The orbital period distribution of CVs is
a probability of 90 per cent as the threshold for tentative periods known to exhibit a minimum at ∼82 min and a gap between ∼2–3 h
returned by the GL algorithm. (Knigge, Baraffe & Patterson 2011). CVs typically have an orbital
period shorter than 10 h in order to meet the Roche lobe filling
condition, but in rares cases which involve an evolved donor, the
3.2 Application to the Chandra & eROSITA data orbital period can be significantly longer (Kalomeni et al. 2016). The
For each Chandra source, we extract the 0.5–8 keV counts within the second and third period ranges cover these characteristic periods,
90 per cent ECR of individual observations to form a time series as whereas the first range probes the spin period of fast rotating WDs in
input to the GL algorithm. In the crowding cluster core, the default 90 IPs. We note that ABs and LMXBs, as well as CVs in rare cases, can
per cent ECR of some neighbouring sources overlap; for such cases, have an orbital period longer than 50 ks. However, the current X-ray
we adopt the 75 per cent ECR for source count extraction. Since the data of 47 Tuc, with individual exposures ࣠70 ks (Table 1), are not
GL algorithm manipulates each photon arrival time to evaluate the optimal for detecting such long periodic signals without facing false
probability of periodic variation against a constant model, there is alarms due to long-term aperodic variability often present in these
no need to separately account for the background level, which is X-ray sources (see more discussions below).
absorbed into the presumed constant. Nevertheless, we estimate the As for the eROSITA sources, time series are extracted from within
background level for each source by extracting counts from within the 75 per cent ECR, which is sufficiently large to include most source
a concentric annulus with inner-to-outer radii of 2–4 times the 90 counts and also sufficiently small to avoid source overlapping, given
per cent ECR, masking any pixel falling within two times the 90 the large PSF of eROSITA compared to that of Chandra. It turns
per cent ECR of neighbouring sources. out, however, most eROSITA sources have a significantly varied 75
By design, the GL algorithm folds the time series at trial fre- per cent ECR from observation to observation, due to their varied off-
quencies (periods). The resolution and range of searched period axis positions in the seven observations, which have largely separated
is compromised between efficiency and computational power. Fol- aimpoint positions (Fig. 1 and Table 1). The increased ECR at large
lowing Bao & Li (2020), we restrict our search in three period off-axis angles necessarily captures more background counts into the
ranges: (100, 3000), (3000, 10000), and (10000, 50000) s, with a source aperture, resulting in an artificially time-variable light curve
frequency resolution of 10−7 , 10−8 , and 10−9 Hz, respectively. Given (see Fig. 2 for an example). We note that for most of the Chandra
the timespan of ∼5 × 108 s between the first and last Chandra sources, the background level is too low to cause a variable light
observations (Table 1), the chosen frequency resolutions are optimal curve. As noted by Bao & Li (2020), a light curve with substantial
for an efficient search of periodic signals. Since the GL algorithm aperiodic variability fed to the GL algorithm may result in a fake
only determines the most probable period, once a tentative period is signal, typically at a period comparable to the time-scale of the

MNRAS 521, 4257–4276 (2023)

 Periodic X-ray sources in 47 Tuc 4261

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

Figure 2. Upper left: The eROSITA 0.2–5 keV light curve of Src-No.481 binned by 1000 s. The grey vertical-dashed lines label the gap between observations,
whereas the green strip represents the local background, the width of which represents 1σ Poisson error. Upper right: Phase-folded light curve at the period
determined by the GL algorithm. The green-dashed line represents the mean count rate. Lower left: The eROSITA background-subtracted 0.2–5 keV light curve,
overlaid by a sinusoidal curve at the most probable period determined by the LS periodogram. Lower right: Normalized LS periodogram of Src-No.481. Three
false alarm levels are denoted by the coloured dotted lines.

variation, if too many photons happened to fall within a small fraction Given the expectation that 47 Tuc may host abundant NS-LMXBs,
of phase bins. Thus, we only analyse time series from the seven we have also extended our period searching to the range of (10,
individual observation. For the four ∼25 ks-long observations, we 100) s, with a resolution of 10−6 Hz, to probe pulsating signal from
choose (100, 3000) and (3000, 10000) s as the period searching range, an NS, However, no significant signal in this period range is found
with a frequency resolution of 10−7 Hz, while for the three ∼8 ks- in either Chandra or eROSITA sources.
long observations, only the range of (100, 3000) s is searched for
periodic signals. We also apply the GL algorithm to the combined
4 P E R I O D S E A R C H I N G R E S U LT S
time series from all seven observations to search for signals in the
range of (10000, 50000) s.
4.1 Result of Chandra sources
To provide a reliability check of any long-period signal reported
by the GL algorithm on the eROSITA data, we further apply the From the time series of the 537 Chandra sources, the GL algorithm
generalized Lomb-Scargle (hereafter LS) periodogram (Lomb 1976; reports 41 candidate periodic signals having a probability greater
Scargle 1982), with a normalization of sample variance following than 90 per cent. However, some of these signals may be spurious
Zechmeister & Kürster (2009). The LS periodogram probes periodic due to one of the following reasons:
signal from a background-subtracted light curve, hence it suffers
little from the aforementioned effect of strongly varied PSF. The (i) The Chandra/ACIS operates in a dithering pattern with a period
LS periodogram has its own drawback, in the sense that it does of 706.96 s in pitch and 999.96 s in yaw4 to distribute the photons
not perform well with interrupted light curves having wide gaps in over more CCD pixels. Any signals detected at these two periods or
between. Nevertheless, the four 25-ks eROSITA observations were their harmonics to within 1 per cent are considered spurious. We thus
taken nearly continuously. Therefore, we apply the LS periodogram exclude nine such spurious signals, all found in sources located near
to background-subtracted light curves extracted from these four CCD gaps or edges.
(ii) As mentioned before, an aperiodically variable light curve may
observations, which have a combined length of ∼100 ks. The
fool the GL algorithm to report a false period. For each source with
background count rate in each of the four observations is estimated
a candidate period, we inspect its light curve to identify short-term
from the smoothed background image generated by the eSASS tool
(i.e. intra-observation) flares. The situation happens in four sources,
erbackmap3
producing six fake periodic signals mainly around the flare duration
We apply the false alarm probability proposed by Baluev (2008)
or the length of the affected observation. Moreover, for sources
to assess the significance of the peak in the LS periodogram (i.e.
exhibiting strong long-term (i.e. inter-observation) variations, we
a tentative period), the robustness of which was demonstrated by
repeat the period search in two subsets of the light curve: one covering
Bao & Li (2022). A valid detection is defined as having a false alarm
only the high state (defined as the observation[s] with the highest
probability lower than 0.27 per cent, i.e. a 3σ significance.
photon flux) and the other excluding the high state. If the periodic

3 https://erosita.mpe.mpg.de/edr/DataAnalysis/erbackmap doc.html 4 https://cxc.harvard.edu/ciao4.4/why/dither.html

MNRAS 521, 4257–4276 (2023)

4262 T. Bao, Z. Li and Z. Cheng
signal could not be reproduced in either subset, we consider it a false periodogram, along with a sinusoidal curve with P = 14366.89 s,
detection. In this way we exclude seven tentative periods, all above which provides a reasonable characterization to the observed light
10 h. curve, in particular being able to match all but one of the alternating
peaks. This suggests that the 7–8 ks periodic signal found in the
Moreover, we note a well known ambiguity in distinguishing the
individual observations is more likely due to a harmonic at half of
true period and its harmonics and subharmonics, i.e. integer division
the true period, which is taken to be 14388.49 s. It is noteworthy that
or multiplication of the true period5 . In principle, the light curve can
although this source is also detected by Chandra (Seq.503 in Cheng
tell the true period only when we fully understand the mechanism
et al. 2019a), the number of ACIS-detected counts is insufficient to
behind the periodic variability. For example, polars often exhibit
recover the periodic signal found in the eROSITA data.
single-peaked structure in their light curves due to the accretion
As illustrated by Fig. 1, all 17 Chandra sources with a periodic
dominated by one of the two magnetic poles. In this case, the
signal are located within the central 1. 7 region of 47 Tuc, where
harmonics can be recognized by a double-peaked shape from the
eROSITA can barely resolve individual sources due to source crowd-
phase-folded light curve. On the other hand, a double-peaked shape
ing. Nevertheless, for the Chandra periodic sources, it is still possible
modulated by the true period may exist in the light curve of IPs,
to confirm the periodic signal with the eROSITA data, provided that

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

in which case the accretion-induced X-ray emission can arise from
background noise does not lead to a complete dilution. Hence, we
both magnetic poles (Bao & Li 2020). With this caveat in mind,
extract eROSITA time series from the positions of the 17 Chandra
we have inspected the likelihood of the detected periods being (sub-
sources within a 50 per cent ECR and run the GL algorithm. Only
)harmonics of the true period on a best-effort basis, assisted with
one source, Seq.162, with a ∼31 200 s signal, is recovered by the
literature information typically derive from optical and ultraviolet
eROSITA data. Fig. 4 shows the phase-folded light curves from both
observations (see details in Section 5.2). This has led us to recognize
telescopes, in which the substantially higher background level of
that one periodic signal, found in Seq.229 (source sequence number
eROSITA is obvious. The periodic signal is clearly due to an eclipse,
following Cheng et al. 2019a), is in fact an integer division of the true
during which the Chandra-detected flux completely vanishes. The
period at 95371.20 s, which was previously unambiguously identified
eROSITA light curve, after accounting for its different chronological
from orbital eclipse in this source (Knigge et al. 2003). Due to
zero-point6 with respect to that of Chandra, finds perfect phase-
the aforementioned limitation in the period searching ranges, the
matching with the Chandra light curve, despite a separation of
GL algorithm originally reports a period at about one-third of the
over 4720 phase cycles between the two. This consistency lends
true value. Once we expand the searching range, the GL algorithm
strong support to the validity of our period searching process with
recovers the true period, which is clearly due to orbital eclipse (Fig.
the eROSITA data.
3). We find no strong evidence for a confusing (sub)-harmonic in the
To summarize, we have identified 18 X-ray sources with a total of
other sources.
20 periodic signals, based on the data of the two telescopes. Phase-
After the above filtering, there remain 19 valid periodic signals
folded light curves of these sources, except for Src-No.481 (Fig. 2)
found among 17 Chandra sources, among which two sources each
and Seq.162 (Fig. 4), are provided in Fig. 3.
exhibit dual periods. The basic information of these sources are
summarized in Table 2, in the order of increasing periods. All but
two periods have a GL probability greater than 99 per cent (all above 4.3 Potential caveat due to red noise
96 per cent). The positions of these sources are indicated in the right-
Accretion-powered systems, such as CVs, LMXBs, and active galac-
hand panel of Fig. 1. Classification of these sources will be addressed
tic nuclei (AGNs) are known to exhibit aperiodic variability on a wide
in Section 5.
range of time-scales (Linnell 1950; Rappaport, Doxsey & Zaumen
1971; Oda et al. 1974). The so-called red noise, a main component
4.2 Result of eROSITA sources of aperiodic variation, may cause false periodic signal especially at
lower frequencies (Warner 1989). We estimate the probability of the
As stated in Section 3.2, the application of the GL algorithm to the
GL algorithm being false-alarmed by aperiodic variation, using a
eROSITA data is restricted to either single observations for periodic
similar procedure as in Bao & Li (2022). Specifically, we first adopt
signals shorter than 10 ks, or the time series combining all seven
an analytical model of the source power spectrum, which has the
observations for periodic signals between 10–50 ks. The latter case
form of,
also requires confirmation with the LS periodogram. 
In the first case, after filtering spurious signals due to strong  4 −1/4
ν
aperiodic variability, only one source, Src-No.481 in Saeedi et al. P (ν) = N ν −1 1 + + CP , (2)
ν0
(2022), is identified by the GL algorithm, which exhibits a periodic
signal at a period of ∼6792 and ∼7825 s in the observation of ObsID or
700 163 and ObsID 700013, respectively. A periodic signal around P (ν) = N ν −α + Cp . (3)
7 ks is also evident in the other two 25-ks observations, but with a
lower significance. In the latter case, only one source, again Src- equation (2) was proposed by Revnivtsev et al. (2010) for describing
No.481, is identified by both GL and LS to exhibit a period of the power spectra of CVs (Balman & Revnivtsev 2012), while
14388.49 and 14366.89 s, respectively. The upper panel of Fig. 2 equation (3) is suitable for the power spectra of LMXBs. Here N
displays the raw light curve fed to the GL algorithm, along with the is the normalization factor, α is the spectral index, and CP represents
phased-folded light curve according to P = 14388.49 s, and the lower the Poisson noise determined by the mean photon flux of the source.
panel displays the background-subtracted light curve fed to the LS The break frequency ν0 is usually 1 mHz for CVs and cannot be
well constrained by the current data due to the presence of Poisson
5 Itis noteworthy that harmonics are conventionally defined in frequency
space. For convenience and not losing clarity, throughout this work we use 6 https://erosita.mpe.mpg.de/edr/DataAnalysis/prod descript/header-keywo
the term harmonics on the period space. rds edr.html

MNRAS 521, 4257–4276 (2023)

 Periodic X-ray sources in 47 Tuc 4263

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

Figure 3. The 0.5–8 keV phase-folded light curve at the modulation period. The green-dashed line represents the mean count rate, whereas the green strip
represents the local background, the width of which represents 1σ Poisson error.

MNRAS 521, 4257–4276 (2023)

 4264

Table 2. Basic information of the periodic X-ray sources in 47 Tuc.

MNRAS 521, 4257–4276 (2023)

Seq RA DEC R P σP Prob. Popt C CB FAP Eclipse Class Note
(◦ ) (◦ ) arcsec s per cent s s counts per cent
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)

217 6.01780 −72.08281 8.5 205.02 0.02 0.9999 – 33 611 144.7 0.0 No LMXB W42, X9, UV
T. Bao, Z. Li and Z. Cheng

414† 6.06625 −72.07680 49.9 3846.15 0.02 0.9982 – 2620 20.8 0.0 No CV W2, X13, UV
185 6.01063 −72.08649 23.6 8517.89 0.10 0.9987 – 966 35.8 0.0 Yes CV W53, UV
414‡ 6.06625 −72.07680 49.9 8646.78 <0.01 1.0000 8649.00 2620 20.8 14.3 Yes CV W2, X13, UV
366 6.04485 −72.07383 35.7 10311.94 1.17 1.0000 10304.93 1075 16.1 1.7 Yes CV W8, UV
423 6.07077 −72.07424 58.1 10384.22 0.35 0.9986 20779.20 2231 17.3 0.0 Yes CV W1, UV
232† 6.02075 −72.08095 3.4 11112.50 0.68 0.9999 – 1664 132.9 0.0 Yes LMXB W37
263 6.02533 −72.08032 3.9 11486.36 0.66 0.9662 11486.02 569 70.4 0.0 No LMXB W29, MSP
e481 6.18225 −72.13861 270.9 14366.89 1.96 1.0000 – 784 372.6 43.4 No MSP Seq.503
331 6.03540 −72.08345 15.2 15232.63 1.44 0.9849 15240.96 828 20.2 0.0 Yes CV W15, UV
273 6.02666 −72.07863 10.1 16824.25 0.16 1.0000 13824.00 6218 30.2 95.6 No CV W27, X10, UV
317 6.03316 −72.08192 10.8 23583.11 1.48 0.9945 – 629 22.6 0.0 No CV W20
162 6.00405 −72.08147 21.7 31200.27 0.17 1.0000 – 28 020 69.3 100.0 Yes LMXB W58, X5
252 6.02362 −72.08036 3.3 44642.86 1.28 0.9999 22947.84 1001 152.5 0.6 No CV W32
232‡ 6.02075 −72.08095 3.4 44883.30 1.27 1.0000 – 1664 132.9 65.2 Yes LMXB W37
283 6.02860 −72.08681 20.7 45310.38 2.14 0.9999 34029.50 287 28.2 0.4 No AB W26
290 6.029851 −72.09607 53.7 46082.95 1.88 0.9912 – 1272 13.7 1.2 Yes CV W25, UV
198 6.01452 −72.08482 16.2 46151.01 1.56 1.0000 45835.20 3517 47.7 0.8 No AB W47, E8
312 6.03259 −72.07823 14.8 48780.49 0.15 0.9999 22242.82 1618 22.0 10.4 No CV W23, UV
229 6.02057 −72.08210 4.5 95731.20 1.93 1.0000 95731.20 1638 244.1 3.9 Yes CV W36, AKO-9, UV
Note. (1) Source sequence number taken from Cheng et al. (2019a), except for the eROSITA source e481, which is from Saeedi et al. (2022). The same source with dual periods is marked by † and ‡. (2) and (3)
Right Ascension and Declination (J2000) of the source centroid. (4) The projected distance from the cluster centre. (5) The modulation period determined by the GL algorithm. Newly detected periods in the X-ray
band are highlighted in bold face. (6) The estimated relative uncertainty of the period, expressed in percentage. (7) The GL probability. (8) Period found in the optical and reported by E03a, Albrow et al. (2001),
or RS18. (9) The number of total counts in the 0.5–8 keV band. (10) The number of estimated background counts. (11) False alarm probability due to potential red noise, as described in Section 4.3. (12) The
presence/absence of eclipsing behaviour. (13) Tentative source classification. (14) Notes for alias (mainly from Grindlay et al. 2001), possible UV counterpart RS18 and possible MSP counterpart.

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

 Periodic X-ray sources in 47 Tuc 4265

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

Figure 5. The power spectral distribution (blue histogram), normalized to
(rms/mean)2 per Hz of source Seq.312 in ObsID 2738. The best-fit model
and 1σ error are presented by the black solid line and the grey strip, while
the theoretical Poisson noise level is marked by the orange-dashed line.

algorithm, with the same period searching range as applied in

Section 3.2. A group of 1000 simulated time series are produced
for each source to evaluate the false alarm probability (FAP) resulted
from red noise, i.e. the fraction of these 1000 time series leading to a
Figure 4. Upper panel: The phase-folded light curve of source Seq.162, false detection. We define a false detection as a signal reported with
from the Chandra observations. Lower panel: The phase-folded light curve a GL probability above 0.99, which is consistent with the majority of
at the same period from the eROSITA observations. The green-dashed line the deteted periodic signals (Table 2). To be conservative, we refer
represents the mean count rate, whereas the green strip denotes the local to a ‘global’ FAP, that is, all reported signals falling in a given period
background, the width of which represents 1σ Poisson error. The phase of searching range are taken as false detections. Except for Seq.414 and
both light curves has been registered to the common time-zero point of MJD Seq.232, both are detected with dual periods in the same range, we
= 51543.875.
divide the given period searching range to two parts to product FAPs
for each periods.
noise. Thus, we fix ν0 at 1 mHz, which has little effect in the results. The resultant FAP for each detected periodic signal is reported in
To mitigate the potential effect of interrupted observations in the Table 2. A low FAP (4 per cent) is found for most periodic signals,
Fourier analysis, we take the longest Chandra observation, i.e. ObsID indicating that they are unlikely false detections due to red noise. Six
2738, to characterize the power spectrum for each of the 18 periodic signals (Src.-No.481, Seq.273, Seq.162, Seq.312, the longer period
sources. We have also fitted the power spectrum using the other long of Seq.414, and the longer period of Seq.232), on the other hand,
observations (ObsIDs 2735, 2736, 2737), but found no significant have an FAP above 10 per cent. However, it should be emphasized
difference for most sources. The only exceptions are Seq.162 and that a high FAP is not a sufficient condition for a fake periodic signal.
Seq.273, which show a varying power spectrum among the different For instance, Seq.162 has the highest FAP (100 per cent) due to its
observations. Nevertheless for uniformity, we stick with ObsID 2738 strong red noise, which results in fake signals at low frequencies
to constrain the power spectrum of these two sources, which has little (long periods), but its detected period is clearly a true period arising
effect in the following estimation of the false alarm probability. We from orbital eclipse (Fig. 4). Similarly, the phase-folded light curve of
have also masked a narrow frequency window around the detected the longer period of both Seq.414 and Seq.232 exhibits an eclipsing
period of a given source. As for Src-No.481, we characterize its behaviour, thus the high FAP does not preclude the reality of these
power spectrum using all four long eROSITA observations, because two periods. In Appendix A, using simulated light curves, we further
the observation gap is negligibly small. demonstrate that even a moderate periodic variation can be readily
The power spectrum of a given source is fitted (with equation (2) or detected by the GL algorithm against strong red noise.
equation (3) dependent on the preferred source class; see Section 5), The remaining three sources with a high FAP do not exhibit
using the maximum likelihood function discussed in Vaughan (2010) an eclipse that otherwise ascertain the reality of the detected
and the Markov Chain Monte Carlo approach (with the python emcee period. However, an inspection of the inter-observation light curve
package, Foreman-Mackey et al. 2013) to determine the best-fit of Src-No.481 clearly identifies a peak-to-peak periodic variation
parameters and errors. Then we simulate the time series based on (Fig. 2), which is captured by both the GL algorithm and the
the best-fit model, following the method proposed by Timmer & LS periodogram. Similarly, the inter-observation light curves of
Koenig (1995). An example of the fitted power spectrum is shown in Seq.273 and Seq.312 clearly reveal periodic variations (see Fig. B1).
Fig. 5. Therefore, the detected period in these three sources are probably
For each source, a photon time series is simulated for the same genuine despite their substantial red noise and the relatively high
epochs of the actual observations, which are then fed to the GL FAP resulted. Nevertheless, we caution that a periodic candidate

MNRAS 521, 4257–4276 (2023)

4266 T. Bao, Z. Li and Z. Cheng
reported by the GL algorithm with a high, simulation-based FAP 20 counts and a signal-to-noise ratio (S/N) greater than 2 per bin.
could be a spurious detection caused by substantial aperoidic noise. The resultant spectra are analysed using XSPEC v12.12.0.
Moreover, considering the large number of sources searched for Anticipating that most, if not all, of the periodic sources are
periodic signals, a substantial possibility cannot be ruled out that LMXBs, CV or ABs, we adopt a fiducial phenomenological spectral
some of the periodic signals with a moderate FAP are also spurious. model, which consists of a bremsstrahlung continuum and an
Additional information about their reality (e.g. eclipsing behaviour unknown line-of-sight absorption (tbabs in XSPEC) to fit the spectra.
or optically-confirmed period), as argued for the above sources, are In principle, the X-ray spectra of CVs and ABs are dominated by
warranted. a collisionally ionized plasma, and thus expected to show metal
Our approach in the above has focused on the actual observational emission lines (Xu, Wang & Li 2016). However, essentially all 18
interval to ensure that the variability among different observations sources exhibit a featureless spectrum, which is understood as due
are preserved. This neglects potential red noise leakage from lower to the moderate spectral S/N and the fact that 47 Tuc has a very
frequencies (i.e. on time-scales longer than the length of the observa- low metallicity, [Fe/H] = −0.76 (Harris 1996; 2010 ed.). It turns out
tions), which might result in an underestimation of the FAP. To assess that the absorbed bremsstrahlung model provides a reasonable fit to
this potential effect, we expand the simulated time interval for those all spectra except for five cases, for which we employ a different

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

sources with a long period (P  6 h), which are more likely to be spectral model. Seq.263, Seq.273, and Seq.252 are better modelled
affected by red noise leakage. For each of these sources, we simulate by a two-bremsstrahlung model. Seq.229, known as a WD accreting
long light curves from the assumed power spectrum that extends to from a subgiant companion (Knigge et al. 2003), is well-fitted by
very low frequencies, and then randomly extract a segment to match an absorbed cooling flow model (mkcflow in Xspec; Mushotzky &
the actual observational length based on which we re-evaluate the Szymkowiak 1988) with an intrinsic partial absorption (pcfabs in
FAP. It turns out that the FAP of these long-period sources does not Xspec; Mukai et al. 2003). Seq.217, suggested to be a BH-LMXB
become significantly higher. This is consistent with the suggestion (Bahramian et al. 2017), shows the most peculiar spectrum with the
by Vaughan et al. (2003) that the amount of red noise leakage from highest S/N among all sources. Instead of a detailed modelling, we
low frequencies is usually negligible for a flat power spectrum (with simply adopt the best-fit model from Bahramian et al. (2017), which
a slopes α < 1.5), which is the case for most of our long-period consists of an optically-thin plasma (vmekal in Xspec) and a disc
sources. reflection component (pexrav in Xspec) to describe this spectrum.
The spectra and best-fit models of the 18 periodic sources are
shown in Fig. 6. Spectral fit results are summarized in Table 3,
4.4 Uncertainty of the derived periods including an unabsorbed 0.5–8 keV luminosity based on the best-fit
model and an assumed distance of 4.5 kpc for 47 Tuc (Harris 1996;
The GL algorithm provides the probability of periodic variation by
2010 ed.).
integrating odds ratio over the period searching range, identifying the
period (frequency) as that with the highest odds ratio. However, the
frequency-dependent odds ratio itself does not quantitatively reflect 5.2 Classifying the periodic sources
the uncertainty of the detected period, which should be related to
Now, we attempt to classify the periodic X-ray sources based
the statistical noise of a given time series. To provide a quantitative
primarily on their temporal and spectral properties. In principle,
estimate of the uncertainty in the derived period, we have generated
the nature of the periodic X-ray sources can also be inferred from
simulated time series of periodic variations (using both sinusoidal
their optical/UV counterparts. However, it is notoriously difficult
and piecewise functions; see Appendix A and section 3.3 in Bao &
to distinguish the genuine counterpart among the numerous GC
Li 2020) with an additional red noise, which are fed to the GL
sources, especially in the dense core. Nevertheless, we can rely on
algorithm to yield a period detection. The input period, the variation
the periodicity determined by multiwavelength studies, as a further
amplitude, and the total length are approximately the same as the
confirmation of both the X-ray periodic signal and the suggested
actual cases. Then, for a given detected period, we take the standard
optical/UV counterpart. For this purpose, we mainly consult Albrow
deviation of 100 detections based on the simulated light curves to
et al. (2001) and E03a, E03b, which conducted systematic timing
characterize the relative uncertainty, σ P , as presented in Table 2. We
analysis for many tical/UV sources in the core of 47 Tuc, and found
caution that this uncertainty should be treated as an approximation
a number of candidate periods, based on high-resolution, multi-epoch
rather than a mathematically rigorous error.
HST observations.
Our classification and reasoning for each source are in order below.
Seq.217 (X9 or W42): This source was first suggested to be an IP
5 S P E C T R A L P RO P E RT I E S A N D S O U R C E
upon the detection of a 218-s periodic modulation in the X-ray light
C L A S S I F I C AT I O N
curve (Grindlay et al. 2001), considered to be a spin modulation.
Here based on much deeper Chandra observations, we detect a
5.1 X-ray spectral properties of the periodic sources
periodic signal at P = 205.02 s, with a very small uncertainty (0.02
Before trying to classify the detected periodic sources, we examine per cent) that makes it statistically distinct from the 218-s period.
their spectral properties, which can hold important information about Unfortunately, Grindlay et al. 2001 did not provide any detail on
their nature. The source and background spectra are extracted from how the 218-s signal was determined. Nor could we reproduce this
the aperture described in Section 3.2, by using the CIAO tool signal using the same observations (in a total exposure of 70 ks) with
specextract, or srctool for the eROSITA source Src-No.481. Two the GL algorithm. The phase-folded at 205 s light curve shows a sine-
sources, Seq.232 and Seq.198, are caught by some observations to like shape with a mild (∼10 per cent) fractional variability. Recently,
exhibit a high or bursting state. Thus, we extract two spectra for each Miller-Jones et al. (2015) argued that the high X-ray luminosity
of these two sources to represent the quiescent and bursting states. (1033 erg s−1 ), hard X-ray spectrum, and high radio-to-X-ray flux
Further, the spectra are adaptively binned over 0.5–8 keV (0.2–5 keV ratio of X9 are inconsistent with a CV, but are more compatible
for the eROSITA spectrum of Src-No.481) to achieve a minimum of with a BH-LMXB, or possibly a transitional MSP (deemed unlikely

MNRAS 521, 4257–4276 (2023)

 Periodic X-ray sources in 47 Tuc 4267

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

Figure 6. Source spectra with the best-fit models. The source sequence number and spectral model are denoted in each panel. Two sources, Seq.232 and
Seq.198, have two sets of spectra, respectively, from the quiescent (black) and bursting (red) states.

MNRAS 521, 4257–4276 (2023)

4268 T. Bao, Z. Li and Z. Cheng
Table 3. X-ray spectral properties of the periodic sources. (Knigge et al. 2008; Miller-Jones et al. 2015). Further, studies are
warranted to understand its true nature.
Seq NH Tb χ 2 (d.o.f) L0.5-8 Seq.414 (X13 or W2): This source was previously reported to
1022 cm−2 keV 1031 erg s−1 have multiple optical periods (2.2, 5.9, and 8.2 h) and an X-ray
(1) (2) (3) (4) (5) period of 6.287 h (E03b). The latter signal is not recovered in our
217 0.15+ 0.02
−0.02
a 0.17+0.01
−0.02 1.30 (229) 228+ 16
−12
analysis based on the total Chandra data set, nor is it found by the
414 0.10+ 0.03
7+ 2
1.04 (133) 12.5+ 0.4 GL algorithm in the single observation studied by E03b. Hence,
−0.03 −1 −0.4
+5 we consider this signal a false alarm. Also all the reported optical
185 0.023 (fixed) 9−3 0.86 (80) 3.7+ 0.3
−0.3 periods are suggested to be false alarms by recent work (RS18). We
366 0.7+ 0.1
−0.1 40 (fixed) 1.05 (102) 8.7+ 0.4
−0.5 detect two periods with this source, a shorter period at 3846.15 s and
+0.03
423 0.04−0.03 4.6+ 1.2
−0.8 1.11 (113) 8.6+ 0.3
−0.3 a longer period at 8646.78 s. It is obvious from the phase-folded
232 0.22+ 0.09
−0.08
+0.2
0.8−0.2 1.08 (84) 3.0+ 0.2
−0.2 light curve that the latter arises from an orbital eclipse. Recent
232† 0.06+ 0.06
−0.05 0.55+ 0.11
−0.09 1.08 (84) 14.6+ 0.9
−0.9
work by Israel et al. (2016) also found the same eclipsing period
263 0.023 (fixed) b 0.4+0.3 (40, fixed) 1.27 (49) 2.1+ 0.2 at ∼8649 s from essentially the same Chandra dataset. Notably,

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

−0.2 −0.2
e481 0.08+ 0.08
0.06+ 0.01
1.00 (16) 0.7+ 0.3 a second shallower dip is present in the phase-folded light curve,
−0.04 −0.01 −0.2
+0.2 +10 which is separated from the main dip by ∼1/2 phase. This raises
331 0.7−0.1 13−4 0.70 (78) 6.4+ 0.4
−0.4
b 0.26+0.02 (20+25 )
the possibility that the true period is 50 per cent less. On the other
273 0.023 (fixed) −0.02 −8 1.18 (141) 21.9+ 0.8
−0.8 hand, RS18 found an eclipsing behaviour in the optical band at the
317 0.17+ 0.07
−0.06 1.6+ 0.5
−0.4 0.72 (44) 2.0+ 0.1
−0.1 period of ∼8649 s given by Israel et al. (2016). Hence, we adopt
+0.01
162 0.16−0.01 0.67+ 0.02
−0.02 1.44 (109) 102+
−1
1
8646.78 s as the true period. The period of 3863.77 s is likely due
252 0.023 (fixed) b 0.5+0.2 (19+35 ) 0.78 (77) 3.9+ 0.2 to spin modulation, which is supported by the sine-like shape of the
−0.2 −9 −0.2
283 0.2+ 0.1
−0.1 0.9+ 0.4
−0.3 1.02 (26) 0.93+ 0.09
−0.09
corresponding phase-folded light curve. Therefore, we classify this
290 +0.03
0.05−0.03 4.4+ 1.0
0.90 (85) 6.4+ 0.3 source as an IP. The presence of a UV counterpart and the relatively
−0.4 −0.2
high luminosity LX ∼ 1.2 × 1032 erg s−1 are consistent with this
198 0.15+ 0.03
−0.03
+0.4
2.5−0.3 1.02 (210) 8.5+ 0.4
−0.4
+0.04
classification.
198† 0.09−0.03 19+−8
22
1.02 (210) 57+
−3
3
Seq.185 (W53): The period of 8517.89 s is newly discovered. No
312 0.19+ 0.04
−0.04
+0.3
2.1−0.3 0.78 (83) 6.1+ 0.3
−0.4 periodic signal was found in the optical/UV observations (E03b,
c 11+3 10.4+
229 0.023 (fixed) −2 1.08 (54) 0.6
−0.6 RS18). This source is most likely a CV, indicated by its LX ∼ 3.7 ×
Note. (1) Source sequence number taken from Cheng et al. (2019a), except 1031 erg s−1 , and a UV counterpart with a blue colour. The phase-
for the eROSITA source e481, which is from Saeedi et al. (2022). †Spectra folded light curve exhibits a wide (∼1/4 phase) but relatively shallow
at the bursting state. (2) Line-of-sight absorption column density fixed at the dip, resembling the X-ray (1–9 keV) light curve of the classical polar
Galactic foreground value if the spectrum provides no strong constraint. (3) AM Her (Heise et al. 1985). In this case, the period can also be
The bremsstrahlung temperature fixed at a value of 40 keV if the spectrum understood as an orbital period. The spectrum of this source also
provides no strong constraint. a Model for this spectrum is not fitted, but appears softer (Tb ∼ 9 keV) than typical IP spectra.
is taken from Bahramian et al. (2017); b Fitted by a two-temperature model; Seq.366 (W8): The period of 10311.94 s for this source is newly
c Shock temperature of the cooling flow model. See text for details. (4) χ 2 and
discovered in the X-ray band. The phase-folded light curve clearly
degree of freedom of the best-fit model. (5) 0.5–8 keV unabsorbed luminosity
for a distance of 4.5 kpc, corrected for the enclosed-energy fraction. Quoted
reveals a deep and narrow eclipse, unambiguously pointing to
errors are at the 90 per cent confidence level. an orbital period. The same period and eclipsing behaviour were
detected in the optical band, but not in the UV band, by E03b.
This source is also likely a CV, for its LX ∼ 8.7 × 1031 erg s−1 and
UV counterpart with a blue colour. The plasma temperature (Tb )
from the X-ray spectral and variability properties). A period of is not well constrained for this source due to its relatively hard
∼28.18 and ∼27.2 min was detected, though at a low significance by X-ray spectrum and insufficient sensitivity of Chandra at energies
Chandra X-ray observations (Bahramian et al. 2017) and HST far-UV above 8 keV. This suggests that the source is an IP, although no spin
observations (Tudor et al. 2018), respectively, which were suggested modulation is found from the X-ray data. We have fixed Tb at 40 keV,
to be the orbital period of the binary system. This periodic signal was typical for IPs when their hard X-ray (up to tens of keV) spectra are
apparent only in Chandra/ACIS observations taken in 2002 and only available (Xu et al. 2016) to derive LX .
in the 0.5–0.7 keV light curve. Hence, our default period search is Seq.423 (W1): The period of 10390.33 s is newly discovered in the
not expected to find this signal. Nevertheless, when applying the GL X-ray band, along with a subharmonic at twice this value. These two
algorithm to the 0.5–0.7 keV light curve in the 2002 Chandra/ACIS periods were also detected in the optical band by E03a who favoured
observations (ObsID 2735, 2736, 2737, 2738), we are able to recover the longer period as the orbital period. The phase-folded X-ray light
a periodic signal at ∼1692 s with a GL probability of 95 per cent. curve at 10390.33 s exhibits both a dip, likely due to eclipse, and a
We find it hard to accommodate the dual period with either an IP narrow peak about 1/2 phase ahead of the dip. When folded at the
or a BH-LMXB. In the former case, the longer period of ∼1692 s is subharmonic period, the two nearly identical dips appear. Hence, we
still too short to be compatible with the orbital period. Moreover, an favour the shorter period, i.e. 10390.33 s, as the true orbital period.
orbital modulation is unlikely to be transient. In the latter case, the We note that E03b favoured the longer period not because of a
longer period of ∼1692 s might be due to orbital modulation, but the higher significance in the signal itself, but rather due to an empirical
shorter period (∼205 s) is still too long to be associated with spin relationship between the absolute magnitude of the donor and the
modulation of the BH. Nevertheless, we follow Miller-Jones et al. orbital period, given that the source is magnetic CV.
(2015) to tentatively classify this source as an LMXB, chiefly due to In view of the relatively high LX ∼ 8.6 × 1031 erg s−1 and the
the presence of a radio counterpart. The presence of strong carbon UV counterpart, we classify this source as a CV, noting that its soft
lines and lack of H or He lines suggest a CO WD as the donor star spectrum (Tb ∼ 4.6 keV) makes it unlikely an IP.

MNRAS 521, 4257–4276 (2023)

 Periodic X-ray sources in 47 Tuc 4269
Seq.232 (W37): This is another source exhibiting two periods. The (RS18). With an X-ray luminosity ∼6.4 × 1031 erg s−1 and a UV
shorter period, 11112.50 s, is detected only in ObsID 2735, which counterpart, this source is likely a CV. The best-fit Tb ∼ 13 keV is
captured an outburst with an X-ray luminosity (∼1.5 × 1032 erg s−1 ) compatible with either magnetic CVs or non-magnetic CVs (Xu et al.
a few times higher than the quiescent level. The same period was 2016).
reported by Heinke, Grindlay & Edmonds (2005a), who suggested Seq.273 (X10 or W27): This source was first suggested to be a CV
W37 to be a quiescent LMXB, although no firm optical counterpart upon the detection of a ∼3.8 h period in the X-ray band (Grindlay
was identified (E03b). The phase-folded light curve clearly shows et al. 2001). Here, based on the much deep Chandra data, we detect a
a deep eclipse, indicating an orbital period. A longer period of periodic signal at 16824.25 s (4.67 h), the same value reported before
44883.30 s is newly detected in the quiescent state, i.e. by excluding by Heinke et al. (2005b). The 3.8-h signal could not be recovered
ObsID 2735. Interestingly, this period is consistent with four times both in our work and in Heinke et al. (2005b). The phase-folded light
the shorter period to within the statistical uncertainty given in Table 2, curve has a shape reminiscent of the ‘one-pole’ behaviour of polars
and the corresponding phase-folded light curve also exhibits an with the largest fractional variation (∼70 per cent) among all 18
eclipse-like behaviour. One possibility is that the longer period is periodic sources. The HST observations identified a UV counterpart
a superorbital period, for instance, caused by the precession of a but found no periodic modulation (RS18). Interestingly, a ∼18.416 h

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

tilted or warped accretion disc. We caution that the longer period is periodic modulation has been found in the GeV gamma-ray light
more affected by the substantial variability in this source, rendering curve of 47 Tuc by Fermi observations (Zhang et al. 2020). Zhang
it a potential spurious signal (Section 4.3). Further observations are et al. (2020) suggested that X10 as a MSP is responsible for the GeV
required to confirm this signal and to clarify its relation with the periodic modulation, although Fermi observations lack the angular
shorter period. The spectra appear very soft at both the bursting resolution to directly resolve X10. However, here we find that the
and quiescent states, with Tb ∼ 0.6 and 0.8 keV, respectively. This X-ray spectrum of X10 is well characterized by a two-temperature
is consistent with thermal X-ray radiation from the surface of an model, with a low-temperature of ∼0.26 keV and a high-temperature
accreting neutron star, as also suggested by Heinke et al. (2005a). of ∼20 keV. This is consistent with typical polar spectra consisting
Hence, we also classify this source as an LMXB. of a soft component from the blackbody-like WD surface and a hard
Seq.263 (W29): This source is known as a redback system with component from the multitemperature accretion column (Ramsay
a MSP accreting from a main-sequence companion (Edmonds et al. et al. 2004). Hence, the X-ray spectrum, the phase-folded light curve,
2002). Periodic variation at ∼3.19 h in the optical and X-rays were the X-ray luminosity (∼2.2 × 1032 erg s−1 ), and the UV counterpart,
found by Bogdanov, Grindlay & van den Berg (2005), understood as all point to a polar classification for this source.
a changing view of the intrabinary shock between the pulsar wind Seq.317 (W20): The period of 23583.11 s is newly discovered in
and the stellar wind from its companion, as described in Hebbar et al. the X-ray band. No periodicity was found by HST observations. This
(2021). The radio counterpart, first detected by Camilo et al. (2000), source was suggested to be a CV due to the identification of a possible
using the Parkes radio telescope with a pulsating period of 2.35 ms bright optical counterpart (E03a), while no bright UV counterpart
and an orbital period of 3.2 h, exhibits eclipsing-like behaviour in the was found that could indicate the presence of an accreting WD
radio light curve for ∼25 per cent of its orbit. Here, we confirm this (RS18). The phase-folded light curve, very similar to that of W27,
orbital behaviour at a period of 11486.36 s. The spectrum is fitted by resembles the typical single-peaked light curve of polars. On the other
a two-temperature model, with a low temperature of ∼0.4 keV and hand, its soft X-ray spectrum (with Tb ∼ 1.6 keV) and moderate X-
a high temperature fixed 40 keV, and with one of the lowest X-ray ray luminosity of ∼2.0 × 1031 erg s−1 are more suggestive of a DN.
luminosities (∼2.1 × 1031 erg s−1 ). Nevertheless, it seems reasonable to classify this source as a CV.
Src-No.481: This is the only periodic source identified from the Seq.162 (X5 or W58): We find a period of 31200.27 s for this
eROSITA data and the only one located outside the half-light radius. source, which is consistent with the periodic signal reported by
The source was suggested to be a MSP by Saeedi et al. (2022), Heinke et al. (2003) based on early Chandra observations. The phase-
based on its soft X-ray spectrum and a potential main-sequence folded light curve clearly shows an eclipse (Fig. 4), thus pointing to
counterpart. However, there is no radio detection at the source an orbital period. The soft X-ray spectrum (with Tb ∼ 0.67 keV, but
position in a recent deep radio continuum survey with the Australia also with significant excess at energies above 5 keV) and a relatively
Telescope Compact Array (Tudor et al. 2022), which gives a 5σ high luminosity (LX ∼ 1 × 1033 erg s−1 ) suggest that this source is
upper limit of 16 μJy beam−1 at 7.25 GHz. The phase-folded light likely an LMXB.
curve exhibits a single-peaked structure (Fig. 2), which is consistent Seq.252 (W32): The source was identified with an optical period
with the accretion being dominated by one of the two magnetic poles. of ∼6.37 h in Albrow et al. (2001). Here, the X-ray data reveals a new
However, the period of 14388.49 s is probably too long for the spin period at 44642.86 s (12.40 h) consistent with twice the optical period
period of a neutron star, but can be compatible with a WD spin. within the statistical uncertainty. The phase-folded light curve also
The latter case, if true, implies that this source is a polar, which appears sine-like. A bright, blue, and marginally variable counterpart
often exhibits the one-pole behaviour (Bao & Li 2020). On the other was suggested by the HST data (E03b). The X-ray spectrum is
hand, the source spectrum is extremely soft, having a best-fit Tb ∼ well fitted by a two-temperature model, with a low-temperature of
0.06 keV, which is more consistent with a MSP (Gentile et al. 2014; ∼0.5 keV and a high-temperature of ∼19 keV. These together with a
Zhao & Heinke 2022) rather than a polar (Xu et al. 2016). However moderate X-ray luminosity ∼4 × 1031 erg s−1 lead us to tentatively
in this case, it is hard to reconcile the shape of the phase-folded classify the source as a CV.
light curve with orbital modulation. With the above concerns, we Seq.283 (W26): The X-ray period of 45310.38 s (12.59 h) is newly
tentatively classify this source as a rotation-powered MSP having a discovered. A period of 9.45 h was previously suggested by optical
mean X-ray luminosity ∼7 × 1030 erg s−1 , the lowest among all 18 observations, which cannot be recovered by the Chandra data. This
periodic sources. source is classified as a coronally active binary (AB) based on a
Seq.331 (W15): The phase-folded light curve at 15232.63 s clearly possible optical variable counterpart, which has a colour consistent
shows an eclipse, consistent with previous optical observations with a main-sequence star or a subgiant (RS18). The long period, low

MNRAS 521, 4257–4276 (2023)

4270 T. Bao, Z. Li and Z. Cheng
X-ray luminoisity ∼9.3 × 1030 erg s−1 , soft spectrum (Tb ∼ 0.9 keV), particular, 9 of the 11 CVs have also been identified as CVs in
and the lack of a UV counterpart support an AB classification, previous work, mainly due to their UV colour (RS18).
although no clear evidence can be inferred from the phase-folded
light curve.
Seq.290 (W25): The X-ray period of 46082.95 s is newly discov- 5.3 Remarks on the X-ray-detected periods
ered. A shallow dip seen in the phase-folded light curve may be E03a reported 36 periodic candidates based on multi-epoch HST
attributed to eclipsing at a moderate inclination angle, pointing to observations (see table 3 therein, excluding the marginal candidate
an orbital period. This source was previously classified as a CV but W120), among which eight were classified as a CV. Among them, five
with no optically determined period (E03a). The colour of its UV sources are also classified as a CV by us (Table 2): W8 [=Seq.366],
counterpart suggests a hot, relatively low-mass WD (RS18), and the W15 [=Seq.331], and W36 [=Seq.229] have an X-ray period nearly
strong aperiodic variability in the UV supports the CV classification. identical to the optical period; W1 [=Seq.423] has an X-ray period
The soft spectrum (with Tb ∼ 4.4 keV) and the moderate X-ray one-half of the optical period; W27 [=Seq.273] has an X-ray period
luminosity ∼6.4 × 1031 erg s−1 are more suggestive of a DN. The ∼20 per cent longer than the optical period. For the remaining
observation of its NUV transient-like variability, which is likely DN

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

three periodic CV candidates (W2, W21, and W34) in E03a, we
outbursts, provides a strong evidence for its DN nature (Modiano, cannot reproduce the optical period. In particular, as mentioned in
Parikh & Wijnands 2020). Section 5.2, W2 [=Seq.414] has an X-ray period identical to the
Seq.198 (W47 or E8): The source has been identified as an AB optically-confirmed period (RS18). The cases of W21 and W34 being
with an optical period of 12.7 h, in the absence of blue excess in a CV have both been challenged by recent work (RS18).
the HST images (Edmonds et al. 1996; Grindlay et al. 2001). With In addition to those CV candidates reported by E03a, W32
the deep Chandra data, we confirm the periodic signal at 46151.01 s [=Seq.252] and W23 [=Seq.312] are classified by us as a CV, and
(12.82 h), based on the time series excluding the one observation their X-ray period are about twice longer than the optical period
(ObsID 2737) contaminated by flare-like variations. The phase- found by Albrow et al. (2001) and E03b, respectively. For the other
folded light curve is sine-like but with rather moderate fractional three CVs classified by us (W53, W20, W25), no periodic signal
variation. The spectrum of the quiescent state has Tb ∼ 2.5 keV and has been reported from either optical or UV observation. Among
LX ∼ 8.5 × 1031 erg s−1 , which are not unseen for ABs (Xu et al. the non-CV periodic sources reported by E03a, one source, W29
2016). The spectrum at the bursting state is much harder and has a [=Seq.263], is classified as a redback and has a nearly identical
luminosity increased by a factor of ∼7. We classify this source as an period in the X-ray and optical bands. Two sources are classified
AB. as an AB: W47 [=Seq.198] has a nearly identical period in the X-
Seq.312 (W23): This source was originally suggested to be an ray and optical bands, while W26 [=Seq.283] has an X-ray period
AB exhibiting an optical period of ∼6.18 h by E03b, while recent ∼30 per cent longer than the optical period. The remaining non-CV
work (RS18) classified it as a CV based on a blue-coloured UV periodic sources in E03a were classified as ABs, but they are either
counterpart and Hα emission. Here, we identify a new periodic too faint to be detected in the X-ray or with long orbital periods
signal at 48780.49 s (11.24 h), with its phase-folded light curve beyond our period-searching range.
sharing strong resemblance with magnetic CVs, i.e. having a To summarize, 6 of the 11 CVs classified in this work show the
single-peaked structure. Significant variability was found in its UV same period in the X-ray and optical bands, which can be safely taken
light curve, by up to 2 mag on a time-scale of hours. Thus, we as the orbital period. This is particularly true for Seq.414, Seq.366,
classify this source as a CV, associating the period with spin- Seq.423, Seq.331, and Seq.229, the phase-folded light curve of which
orbital modulation. We note that its soft spectrum (Tb ∼ 2.1 keV), clearly exhibits an eclipse. The X-ray period of the other five CVs
moderate luminosity (∼6.1 × 1031 erg s−1 ), and phase-folded light is not detected in the optical or UV, but can also be taken as the
curve, similar to that of W20 are more suggestive of a polar. We orbital period. Support for this comes from the sign of an eclipse
note, however, such a period is unusually long compared to the (Seq.185 and Seq.290) or the sign of spin-orbit modulation (Seq.273,
known orbital periods (identical to the spin period) of polars in Seq.317, and Seq.312). One LMXB (Seq.263) and one AB (Seq.198)
the solar neighbourhood (Ritter & Kolb 2003). To our knowledge, also exhibit the same period in the X-ray and optical bands, which
only one system, V479 Andromedae having a similarly long orbital is most likely the orbital period. The X-ray period detected in the
period, (P ∼ 14.25 h), has been suggested to be a polar by its other four candidate LMXBs is not seen in the optical band. Among
strong He II line in the optical spectrum (González-Buitrago et al. them, two LMXBs (Seq.232 and Seq.162) clearly exhibit an eclipse
2013). in the phase-folded light curve, unambiguously pointing to an orbital
Seq.229 (W36 or AKO-9): This source is a spectroscopically period. On the other hand, the nature of Seq.217 and Src-No.481
confirmed long-period eclipsing CV system based on HST observa- remain elusive, so does their X-ray period. The X-ray period of
tions (Edmonds et al. 1996; Knigge et al. 2003). As mentioned in Seq.283, another candidate AB, is not seen in the optical, either, but
Section 3.1, originally we detect a periodic signal at one-third of the is also tentatively taken as the orbital period.
suggested orbital period, mainly due to the limit in our default period-
searching range. The true orbital period (95731.20 s or ∼26.59 h) is
recovered once the searching range is relaxed. The phase-folded light 6 DISCUSSION
curve clearly shows the eclipsing behaviour. The long orbital period
requires a subgiant as the companion star, which was confirmed based The periodic X-ray sources identified in previous sections can
on HST UV spectroscopy and photometry (Knigge et al. 2003). The provide an important diagnosis to the underlying close binary popu-
X-ray spectrum with Tb ∼ 11 keV and LX ∼ 1 × 1032 erg s−1 are lations in 47 Tuc, which in turn allows for a meaningful comparison
compatible with a DN. with the field populations, in particular CVs. Nevertheless, it is
In short, we have classified to a varied degree of confidence five necessary to first examine the purity and completeness of the periodic
LMXBs, eleven CVs, and two ABs (Table 2) based on their X-ray sources (Section 6.1), before implications on their demography and
timing and spectral properties and multiwavelength counterparts. In origin can be addressed (Section 6.2).

MNRAS 521, 4257–4276 (2023)

 Periodic X-ray sources in 47 Tuc 4271
6.1 Purity and completeness of the periodic X-ray sources 6.2 Orbital period distribution and origin of CVs
It is possible that some of the periodic X-ray sources are background Now, we turn to the orbital period distribution of the X-ray sources,
AGNs, which are known to exhibit in their X-ray emission quasi- focusing on the CV candidates. Fig. 7 displays the histogram of the
periodic oscillations (QPOs) as well as the more dramatic quasi- putative orbital period of the 11 CVs in 47 Tuc. For comparison,
periodic eruptions (QPEs) in the frequency range of 10−4 –10−2 Hz we also show the orbital period distributions of CVs in the solar
(Pan et al. 2016; Carpano & Jin 2018; Zhang et al. 2018; Miniutti neighbourhood from Ritter & Kolb (2003) and twenty CVs in the
et al. 2019; Song et al. 2020; Arcodia et al. 2021). Recently, Bao & Galactic bulge (i.e. the LW) from Bao & Li (2020). The sample
Li (2022) conducted a systematic search for AGN QPOs from the of solar neighbourhood CVs (red-dashed histogram) has a much
7-Ms Chandra Deep Field-South (CDF-S; Luo et al. 2017), which larger size than the bulge (blue solid histogram) and 47 Tuc (green
consists of ∼1000 independent X-ray sources. Using both the LS filled and hatched histogram) samples, which clearly exhibits the
periodogram and the GL algorithm, Bao & Li (2022) found no two well known features, namely, the period gap between ∼2–3 h
statistically significant periodic signals that persist over the entire and the period minimum at ∼82 min, as marked in Fig. 7. Notably,
7-Ms exposures, or over any of the four subsets having an effective the Galactic bulge CVs are almost all found within or below the

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

exposure of 1–3-Ms. From the empirical log N–log S distribution of period gap, having a longest period of 3.33 h. According to Bao & Li
cosmic X-ray background (CXB), Cheng et al. (2019a) estimated (2020), this is not due to selection bias but can be understood as an
the number of background AGNs to be ࣠4.8 within a projected age effect in the sense that the Galactic bulge CVs are more evolved
radius of 1 arcmin from the centre of 47 Tuc, i.e. within which all the systems having substantially shrunk orbits compared to their solar
detected periodic sources are located. Even the much larger eROSITA neighbourhood counterparts.
FoV, when compromised by its moderate sensitivity, is expected to Several interesting aspects of the 47 Tuc sample are revealed in
contain ࣠200 background AGNs. A simple scaling from the non- this diagram:
detection of QPOs among the ∼1000 CDF-S sources suggests that
the possibility of having an AGN QPO in the field of 47 Tuc is
negligibly small. The moderately high Galactic latitude of 47 Tuc (b (i) No CV is found to have a period below 2.15 h, the lower bound
≈ −44◦. 9) also implies that contamination by foreground sources is of the period gap (Knigge et al. 2011). This is in stark contrast with the
negligible. Cheng et al. (2019a) also found negligible contamination bulge and solar neighbourhood CVs, which have 75 ± 25 per cent
by the Small Magellanic Cloud, which is an extended background (15 of 20) and 69 ± 3 per cent (769/1116) found below the gap,
object near the line of sight toward 47 Tuc. Thus it is safe to conclude respectively. Here, the quoted errors are of 1σ Poisson errors.
that all 18 periodic X-ray sources are physically associated with (ii) Long period CVs, i.e. those having a period beyond the upper
47 Tuc. bound of the gap (3.18 h), contribute to 64 ± 30 per cent (7/11) of
Completeness of the periodic sources is less straightforward to the 47 Tuc sample. This is significantly higher than the fraction of
assess. For this purpose, we contrast with the work of (Bao & Li 5 ± 4 per cent (1 of 20) and 23 ± 2 per cent (254 of 1116) in the
2020), who found 23 periodic sources among 847 X-ray sources in bulge and solar neighbourhood sample, respectively. Moreover, 4 of
the Limiting Window (LW), based on 1-Ms Chandra exposure of the 11 CVs in 47 Tuc have an orbital period above 12 h, which is
the Limiting Window (LW) and the GL algorithm. The point-source rarely seen even in the much larger sample of solar neighbourhood
sensitivity limit of the LW reaches LX ∼ 1030 erg s−1 (Revnivtsev CVs (∼1 per cent, 13 of 1116).
et al. 2010), which is only slightly high than that of 47 Tuc (iii) 36 ± 20 per cent (4 of 11) of the 47 Tuc sample is found with
(LX ∼ 4 × 1029 erg s−1 ). Moreover, the detected periodic sources the gap, while this fraction is 20 ± 10 per cent (4 of 20) and only
typically have a flux significantly above the sensitivity limit. Thus, 8 ± 1 per cent (93 of 1116) in the bulge and solar neighbourhood
it is statistically meaningful to compare the two samples. Bao & sample, respectively.
Li (2020) assessed the completeness of periodic signal detection Fig. 8 further depicts the CV orbital period versus the X-ray
with the GL algorithm by simulating a large set of X-ray light luminosity (upper panel) and the projected radius from the centre of
curves, which have the characteristic shape of either a sine function 47 Tuc (lower panel). For completeness, sources classified as LMXBs
or an eclipse and cover the ranges of orbital period, fractional and ABs are included, except for Seq.217 (X9) and Src-No.481, the
variation amplitude, and source counts typical of the LW sources X-ray period of which is of an unclear origin (Section 5.2). For
(mostly CVs and ABs). They found that for a sinusoidal light Seq.232 and Seq.198, the luminosity at the quiescent state is plotted.
curve, the GL detection completeness is generally 20 per cent for The Galactic bulge CVs with an identified orbital period (Bao &
source counts C  100 and exceeds ∼50 per cent for C  300; Li 2020) are also included in the upper panel for comparison. The
the detection completeness rapidly rises to 90 per cent at high 0.5–8 keV luminosities of these sources have been derived from the
variation amplitudes (60 per cent). For an eclipsing light curve, best-fit spectral model in Bao & Li (2020). In the lower panel, we
the GL detection completeness is lower for the same number of use a blue-shaded area to represent the effective eROSITA FoV, with
counts and exceeds 50 per cent only with C  800. In both cases, an inner radius of Rin = 1. 7 and an outer radius of Rour = 42 arcmin.
the detection completeness increases with increasing period. These The latter is coincident with the cluster’s tidal radius (Harris 1996).
simulation results can be borrowed to evaluate the present case, since The half-light radius (rh = 3. 17) and core radius (rc = 0. 36) of 47
the 47 Tuc sources have quite similar ranges of periods and source Tuc are also denoted. From Fig. 8, an additional aspect of the 47 Tuc
counts, but also with an on-average lower background. We estimate CVs can be stated:
that C ∼ 100, corresponding to LX ∼ 3 × 1030 erg s−1 in 47 Tuc can (iv) More than half (6 of 11) of the CVs with an identified period
be taken as the detection limit for a typical periodic signal, whereas are located within the core radius, which include six of the seven
C  300, corresponding to LX  1031 erg s−1 can be regarded as the long-period CVs. On the other hand, all four CVs within the period
completeness limit. This is consistent with the lowest luminosity of gap are located outside the core radius. In the meantime, no CVs
9 × 1030 erg s−1 found among the periodic sources (excluding the with an identified period are found outside R = 1 arcmin, or about
eROSITA source Src-No.481). one-third of the half-light radius.

MNRAS 521, 4257–4276 (2023)

4272 T. Bao, Z. Li and Z. Cheng

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

Figure 7. The orbital period distribution of the 47 Tuc CVs (cyan-filled histogram), in comparison with that of field CVs (red-dotted histpgram) and galactic
bulge CVs (blue histpgram). Period gap and period minimum are delineated by a pair of vertical orange-solid lines, and a vertical grey-dashed line, respectively,
the values of which are taken from Knigge et al. (2011). The bottom panel shows the cumulative distribution.

The above outlined behaviour of the CV orbital periods have (Ritter & Kolb 2003) to 36 per cent (Pala et al. 2020) has been
important implications about the origin of present-day CVs in 47 estimated.
Tuc. For (i), the lack of short-period CVs appears contradictory to On the other hand, period bouncers as low-luminosity systems
the theoretical expectation that short-period CVs, especially period are unlikely to be detected with the current Chandra data. This can
bouncers, should dominate the present-day CV population in GCs be inferred from the flux-limited X-ray surveys conducted by Preto-
(Belloni et al. 2016), which essentially reflects the fact that the rius & Knigge (2012), which suggest that period bouncers would have
evolutionary time-scale of CVs is short compared to the age of LX  1030 erg s−1 , according to the empirical luminosity distribution
GCs. We emphasize that this paucity of short-period CVs is not of non-magnetic CVs. CVs with such low X-ray luminosities may
solely a selection effect, since the simulations presented in Bao & still be detected as an X-ray source, but the small number of counts
Li (2020) indicate that the current Chandra data are sensitive to (࣠a few tens) they produce are insufficient for the GL algorithm to
periodic signals between 1–2 h and of an X-ray luminosity down reveal their orbital period.
to ∼1031 erg s−1 . Indeed, most LW CVs with an identified orbital For (ii), the high fraction of long-period CVs in 47 Tuc is also
period are found below the gap and have LX  2 × 1031 erg s−1 . at odds with the GC age. The low fraction of long-period CVs in
Thus the paucity of short-period CVs in 47 Tuc suggests a CV the Galactic bulge compared to the field has been explained by an
population different from those in the Galactic bulge and solar age effect (Bao & Li 2020). Given the age of GCs, the intuitive
neighbourhood. We note that the majority of the 20 LW CVs expectation is that 47 Tuc would have a similar or even lower fraction
were classified as a magnetic CV (IP or polar), mainly based of long-period CVs compared to the bulge, which, however, is not
on the ‘one-peaked’ behaviour in their phase-folded light curve observed. We emphasize that three of the four 47 Tuc CVs with an
(Bao & Li 2020). Bao & Li (2020) further demonstrated that the identified period above 12 h have a bright, variable UV counterpart,
detectability of orbital modulation for DNe, typically through an arguing strongly for their CV identity. While only one of them
eclipse, is generally much lower than the detectability for magnetic (Seq.229) shows a clear sign of orbital eclipse, such long periods
CVs. This is consistent with the analysis of field CVs using ASCA cannot be attributed to the WD spin period and are most probably the
observations, which found that 11 of 16 IPs exhibit an X-ray orbital orbital period. Given the wide orbits, the donor should be a subgiant
modulation (Parker, Norton & Mukai 2005), while only 4 of 34 in order to fill the Roche lobe. The progenitor mass of the donors
DNe do so (Baskill, Wheatley & Osborne 2005). This implies that in these long period CVs should be higher than the main-sequence
the non-detection of short-period CVs might be attributed to an turn-off mass, ∼0.9 M , of 47 Tuc (Hesser et al. 1987; Bergbusch &
intrinsically lower fraction of magnetic CVs among all CVs in 47 Stetson 2009), and in turn requires a more massive WD. This strongly
Tuc, compared to the field population, in which a fraction of 23 suggests a dynamical origin for these systems (e.g. via exchange or

MNRAS 521, 4257–4276 (2023)

 Periodic X-ray sources in 47 Tuc 4273

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

Figure 8. Upper panel: The 0.5–8 keV X-ray Luminosity versus orbital period, for CVs (red triangles), ABs (purple circles), and LMXBs (green stars) identified
in 47 Tuc. Galactic bulge CVs (black triangles) with an identified orbital period are plotted for comparison. Low panel: The project distance from the centre
of 47 Tuc versus orbital period. The core radius (rc ) and half-light radius (rh ) are denoted by the orange and blue-dashed line, respectively. The horizontal
blue-shaded area marks the eROSITA field-of-view, which extends to the cluster’s tidal radius. In both panels, the period gap and period minimum of CVs, with
values taken from Knigge et al. (2011), are denoted by the yellow strip and the grey vertical line, respectively.

tidal capture), which is consistent with the fact that three of the four not well understood. For example, the observed mass transfer rate of
systems are located within the core radius (Fig. 8), where the stellar KIC5608384, is a factor of 20 lower than predicted by a MESA model
dynamical encounter rate is highest. A possible formation channel is (Yu et al. 2019). The presence of long-period CVs with relatively low
predicted by the simulations of Shara & Hurley (2006) (see model X-ray luminosities in a massive GC invites further investigation.
CV1 in Table 2 therein), in which weak orbital perturbations force As for (iii), finding a significant fraction of CVs within the
the mass transfer start when the donor is still on the subgiant branch, period gap is even more unexpected. Theoretically, non-magnetic
having Porb ∼ 11.5 h. The CV phase of this case lasts for ∼16 Myr, CVs are not able to sustain mass transfer within the period gap,
until Porb reduces to ∼6.5 h, allowing for the onset of the common since the contraction of the donor on a thermal time-scale is
envelope (CE) phase. Generally, the exchange interaction is most faster than the orbital shrinking on the much longer gravitational
likely to produce CVs with donor mass greater than 0.7 M , and the radiation time-scale, making the system detached (Ritter 1988). Two
mass-transfer phase is short-lived, on the order of 10–100 Myr. This possible explanations may be considered for these systems. The first
means that these long-period CVs formed only recently. possibility is that they are magnetic CVs, thus their mass transfer is
The fact that these long-period (Porb > 10 h) CVs have only controlled by the magnetic field of the WD and remains substantial
moderate X-ray luminosities (LX ࣠ 1032 erg s−1 ) are also remarkable. across the gap (Garraffo et al. 2018). We note that this could be the
In fact they are no more luminous than their short-period counter- case for the four LW CVs found within the gap (Fig. 8). In 47 Tuc,
parts, whereas standard evolution models of CVs generally predict a magnetic CV is highly likely the case for Seq.414, which exhibits
that the mass transfer rate is more than one order of magnitude a dual-period and is best understood as an IP (Section 5.2). The
higher in long-period CVs (Kalomeni et al. 2016), which should hard X-ray spectrum of Seq.366 is also suggestive of an IP, while
lead to a higher LX . In view of the combined properties of a long the light curve of Seq.185 is suggestive of a polar (Section 5.2). The
orbital period and an evolved donor, these sources are analogous to problem with this explanation is that we should expect to detect other
several field CVs including 1RXH J082623.6-505741 (Porb ∼ 10.4 h, magnetic CVs below the gap in view that short-period CVs make up
Sokolovsky et al. 2022), EY Cyg (Porb ∼ 11 h, Echevarrı́a et al. 2007; 48 per cent of all magnetic CVs in the solar neighbourhood (Ritter &
Nabizadeh & Balman 2020), CXOGBS J175553.2-281633 (Porb ∼ Kolb 2003). However, none is observed, which, as we argue in the
10.3 h, Gomez et al. 2021), and KIC5608384 (Porb ∼ 9 h, Yu et al. above may be understood as due to a low occupation fraction of
2019). The low mass transfer rates inferred for these binaries are still magnetic CVs. This leads us to consider the second possibility, that

MNRAS 521, 4257–4276 (2023)

4274 T. Bao, Z. Li and Z. Cheng
the periodic CVs, the bright group and the faint group, respectively.
Since the bright group should mainly consist of CVs and quiescent
LMXBs, the effect of mass-segregation alone is unlikely to explain
the steeper profile of the periodic sources. We suggest that the latter
reflect a subgroup of CVs having recently formed out of two-body or
multibody interactions inside the cluster core, which would naturally
follow a steep radial profile.
The lack of periodic sources outside one-third of the half-radius,
expect for the unclassified source Src-No.481, deserves some re-
marks. We note that this apparent lack is not due to an insufficient
sensitivity. First, the Chandra data has a nearly constant detection
sensitivity out to rh (Cheng et al. 2019a), meaning that any periodic
source having similar properties with those detected should be found
by the GL algorithm with an equal probability. However, none of

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

the 156 sources located between 1/3 rh to 1 rh exhibits a significant
periodic signal. Second, the eROSITA observations have a source
detection sensitivity of 5 × 1030 erg s−1 out to a radius of at least 18. 8
(Saeedi et al. 2022). Although ideally one would use simulations to
determine the precise sensitivity of the eROSITA data for periodic
signal detection, we may empirically estimate that this sensitivity
limit for periodic signals is about one order of magnitude higher
than the source detection limit, i.e. 5 × 1031 erg s−1 . We note that the
Figure 9. Radial surface density profiles of the periodic CVs (black), bright luminosity of Src-No.481 is significantly below this limit, due to its
X-ray sources (LX  5 × 1030 erg s−1 , green), faint X-ray sources (LX unusual softness, but otherwise supports our conservative estimate.
࣠ 5 × 1030 erg s−1 , red). The CXB contribution has been subtracted. The
After accounting for the CXB contribution, we further estimate that
coloured dash–dotted lines represent the best-fit power-law function of three
profiles out to the core radius. The half-light radius and core radius are mark
there are ∼80–90 sources (∼20 have LX > 5 × 1031 erg s−1 ), out of
by the blue and orange lines. a total of 237 eROSITA sources lying between R = 1. 7 and 18. 8,
which should belong to 47 Tuc. Therefore, even with a 10 per cent
GL detection efficiency on average, we should expect to find 8–9
is, these systems are dynamically formed (e.g. via exchange), or their periodic sources, while only one is actually observed. This, along
progenitor binaries have been dynamically altered, such that they are with the Chandra non-detection of periodic sources outside R =
semidetached when crossing the period gap. We remind that these 1 arcmin suggests an intrinsic paucity of bright CVs at large radii.
four CVs are all currently located outside the core radius. Therefore, Since bright CVs and their progenitor binaries should be substantially
if they indeed formed via dynamical encounters, the encounter most heavier than the background stars, they could have sunk to the cluster
likely took place inside the core and must have provided the resultant core. This is in accord with the abrupt drop of the surface density
binary with a sufficiently large radial velocity to escape from the core. profile of the bright group at R  2 arcmin (Fig. 9), or a ‘dip’ identified
Finding such CVs outside the core is reasonable, since the two-body by Cheng et al. (2019a). On the other hand, CVs recently formed
relaxation time at the half-light radius of 47 Tuc (Trel ∼ 3.5 Gyr, in the cluster core via dynamical interactions may have received a
Belloni et al. 2019) is much longer than the CV evolutionary time substantial kick velocity that could take them to the cluster outskirt.
crossing the gap (∼0.4–1.3 Gyr) (Knigge et al. 2011). Indeed, the MOCCA simulation of Belloni et al. (2019) predicts that
For (iv), as already discussed in the above, the locations of the on average about half of the entire population of detectable CVs
CVs provide a strong hint for their possible dynamical origin. We are currently located outside the half-light radius. This discrepancy
further construct the radial surface density profile for the 11 periodic between the observation and simulation deserves further study.
X-ray sources classified as a CV, as shown in Fig. 9 (black squares).
For comparison, we also plot the radial surface density profiles of
all the Chandra-detected X-ray sources within the half-light radius 7 S U M M A RY
of 47 Tuc, excluding the periodic sources. Following Cheng et al. We have presented a systematic search for X-ray periodic signals
(2019a), these sources are divided into two groups: the bright group among 537 sources resolved by Chandra and 888 sources resolved
(LX  5 × 1030 erg s−1 , green dots) and the faint group (LX ࣠ by eROSITA in the field of 47 Tuc.
5 × 1030 erg s−1 , red dots). Contamination by the CXB sources has Our main results include:
also been statistically subtracted (see Fig. 5 in Cheng et al. 2019a for
detail). Cheng et al. (2019a) showed that both the bright and faint (i) Using the Gregory-Loredo algorithm, we have detected 20 X-
groups are more centrally concentrated than the normal stars in 47 ray periodic signals from 18 independent sources, among which 17
Tuc, which can be understood as the effect of mass-segregation, i.e. are detected by Chandra and one by eROSITA. Fourteen of the 20
the X-ray sources as close binaries are on-average heavier than the periodic signals are newly discovered.
single stars, and are more likely to sink into the cluster core through (ii) We have tentatively classified the 18 periodic sources into 11
two-body relaxation. Remarkably, Fig. 9 shows that the periodic CVs, 4 LMXBs, 1 MSP, and 2 ABs based primarily on their X-ray
CVs have an even steeper radial distribution than either the bright temporal and spectral properties, as well as optical/UV information,
group or the faint group, which is most prominent within the core when available.
radius. To demonstrate this more quantitatively, we fit the three radial (iii) Despite a sample subject to potential selection bias and small-
profiles out to the core radius using a power-law function, finding a number statistics, the orbital period distribution defined by the 11
slope of −1.6 ± 0.3, −0.8 ± 0.1, and −0.53 ± 0.04 (1σ errors) for CVs in 47 Tuc is significantly different from that of the Galactic

MNRAS 521, 4257–4276 (2023)

 Periodic X-ray sources in 47 Tuc 4275
bulge and solar neighbourhood CVs. The paucity of short-period CVs Cheng Z., Mu H., Li Z., Xu X., Wang W., Li X., 2020a, ApJ, 892, 16
below the period gap might be attributed to a high occupation fraction Cheng Z., Li Z., Wang W., Li X., Xu X., 2020b, ApJ, 904, 198
of non-magnetic CVs in 47 Tuc, for which the orbital modulation is Clark G. W., 1975, ApJ, 199, L143
more difficult to detect by the GL algorithm. In the meantime, both Echevarrı́a J., Michel R., Costero R., Zharikov S., 2007, A&A, 462, 1069
Edmonds P. D., Gilliland R. L., Guhathakurta P., Petro L. D., Saha A., Shara
the overabundance of long-period CVs with a subgiant donor and
M. M., 1996, ApJ, 468, 241
a substantial fraction of CVs found within the period gap provide
Edmonds P. D., Gilliland R. L., Camilo F., Heinke C. O., Grindlay J. E., 2002,
strong evidence for a group of CVs having recently formed via ApJ, 579, 741
dynamical interactions. Edmonds P. D., Gilliland R. L., Heinke C. O., Grindlay J. E., 2003a, ApJ,
(iv) The steep radial surface density profile of the periodic CVs 596, 1177
provides further evidence for a group of CVs having recently formed Edmonds P. D., Gilliland R. L., Heinke C. O., Grindlay J. E., 2003b, ApJ,
out of dynamical interactions inside the cluster core. In the meantime, 596, 1197
despite sufficient sensitivity of the X-ray data, no periodic CVs are Foreman-Mackey D., Hogg D. W., Lang D., Goodman J., 2013, PASP, 125,
detected outside one-third of the half-light radius and up to the 306
cluster’s tidal radius. Garraffo C., Drake J. J., Alvarado-Gomez J. D., Moschou S. P., Cohen O.,

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

In future work, we plan to extend the search for periodic X-ray 2018, ApJ, 868, 60
Gentile P. A. et al., 2014, ApJ, 783, 69
sources to other Galactic GCs, which will be valuable for a better
Gomez S. et al., 2021, MNRAS, 502, 48
understanding of the periodic features found in 47 Tuc and, more González-Buitrago D., Tovmassian G., Zharikov S., Yungelson L., Miyaji T.,
generally, in the dynamical environment of GCs. Echevarrı́a J., Aviles A., Valyavin G., 2013, A&A, 553, A28
Gregory P. C., Loredo T. J., 1992, ApJ, 398, 146
Grindlay J. E., Heinke C., Edmonds P. D., Murray S. S., 2001, Science, 292,
AC K N OW L E D G E M E N T S
2290
This work is supported by the National Natural Science Foundation Haggard D., Cool A. M., Davies M. B., 2009, ApJ, 697, 224
of China (grant 12225302) and the science research grants from Harris W. E., 1996, AJ, 112, 1487
the China Manned Spaced Project (CMS-CSST-2021-B02, CMS- Hebbar P. R., Heinke C. O., Kandel D., Romani R. W., Freire P. C. C., 2021,
MNRAS, 500, 1139
CSST-2021-B11). B.T. acknowledges support by the Postgradu-
Heggie D. C., 1975, MNRAS, 173, 729
ate Research & Practice Innovation Program of Jiangsu Province Heggie D., Hut P., 2003, The Gravitational Million-Body Problem: A
(KYCX22 0106). Z. C. acknowledges support by The National Multidisciplinary Approach to Star Cluster Dynamics. Cambridge Univ.
Natural Science Foundation of China (grant 12003017). The authors Press, Cambridge
wish to thank the anonymous referee for helpful comments that Heinke C. O., 2010, in Kalogera V., van der Sluys M.eds, AIP Conf.
improve our work, Ziteng Wang for valuable discussions, and Lulu Proc. Vol. 1314, International Conference on Binaries: in celebra-
Xing for help with the graphics. tion of Ron Webbink’s 65th Birthday. Am. Inst. Phys., New York,
p. 135
Heinke C. O., Grindlay J. E., Lloyd D. A., Edmonds P. D., 2003, ApJ, 588,
DATA AVA I L A B I L I T Y 452
Heinke C. O., Grindlay J. E., Edmonds P. D., 2005a, ApJ, 622, 556
The data underlying this article will be shared on reasonable request
Heinke C. O., Grindlay J. E., Edmonds P. D., Cohn H. N., Lugger P. M.,
to the corresponding author. Camilo F., Bogdanov S., Freire P. C., 2005b, ApJ, 625, 796
Heinke C. O., Wijnands R., Cohn H. N., Lugger P. M., Grindlay J. E., Pooley
D., Lewin W. H. G., 2006, ApJ, 651, 1098
REFERENCES
Heinke C. O. et al., 2020, MNRAS, 492, 5684
Albrow M. D., Gilliland R. L., Brown T. M., Edmonds P. D., Guhathakurta Heise J., Brinkman A. C., Gronenschild E., Watson M., King A. R., Stella L.,
P., Sarajedini A., 2001, ApJ, 559, 1060 Kieboom K., 1985, A&A, 148, L14
Arcodia R. et al., 2021, Nature, 592, 704 Hesser J. E., Harris W. E., Vandenberg D. A., Allwright J. W. B., Shott P.,
Bahramian A. et al., 2017, MNRAS, 467, 2199 Stetson P. B., 1987, PASP, 99, 739
Balman S¸ ., Revnivtsev M., 2012, A&A, 546, A112 Hills J. G., 1975, AJ, 80, 809
Baluev R. V., 2008, MNRAS, 385, 1279 Israel G. L., Esposito P., Rodrı́guez Castillo G. A., Sidoli L., 2016, MNRAS,
Bao T., Li Z., 2020, MNRAS, 498, 3513 462, 4371
Bao T., Li Z., 2022, MNRAS, 509, 3504 Ivanova N., Heinke C. O., Rasio F. A., Taam R. E., Belczynski K., Fregeau
Baskill D. S., Wheatley P. J., Osborne J. P., 2005, MNRAS, 357, 626 J., 2006, MNRAS, 372, 1043
Bassa C. G., Pooley D., Verbunt F., Homer L., Anderson S. F., Lewin W. H. Kalomeni B., Nelson L., Rappaport S., Molnar M., Quintin J., Yakut K., 2016,
G., 2008, A&A, 488, 921 ApJ, 833, 83
Belloni D., Giersz M., Askar A., Leigh N., Hypki A., 2016, MNRAS, 462, Katz J., Horwitz G., Klapisch M., 1975, ApJ, 199, 307
2950 Knigge C., 2012, Mem. Soc. Astron. Italiana, 83, 549
Belloni D., Giersz M., Rocha-Pinto H. J., Leigh N. W. C., Askar A., 2017, Knigge C., Zurek D. R., Shara M. M., Long K. S., Gilliland R. L., 2003, ApJ,
MNRAS, 464, 4077 599, 1320
Belloni D., Giersz M., Rivera Sandoval L. E., Askar A., Ciecielåg P., 2019, Knigge C., Dieball A., Maı́z Apellániz J., Long K. S., Zurek D. R., Shara M.
MNRAS, 483, 315 M., 2008, ApJ, 683, 1006
Bergbusch P. A., Stetson P. B., 2009, AJ, 138, 1455 Knigge C., Baraffe I., Patterson J., 2011, ApJS, 194, 28
Bogdanov S., Grindlay J. E., van den Berg M., 2005, ApJ, 630, 1029 Lan S.-H., Kong A. K. H., Verbunt F., Lewin W. H. G., Bassa C., Anderson
Camilo F., Lorimer D. R., Freire P., Lyne A. G., Manchester R. N., 2000, S. F., Pooley D., 2010, ApJ, 712, 380
ApJ, 535, 975 Linnell A. P., 1950, Harv. Coll. Obs. Circ., 455, 1
Carpano S., Jin C., 2018, MNRAS, 477, 3178 Lomb N. R., 1976, Ap&SS, 39, 447
Cheng Z., Li Z., Xu X., Li X., 2018, ApJ, 858, 33 Lu T.-N., Kong A. K. H., Bassa C., Verbunt F., Lewin W. H. G., Anderson S.
Cheng Z., Li Z., Li X., Xu X., Fang T., 2019a, ApJ, 876, 59 F., Pooley D., 2009, ApJ, 705, 175
Cheng Z., Li Z., Fang T., Li X., Xu X., 2019b, ApJ, 883, 90 Luo B. et al., 2017, ApJS, 228, 2

MNRAS 521, 4257–4276 (2023)

4276 T. Bao, Z. Li and Z. Cheng
Maxwell J. E., Lugger P. M., Cohn H. N., Heinke C. O., Grindlay J. E., Budac P = 48780.49 s, i.e. nearly identical to the actual case of Seq.312,
S. A., Drukier G. A., Bailyn C. D., 2012, ApJ, 756, 147 as trials. The phase φ is fixed at zero since it has no effect on the
Miller-Jones J. C. A. et al., 2015, MNRAS, 453, 3918 detection rate. The red noise is based on the best-fit power spectral
Miniutti G. et al., 2019, Nature, 573, 381 distribution for Seq.312 (Fig. 5). We follow the method described in
Modiano D., Parikh A. S., Wijnands R., 2020, A&A, 634, A132
Section 4.3 to generate a group of 1000 time series, combing the red
Mukai K., Kinkhabwala A., Peterson J. R., Kahn S. M., Paerels F., 2003, ApJ,
noise and the sinusoidal variation. These time series are then fed to the
586, L77
Mushotzky R. F., Szymkowiak A. E., 1988, in Fabian A. C.ed., Proc. NATO
GL algorithm over the searching range of (10 000, 50 000) s. It turns
Advanced Study Institute (ASI) Series C, Vol. 229, Cooling Flows in out that a periodic signal is detected in all 1000 simulated time series
Clusters and Galaxies. Kluwer, Dordrecht, p. 53 with a reported period within 0.1 per cent of the input period and a
Nabizadeh A., Balman S¸ ., 2020, Adv. Space Res., 66, 1139 GL probability greater than 0.99. This exercise thus demonstrates the
Oda M., Takagishi K., Matsuoka M., Miyamoto S., Ogawara Y., 1974, PASJ, efficiency of the GL algorithm in detecting periodic signals against
26, 303 substantial red noise.
Pala A. F. et al., 2020, MNRAS, 494, 3799
Pan H.-W., Yuan W., Yao S., Zhou X.-L., Liu B., Zhou H., Zhang S.-N., 2016,

Downloaded from https://academic.oup.com/mnras/article/521/3/4257/7081359 by guest on 09 April 2024

ApJ, 819, L19 A P P E N D I X B : I N T E R - O B S E RVAT I O N L I G H T
Parker T. L., Norton A. J., Mukai K., 2005, A&A, 439, 213 C U RV E S O F S E Q . 3 1 2 A N D S E Q . 2 7 3 S H OW I N G
Pooley D., Hut P., 2006, ApJ, 646, L143 P E R I O D I C VA R I AT I O N S
Pooley D. et al., 2002, ApJ, 569, 405
Pooley D. et al., 2003, ApJ, 591, L131 The long-term light curve of the eROSITA source Src-No.481, as
Predehl P. et al., 2021, A&A, 647, A1 shown in Fig. 2, exhibits quite regular periodic variation. Thanks
Pretorius M. L., Knigge C., 2012, MNRAS, 419, 1442 to the very small observation gaps, the periodicity is appreciable to
Ramsay G., Cropper M., Mason K. O., Córdova F. A., Priedhorsky W., 2004, the naked eye in this nearly continuous light curve. We perform a
MNRAS, 347, 95 similar exercise for Seq.312 and Seq.273. Despite the more irregular
Rappaport S., Doxsey R., Zaumen W., 1971, ApJ, 168, L43 cadence of the Chandra observations, we manage to construct the
Revnivtsev M. et al., 2010, A&A, 513, A63 inter-observation light curves from several long observations after
Ritter H., 1988, A&A, 202, 93
eliminating the observation gaps with an integer multiplication of the
Ritter H., Kolb U., 2003, A&A, 404, 301
detected period. As shown in Fig. B1, despite substantial aperiodic
Rivera Sandoval L. E. et al., 2018, MNRAS, 475, 4841
Saeedi S. et al., 2022, A&A, 661, A35
variability in both sources, their inter-observation light curve exhibits
Scargle J. D., 1982, ApJ, 263, 835 a peak-to-peak variation following the detected period, as in the case
Servillat M. et al., 2008, A&A, 490, 641 of Src-No.481. This strongly suggests that the detected periods are
Shara M. M., Hurley J. R., 2006, ApJ, 646, 464 intrinsic, rather than a fake signal caused by the red noise.
Sokolovsky K. V. et al., 2022, ApJ, 934, 142
Song J. R. et al., 2020, A&A, 644, L9
Timmer J., Koenig M., 1995, A&A, 300, 707
Tudor V. et al., 2018, MNRAS, 476, 1889
Tudor V. et al., 2022, MNRAS, 513, 3818
Vaughan S., 2010, MNRAS, 402, 307
Vaughan S., Edelson R., Warwick R. S., Uttley P., 2003, MNRAS, 345, 1271
Warner B., 1989, Inf. Bull. Var. Stars, 3383, 1
Xu X.-J., Wang Q. D., Li X.-D., 2016, ApJ, 818, 136
Yu Z. et al., 2019, MNRAS, 489, 1023
Zechmeister M., Kürster M., 2009, A&A, 496, 577
Zhang P.-f., Zhang P., Liao N.-h., Yan J.-z., Fan Y.-z., Liu Q.-z., 2018, ApJ,
853, 193
Zhang P.-F., Zhou J.-N., Yan D.-H., Yan J.-Z., Fan Y.-Z., Fang J., Zhang L.,
2020, ApJ, 904, L29
Zhao J., Heinke C. O., 2022, MNRAS, 511, 5964

A P P E N D I X A : A D D I T I O N A L C O N S I D E R AT I O N
ON RED NOISE
As discussed in Section 4.3, several periodic sources are accompa-
nied with substantial red noise, which may give rise to a relatively Figure B1. The Chandra 0.5–8 keV long-term light curve of Seq.312 (upper
high false detection probability. However, the presence of red noise panel) and Seq.273 (lower panel), covering several long observations that
together reveal the periodic variation (with a sinusoidal curve overlaid to
does not preclude the detection of true periodic variation. This can
guide the eye). Each set of coloured data points are from a single ObsID (as
be demonstrated as follows. labelled) and have a bin size of 7000 s for Seq.312 and 1500 s for Seq.273.
Specifically, we consider the possibility of detecting a periodic The yellow strips mark the eliminated gap (an integer multiplication of the
signal, which assumes an intrinsic sinusoidal form in the presence of period) between two consecutive observations.
red noise. The periodic variation is defined as f(t) = A × λ sin (ωt +
φ), where λ represents the mean source count rate, ω = 2π/P is the
This paper has been typeset from a TEX/LATEX file prepared by the author.
frequency, and A is the variation amplitude. We adopt A = 0.5 and

MNRAS 521, 4257–4276 (2023)