The Astrophysical Journal, 816:60 (9pp), 2016 January 10 doi:10.

3847/0004-637X/816/2/60
© 2016. The American Astronomical Society. All rights reserved.

SPECTRAL AND TEMPORAL PROPERTIES OF THE ULTRA-LUMINOUS X-RAY PULSAR IN M82 FROM 15
YEARS OF CHANDRA OBSERVATIONS AND ANALYSIS OF THE PULSED EMISSION USING NuSTAR
Murray Brightman1, Fiona Harrison1, Dominic J. Walton1,2, Felix Fuerst1, Ann Hornschemeier3,4, Andreas Zezas5,6,7,
Matteo Bachetti8, Brian Grefenstette1, Andrew Ptak3,4, Shriharsh Tendulkar1, and Mihoko Yukita3,4
1
Cahill Center for Astrophysics, California Institute of Technology, 1216 East California Boulevard, Pasadena, CA 91125, USA
2
Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109, USA
3
NASA Goddard Space Flight Center, Code 662, Greenbelt, MD 20771, USA
4
The Johns Hopkins University, Homewood Campus, Baltimore, MD 21218, USA
5
Physics Department, University of Crete, Heraklion, Greece
6
Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138, USA
7
Foundation for Research and Technology-Hellas, 71110 Heraklion, Crete, Greece
8
INAF/Osservatorio Astronomico di Cagliari, via della Scienza 5, I-09047 Selargius (CA), Italy
Received 2015 July 21; accepted 2015 November 18; published 2016 January 5

ABSTRACT
The recent discovery by Bachetti et al. of a pulsar in M82 that can reach luminosities of up to 1040 erg s−1, a factor
of ∼100 times the Eddington luminosity for a 1.4 Me compact object, poses a challenge for accretion physics. In
order to better understand the nature of this source and its duty cycle, and in light of several physical models that
have been subsequently published, we conduct a spectral and temporal analysis of the 0.5–8 keV X-ray emission
from this source from 15 years of Chandra observations. We analyze 19 ACIS observations where the point-spread
function (PSF) of the pulsar is not contaminated by nearby sources. We fit the Chandra spectra of the pulsar with a
power-law model and a disk blackbody model, subjected to interstellar absorption in M82. We carefully assess for
the effect of pile-up in our observations, where four observations have a pile-up fraction of >10%, which we
account for during spectral modeling with a convolution model. When fitted with a power-law model, the average
photon index when the source is at high luminosity (LX>1039 erg s−1) is Γ=1.33±0.15. For the disk
blackbody model, the average temperature is Tin=3.24±0.65 keV, the spectral shape being consistent with
other luminous X-ray pulsars. We also investigated the inclusion of a soft excess component and spectral break,
finding that the spectra are also consistent with these features common to luminous X-ray pulsars. In addition, we
present spectral analysis from NuSTAR over the 3–50 keV range where we have isolated the pulsed component. We
find that the pulsed emission in this band is best fit by a power-law with a high-energy cutoff, where
+5
Γ=0.6±0.3 and EC = 14- 3 keV. While the pulsar has previously been identified as a transient, we find from
our longer-baseline study that it has been remarkably active over the 15-year period, where for 9/19 (47%)
observations that we analyzed, the pulsar appears to be emitting at a luminosity in excess of 1039 erg s−1, greater
than 10 times its Eddington limit.
Key words: galaxies: individual (M82) – stars: neutron – X-rays: binaries

1. INTRODUCTION High B-field (B1012 G) accreting pulsars have been

observed to emit in excess of their isotropic Eddington
The discovery of coherent pulsations with a period of 1.37 s
in the X-ray emission of M82 by NuSTAR (Bachetti et al. 2014, luminosities. Magnetic fields allow pulsars to exceed their
hereafter B14), shown to be associated with an ultraluminous Eddington luminosities by funneling the accreting material
X-ray source (ULX) that is known to reach luminosities of along the magnetic field lines onto the magnetic poles of the
1040 erg s−1 (with the assumption that the source radiates neutron star, with the X-ray emission radiating out the sides of
isotropically), is a challenge to accretion physics and has fueled the accretion column. Observational evidence that showed the
speculation as to the nature of this source (Christodoulou et al. pulsar SMC X-1 to be super-Eddington motivated calculations
2014; Lyutikov 2014; Dall’Osso et al. 2015; Ekşi et al. 2015; of radiative transfer in the presence of these magnetic fields by
Fragos et al. 2015; Kluźniak & Lasota 2015; Mushtukov et al. Basko & Sunyaev (1975). The authors calculated the limiting
2015; Shao & Li 2015). Since the pulsations are almost luminosity of these systems, showing that this depends strongly
certainly produced by a rapidly spinning magnetized neutron on the geometry of the accretion channel (Basko &
star with a mass of ∼1.4 Me, the observed peak X-ray Sunyaev 1976). Dall’Osso et al. (2015) consider the observa-
luminosity is 100 times the system’s Eddington limit. tional properties of the pulsar in M82 and numerically solve the
From B14, the neutron star orbits its companion with a 2.5- torque equation with respect to the scenario in which matter is
day period that is close to circular, with a projected semimajor funneled along the magnetic field lines and argue in favor
axis of 22.225 light s (6.66×106 km) and a companion star of a high magnetic field strength (B∼1013 G) to explain the
with a minimum mass of 5.2 Me. A linear spin-up is also variation in Ṗ and the pulsar’s high luminosity. In this case
observed from the pulsations during the NuSTAR observations, the strength of the magnetic field would disrupt the accretion
with a pulse derivative P˙  -2 ´ 10-10 s s−1 that varies from disk at much larger radii ((80–90)RNS), causing the disk
observation to observation. The pulse profile is also close to temperature to decrease. This would be difficult to observe,
sinusoidal (B14). Any theoretical model must be able to however, since the X-ray emission from the accreting pulsar is
account for all of these properties. dominated by the base of the accretion column near the neutron

1
The Astrophysical Journal, 816:60 (9pp), 2016 January 10 Brightman et al.

star’s surface. A magnetic field of B∼10 G or stronger was

13
The goal of this paper is to better understand the duty cycle
also favored for the ULX pulsar by Ekşi et al. (2015), also of the source and its spectral characteristics and discuss these
based on calculations of torque equilibrium. Magnetic fields of characteristics with respect to theoretical models and other
these strengths have the power to reduce the Compton luminous pulsars. While only Chandra data can be used to
scattering cross-section, thus increasing the critical accretion spatially resolve the pulsar, NuSTAR, with its timing capabil-
luminosity. Dall’Osso et al. (2015) calculate that for 30 keV ities, allows it to temporally isolate the pulsed component due
photons, a B∼1013 G field would decrease the scattering to the coherent pulsations emitted by the source. We carry out
cross-section by a factor of ∼50. spectral analysis of the source in the 0.5–8 keV band using
Conversely however, Kluźniak & Lasota (2015) argue for a archival Chandra data and spectroscopy in the 3–50 keV band
low magnetic field strength (B < 109 G). They base their using NuSTAR. In Section 2 we describe the Chandra data and
conclusions on the ratio of spin-up rate to luminosity (10−50 analysis and in Section 3 we describe the NuSTAR data and
(erg s)−1), which is an order of magnitude lower than typical analysis. In Section 4 we present our results and in Section 5
X-ray pulsars. These authors argue that a disk truncated at large we discuss our findings with respect to other luminous X-ray
radii would not provide the required lever arm to power the pulsars and their implications for theoretical models. A distance
observed spin-up. Another interpretation of the magnetic field of 3.3 Mpc to M82 is assumed throughout (Foley et al. 2014).
by Christodoulou et al. (2014) implies that the magnetic field is
in fact typical of pulsars at ∼1012 G and that the observed 2. CHANDRA DATA AND ANALYSIS
luminosity can be accounted for by geometric beaming.
However, the near-sinusoidal shape of the pulse profile In total, M82 has been observed 28 times by Chandra, first
suggests that strong beaming does not occur, and furthermore, on 1999 September 20 with ACIS-I and most recently on 2015
Eddington ratios of ∼100 are difficult to reconcile with this January 20 with ACIS-S, covering more than 15 years of
scenario. It is clear from these arguments that further activity in the galaxy. The data contain a range of exposure
observational evidence, such as the duty cycle of the source times for each obsID, ranging from 2 to 120 ks, taken with all
and X-ray spectral properties, are needed to gain more insights three instruments, ACIS-I, ACIS-S, and HRC, all without the
into the nature of the source and its evolution. use of the gratings. Furthermore, the galaxy has been placed at
The ULX associated with the pulsar was first resolved by a mixture of on-axis and off-axis angles, the off-axis angles
Chandra HRC in 1999 October (Matsumoto et al. 2001) and being used to mitigate pile-up from the bright X-ray sources by
designated CXOM82 J095551.1+694045. This source is now spreading out the counts over a wider area of the detector. This
known to be the second most luminous X-ray source in M82 has often been combined with sub-array readout, typically 1/8
of one ACIS-S chip, also used to mitigate pile-up by reducing
(Feng & Kaaret 2007; Kong et al. 2007) after the ULX M82
frame times from 3.2 to 0.4 s. These observational data are
X-1 (CXOU J095550.2+694047) and thus is commonly
summarized in Table 1.
referred to as M82 X-2. Both Feng & Kaaret (2007) and Kong
We also present images of the central 20″×20″ region of
et al. (2007) use the Chandra data on M82 to study this source,
M82 from each of the 28 obsIDs, centered on X-2 in Figure 1.
noting its high X-ray luminosity and its month-timescale
This figure also illustrates the range of Chandra data available
variability, identifying it as a transient. As for associations at for this galaxy. It is clear from this figure that not all of the data
other wavelengths, Kong et al. (2007) found that X-2 is can be used to study X-2, as in many cases the PSF of X-2 is
coincident with the position of a star cluster seen in a near- blended with the PSF of two nearby sources to the south (i.e.,
infrared Hubble Space Telescope NICMOS F160W image, also obsIDs 378, 379, 380, 6097, 8190, and 10027).
associated with the radio source 42.21+59.2 from McDonald We carry out analysis on level 2 event data that has
et al. (2002), identified as an H II region. Furthermore, Gandhi undergone standard data processing by the Chandra X-ray
et al. (2011) presented high-resolution mid-infrared imaging of Center (v8.4.4 for obsID 10025; v8.4.5 for 2933, 10026,
the center of M82 and tentatively assigned their source #11 as 10542, 10543, 10544, 10545, 10925, 11104, 11800, 13796;
a counterpart to X-2, based on its proximity to the position of v8.5.1.1 for 5644, 6361, 15616; v10.2.1 for 16580; v10.3.1 for
the radio source. M82 X-1, which generally dominates the 17578; and v10.3.3 for 16023). We reprocess obsIDs 361 and
X-ray emission of the galaxy, is a candidate for an intermediate 1302 using the CIAO (v4.7, CALDB v4.6.5) script CHANDRA_R-
mass black hole, based on its extreme X-ray luminosity, which EPRO. We extract the spectrum of the source for all obsIDs
reach up to 1041 erg s−1 (e.g., Kaaret et al. 2006) and the where the PSF of X-2 is not blended with those of nearby
detection of twin-peaked quasi-periodic oscillations at frequen- sources. Since our aim is to carry out spectral fitting, we do not
cies of 3.3 and 5.1 Hz (Pasham et al. 2014). M82 X-1 and use the three obsIDs of HRC data due to the limited spectral
X-2 are separated by only 5″ and thus only resolvable by capabilities of this instrument. After excluding six obsIDs
Chandra. where the PSF of X-2 is blended and the three obsIDs of HRC
Since the results presented in Feng & Kaaret (2007) and data, our data set for spectral extraction consists of 19 ACIS
Kong et al. (2007), M82 has been observed by Chandra on 18 observations. We use the CIAO tool SPECEXTRACT to extract the
additional occasions, including the Chandra data used in B14 spectra, which produces source and background spectra and
to identify the source of the ultraluminous pulsations. Feng & redistribution matrices and auxiliary response files (RMF and
Kaaret (2007) and Kong et al. (2007) identify X-2 as a ARFs). For on-axis observations, we use circular regions of
transient, however, B14 found that the source retains its high radius 1 2 centered on the source. A radius of 1 2 encircles
luminosity seven years later. If X-2 persists at luminosities of 85% of the energy at 6 keV for an on-axis point source. This
∼1040 erg s−1, assuming a mass-to-energy conversion effi- radius increases by a factor of ∼2 for off-axis angles of ∼4′,
ciency of unity, the neutron star will grow at a rate of and the PSF elongates into an elliptical shape. Therefore, for
∼2×10−7Me yr−1, meaning it will collapse into a black hole off-axis observations, we use elliptical regions with 2″ and 1″
within ∼10 million years. semimajor and semi-minor axes, respectively. For observations

2
The Astrophysical Journal, 816:60 (9pp), 2016 January 10 Brightman et al.

Table 1
Chandra Observational Data

ObsID Date Instrument Exposure Pile-up Fraction Notes

(ks)
(1) (2) (3) (4) (5) (6)
361 1999 Sep 20 ACIS-I 33.7 <1% on-axis
1302 1999 Sep 20 ACIS-I 15.7 <1% on-axis
1411* 1999 Oct 28 HRC 54.0 L L
378* 1999 Dec 30 ACIS-I 4.2 L off-axis, sub-array, X-2 PSF blended
379* 2000 Mar 11 ACIS-I 9.1 L off-axis, sub-array, X-2 PSF blended
380* 2000 May 07 ACIS-I 5.1 L off-axis, X-2 PSF blended
2933 2002 Jun 18 ACIS-S 18.3 >10% on-axis
6097* 2005 Feb 04 ACIS-S 58.2 L off-axis, sub-array, X-2 PSF blended
5644 2005 Aug 17 ACIS-S 75.1 <5% on-axis, sub-array
6361 2005 Aug 18 ACIS-S 19.2 <5% on-axis, sub-array
8189* 2007 Jan 09 HRC 61.6 L L
8505* 2007 Jan 12 HRC 83.6 L L
8190* 2007 Jun 02 ACIS-S 5.8 L off-axis, X-2 PSF blended
10027* 2008 Oct 04 ACIS-S 20.2 L off-axis, X-2 PSF blended
10025 2009 Apr 17 ACIS-S 19.2 <1% off-axis, sub-array
10026 2009 Apr 29 ACIS-S 18.7 <1% off-axis
10542 2009 Jun 24 ACIS-S 120 <1% on-axis
10543 2009 Jul 01 ACIS-S 120 <1% on-axis
10925 2009 Jul 07 ACIS-S 45.1 <1% off-axis
10544 2009 Jul 07 ACIS-S 74.5 <1% off-axis
11104 2010 Jun 17 ACIS-S 10.1 >10% on-axis
11800 2010 Jul 20 ACIS-S 17.1 <1% off-axis
10545 2010 Jul 28 ACIS-S 96.3 ∼5% off-axis
13796 2012 Aug 09 ACIS-S 20.1 >10% on-axis
15616 2013 Feb 24 ACIS-S 2.1 <5% on-axis, short observation (2 ks)
16580 2014 Feb 03 ACIS-S 47.5 >10% on-axis
17578 2015 Jan 16 ACIS-S 10.1 ∼1% off-axis, sub-array
16023 2015 Jan 20 ACIS-S 10.1 <10% on-axis

Note. Details of the 28 Chandra observations of M82 taken from 1999 to 2015, ordered by date. Column (1) gives the obsID (* indicates that this observation was not
used in our investigation due to blended point-spread functions (PSFs) or HRC data), column (2) gives the date of the observation, column (3) gives the instrument
used, column (4) gives the total exposure time in ks, column (5) gives an estimate of the pile-up fraction described in Section 2 and column (6) gives details about if
the observation was taken off-axis and/or with a sub-array of pixels in order to mitigate the effect of pile-up. We also note if the PSF of X-2 is blended with nearby
sources.

where X-2 is faint or absent in the image, we center these enough signal (∼20 counts ks−1) to undertake the background
regions at the same positions relative to the bright persistent extraction. We use this background region for all obsIDs.
source X-1 as for the obsIDs where X-2 is clearly detected. We A major effect that must be taken into account before we can
also use smaller, 0 8 regions for these observations to avoid perform spectral fitting is pile-up. Pile-up occurs when more
contamination from the two nearby sources to the south. than one photon lands on the same pixel between two readouts
In order to carry out background extraction, we initially of the CCD. In this case multiple photons are counted as a
considered a large circular region outside the galaxy on the single event with increased energy, or they are not counted at
same chip as the galaxy; however, we find that the spectra all (in the case of grade migration). Pile-up therefore affects
resulting from the subtraction of this background contain a both the inferred flux and spectral shape. The standard readout
prominent residual soft component. When extracting the time for the ACIS detectors is 3.2 s, and with a count rate that
spectra of the diffuse emission (see Ranalli et al. 2008) from can get up to ∼1 count s−1, the effect of pile-up on X-2
observations cannot be neglected. As mentioned above, the
a region nearby X-2, we determine that the soft component
effect of pile-up can be reduced by the use of a sub-array of
likely results from this diffuse emission. Thus we use the
pixels that reduces the readout time to 0.4 s for the 1/8 ACIS-S
spectrum of the diffuse emission for background subtraction.
chip, or by moving the source off-axis, which spreads the
Figure 2 shows the Chandra image from a 75 ks on-axis photons over several pixels. Off-axis observations have the
exposure of M82 (obsID 5644), where the scaling enhances the drawback that the PSF of close sources can become blended.
diffuse emission. This reveals the various structures present in To estimate the level of pile-up in our data, we use the CIAO
the diffuse emission around X-2, which is particularly strong to tool PILEUP_MAP, which outputs an image of counts per ACIS
the southeast. In our choice of background region, we aim to frame, which can then be used to estimate a pile-up fraction.9
avoid the brightest diffuse emission, which is too far from the For off-axis and sub-array observations, the count rate per
source to contribute to its background, plus we aim to avoid the frame is typically less than 0.1, which corresponds to a pile-up
point sources to the south and the PSF of X-1 to the west. This fraction of <5%. For on-axis observations with full CCDs, the
results in a relatively small 7 6×1 9 rectangular extraction
9
region to the northeast of X-2, which, though small, contains http://cxc.harvard.edu/csc/memos/files/Davis_pileup.pdf

3
The Astrophysical Journal, 816:60 (9pp), 2016 January 10 Brightman et al.

Figure 1. 20″×20″ 0.5–8 keV Chandra images of the central region of M82, consisting of 28 obsIDs taken over the 15-year period 1999 to 2015. The obsIDs are
written in the top left of each image. Images are centered on the position of the ultraluminous pulsar, X-2, which is marked with white crosshairs. The data were taken
with a variety of instruments, exposure times, off-axis angles, and detector sub-arrays, the details of which are given in Table 1. From the top left to the bottom right,
the images are ordered by epoch, starting in 1999 September and ending in 2015 January. The green circle to the northeast of X-2 marks the radio kinematic center of
the galaxy from Weliachew et al. (1984). North is to the top of the image and east is to the left.

are >10%, 2933, 11104, 13796, and 16580. The effect of pile-
up can be accounted for in spectral fitting for cases where pile-
up is not too strong, which we judge to be the case here, using a
convolution model based on an algorithm presented in
Davis (2001).
We use XSPEC (v12.8.2) to carry out spectral fitting of X-2.
We group the spectra with a minimum of 20 counts per bin
using the HEASARC tool GRPPHA using the χ2 fit statistic and
carrying out background subtraction. There are not enough
counts in the spectra extracted from obsID 1302 and 15616 for
χ2 fitting, thus we group the spectrum to have a minimum of 1
count per bin and use the Cash fit statistic. We fit the spectra in
the energy range 0.5–8 keV with two models, a power-law
(powerlaw) model and a disk blackbody (diskbb) model,
both of which are subjected to photoelectric absorption local to
the source (zwabs).
Despite subtracting off the diffuse background close to the
position of X-2, we find that soft excess emission still persists
over the above models. It is possible that this is residual diffuse
emission that is not accounted for by the background
subtraction; however, it may also originate from optically thin
Figure 2. Figure of obsID 5644 showing the source (circular) and background plasma that has been photoionized by the pulsar, evidence for
(rectangular) spectral extraction regions. The scaling has been modified in this
figure with respect to Figure 1 to emphasize the diffuse emission. The
which has been found around other ULXs such as Holmberg II
background region was chosen to contain the diffuse emission, which was as X-1 (Dewangan et al. 2004), NGC 7424 ULX 2 (Soria
local to the source as possible while also excluding nearby point sources and et al. 2006), and NGC 5408 X-1 (Strohmayer et al. 2007). We
the PSF of X-1. North is up and east is left. account for this soft excess component in any case by adding
the collisionally ionized diffuse gas model, apec. While many
counts per frame are many times higher, with typical pile-up X-ray pulsars indeed present soft X-ray excesses that are
fractions of >10%. We list these estimates in Table 1. We thought to be produced by the reprocessing of the hard X-rays
identify four observations where the estimated pile-up fractions in the inner region of the accretion disk (see, e.g., Hickox

4
The Astrophysical Journal, 816:60 (9pp), 2016 January 10 Brightman et al.
−2
et al. 2004), the absorption in this system (NH∼3×10 cm )22
fewer soft X-ray photons than the powerlaw model. The
makes it unlikely that the excess we observe is intrinsic to X-2. redshift is fixed at 0.00067 for all model components.
Similarly, we find that if we subject the apec model to the Uncertainties on the free parameters are calculated using the
same absorption that is applied to the powerlaw or diskbb Δχ2 = 4.61 criterion, which corresponds to 90% confidence
models, it cannot account for the soft excess. Furthermore, level for two interesting parameters.
adding absorption to this component that is independent from
the absorption of the pulsar does not produce an improvement 3. NuSTAR DATA AND ANALYSIS
in χ2. We conclude that the soft excess most likely originates
from foreground material in the line of sight that is sufficiently NuSTAR (Harrison et al. 2013) observed the M82 field seven
outside the bulk of the diffuse gas that it is not absorbed. times between 2014 January 23 and March 06, as described
The absorption column and Γ (or Tin) are degenerate given a in B14, for a total exposure of 1.91 Ms during which the
limited bandpass (i.e., a hard spectrum can be fitted with a steep pulsations were detected. A Chandra exposure (47.5 ks, obsID
Γ and high NH or a flat Γ and low NH). For this reason we 16580) overlapped with the NuSTAR observation during a
simultaneously fit for NH across all epochs, which is thus period where the pulsations were present. While B14 presented
timing and photometric analysis of the pulsar, we aim here to
driven by the spectra with the greatest counts. The parameters
present for the first time some of its spectral characteristics
of the powerlaw or diskbb models are free for each epoch.
above 10 keV by isolating the pulsed component. For this we
The absorption in the spectrum is attributed to the interstellar
used the NuSTAR data analysis software (NuSTARDAS)
medium in M82 rather than being local to the neutron star, and version 1.2.0 and NuSTAR CALDB version 20130509 with
thus is not expected to change on the timescales that we are the standard filters to obtain good time intervals, excluding the
considering. We confirm this explicitly by fitting the spectra periods where the source was occulted by the Earth or was
individually, with free NH parameters, and find no evidence for transiting through the South Atlantic Anomaly. We used the
a variable absorber. Furthermore, as we attribute the soft excess pulsar ephemeris described in B14 to extract “pulse-on” and
to diffuse emission, we also simultaneously fit for the “pulse-off” spectra. The pulse-on spectrum is defined to be the
temperature of the apec model. We leave the normalizations brightest 25% of the pulse profile, while the pulse-off is defined
of this model free since the small circular source extraction to be the faintest 25% of the pulse profile. We then subtract the
regions for on-axis observations and the larger elliptical pulse-off spectrum from the pulse-on spectrum to obtain the
extraction regions for off-axis observations contain differing pulsed spectrum, which we model with some simple models in
amounts of the diffuse emission. order to characterize the data and to facilitate comparison with
For both model cases we test the effect of adding the other well-studied pulsars. Due to the triggered readout of the
multiplicative pileup model. In these cases, the frame time NuSTAR detectors, pile-up is not an issue at the flux levels of
parameter is set as dependent on the size of the sub-array used. M82. The NuSTAR data were rebinned to a signal-to-noise of 3,
We set the PSF fraction (not the fraction of the PSF included in providing a significant signal up to ∼40–50 keV. We fit the
the extraction region but the fraction of counts in the region pulsed spectrum with models consisting of an absorbed power-
which are from the point source whose pile-up is being law continuum both with and without an exponential cutoff
modeled) to 95%. The only parameter left free when using this (cutoffpl), both of which are subjected to interstellar
model is α, which is the grade morphing parameter. The absorption, where we fix the NH to 3×1022 cm−2, as
parameter α is related to the grade migration function, determined from the Chandra analysis.
G = a p - 1, where p is the number of piled photons. We
consider four factors when deciding whether to include this 4. RESULTS
model in the spectral fit. We visually inspect the spectrum for a
telltale turn-up at high energies. We use the estimated pile-up The results of the spectral fits using both the power-law and
fraction from the PILEUP_MAP tool and consider if the observation disk blackbody models are given in Table 2. The fit statistic,
was taken off-axis and/or with a sub-array. Generally, on-axis χ2, for the combined fits with the power-law model is 1439.89
observations show pile-up fractions greater than 10% and we with 1315 degrees of freedom (cr2 =1.09), whereas for the
disk blackbody model it is χ2=1323.03 with 1315 degrees of
thus include the pileup model in the fit. Lastly, if the pileup
freedom (cr2 =1.01). Comparing these fit statistics suggests
model is included and the best-fit α parameter converges on a
that the disk blackbody model is overall the best model.
small number or zero, which is considered unphysical, we
However, at the lowest luminosities (∼1038 erg s−1), the
remove the pileup model from the fit.
temperature of the disk blackbody model is not well-
For the powerlaw model, the free parameters are the
constrained and hits the upper bound of 10 keV imposed for
photon index, Γ, and the normalization. We allow Γ to vary spectral fitting during the error calculations. In some cases the
between −3 and 10. For the diskbb model, the free error calculations fail. This is most likely due to the low
parameters are Tin, the temperature of the inner disk in keV, number of counts at these luminosities.
and the normalization. We restrict the disk temperature to The intrinsic luminosity estimates (corrected for absorption)
<10 keV. For both models the normalization of the apec between the two models differ due to both the diverging
model is a free parameter. As described above, the temperature spectral shape above 8 keV (luminosity measurements are
of this component was fit for simultaneously across all epochs extrapolated to 10 keV for comparison with previous works)
+0.28 +0.38
and found to be 0.44- 0.25 keV and 0.58-0.17 keV for the and in the soft X-ray band, plus the differing pile-up estimates
powerlaw and diskbb models, respectively. The NH of the caused by the difference in spectral shape of the models. The
zwabs model was also fitted simultaneously and found to be disk blackbody model gives a systematically lower 0.5–10 keV
+0.15 −2 +0.10 −2
0.14 ´ 10 cm 0.09 ´ 10 cm , respectively.
3.4- 22 and 2.8- 22
intrinsic luminosity, typically 25% lower. Nonetheless, our
The diskbb model requires less absorption since it predicts analysis shows that X-2 is frequently observed to be emitting

5
The Astrophysical Journal, 816:60 (9pp), 2016 January 10 Brightman et al.

Table 2
Spectral Fitting Results

ObsID Date Power-law Model Parameters Disk Blackbody Model Parameters

αpl Γpl LX,pl αdisk Tin,disk LX,disk
(×1038 erg s−1) (keV) (×1038 erg s−1)
(1) (2) (3) (4) (5) (6) (7) (8)
+1.30 +0.6 +7.36 +0.6
361 1999 Sep 20 - 1.30- 1.95 1.0- 0.4 - 2.64- 1.56 0.8- 0.4
+1.53 +2.0
1302 1999 Sep 20 - 1.71-1.71 1.1-0.5 - - -
+3.1 +0.21 +7 +0.14 +0.75 +142
2933 2002 Jun 18 0.69- 0.39 1.55- 0.20 69- 6 0.22- 0.11 2.46- 0.44 579- 97
+0.05 +2 +0.20 +2
5644 2005 Aug 17 - 1.42- 0.05 115- 2 - 2.86- 0.17 92- 2
+0.11 +4 +0.59 +5
6361 2005 Aug 18 - 1.34- 0.11 102- 4 - 3.16- 0.41 83- 4
+0.90 +7.3 +8.22 +1.8
10025 2009 Apr 17 - 2.41- 1.11 7.3- 3.0 - 1.78-0.84 3.6- 1.0
+2.14 +13.6
10026 2009 Apr 29 - 0.14- 0.14 2.3- 1.6 - - -
+0.28 +0.5 +0.67 +0.1
10542 2009 Jun 24 - 1.99- 0.28 3.5- 0.4 - 1.94- 0.40 2.3- 0.2
+0.57 +0.2 +4.79 +0.3
10543 2009 Jul 01 - 1.20- 0.64 1.4- 0.2 - 5.21- 3.03 1.2- 0.3
+0.68 +0.7 +2.58 +0.5
10925 2009 Jul 07 - 1.08-0.76 2.6- 0.5 - 7.42- 7.42 2.3- 0.7
+0.91 +0.4 +0.99 +0.3
10544 2009 Jul 07 - 1.00- 1.07 1.1-0.3 - 9.01- 9.01 1.1-0.4
+0.29 +0.29 +29 +0.33 +5.77 +180
11104 2010 Jun 17 0.20- 0.17 1.14-0.17 197- 64 0.27- 0.24 4.23- 1.46 143- 40
+1.25 +7.1 +8.45 +1.2
11800 2010 Jul 20 - 2.51- 1.74 4.1- 2.4 - 1.55- 0.88 1.9- 0.8
+0.09 +1 +0.51 +1
10545 2010 Jul 28 - 1.34-0.09 36- 1 - 3.22- 0.38 30- 1
+0.14 +0.19 +0 +0.12 +1.38 +55
13796 2012 Aug 09 0.24- 0.09 1.15- 0.10 227- 55 0.25- 0.09 4.17-0.83 197- 50
+2.48 +20.5
15616 2013 Feb 24 - 1.60- 3.18 4.6- 2.9 - - -
+0.12 +0.09 +61 +0.20 +0.70 +43
16580 2014 Feb 03 0.27- 0.06 1.27- 0.09 227- 49 0.27- 0.05 3.33- 0.51 210- 77
+0.22 +4 +0.78 +4
17578 2015 Jan 16 - 1.55-0.23 52- 4 - 2.36- 0.45 39- 4
+0.27 +678 +3.55 +11
16023 2015 Jan 20 - 1.28- 0.28 76- 38 - 3.37- 1.04 36- 6

Note. Results of the X-ray spectral fitting of the 19 obsIDs where reliable spectral information on X-2 could be extracted, fitted with the absorbed power-law and
absorbed disk blackbody models. Column (1) gives the obsID, column (2) gives the date of the observation, column (3) gives the grade morphing parameter of the
pile-up model when convolved with the power-law model, if this is used in the fit (“-” indicates that the pile-up model was not used), column (4) gives the photon
index of the power-law model, column (5) gives the unabsorbed 0.5–10 keV luminosity of the power-law model. Column (6) gives the grade morphing parameter of
the pile-up model, when convolved with the disk blackbody model, if this is used in the fit (“-” indicates that the pile-up model was not used), column (7) gives the
disk temperature in keV. “u” indicates that this parameter hit the upper limit of 10 in the spectral fit and “l” indicates it hit the lower limit of 0. “-” indicates where the
error calculation fails. Column (8) gives the unabsorbed 0.5–10 keV luminosity of the disk blackbody model.

Figure 3. Long-term activity of X-2 over the 15 years of Chandra observations showing that the source is frequently observed to be radiating at many times its
Eddington limit. Vertical lines indicate the 90% confidence range on the measured 0.5–10 keV luminosity, which is calculated from the power-law model assuming a
distance to M82 of 3.3 Mpc. Dark blue squares show observations that were taken off-axis and/or with a sub-array of pixels to mitigate the effect of pile-up. Light
blue squares are observations taken on-axis with the full array of CCDs. The horizontal lines show the Eddington limiting luminosity for a 1.4 Me object, along with
10 and 100 times this.

well above its Eddington luminosity, assuming a typical NS transient, as identified by Feng & Kaaret (2007) and Kong
mass of 1.4 Me, regardless of the model used. This is et al. (2007).
illustrated in Figure 3, which shows intrinsic (unabsorbed) LX We also note the period of extreme flux variability in 2010
against time. For 9/19 (47%) observations that we analyzed, where the source drops almost two orders of magnitude in
we found that X-2 emits at a luminosity in excess of brightness in the space of a month. The exposure times of these
1039 erg s−1 and is thus relatively persistent rather than two observations (11104 and 11800) are too short to reveal any

6
The Astrophysical Journal, 816:60 (9pp), 2016 January 10 Brightman et al.

Figure 4. Light curves of X-2 for obsIDs 5644 and 10545, with 2 ks bins.

significant variability during the observations, however. Of the

longer observations available, obsIDs 5644 (75 ks) and 10545
(96 ks) are the least affected by pile-up due to a sub-array of
pixels used in the former and X-2 lying at off-axis angles in the
latter. The light curves of these observations are shown in
Figure 4. While X-2 does not show much variability during
obsID 5644, it does show variability of up to a factor of two
on ks-timescales during obsID 10545. Figure 5. 75 ks Chandra spectrum of X-2 taken in 2005 (obsID 5644), taken
We also examine the spectra of X-2 for these two long on-axis but with a sub-array of pixels to reduce pile-up. The top panel shows
the unfolded spectrum fitted with an absorbed power-law, plotted with a dashed
observations in order to gain further insights into the source, line. The dotted line shows the apec model used to model the excess diffuse
shown in Figures 5 and 6. During obsID 5644 X-2 was background. The middle panel shows the data-to-model ratio of this fit. The
observed to be near its peak luminosity of ∼1040 erg s−1, while bottom panel shows the data-to-model ratio of a fit with the disk blackbody
during obsID 10545 X-2 was at a lower luminosity of model.
∼4×1039 erg s−1. In Figure 5 we show the data-to-model
ratios of these spectra when fitted by the power-law and disk
blackbody models. In neither case do the ratios show any
deviations indicative of a bad fit. Furthermore, the χ2 values are
comparable between the models.
Figure 7 shows the relationship between the spectral
parameters Γ of the power-law model and Tin of the disk
blackbody model with respect LX. For LX>1039 erg s−1 the
mean Γ=1.33±0.15 (1σ), whereas the mean
Tin=3.24±0.65. Below 1039 erg s−1, Γ and Tin are not
well-constrained and show a large spread in values.
In addition to the Chandra spectral analysis, we have
conducted NuSTAR pulse-phased spectroscopy of the pulsar in
the 3–50 keV range, as described in Section 3. We fit the pulsed
spectrum with a power-law model and a model with an
exponential cutoff (cutoffpl in XSPEC), both of which are
subjected to photoelectric absorption. We find that the
power law with a cutoff is significantly preferred (Δχ2=20
for one additional free parameter). The fit is excellent
(χ2/dof = 132/126) and we find the photon index of the
pulsed component to be Γ=0.6±0.3, with a high-energy
+5
cutoff, EC = 14- 3 keV. The average pulsed flux in this band is
−12
5.7±0.4×10 erg cm−2 s−1, corresponding to a luminos-
ity of 7.5×10 erg s−1 at 3.3 Mpc. We present the pulsed
39
Figure 6. 96 ks Chandra spectrum of X-2 taken in 2010 (obsID 10545), taken
spectrum of M82 X-2 in Figure 8, unfolded through the off-axis to reduce pile-up. The top panel shows the unfolded spectrum fitted
instrumental response, assuming the cutoff power-law model. with an absorbed power-law, plotted with a dashed line. The dotted line shows
the apec model used to model the excess diffuse background. The middle
panel shows the data-to-model ratio of this fit. The bottom panel shows the
data-to-model ratio of a fit with the disk blackbody model.
5. COMPARISON OF THE M82 PULSAR WITH OTHER
LUMINOUS X-RAY PULSARS AND IMPLICATIONS FOR of magnitude brighter than these sources, a comparison with
THEORETICAL MODELS them is valuable. The X-ray spectra of these sources are
Other examples of luminous (>1038 erg s−1) X-ray typically fitted with a power-law model with Γ=0.5–1.5,
pulsars include SMC X-1, LMC X-4, GRO J1744–28, subjected to absorption along the line of sight, and in some
RX J0059.2–7138, XTE J0111.2–7317, and A0538–66. cases with a high-energy cutoff with energies ranging from
Although M82 X-2 can reach luminosities that are an order 5 to 30 keV.

7
The Astrophysical Journal, 816:60 (9pp), 2016 January 10 Brightman et al.

law where Γ=0.91±0.03 and the cutoff energy is

+1.4
5.5- 0.5 keV. The spectrum of LMC X-4 did not require a
cutoff, and could be reproduced by a plain power law with
Γ=0.69±0.04. The other examples of super-Eddington
pulsars, GRO J1744–28, RX J0059.2–7138, XTE
J0111.2–7317, and A0538–66 are also described well by cutoff
power laws or power-law models, both with similar parameters
(see Skinner et al. 1982; Nishiuchi et al. 1999; Yokogawa et al.
2000; Sidoli et al. 2015; Younes et al. 2015, respectively).
The phase-averaged 0.5–8 keV X-ray spectral properties of
M82 X-2 are very similar to these other sources, with
Γ=1.33±0.15 for a power-law model with no cutoff. The
long Chandra observation obsID 5644 where the source is
caught in an ultraluminous state, but where pile-up is
negligible, offers us the opportunity to test for the presence
of a cutoff and if any constraints can be placed on it. Fitting
+0.68
this spectrum with this model yields G = 0.70- 0.65 and
+50.9
EC = 6.19-2.9 keV, where χ =389.87 for 338 degrees of
2

freedom. The fit with the power law without a cutoff gives

χ2=395.66 for 339 degrees of freedom, thus the addition of
the cutoff yields an improvement in χ2 of only six for the
addition of one parameter, and the cutoff energy is not well-
constrained in the Chandra data alone.
NuSTAR, however, can measure the cutoff due to its
sensitivity above 10 keV. From analysis of the pulsed
component in the 3–50 keV range we find that the pulsed
spectrum is best fitted by a power law with a high-energy
+5
Figure 7. Figures showing (top) the relationship between Γ and LX of X-2 cutoff, where Γ=0.6±0.3 and EC = 14- 3 keV. These
when fitted with a power-law model and (bottom) the relationship between Tin values are very similar to the phase-averaged spectrum
and LX when fitted with a disk blackbody model. Dark blue squares show described above, albeit with much better constraints owing to
observations that were taken off-axis and/or with a sub-array of pixels to
mitigate the effect of pile-up. Light blue squares are observations taken on-axis
the high-energy sensitivity of NuSTAR. The similarity to the
with the full array of CCDs. In some cases, at the lowest luminosities, the best- phase-averaged spectrum from Chandra indicates that the
fit disk temperature is off the scale of this figure and with large unconstrained pulsed spectrum dominates the emission of the pulsar. It is also
values. These are indicated with vertical lines in the figure not associated with a interesting to note that the cutoff energy of the pulsed
square.
component is higher than that observed in other ULXs where
the nature of the accretors remains unknown, whose spectra
typically cutoff at 6–8 keV (e.g., Bachetti et al. 2013; Walton
et al. 2013; Mukherjee et al. 2015; Rana et al. 2015).
Dall’Osso et al. (2015) note the empirical relationship
between the energy of the cyclotron resonance features, Ecyc, of
four X-ray pulsars and the cutoff energy in their spectra, EC
(see Makishima et al. 1990), where Ecyc=(1.4–1.8)×EC.
Many of the theoretical modeling papers that have aimed to
explain the nature of the ultraluminous pulsar have presented
different scenarios for the strength of its magnetic field. Since
Ecyc is directly proportional to the strength of the magnetic
field, then EC could potentially yield information about the
magnetic field strength. Following this, the 14 keV cutoff that
we measure in the pulsed spectrum implies a B∼1012 G
magnetic field. However, we note that there are exceptions to
the above relation; for example, KS1947+319, where
Ecyc = 12 keV and EC = 22 keV (Fürst et al. 2014).
Dall’Osso et al. (2015) also discuss the variability exhibited
Figure 8. NuSTAR 3–50 keV unfolded pulsed spectrum of M82 X-2, fitted
by X-2, which can be explained by relatively small changes in
+5
with a power law, Γ=0.6±0.3, with a high-energy cutoff, EC = 14-3 keV.
the mass accretion rate in the presence of a strong magnetic
The average pulsed flux in this band is 5.7±0.4×10−12 erg cm−2 s−1, field, whereby the source at low luminosities enters the
corresponding to a luminosity of 7.5×1039 erg s−1 at 3.3 Mpc. propeller regime. We note, however, that the minimum
luminosity at which accretion is possible in the presence of a
Paul et al. (2002) fit the phase-averaged spectra of SMC X-1 B=1013 G magnetic field, a strength which they favor, with a
and LMC X-4 in the 0.7–10 keV band using ASCA data. The 1.37 s period, is ∼2×1039 erg s−1 (see Stella et al. 1986).
spectrum of SMC X-1 was modeled well with a cutoff power Considering that we observe X-2 at much lower luminosities,

8
The Astrophysical Journal, 816:60 (9pp), 2016 January 10 Brightman et al.

this argues against such a strong magnetic field given the the Smithsonian Astrophysical Observatory. This work also
discussion above. made use of data from the NuSTAR mission, a project led by
Concerning the duty cycle of X-2, we have found that the the California Institute of Technology, managed by the Jet
source is more persistent than previously reported by Feng & Propulsion Laboratory, and funded by NASA. We thank the
Kaaret (2007) and Kong et al. (2007), due to the longer NuSTAR Operations, Software, and Calibration teams for
baseline of our investigation and larger data set. For 9/19 support with the execution and analysis of these observations.
(47%) observations that we analyzed, we found that X-2 emits This research has made use of the NuSTAR Data Analysis
at a luminosity in excess of 1039 erg s−1. Luminosities of Software (NuSTARDAS) jointly developed by the ASI Science
∼1039–1040 erg s−1 imply that the neutron star is growing at a Data Center (ASDC, Italy) and the California Institute of
rate of ∼2×10−8–10−7Me yr−1, assuming isotropic emission Technology (USA). A.Z. acknowledges funding from the
and a mass-to-energy conversion efficiency of unity, meaning it European Research Council under the European Union’s
will collapse into a black hole within ∼10–100 million years. Seventh Framework Programme (FP/2007–2013)/ERC grant
These results could have important implications for the agreement No. 617001.
formation and growth of supermassive black holes, theoretical Facilities: CXO (ACIS), NuSTAR
modeling of which often employs an early super-critical
accretion phase to explain the masses of the supermassive
black holes found in quasars at z∼7 (e.g., Volonteri & REFERENCES
Rees 2006; Mortlock et al. 2011). Low-mass X-ray binaries
Bachetti, M., Harrison, F. A., Walton, D. J., et al. 2014, Natur, 514, 202
have also been proposed to be a potential source of ionizing Bachetti, M., Rana, V., Walton, D. J., et al. 2013, ApJ, 778, 163
radiation for heating the intergalactic medium during the epoch Basko, M. M., & Sunyaev, R. A. 1975, A&A, 42, 311
of reionization (Fragos et al. 2013), which may indeed have a Basko, M. M., & Sunyaev, R. A. 1976, MNRAS, 175, 395
contribution from a non-negligible population of ultraluminous Christodoulou, D. M., Laycock, S. G. T., & Kazanas, D. 2014,
arXiv:1411.5434
pulsars. Dall’Osso, S., Perna, R., & Stella, L. 2015, MNRAS, 449, 2144
Davis, J. E. 2001, ApJ, 562, 575
6. SUMMARY AND CONCLUSIONS Dewangan, G. C., Miyaji, T., Griffiths, R. E., & Lehmann, I. 2004, ApJL,
608, L57
In this paper we have conducted a temporal and spectral Ekşi, K. Y., Andaç, İ. C., Çıkıntoğlu, S., et al. 2015, MNRAS, 448, L40
analysis of the 0.5–8 keV X-ray emission from the ultralumi- Feng, H., & Kaaret, P. 2007, ApJ, 668, 941
nous X-ray pulsar M82 X-2 from 15 years of Chandra Foley, R. J., Fox, O. D., McCully, C., et al. 2014, MNRAS, 443, 2887
observations and pulse-phased spectroscopy in the 3–50 keV Fragos, T., Lehmer, B. D., Naoz, S., Zezas, A., & Basu-Zych, A. 2013, ApJL,
776, L31
band from NuSTAR data. Our main findings are as follows. Fragos, T., Linden, T., Kalogera, V., & Sklias, P. 2015, ApJL, 802, L5
Fürst, F., Pottschmidt, K., Wilms, J., et al. 2014, ApJL, 784, L40
1. When fitted with a power-law model, the average Gandhi, P., Isobe, N., Birkinshaw, M., et al. 2011, PASJ, 63, 505
photon index for epochs where LX>1039 erg s−1 in Harrison, F. A., Craig, W. W., Christensen, F. E., Hailey, C. J., &
the 0.5–8 keV band is Γ=1.33±0.15. For the disk Zhang, W. W. 2013, ApJ, 770, 103
blackbody model, the average temperature is Hickox, R. C., Narayan, R., & Kallman, T. R. 2004, ApJ, 614, 881
Tin=3.24±0.65 keV. This spectral shape is consistent Kaaret, P., Simet, M. G., & Lang, C. C. 2006, ApJ, 646, 174
Kluźniak, W., & Lasota, J.-P. 2015, MNRAS, 448, L43
with other luminous X-ray pulsars. We also investigated Kong, A. K. H., Yang, Y. J., Hsieh, P.-Y., Mak, D. S. Y., & Pun, C. S. J. 2007,
the inclusion of a soft excess component and spectral ApJ, 671, 349
break finding that the spectra are also consistent with Lyutikov, M. 2014, arXiv:1410.8745
these features common to luminous X-ray pulsars. Makishima, K., Mihara, T., Ishida, M., et al. 1990, ApJL, 365, L59
Matsumoto, H., Tsuru, T. G., Koyama, K., et al. 2001, ApJL, 547, L25
2. The pulsed emission of X-2 in the 3–50 keV band from McDonald, A. R., Muxlow, T. W. B., Wills, K. A., Pedlar, A., &
NuSTAR data is best fitted by a power law with a high- Beswick, R. J. 2002, MNRAS, 334, 912
+5
energy cutoff, where Γ=0.6±0.3 and EC = 14- 3 keV Mortlock, D. J., Warren, S. J., Venemans, B. P., et al. 2011, Natur, 474, 616
39 −1 Mukherjee, E. S., Walton, D. J., Bachetti, M., et al. 2015, arXiv:1502.01764
with a luminosity of 7.5×10 erg s .
3. Our results show that X-2 has been remarkably active Mushtukov, A. A., Suleimanov, V. F., Tsygankov, S. S., & Poutanen, J. 2015,
arXiv:1506.03600
over the 15-year period considered. We find that for 9/19 Nishiuchi, M., Koyama, K., Maeda, Y., et al. 1999, ApJ, 517, 436
(47%) observations that we analyzed, the pulsar appears Pasham, D. R., Strohmayer, T. E., & Mushotzky, R. F. 2014, Natur, 513, 74
to be emitting at a luminosity in excess of 1039 erg s−1, Paul, B., Nagase, F., Endo, T., et al. 2002, ApJ, 579, 411
which is greater than 10 times its Eddington limit. Rana, V., Harrison, F. A., Bachetti, M., et al. 2015, ApJ, 799, 121
Luminosities of ∼1039–1040 erg s−1 imply that the Ranalli, P., Comastri, A., Origlia, L., & Maiolino, R. 2008, MNRAS,
386, 1464
neutron star is growing at a rate of ∼2×10−8– Shao, Y., & Li, X.-D. 2015, ApJ, 802, 131
10−7Me yr−1 and is expected to collapse into a black Sidoli, L., Palombara, N. L., Esposito, P., Tiengo, A., & Mereghetti, S. 2015,
hole within ∼10–100 million years. MNRAS, 449, 3710
Skinner, G. K., Bedford, D. K., Elsner, R. F., et al. 1982, Natur, 297, 568
Soria, R., Kuncic, Z., Broderick, J. W., & Ryder, S. D. 2006, MNRAS,
This work made significant use of archival observations 370, 1666
made by the Chandra X-ray observatory, for which we thank Stella, L., White, N. E., & Rosner, R. 1986, ApJ, 308, 669
the builders and operators, as well as the software package Strohmayer, T. E., Mushotzky, R. F., Winter, L., et al. 2007, ApJ, 660, 580
CIAO. The data were obtained from the High Energy Volonteri, M., & Rees, M. J. 2006, ApJ, 650, 669
Walton, D. J., Fuerst, F., Harrison, F., et al. 2013, ApJ, 779, 148
Astrophysics Science Archive Research Center (HEASARC), Weliachew, L., Fomalont, E. B., & Greisen, E. W. 1984, A&A, 137, 335
which is a service of the Astrophysics Science Division at Yokogawa, J., Paul, B., Ozaki, M., et al. 2000, ApJ, 539, 191
NASA/GSFC and the High Energy Astrophysics Division of Younes, G., Kouveliotou, C., Grefenstette, B. W., et al. 2015, ApJ, 804, 43