MNRAS 000, 1–12 (2019) Preprint 6 November 2020 Compiled using MNRAS LATEX style file v3.

The study of multi-peaked type-I X-ray bursts in the neutron-star

low mass X-ray binary 4U 1636−536 with RXTE

Chen Li1,2,3,4★ , Guobao Zhang1,2,3,4 †, Mariano Méndez 5 , Jiancheng Wang1,2,3,4 ,

and
1
Ming Lyu6,7
Yunnan Observatories, Chinese Academy of Sciences, Kunming 650216, People’s Republic of China
2 Key Laboratory for the Structure and Evolution Celestial Objects, Chinese Academy of Sciences, Kunming 650216, People’s Republic of China
3 Center for Astronomical Mega-Science, Chinese Academy of Sciences, Beijing 100012, People’s Republic of China
arXiv:2011.02736v1 [astro-ph.HE] 5 Nov 2020

4 University of Chinese Academy of Science, Beijing 100049, People’s Republic of China

5 Kapteyn Astronomical Institute, University of Groningen, P.O. Box 800, NL-9700 AV Groningen, The Netherlands
6 Department of Physics, Xiangtan University, Xiangtan, Hunan 411105, People’s Republic of China
7 Key Laboratory of Stars and Interstellar Medium, Xiangtan University, Xiangtan, Hunan 411105, People’s Republic of China

Accepted XXX. Received YYY; in original form ZZZ

ABSTRACT
We have found and analysed 16 multi-peaked type-I bursts from the neutron-star low mass
X-ray binary 4U 1636−53 with the Rossi X-ray Timing Explorer (RXTE). One of the bursts is
a rare quadruple-peaked burst which was not previously reported. All 16 bursts show a multi-
peaked structure not only in the X-ray light curves but also in the bolometric light curves. Most
of the multi-peaked bursts appear in observations during the transition from the hard to the
soft state in the colour-colour diagram. We find an anti-correlation between the second peak
flux and the separation time between two peaks. We also find that in the double-peaked bursts
the peak-flux ratio and the temperature of the thermal component in the pre-burst spectra are
correlated. This indicates that the double-peaked structure in the light curve of the bursts may
be affected by enhanced accretion rate in the disc, or increased temperature of the neutron star.
Key words: stars: individual: 4U 1636−53: binaries - X-rays: bursts

1 INTRODUCTION bursts with triple-peaked structure have also been observed in 4U

1636−53 (van Paradijs et al. 1986; Zhang et al. 2009). While inves-
Thermonuclear (Type I) X-ray bursts show a sudden increase in
tigating the cooling phase of X-ray bursts in 4U 1636−53, Zhang
X-ray intensity, becoming ∼ 10 − 100 times brighter than the per-
et al. (2011) reported 12 double-peaked bursts and found that most
sistent level, triggered by unstable ignition of accreted fuel on the
of them appeared at the vertex of colour-colour diagram. Recently,
surface of an accreting neutron star (NS) in low mass X-ray binaries
there were two new observations of double-peaked burst, one in
(LMXBs) (Galloway et al. 2008; Galloway & Keek 2017). Type I
the soft spectral state in 4U 1608−52 (Jaisawal et al. 2019), and
X-ray bursts were first detected in 1975 in the binary 3𝐴1820 − 30
another one in SAX J1808.4−3658 (Bult et al. 2019), both using
in the globular cluster NGC 6624 (Grindlay 1976); subsequently a
the Neutron Star Interior Composition Explorer (NICER).
growing population of bursters has been observed by different X-
The double-peaked structures in the burst light curve can be
ray satellites (Galloway & Keek 2017). In a typical X-ray burst, the
separated into two groups. The first one consists of bursts with
light curve shows a single-peaked profile with a fast rise (∼ 1 − 5
a double-peaked profile in X-rays but a single-peaked profile in
s) and an exponential decay within 10 − 100 s (Lewin et al. 1993;
the bolometric lightcurve, generally accompanied by photospheric
Strohmayer & Bildsten 2006; Galloway et al. 2008).
radius expansion (PRE), where the flux of the burst reaches the
Besides the single-peaked normal bursts, multi-peaked bursts
Eddington luminosity. In this case, the temperature of the photo-
have also been reported in previous studies. Double-peaked bursts
sphere temporarily shifts out of the instrument passband, causing
have been reported in several NS-LMXBs, e.g., 4U 1608−52 (Pen-
an apparent dip in the observed X-ray light curve (Paczynski 1983).
ninx et al. 1989), GX 17+2 (Kuulkers et al. 2002), 4U 1709−267
The other group consists of bursts that have a double-peaked profile
(Jonker et al. 2004) and MXB 1730−335 (Bagnoli et al. 2014). With
both in X-rays and the bolometric lightcurve. Most of these bursts
the Rossi X-ray Timing Explorer (RXTE), Watts & Maurer (2007)
have low peak flux, although PRE bursts with double-peaked pro-
analyzed 4 double-peaked bursts in 4U 1636−53. Still more rare,
files both in X-rays and bolometric luminosity have been recently
observed with NICER (Jaisawal et al. 2019; Bult et al. 2019).
★ E-mail: [email protected] Several theoretical models have been proposed to explain the
† E-mail: [email protected] double-peaked bursts. Fujimoto et al. (1988) proposed a model of

© 2019 The Authors

2 Chen Li et al.
stepped thermonuclear energy generation due to shear instabilities (Strohmayer et al. 1998a,b). 4U 1636−53 is an Atoll source, and as
in the fuel on the NS surface. Melia & Zylstra (1992) suggested the source moves in the colour-colour diagram (hereafter CCD) from
that the double-peaked bursts are due to the scattering of the X-ray the top right to bottom right, the accretion rate gradually increases,
emission by material evaporated from the disk during the burst. with a transition from the Island to the Banana state (Hasinger & van
These models, however, can not reproduce the observed double- der Klis 1989). The single peak bursts show a uniform distribution in
peaked profiles in the light curve, black-body temperature and ra- the CCD (Zhang et al. 2011). About a dozen double-peaked and two
dius (Bhattacharyya & Strohmayer 2006a). Fisker et al. (2004) sug- triple-peaked X-ray bursts have been discovered from this source
gested that a waiting point impedes the nuclear reaction flow and using different satellites (Sztajno et al. 1985; van Paradijs et al.
causes a stepped release of thermonuclear energy, but this idea has 1986; Lewin et al. 1987; Bhattacharyya & Strohmayer 2006a,b;
difficulties in explaining the large dips observed between the two Watts & Maurer 2007; Galloway et al. 2008; Zhang et al. 2009,
peaks (Bhattacharyya & Strohmayer 2006a). The thermonuclear 2011). The large multi-peaked bursts sample makes 4U 1636−53
flame spreading model provided by Bhattacharyya & Strohmayer an ideal source to study the properties and evolution of this kind of
(2006a) suggested that the double-peaked structure is caused by bursts.
high latitude ignition and stalling approaching the equator. This The structure of this paper is organised as follows. In Section
model qualitatively explains the essential features of the light curve 2 we describe the data analysis of our sample. In Section 3 we show
and reproduces the spectral evolution of two double-peaked bursts the results of light curves and spectra. In Section 4 we discuss our
in 4U 1636−53. However, one of the problems of the flame spread- findings in the context of previous theoretical work.
ing model is that it can not explain the triple-peaked bursts (Zhang
et al. 2009). Lampe et al. (2016) in their simulations found that
low accretion rate and high metallicity could affect the burst mor-
phology and produce twin-peaked structure when a large amount of 2 DATA ANALYSIS
hydrogen has been depleted. Recently, Bult et al. (2019) suggested The Rossi X-ray Timing Explorer (RXTE) was launched in 1995,
that the bright double-peaked bursts are due to the local Edding- and operated until 2012 with a circular orbit at an altitude of 580
ton limits associated with the hydrogen and helium layers of the km, correspoding to an orbital period of about 96 min (Bradt et al.
NS envelope. Understanding these mechanisms is important, be- 1993). We analysed all archived data from the proportional Counter
cause current models for multi-peaked X-ray bursts have met with Array (PCA) which is the main instrument onboard RXTE. The PCA
only partial success in explaining their light curves and temperature consists of five collimated proportional counter units (PCUs), which
profiles. are sensitive in the 2−60 keV energy range with an energy resolution
Burst properties in an individual system depend mainly on ac- of ∼ 1 keV at 6 keV (Jahoda et al. 2006). For each observation
cretion rate (Fujimoto et al. 1981; Bildsten 2000; Zhang et al. 2011; we used the Standard2 data (16-s time resolution and 129 energy
Galloway & Keek 2017). For a specific source, given a certain channels) to calculate X-ray colours. We used the Standard1 mode
global accretion rate, the local accretion rate varies with latitude, (only the PCU2) to produce the burst light curves. For the time-
being higher at the equator and lower at high latitude (Cooper & resolved spectral analysis of the bursts, we extracted spectra in 64
Narayan 2007). The ignition latitude depends on the column depth channels from the Event data of all available PCUs.
related to accretion rate to trigger a burst. As the increasing global We studied 336 type I X-ray bursts (as in Zhang et al. 2013)
accretion rate, the lower latitude region firstly reach the critical lo- in LMXB 4U 1636−53 with RXTE and discover 16 multi-peaked
cal accretion rate and the fuel become stable burning, the ignition bursts. The 0.125s bin light curves of these 16 bursts in the 2 − 60
should occur at higher latitude even pole. Concerning the research keV are shown in Figure 1. Following the procedure in Zhang et al.
of Cooper & Narayan (2007) with the thermonuclear spreading (2011) and Zhang et al. (2013), we searched the 1s Standard1 light
model of Bhattacharyya & Strohmayer (2006a), Watts & Maurer curve for burst visually, and considered that the start time of a
(2007) expected to find more double-peaked bursts at higher global burst is when the flux is larger than 3 times the 1𝜎 error of the
accretion rates than the single-peaked bursts. However, Watts & average persistent flux. We find 14 double-peaked bursts (4 bursts
Maurer (2007) presented the analysis of the accretion rate limited has investigated by Watts & Maurer 2007), one triple-peaked burst
of 4 double-peaked bursts and posed a challenge to the above ex- (Zhang et al. 2009) and one quadruple-peaked burst which was not
pectation. In this paper, we collect a large sample to provide a more reported in previous work. We divided the 16 bursts into two classes
complete description of the observational features of multi-peaked according to the number of peaks. All the double-peaked bursts are
bursts and further discuss the relation between accretion rate and classified as Class 1 bursts and the bursts with more than two peaks
multi-peaked structure. are classified as Class 2 bursts.
In the time-resolved spectral analysis, the standard approach To study these bursts in detail, we introduce several parameters
is to fit the X-ray burst spectra by assuming a constant persistent to characterise the burst light curve (see Table 1 and Table 2). In
emission (non-burst component) during the burst (Galloway et al. Table 1, the column 𝑛 is the burst number sorted by the value of
2008). However, recent studies provide evidence of enhanced accre- peak flux ratio, 𝑟 1,2 (see details later). Because most of the data
tion during type-I X-ray bursts (Worpel et al. 2013, 2015). The 𝑓 𝑎 around the peak show a symmetric distribution, we used a Gaussian
method gives improvements in the quality of spectral fit compared function to fit the data around each single peak in each burst light
to the standard approach, and the analysis is sensitive to changes in curve to get the peaking time. The quantity 𝑡 𝑝,1 is the peak time of
the persistent spectrum in the 2.5 − 20 keV (Worpel et al. 2015). In the first peak, 𝛿 is the separation time between the first and second
this paper, we adopt both the standard approach and the 𝑓 𝑎 method peak, and 𝑡 𝑝,2 is the peak time of the second peak. In Table 2,
to analyze our sample of multi-peaked bursts. we list the characteristics of the burst light curve in Class 2 bursts.
The LMXB 4U 1636−53 is one of the best-studied sources of The fitting process of triple-peaked burst is similar to that of the
X-ray bursts. The NS is in a binary system in a 3.8 hr orbit (van double-peaked bursts. About the quadruple-peaked burst, we have
Paradijs et al. 1990) with an 18th magnitude blue star companion a more detailed discussion in Sec 3.5. The quantities 𝑡 𝑝,3 and 𝑡 𝑝,4
(Galloway et al. 2008), and the spin period of the NS is 581 Hz are the peak time of the third and fourth peak, respectively, 𝛿23 and

MNRAS 000, 1–12 (2019)

 multi-peaked bursts in 4U 1636-53 3

5 3
#1 #9
4
3 2
2 1
1
0 0
3
3 #2 #10
2 2

1 1

0 0
3
3 #3 #11
2 2

1 1
Rate (103 counts/s)

Rate (103 counts/s)

0 0
3
4 #4 #12
3 2
2
1
1
0 0
3 #5 2 #13
2
1
1

0 0
3
2 #6 #14
2
1
1

0 0
3
3 #7 #15
2 2

1 1

0 0
3
3 #8 #16
2 2

1 1

00 5 10 15 20 25 30 35 00 5 10 15 20 25 30 35
time(s) time(s)

Figure 1. The light curves in the 2 − 60 keV range of 16 multi-peaked bursts in 4U 1636−53 at 0.125s resolution observed with RXTE. The 14 double-peaked
bursts are defined as Class 1 bursts and are ordered by peak flux ratio (the ratio of the first peak flux to the second peak flux, 𝑟1,2 ). Burst #15 (three peaks) and
burst #16 (four peaks) are defined as Class 2 bursts.

𝛿34 are the separation time between the second and the third peak, We extracted the 64-s interval spectrum prior to a burst as
as well as between the third and the fourth peak, respectively. For persistent emission. We generated the instrument response matrix
all of these parameters we give the 1𝜎 error. using the tool pcarsp and the instrumental background using the
To trace the spectral state of the source when these multi- tool pcabackest in HEAsoft for each spectrum. In this work, we
peaked bursts appear, we made a CCD as shown in Figure 2. We fitted the spectrum in the 3.0 − 20 keV band using XSPEC version
defined the soft colour as the ratio of the count rate in the 3.5 − 6.0 12.10.1 (Arnaud 1996). We added a 0.5% systematic error to the
keV to the count rate in the 2.0 − 3.5 keV bands, and the hard colour pre-burst spectra because of calibration uncertainties. During the
as the ratio of the count rate in the 9.7 − 16.0 keV to the count rate fitting process, we included the effect of interstellar absorption using
in the 6.0 − 9.7 keV bands (Zhang et al. 2009). The colours of the the cross-sections of Balucinska-Church & McCammon (1992) and
source are normalised by the Crab. For the colours of the source solar abundances from Anders & Grevesse (1989), with a fixed
before the burst, we used 64-s of the pre-burst spectrum. In Figure hydrogen column density, 𝑁H , of 0.36 × 1022 cm−2 (Pandel et al.
2, the grey points represent all available observations. The black 2008).
crosses represent all the bursts in this source. The red filled circles
stand for Class 1 bursts, and the blue filled triangles indicate the
Class 2 bursts. We find that most of the multiple-peaked bursts are
located close to the vertex of the CCD.

MNRAS 000, 1–12 (2019)

4 Chen Li et al.

Table 1. Properties of the light curves of Class 1 bursts in 4U 1636−53. The column 𝑛 gives burst number sorted by the value of the peak flux ratio (𝑟1,2 )
which is the ratio of the first peak to the second peak flux. We used a Gaussian function to fit each peak of every burst. The quantify 𝑡 𝑝,1 gives the peak time
of the first peak, 𝛿 is the separation time between the first and the second peak, 𝑡 𝑝,2 is the peak time of the second peak.

Start time (UTC) Obsid 𝑛 𝑟1,2 𝑡 𝑝,1 (s) 𝛿(s) 𝑡 𝑝,2 (s)
2002-02-28 23:42:54.756 60032-05-15-00 #1 0.29±0.04 1.6±0.3 4.0±0.3 5.6±0.1
2010-01-16 01:59:57.756 95087-01-08-00 #2 0.40±0.08 1.1±0.2 5.1±0.3 6.2±0.1
2001-09-05 08:15:05.756 60032-01-09-01 #3 0.61±0.08 3.0±0.4 4.9±0.4 7.9±0.2
2009-02-20 03:43:55.756 94087-01-29-00 #4 0.65±0.10 3.1±0.2 4.9±0.2 8.0±0.1
2001-10-03 00:22:20.756 60032-01-13-01 #5 0.73±0.09 3.1±0.1 4.0±0.1 7.1±0.1
2002-01-15 14:08:16.756 60032-05-08-00 #6 0.75±0.13 4.2±0.2 5.5±0.3 9.8±0.2
2006-02-06 16:21:20.756 91024-01-72-10 #7 0.79±0.11 3.7±0.1 6.3±0.2 10.0±0.1
Class1
2008-05-20 01:15:35.256 93087-01-66-10 #8 0.81±0.20 5.0±0.2 6.6±0.3 11.6±0.2
2002-01-08 12:22:46.256 60032-01-19-000 #9 1.05±0.15 3.6±0.1 6.1±0.2 9.6±0.2
2005-10-09 06:01:21.756 91024-01-12-10 #10 1.10±0.15 3.6±0.2 4.7±0.3 8.4±0.2
2008-11-12 01:18:29.256 93087-01-56-20 #11 1.34±0.23 3.0±0.1 6.4±0.1 9.4±0.1
2006-08-08 16:40:14.256 92023-01-80-00 #12 1.56±0.22 3.0±0.1 6.3±0.1 9.3±0.1
2009-12-21 03:44:05.756 94087-01-83-10 #13 2.24±0.62 3.0±0.1 6.4±0.3 9.4±0.3
2006-08-17 09:24:22.256 92023-01-84-00 #14 2.85±0.74 2.7±0.1 6.4±0.2 9.1±0.2

Table 2. Properties of the light curves of Class 2 bursts in 4U 1636−53. The quantities 𝑡 𝑝,3 and 𝑡 𝑝,4 are the peak time of the third and fourth peak, respectively,
𝛿23 and 𝛿34 are the separation time between the second and the third peak, as well as between the third and the fourth peak, respectively. For the triple-peaked
burst, we use the same method that we used in double-peaked bursts. For the quadruple-peaked burst, we use two GAUSSIAN functions adding two BURS
models to fit all data of the light curve (see detailed discussion in Sec 3.5).

Start time (UTC) Obsid 𝑛 𝑡 𝑝,1 (s) 𝛿(s) 𝑡 𝑝,2 (s) 𝛿23 (s) 𝑡 𝑝,3 (s) 𝛿34 (s) 𝑡 𝑝,4 (s)
2006-12-11 18:23:26.756 92023-01-44-10 #15 3.2±0.1 4.4±0.1 7.6±0.1 4.0±0.2 11.6±0.1 − −
Class2
2010-12-26 20:26:19.756 95087-01-82-10 #16 3.4±0.1 3.7±0.1 7.1±0.1 11.8±0.2 18.9±0.2 2.7±0.2 21.6±0.1

1.2
Class 1 bursts
Class 2 bursts
1.1

1.0
hard colour

0.9

#8
0.8
#5
#3 #4
0.7 #12
#10 #9 #13
#6 #7 #1
0.6 #15 #2
#14
#16 #11

0.5
0.9 1.0 1.1 1.2 1.3 1.4
soft colour(Crab)
Figure 2. Colour-colour diagram (CCD) of all RXTE observations of 4U 1636-53. The colours of 4U 1636−53 are normalized to the colours of Crab. Grey
points represent all available observations with one point per observation. Black crosses stand for all X-ray burst in this source. Red filled circles and blue filled
triangles represent Class 1 and Class 2 bursts, respectively.

MNRAS 000, 1–12 (2019)

 multi-peaked bursts in 4U 1636-53 5
bolometric flux (not the net burst count rate ratio in Figure 1) from
the standard approach to calculate the ratio 𝑟 1,2 .
After subtracting the instrumental background for the time-
6 resolved spectra, we used another model that allows us to vary
the pre-burst spectrum by a free scaling factor (Worpel et al.
5 2013). We used two models to describe the persistent emission:
TBabs*(bbodyrad + powerlaw) and TBabs*(diskbb + powerlaw)
4 in XSPEC. The fit results for the above two models are shown in
tp, 1 (s)

Tables 3 and 4, respectively. In Table 3, 𝑘𝑇bb is the blackbody tem-

3 perature and 𝐾bb is the normalization of the balckbody, 𝛼 is the
power-law index and 𝑁 is the normalization of the power law. The
2
spectral parameters of our best-fitting model are given with 1 𝜎
1 error. In Table 4, 𝑘𝑇dbb is the disk blackbody temperature and 𝐾dbb
is the normalization of the disk balckbody. Comparing the fitting
0 results of the two pre-burst spectra models (see Table 3 and Ta-
7 ble 4), we selected the TBabs*(bbodyrad + powerlaw) as the best
pre-burst model, so we used TBabs*( 𝑓 𝑎 *(bbodyrad + powerlaw) +
bbodyrad) in our analysis (the so-called 𝑓 𝑎 method) to re-fit the net
6 burst spectra. The parameter 𝑓 𝑎 was allowed to vary between −100
and 100 during the fits.
(s)

5 In Tables 3 and 4, 𝐹p represents the persistent unabsorbed flux

in the 2.5 − 25 keV band and 𝛾 is the luminosity of the source before
the burst in Eddington units. We used the unabsorbed 2.5 − 25
4 keV flux, 𝐹p , a bolometric correction factor 𝑐 bol = 1.2 (Galloway
et al. 2008), and the source distance of 5.95 kpc to estimate the
X-ray persistent luminosity, 𝐿 p . In the calculation of Eddington
3 0.5 1.0 1.5 2.0 2.5 3.0 3.5 luminosity we assume that the ratio of the colour temperature to the
r1, 2 effective temperature, 𝑇𝑐 /𝑇𝑒 , is 1.4 (Madej et al. 2004), the NS mass
is 1.4 𝑀  and the hydrogen mass fraction X = 0.7. We note that, in
the model of TBabs*(bbodyrad+powerlaw) 𝑇𝑐 corresponds to 𝑘𝑇bb
Figure 3. The peak time (𝑡 𝑝,1 ) of the first peak (upper panel) and the and in the model of TBabs*(diskbb+powerlaw) 𝑇𝑐 corresponds to
separation time ( 𝛿) between the two peaks (lower panel) against the peak 𝑘𝑇dbb .
flux ratio (𝑟1,2 ) in the sample of the double-peaked bursts of 4U 1636−53.
Grey dashed line shows the ratio 𝑟1,2 = 1. All these data are given in Table
1.
3 RESULTS
3.1 Multi-peaked bursts light curves
We used all available PCUs during the X-ray bursts to produce
time-resolved spectrum. We corrected every spectrum for dead time Figure 1 shows the 14 bursts with double-peaked profiles. The bursts
according to the methods supplied by the RXTE team. Since the display large or gentle dips in their light curves. The first peak can
light curve of bursts decay is quite smooth, to compensate the lower be either weaker/shorter or stronger/longer than the second peak.
count rates, we extracted spectrum over longer intervals in the tail In Figure 3 we show the rise time, 𝑡 𝑝,1 , of the first peak and the
of the bursts. For each multiple-peaked burst, we generated one separation time, 𝛿, of the two peaks as a function of the peak-flux
instrument response matrix using the pcarsp and the instrumental ratio, 𝑟 1,2 . The grey dashed line represents 𝑟 1,2 = 1. The rise time
background using the pcabackest in HEAsoft. of the first peak is in the range of ∼ 1 − 6s, and the separation in
For the burst spectral analysis, we initially used a single- time between the two peaks is in the range of ∼ 3 − 7s. We find that:
temperature blackbody model, TBabs*bbodyrad, to fit the net burst (1) in the case of 𝑟 1,2 < 1, as the peak flux ratio increases, both the
spectra, which is well established as a standard procedure in X-ray rise time of the first peak and the separation in time increase; (2) in
burst analysis (Kuulkers et al. 2002; Galloway et al. 2008). The the case of 𝑟 1,2 > 1, the above parameters do not depend upon peak
model provides the blackbody colour temperature, 𝑘𝑇bb , and the flux ratio.
normalization, 𝐾bb , proportional to the square of the blackbody ra- Besides the triple-peaked burst (burst #15) reported by Zhang
dius of the burst emission surface, and it allows us to estimate the et al. (2009), we also discovered a ∼ 40 s long quadruple-peaked
bolometric luminosity as a function of time assuming a distance burst (burst #16; date of observation: 2006 August 17), which was
of 5.95 kpc (Fiocchi et al. 2006). The bursts bolometric flux are not reported in previous studies.
calculated as
𝑘𝑇bb 4
 
𝐹 = 1.076 × 10−11 𝐾bb ergs cm−2 s−1 , (1) 3.2 Multi-peaked bursts in the CCD
1 keV
Figure 2 shows the position of all multi-peaked bursts in the CCD
where 𝐾bb = 𝑅km 2 /𝐷 2 , 𝑅
10 km is the effective radius of the emitting of 4U 1636−53. The normal, single-peaked, bursts observed with
surface in km, and 𝐷 10 is distance to the source in units of 10 kpc. RXTE are distributed more or less uniformly across the CCD. How-
We defined 𝑟 1,2 as the ratio of the first peak to the second peak flux. ever, except for burst #1 and #8, most of the multiple-peaked bursts
We note that, to reduce the effect of the instrument, we used the are located close to the vertex of the CCD.

MNRAS 000, 1–12 (2019)

6 Chen Li et al.

Table 3. Best-fitting parameters of the persistent emission before the 16 multi-peaked bursts in 4U 1636−53 with the model TBabs*(bbodyrad+powerlaw).
The quantity 𝑘𝑇bb is the blackbody temperature, 𝐾bb is the normalisation of the blackbody, 𝛼 is the power-law index, 𝑁 is the normalisation of the power law,
𝐹p represents the persistent unabsorbed flux in the 2.5 − 25 keV band and 𝛾 is the luminosity of the source before the burst in Eddington units. The spectral
parameters of our best-fitting model are given with 1 𝜎 error.

Obsid 𝑛 𝜒𝜈2 𝑘𝑇bb 𝐾bb 𝛼 𝑁 𝐹p 𝛾

(keV) (10−9 ergs−1 cm−2 ) 𝐿p /𝐿Edd
60032-05-15-00 #1 1.19 1.68±0.03 27.90±1.72 2.72±0.04 3.38±0.25 5.31 13.0%
95087-01-08-00 #2 1.89 1.78±0.06 8.10±0.98 2.71±0.09 1.44±0.19 2.17 5.3%
60032-01-09-01 #3 1.75 1.79±0.04 10.64±0.74 2.92±0.07 2.60±0.27 2.81 6.9%
94087-01-29-00 #4 0.90 1.78±0.06 7.68±0.94 2.84±0.09 1.95±0.25 2.24 5.5%
60032-01-13-01 #5 1.08 1.84±0.04 8.81±0.66 3.03±0.11 2.63±0.38 2.44 6.0%
60032-05-08-00 #6 1.42 1.75±0.06 7.26±1.02 2.62±0.08 1.28±0.15 2.09 5.1%
91024-01-72-10 #7 0.82 1.69±0.05 11.92±1.26 2.77±0.07 1.96±0.22 2.61 6.4%
Class1
93087-01-66-10 #8 2.20 1.57±0.12 10.20±3.99 2.20±0.09 0.81±0.16 2.60 6.4%
60032-01-19-000 #9 1.00 1.82±0.06 6.66±0.75 2.78±0.09 1.75±0.22 2.20 5.4%
91024-01-12-10 #10 1.22 1.79±0.07 5.15±0.74 2.71±0.11 1.11±0.17 1.58 3.9%
93087-01-56-20 #11 1.09 1.79±0.06 8.70±0.96 2.92±0.13 1.90±0.35 2.15 5.3%
92023-01-80-00 #12 0.92 1.81±0.05 8.10±0.79 2.81±0.09 1.99±0.24 2.44 6.0%
94087-01-83-10 #13 1.54 1.87±0.05 8.33±0.07 3.06±0.15 2.66±0.51 2.40 5.9%
92023-01-84-00 #14 2.25 1.87±0.09 6.73±1.07 2.93±0.18 2.12±0.53 2.19 5.4%
92023-01-44-10 #15 1.11 1.86±0.04 5.93±0.46 3.07±0.13 1.94±0.31 1.70 4.2%
Class2
95087-01-82-10 #16 1.38 1.82±0.06 10.02±1.15 3.18±0.10 2.92±0.98 2.34 5.7%

Table 4. Best-fitting parameters of the persistent emission before the 16 multi-peaked bursts in 4U 1636−53 with the model TBabs*(diskbb+powerlaw). The
quantity 𝑘𝑇dbb is the disc blackbody temperature, 𝐾dbb is the normalisation of the disk blackbody. 𝛼 is the power-law index, 𝑁 is the normalisation of the
power law, 𝐹p represents the persistent unabsorbed flux in the 2.5 − 25 keV band and 𝛾 is the luminosity of the source before the burst in Eddington units. The
spectral parameters of our best-fitting model are given with 1 𝜎 error.

Obsid 𝑛 𝜒𝜈2 𝑘𝑇dbb 𝐾dbb 𝛼 𝑁 𝐹p 𝛾

(keV) (10−9 ergs−1 cm−2 ) 𝐿p /𝐿Edd
60032-05-15-00 #1 0.72 2.42±0.05 6.60±0.59 2.62±0.12 1.48±0.29 5.20 12.8%
95087-01-08-00 #2 2.02 2.66±0.13 1.68±0.26 2.80±0.31 0.97±0.38 2.14 5.3%
60032-01-09-01 #3 1.90 2.66±0.05 2.38±0.16 3.32±0.37 2.46±1.10 2.76 6.8%
94087-01-29-00 #4 0.84 2.65±0.11 1.64±0.22 3.02±0.29 1.61±0.55 2.21 5.4%
60032-01-13-01 #5 1.13 2.68±0.04 2.19±0.26 3.94±0.63 4.17±3.42 2.41 5.9%
60032-05-08-00 #6 1.37 2.59±0.14 1.56±0.31 2.62±0.19 0.83±0.22 2.05 5.0%
91024-01-72-10 #7 0.72 2.47±0.10 2.68±0.38 2.81±0.20 1.25±0.33 2.56 6.3%
Class1
93087-01-66-10 #8 2.22 2.31±0.24 2.35±1.24 2.05±0.20 0.47±0.21 2.58 6.3%
60032-01-19-000 #9 1.27 2.74±0.12 1.38±0.18 2.96±0.34 1.45±0.63 2.16 5.3%
91024-01-12-10 #10 1.12 2.70±0.14 1.07±0.18 2.86±0.32 0.84±0.33 1.54 3.8%
93087-01-56-20 #11 0.98 2.63±0.08 1.99±0.22 3.35±0.61 1.67±1.45 2.10 5.2%
92023-01-80-00 #12 0.90 2.72±0.08 1.72±0.16 3.10±0.34 1.75±0.70 2.40 5.9%
94087-01-83-10 #13 1.41 2.71±0.05 2.10±0.30 4.04±0.75 4.41±4.76 2.37 5.8%
92023-01-84-00 #14 2.17 2.78±0.09 1.58±0.35 3.66±0.94 2.94±4.85 2.16 5.3%
92023-01-44-10 #15 1.14 2.69±0.05 1.53±0.19 4.13±0.62 3.62±2.85 1.68 4.1%
Class2
95087-01-82-10 #16 1.45 2.53±0.07 3.03±0.46 5.80±1.59 29.94±108.07 2.37 5.8%

With the lowest peak flux ratio (𝑟 1,2 ∼ 0.3) and highest peak 3.3 The relation between pre-burst spectrum and burst
flux (∼ 5.2 × 10−8 ergs cm−2 s−1 using the standard approach), profile
burst #1 is located at the bottom right, when the source was in the
so-called upper banana branch in the CCD. In the light curve of burst To investigate the relation between the pre-burst spectra and the
#1, the first peak is relative weak (𝑟 1,2 ∼ 0.3) and the whole burst is multi-peaked burst light curve profiles, we compared the spectral
dominated by the second peak. Burst #1 has also the second shortest parameters of the persistent emission of Class 1 bursts with two
rise time (𝑡 𝑝,1 ∼ 1.6 s) of the first peak and shortest separation time properties of the light curve: the peak flux ratio, 𝑟 1,2 , and the sep-
(𝛿 ∼ 4.0 s). aration time of two peaks, 𝛿. Figure 4 shows the spectral param-
eters of the persistent emission (in two models) against the peak
flux ratio. The red dots represent the Class 1 bursts, and the blue
triangles represent Class 2 bursts. The left panels (a) and (b) in Fig-
ure 4 display, respectively, the blackbody temperature, 𝑘𝑇bb , and
Burst #8 is located at the position between the Banana and power-law index, 𝛼, against the peak flux ratio, 𝑟 1,2 , for the model
the Island states in the CCD. Different from burst #1, burst #8 has TBabs*(bbodyrad+powerlaw). The blackbody temperature ranges
the longest rise time (𝑡𝑟 ,1 ∼ 5.0 s) of the first peak and the longest from 1.57 keV to 1.87 keV. There appears to be a positive correla-
separation time (𝛿 ∼ 6.6 s), and a high 𝑟 1,2 (∼ 0.8). tion between the blackbody temperature and the peak flux ratio. In

MNRAS 000, 1–12 (2019)

 multi-peaked bursts in 4U 1636-53 7
order to check that, we firstly fit these data with a constant model (Suleimanov et al. 2011; Zhang et al. 2011). The 𝑓 𝑎 plotted in the
and get 𝜒2 = 24.6 for 13 d.o.f. We then fit these data with a line bottom panel shows values larger than one during the whole burst.
function, and get a slope of 0.07 ± 0.02 with a 𝜒2 = 13.8 for 12 Due to the new parameter 𝑓 𝑎 , the bolometric flux is lower than in
d.o.f. The F-test probability for these two fits is 0.009, indicating the standard approach and has larger error bars. The horizontal solid
that a linear function is slightly better than a constant. line in the bottom panels stands for 𝑓 𝑎 =1. In general, the reduced
The right panels (c) and (d) in Figure 4 show, respectively, the 𝜒2 distribution in the 𝑓 𝑎 method is lower than that in the standard
disk-blackbody temperature, 𝑘𝑇dbb , and power-law index, 𝛼, against approach.
the peak flux ratio, 𝑟 1,2 for the model TBabs*(diskbb+powerlaw). Finally, we investigated the 𝑓 𝑎 time evolution of these 16 multi-
For panel (c) of Figure 4, we do the same analysis as in panel (a), peaked bursts in Figure 8 using the same order as in Figure 1. The
obtaining a 𝜒2 = 31.7 for 13 d.o.f. for a constant model and a horizontal black dashed lines correspond to 𝑓 𝑎 = 1. In all multi-
slope of 0.09 ± 0.03 with a reduced chi-square of 1.64 (𝜒2 /𝑑𝑜 𝑓 = peaked bursts, the 𝑓 𝑎 values vary in the range of ∼ 1 − 7 during the
19.7/12) for the line function. The F-test probability for these two bursting period, which is similar to the range observed by Worpel
fits is 0.019. The above analysis indicates that the peak flux ratio, et al. (2013) in other bursts ( 𝑓 𝑎 ∼ 2 − 10). In general, high values
𝑟 1,2 , is marginally correlated with the temperature of the thermal of 𝑓 𝑎 coincide with large flux values during the burst. We study
component in the pre-burst spectra. the correlation between 𝑓 𝑎 and bolometric flux (e.g. blackbody
temperature and radius) for the 16 multi-peaked bursts using the
method of cross-correlation lags (Peterson et al. 1998; Sun et al.
3.4 Time-resolved burst spectra 2018). We do not find any obvious correlation between 𝑓 𝑎 and
We adopted both the standard approach and the 𝑓 𝑎 method to fit burst spectral parameters, and also we do not find any evidence
the time-resolved burst spectra. An example (burst #5) of the best- of a relation between the centroid of the cross-correlation function
fitting parameters using standard approach is shown in the left panel (time delay between 𝑓 𝑎 and bolometric flux) and the double-peaked
of Figure 5. From the top to the third panel, we display the blackbody structure (e.g. the peak-flux ratio).
temperature, blackbody radius and bolometric flux as a function of .
time. We used a 0.5-s bin light curve in the 2 − 60 keV range in
the fourth panel. The double-peaked structures occur simultane-
ously in the X-ray and bolometric light curves, which indicates that 3.5 The quadruple-peaked X-ray burst
the double-peaked profiles during the X-ray bursts is not due to a In Figure 9, we show the light curve of the quadruple-peaked X-ray
passband effect of the instrument. The time-resolved temperature burst. We use two GAUSSIAN functions adding two BURS models
shows two local maxima, as is also the case in the bolometric flux to fit the quadruple-peaked burst light curve data, where BURS
curve, however, the peaking time in temperature is different from the represents a model to describe the burst component in QDP 1 .
peaking time in bolometric flux curve. After initially growing, the
radius continues increasing following a dip, and then remains more  0 𝑡 < 𝑆𝑇
or less constant. The double-peaked structures are also apparent in 


𝐹 𝑁𝑌 = 𝐵𝑁 ∗ (𝑡 − 𝑆𝑇)/(𝑃𝑇 − 𝑆𝑇) 𝑆𝑇 < 𝑡 < 𝑃𝑇 , (2)
the bolometric flux in the rest of the bursts of Class 1 bursts.  𝐵𝑁 ∗ 𝐸 𝑋 𝑃(−(𝑡 − 𝑃𝑇)/𝐷𝑇)

𝑃𝑇 < 𝑡
We then measured the local peak temperature of the double- 
peaked bursts in the standard approach. The peak temperature ratio, where 𝑡 is time, 𝐹 𝑁𝑌 is the photons per second, 𝑆𝑇 and 𝑃𝑇 represent
𝜃 𝑘𝑇 represents the ratio of the first peak to the second peak temper- the start and peak time of each peak, respectively, and 𝐵𝑁 and 𝐷𝑇
ature. Figure 6 shows the peak temperature ratios against the peak stand for the normalisation and decay factor, respectively.
flux ratios in the 14 double-peaked bursts. The vertical and horizon- In the top panel of Figure 9, we show the fitting results, getting
tal dashed lines represent 𝑟 1,2 =1.0 and 𝜃 𝑘𝑇 =1.2, respectively. When a 𝜒2 = 120.64 for 62 d.o.f. In the bottom, we show the residual of
𝑟 1,2 > 1, the peak temperature ratios are always larger than 1.2. In the fit in units of the error, and the black horizontal line represents
the case of 𝑟 1,2 < 1, the first peak temperature can be higher or the (data-fit)/err = 0. To verify the existence of the third peak, we
lower than the second peak temperature and 𝜃 𝑘𝑇 is always less than also use three components (one GAUSSIAN function adding two
1.2. There appears to be a bimodal distribution in the temperature BURS models) to re-fit the light curve, getting a 𝜒2 = 386.79 for 65
ratios. d.o.f. The F-test probability for these two fits is 1.1 × 10−15 , which
Figure 7 shows the first and second peak flux in the stan- indicates that the third peak is significant. This is the first time that
dard approach against the duration time between two peaks, re- a quadruple-peaked burst is reported in 4U 1636−53.
spectively. The typical PRE peak flux is in the range of (6 − 8) × There are three relatively strong peaks (∼ 3s, ∼ 7s, ∼ 22s for
10−8 ergs cm−2 s−1 in 4U 1636−53 (Lyu et al. 2015). In our the first, second, fourth peak) in the light curve and one weak peak
present sample, there are no PRE events. There is no significant (∼ 19s for the third peak). There is a very long time separation
trend between the first peak flux and the duration time, however, (∼ 12s) between the second and the third peak. The separation
an anti-correlation is present between the second peak flux and the time between the first and last peak in this quadruple-peaked burst
separation time between two peaks. is ∼ 18s, which is similar to the separation (∼ 17s) in the triple-
We show the fits results of burst #5 with the 𝑓 𝑎 method in the peaked burst reported by van Paradijs et al. (1986), but about two
right-hand panel of Figure 5. As in the standard approach, both the times longer than that in the triple-peaked burst (∼ 8s) in Zhang
bolometric flux and the X-ray light curve show a double-peaked et al. (2009).
structure. The evolution of black body temperature and radius in We investigated the time-resolved spectra of the quadruple-
the 𝑓 𝑎 method is similar to that in the standard approach. The peaked burst with two models. In the standard approach (the left-
increased black body radius in the cooling phase may be explained hand panel of Figure 10), both the bolometric light curve and X-
in two ways. One would be the influence of persistent emission
(accretion geometry) in the soft state (Kajava et al. 2014), the other
would be different canonical composition in the NS atmosphere 1 https://heasarc.gsfc.nasa.gov/ftools/others/qdp/qdp.html

MNRAS 000, 1–12 (2019)

8 Chen Li et al.

2.8
1.9

1.8 2.6

kTdbb (keV)
kTbb (keV)

1.7 2.4

1.6 x2=13.81, dof=12 2.2

x2=19.67, dof=12
x2=24.58, dof=13 x2=31.68, dof=13
1.5 (a) (c)
2.0
0.5 1.0 1.5 2.0 2.5 0.5 1.0 1.5 2.0 2.5

7.5
3.3
6.0
3.0

2.7 4.5

2.4 3.0

2.1 (b) 1.5

(d)
0.5 1.0 1.5 2.0 2.5 0.5 1.0 1.5 2.0 2.5
r1, 2 r1, 2

Figure 4. Parameter of the persistent spectrum before the multi-peaked burst in 4U 1636−53. Left panels: blackbody temperature (𝑘𝑇bb ), and power-law index
(𝛼) for the model TBabs*(bbodyrad+powerlaw) against the peak flux ratio (𝑟1,2 ); right panels: the disk blackbody temperature (𝑘𝑇dbb ) and the power-law index
(𝛼) for the model TBabs*(diskbb+powerlaw) against the peak flux (𝑟1,2 ). Red dots and blue triangles represent Class1 and Class 2 bursts, respectively. We use
a linear (red line) and constant (black line) function to fit only the Class 1 bursts data.

ray light curve show four peaks, three pronounced and small one. phenomenon may be explained if the burst consumes less fuel dur-
The total burst fluence is (26.9 ± 3.5) × 10−8 ergs cm−2 and the ing the first peak, such that it is easier to release subsequently the
bolometric first peak flux is (2.0±0.6) ×10−8 ergs cm−2 s−1 . There fuel that is left to trigger the second peak. Therefore, the bursts with
is a ∼ 5s long waiting period from 13s to 18s in the light curve, and higher peak flux ratio need more time to accumulate fuel to release
the bolometric flux is always higher than the persistent emission. the second peak energy. As we know, in the bottom panel of Figure
The effective temperature profile, after an initially rising, decreases 7, we also find an anti-correlation between the second peak flux and
with time and shows three noticeable local maxima. The blackbody the separation time between two consecutive peaks. This indicates
radius increases steadily during the whole burst, which is similar to that the double-peaked bursts with longer peak separation time have
what was observed in the triple-peaked burst in Zhang et al. (2009). a weaker second peak. If we assume that the mass accretion is neg-
In the 𝑓 𝑎 method (the right-hand panel of Figure 10), both ligible during the burst, this anti-correlation indicates that when the
the bolometric flux and X-ray light curve display four peaks as separation time is long, there is less fuel for the second peak. These
in the standard approach. Similar to the double-peaked bursts, the results indicate that a single peak in these double-peaked bursts is
bolometric flux in the 𝑓 𝑎 method is lower than that of the standard not isolated, and the double-peaked profile is affected by the strength
approach due to the variation 𝑓 𝑎 being larger than one during the first and separation time of the two single peaks.
two peaks. The evolution of the blackbody radius and the blackbody Most of the multi-peaked bursts in our sample appear during
temperature in the 𝑓 𝑎 method is similar to that in the standard the transition from the hard to the soft state in the CCD where PRE
approach. bursts are also present (Watts & Maurer 2007; Zhang et al. 2009).
That the PRE and multi-peaked bursts appear in the same state of the
source may provide an important clue to understanding the origin
of the multi-peaked bursts, indicating that the appearance of the
4 DISCUSSION
multi-peaked bursts could be affected by the transition of the source
We analyzed all available RXTE data of the LMXB 4U 1636−53, from the hard to the soft state. If mass accretion rate increases from
and we found 14 double-peaked bursts with complex profiles, the upper right to the lower right in the CCD, there is no apparent
one triple-peaked burst, and one quadruple-peaked burst that had relation between mass accretion rate and the parameter 𝑟 1,2 . We
not been reported before. The bolometric flux light curve of the have checked the correlation between the persistent flux and the
multiple-peaked bursts exhibits a similar profile as the X-ray light double-peaked burst parameters (peak ratio 𝑟 1,2 ; peak separation
curve, which indicates that the multi-peaked structures in 4U 𝛿), but do not find a clear trend between them either.
1636−53 are not due to passband instrumental effect (Jaisawal et al. To further investigate if mass accretion affects the burst light
2019). Between the peaks, the flux of these multi-peaked bursts curve structure, we studied the pre-burst spectra. We find that there
never drops near or below the pre-burst level, which is different is a positive correlation between the peak-flux ratio and the temper-
from the triple bursts in Boirin et al. (2007) and Beri et al. (2019). ature of thermal component in the pre-burst spectra for the double-
We find a marginal positive correlation between the peak- peaked bursts, where we use a blackbody model to fit the burst
separation time and the peak-flux ratio in Class 1 bursts (bottom time-resolved spectra. The soft thermal component of X-ray spec-
panel of Figure 3). This suggests that double-peaked bursts with tra in accreting NS-LMXBs is generally explained by the emission
high peak-flux ratio value have a longer peak separation time. This from the NS surface and the accretion disc. Sanna et al. (2013)

MNRAS 000, 1–12 (2019)

 multi-peaked bursts in 4U 1636-53 9

1.5

1.4
standard approach fa method
3 1.3

2 1.2
kT (keV)

kT
1.1
1
1.0
0
20 0.9
15
R (km)

0.8
10
0.5 1.0 1.5 2.0 2.5 3.0 3.5
5 r1, 2
0
Rate (103count/s) Fb (10 8ergcm 2s 1)

3 Figure 6. The peak temperature ratio (𝜃𝑘𝑇 ) against the peak flux ratio
(𝑟1,2 ) for the double-peaked bursts in 4U 1636−53 obtained from fits using
2
the standard procedure to fit time-resolved spectra of X-ray bursts. The
1 vertical and horizontal dashed lines correspond to the 𝑟1,2 =1.0, 𝜃𝑘𝑇 =1.2
respectively. We note that the local maxima of temperature and flux do not
0 necessarily occur at the same time.
8
6
4
2
0 3
3
2
Fp, 1 (10 8 ergcm 2 s 1)
xv2

1 2
0
5
4
3 1
fa

2
1
0 0 5 10 15 20 25 0 5 10 15 20 25 0
time(s) time(s)
5
Fp, 2 (10 8 ergcm 2 s 1)

3
Figure 5. An example from Obsid 60032-01-13-01 of the spectral evolution
of double-peaked burst spectra in 4U 1636−53 using two burst models. Left 2
panel: standard approach, right panel: 𝑓𝑎 method. From top to bottom we
display the blackbody temperature, the radius of the emitting surface, the 1
bolometric flux, the 2 − 60 keV light curve at 0.5s bin extracted from all
available PCUs, 𝜒𝜈2 , and 𝑓𝑎 . The horizontal solid line in the bottom pannel 0 4.0 4.5 5.0 5.5 6.0 6.5 7.0
represents 𝑓𝑎 =1.
(s)

analyzed the X-ray spectra of six observations of 4U 1636−53 in

different spectral states taken with XMM-Newton and RXTE simul-
Figure 7. The first (𝐹𝑝,1 ) and second (𝐹𝑝,2 ) peak flux using the standard
taneously. They find that the temperature at the inner disc radius is
approach against the separation time ( 𝛿) between the two peaks of 14 double-
∼ 0.1 − 0.8 keV, and the temperature at the NS surface is ∼ 1.4 − 2.0 peaked bursts in 4U 1636−53.
keV. We used a black-body or a disk black-body to depict the en-
hancement of the thermal component. The emission from the NS
surface and inner region of the accretion disc are degenerate in our
persistent spectrum. The high blackbody (or disc blackbody) tem-
perature could be due to an enhanced accretion rate in the disc or
an increased temperature of the NS surface. Since the above discus-

MNRAS 000, 1–12 (2019)

10 Chen Li et al.

2000 data
fit:twoGAUSSIANs+ twoBURSs
1750 peak 1 fit
peak 2 fit
3 #1 1500 peak 3 fit
peak 4 fit

counts/s
2
1
0
1250
1
6
5
#2 1000
4
3
2
750
1
0
500
1
5
4
#3
3 250
2
1 reduced chi-sqr:1.95
0 3
1
5
4
#4 2

(data fit)/err
3
2
1 1
0
1
5 #5 0
4
3
2
1
1
0 2
6
5
#6
4
3
3
2
1
0 4 0 5 10 15 20 25 30 35
1
4
3
#7 time
2
1
0
1
5
4
#8
3
2
1
0 Figure 9. The light curve of the quadruple-peaked burst (burst #16) from
1
2
3 Obsid 95087-01-82-10 of 4U 1636−53. The four peaks can be well described
6
5
#9
4 by two GAUSSIAN functions adding two BURS models in light curve.
3
2
fa

1
0
1
6
5
#10
4
3
2
sion indicates that the double-peaked properties are not affected by
1
0
1
accretion rate, we suggest that the double- peaked bursts with high
2
6
5
#11 peak flux ratio might appear when the NS surface temperature is
4
3
2
high.
1
0 To date, there are several models to explain the unusual double-
1
5 #12
4
3
peaked structure in X-ray bursts. One of them is the thermonuclear
2
1 flame spreading model (Bhattacharyya & Strohmayer 2006a,b),
0
1
4 #13 which succeeded in explaining the large dip in the X-ray light curve
3
2
and reproducing the spectral evolution of double-peaked bursts.
1 Cooper & Narayan (2007) suggested that ignition latitude has a
0
5
4
#14 positive correlation with the accretion rate. Combing these ideas
3
2
1
with the results in Figure 4 and assuming that the temperature of
0
1 the NS is correlated to the accretion rate, the high peak flux ratio
6
5
#15
4
3
of double-peaked bursts would correspond to high latitude ignition
2
1 (Bhattacharyya & Strohmayer 2006a,b).
0
1
6
5
#16 Alternatively, Lampe et al. (2016) investigated the possibil-
4
3
2 ity of the nuclear origin of the double-peaked bursts based on the
1
0
1 model of nuclear waiting points in the rp-process explaining the
time(s)
2
0 5 10 15 20 25 30 35
double-peaked profile (Fisker et al. 2004). They find that for certain
metallicities and low accretion rate, there is an anti-correlation be-
tween the peak-flux ratio (𝑟 1,2 ) and accretion rate. We do not find
that the peak-flux ratio of double-peaked burst decreases when the
source moves in the CCD (see Figure 2) from the top right to the
bottom right, as the accretion rate gradually increases. If the tem-
perature before the burst is propotional to mass accretion rate, the
Figure 8. The time evolution of the 𝑓𝑎 factor of the 16 multi-peaked bursts correlation between the peak flux and the accretion rate does not
in the 4U 1636−53. Black horizontal dashed lines represent 𝑓𝑎 = 1. The agree (see Figure 4) with the simulation of Lampe et al. (2016) .
sequence of the bursts is the same as in Figure 1. The value of 𝑓𝑎 changes In their simulation, as accretion rate increases, the double-peaked
with the bolometric flux and is larger than 1 during the whole burst in all structure shows a distinct two stage, different from the large dip in
cases.
the bolometric flux profile of our sample (See Figure 1).
The newly discovered quadruple-peaked burst in 4U 1636−53
poses a problem to the thermonuclear spreading model of Bhat-
tacharyya & Strohmayer (2006a,b). If this model is applied to ex-

MNRAS 000, 1–12 (2019)

 multi-peaked bursts in 4U 1636-53 11
contains 14 double-peaked, one triple-peaked and one quadruple-
peaked burst; the latter had never been reported before.
(i) Most of the multi-peaked bursts in our sample appear during
3
standard approach fa method the transition from the hard to the soft state in the CCD.
(ii) We find that double-peaked bursts with high peak-flux
2
kT (keV)

ratio appear when the pre-burst temperature is high; the high NS

1 temperature may be due to enhanced accretion rate on to the NS
surface.
0
25 (iii) We find an anti-correlation between the second peak flux
20 and the peak-separation time for double-peaked bursts.
R (km)

15
10 (iv) The quadruple-peaked burst shows a long separation time
5 between the second and the third peak which is difficult to explain
0
3 with current models.
Rate (103count/s) Fb (10 8ergcm 2s 1)

2
(v) We use the 𝑓 𝑎 method to re-analyse these 16 bursts and we
find no evidence that the multi-peaked structure is due to enhanced
1
accretion during the bursts.
0
2

1
ACKNOWLEDGEMENTS
0
3 This research has made use of data obtained from the High En-
ergy Astrophysics Science Archive Research Center (HEASARC),
2
xv2

provided by NASA’s Goddard Space Flight Center and NASA’s As-

1
trophysics Data System Bibliographic Services. We thank Liantao
0 Cheng, Shasha Li for their helpful discussions. G.B. acknowledges
8
6 funding support from the National Natural Science Foundation of
4 China (NSFC) under grant No. U1838116 and Y7CZ181002. M.L.
fa

2
0
is supported by National Natural Science Foundation of China (grant
2
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
No.11803025) and the Hunan Provincial Natural Science Founda-
time(s) time(s) tion (grant No. 2018JJ3483) .

REFERENCES
Figure 10. The spectral evolution of the quadruple-peaked burst (burst Anders E., Grevesse N., 1989, Geochimica et Cosmochimica Acta, 53, 197
#16) from Obsid 95087-01-82-10 of 4U 1636−53 using both the standard Arnaud K. A., 1996, XSPEC: The First Ten Years. p. 17
approach and the 𝑓𝑎 method. The explanations of these panels are similar to Bagnoli T., in’t Zand J. J. M., Patruno A., Watts A. L., 2014, MNRAS, 437,
footnote in the Figure 5. The horizontal solid line in the bottom panel stands 2790
for 𝑓𝑎 =1. Balucinska-Church M., McCammon D., 1992, ApJ, 400, 699
Beri A., et al., 2019, MNRAS, 482, 4397
Bhattacharyya S., Strohmayer T. E., 2006a, ApJ, 636, L121
plain the four peaks, the burning front not only needs to stall three Bhattacharyya S., Strohmayer T. E., 2006b, ApJ, 641, L53
time, but it also needs a longer "waiting" period between the second Bildsten L., 2000, AIP Conference Proceedings, 522, 359
and third peak than the separation time in the double-peaked burst. Boirin L., Keek L. Méndez M., Cumming A., in’t Zand J. J. M., Cottam J.,
Similar to normal single-peaked bursts, we find that the 𝑓 𝑎 Paerels F., Lewin W. H. G., 2007, A&A, 465, 559
value is large when the flux is high during the multi-peaked bursts. Bradt H. V., Rothschild R. E., Swank J. H., 1993, A&AS, 97, 355
We investigated the relation between 𝑓 𝑎 and the burst spectral pa- Bult P., et al., 2019, ApJ, 885, L1
Cooper R. L., Narayan R., 2007, ApJ, 657, L29
rameters, but do not find any obvious correlation between them. The
Fiocchi M., Bazzano A., Ubertini P., Jean P., 2006, ApJ, 651, 416
enhanced 𝑓 𝑎 can be interpreted as changes in the accretion flow rate
Fisker J. L., Thielemann F.-K., Wiescher M., 2004, ApJ, 608, L61
by Poynting-Robertson drag (Walker 1992). However, it is difficult Fujimoto M. Y., Hanawa T., Miyaji S., 1981, ApJ, 247, 267
to explain the dip between peaks by reducing the accretion. Fujimoto M. Y., Sztajno M., Lewin W. H. G., van Paradijs J., 1988, A&A,
The double-peaked structure could be due to the influence of 199, L9
the variation of the accretion geometry, but we should observe more Galloway D. K., Keek L., 2017, arXiv e-prints, p. arXiv:1712.06227
double-peaked bursts to test this hypothesis. We note also that it is Galloway D. K., Muno M. P., Hartman J. M., Psaltis D., Chakrabarty D.,
hard to explain the existence of triple or quadruple-peaked profiles 2008, ApJS, 179, 360
with the variation of the geometry. Grindlay J. E., 1976, Comments on Astrophysics, 6, 165
Hasinger G., van der Klis M., 1989, A&A, 225, 79
Jahoda K., Markwardt C. B., Radeva Y., Rots A. H., Stark M. J., Swank
J. H., Strohmayer T. E., Zhang W., 2006, ApJS, 163, 401
5 SUMMARY Jaisawal G. K., et al., 2019, ApJ, 883, 61
Jonker P. G., Galloway D. K., McClintock J. E., Buxton M., Garcia M.,
We found 16 bursts with multi-peaked structure by investigating 336 Murray S., 2004, MNRAS, 354, 666
type I X-ray bursts in LMXB 4U 1636−53 with RXTE. Our sample Kajava J. J. E., et al., 2014, MNRAS, 445, 4218

MNRAS 000, 1–12 (2019)

12 Chen Li et al.
Kuulkers E., Homan J., van der Klis M., Lewin W. H. G., Méndez M., 2002, burst#6 burst#6
6000 6000
A&A, 382, 947 the peaked data
fit: two Gaussian functions
the peaked data
fit: one Gaussian function
5000 5000
Lampe N., Heger A., Galloway D. K., 2016, ApJ, 819, 46

counts/s

counts/s
4000 4000
Lewin W. H. G., Penninx W., van Paradijs J., Damen E., Sztajno M., Truem- 3000 3000

per J., van der Klis M., 1987, ApJ, 319, 893 2000 2000
1000 1000
Lewin W. H. G., van Paradijs J., Taam R. E., 1993, Space Sci. Rev., 62, 223 6
reduced chi-sqr:6.67
7.5
reduced chi-sqr:15.24
5.0
Lyu M., Méndez M., Zhang G., Keek L., 2015, MNRAS, 454, 541

fit)/err

fit)/err
4
2.5
2 0.0
Madej J., Joss P. C., Różańska A., 2004, ApJ, 602, 904
0 2.5
Melia F., Zylstra G. J., 1992, ApJ, 398, L53 5.0

(data

(data
2
7.5
Paczynski B., 1983, ApJ, 267, 315 4 10.0
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
Pandel D., Kaaret P., Corbel S., 2008, ApJ, 688, 1288 time time
Penninx W., Damen E., Tan J., Lewin W. H. G., van Paradijs J., 1989, A&A,
208, 146 burst#8 burst#8
2000 the peaked data 2000 the peaked data
Peterson B. M., Wanders I., Horne K., Collier S., Alexander T., Kaspi S., 1750 fit: two Gaussian functions 1750 fit: one Gaussian function
1500 1500

counts/s

counts/s
Maoz D., 1998, PASP, 110, 660 1250 1250
1000 1000
Sanna A., Hiemstra B., Méndez M., Altamirano D., Belloni T., Linares 750 750
500 500
M., 2013, Monthly Notices of the Royal Astronomical Society, 432, 250 250
4 reduced chi-sqr:2.85 8 reduced chi-sqr:10.08
1144–1161

(data fit)/err

(data fit)/err
3 6
Strohmayer T., Bildsten L., 2006, New views of thermonu- 2 4
1 2
clear bursts. Cambridge University Press, p. 113–156, 0 0
1 2
doi:10.1017/CBO9780511536281.004 2 4
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
Strohmayer T. E., Zhang W., Swank J. H., White N. E., Lapidus I., 1998a, time time
The Astrophysical Journal, 498, L135
Strohmayer T. E., Zhang W., Swank J. H., Lapidus I., 1998b, The Astro- Figure A1. The light curves of burst #6 and #8 of 4U 1636−53. The peaking
physical Journal, 503, L147 time data of these two bursts can be well described by two Gaussian function.
Suleimanov V., Poutanen J., Revnivtsev M., Werner K., 2011, ApJ, 742, 122
Sun M., Grier C. J., Peterson B. M., 2018, PyCCF: Python Cross Correlation
Function for reverberation mapping studies (ascl:1805.032)
Sztajno M., van Paradijs J., Lewin W. H. G., Trumper J., Stollman G., Pietsch
W., van der Klis M., 1985, ApJ, 299, 487
Walker M. A., 1992, ApJ, 385, 642
Watts A. L., Maurer I., 2007, A&A, 467, L33
Worpel H., Galloway D. K., Price D. J., 2013, ApJ, 772, 94
Worpel H., Galloway D. K., Price D. J., 2015, ApJ, 801, 60
Zhang G., Méndez M., Altamirano D., Belloni T. M., Homan J., 2009,
MNRAS, 398, 368
Zhang G., Méndez M., Altamirano D., 2011, MNRAS, 413, 1913
Zhang G., Méndez M., Belloni T. M., Homan J., 2013, MNRAS, 436, 2276
van Paradijs J., Sztajno M., Lewin W. H. G., Trumper J., Vacca W. D., van
der Klis M., 1986, MNRAS, 221, 617
van Paradijs J., et al., 1990, A&A, 234, 181

APPENDIX A: DOUBLE-PEAKED EVENTS IN OUR

SAMPLE
Some bursts appear to be flat after the onset followed by a clear
peak, such as burst #6 and #8. In order to verify the significance
of the double-peaked structure of these bursts, we use either one
or two Gaussian functions to fit the bursts around the peaking time
data.
For burst #6, at the top left panel of A1, we show the fitting
results of the two Gaussian functions, yielding 𝜒2 = 113.3 for 17
d.o.f, and at the top right panel of A1, we show the fitting results
with one Gaussian function, yielding 𝜒2 = 304.8 for 20 d.o.f. The
F-test probability for these two fits is 0.0006, which indicates that
the double-peaked structure is significant.
For burst #8, at the bottom left panel of A1, we show the fitting
results of the two Gaussian functions, we yielding 𝜒2 = 59.8 for
21 d.o.f, and at the bottom right panel of A1, we show the fitting
results with one Gaussian function, yielding 𝜒2 = 241.9 for 24
d.o.f. The F-test probability for these two fits is 1.4 × 10−6 , which
also indicates that the double-peaked structure is significant.

This paper has been typeset from a TEX/LATEX file prepared by the author.

MNRAS 000, 1–12 (2019)