X-ray Variability and Evidence for Pulsations from the UniqueRadio Pulsar/X-ray Binary Transition Object FIRST

J102347.6+003841

Anne M. Archibald1

aarchiba@physics.mcgill.ca

Victoria M. Kaspi1

Slavko Bogdanov1

Jason W. T. Hessels2,3

Ingrid H. Stairs4

Scott M. Ransom5

and

Maura A. McLaughlin6

ABSTRACT

We report on observations of the unusual neutron-star binary systemFIRST J102347.6+003841 carried out using the XMM-Newton satellite. This system con-sists of a radio millisecond pulsar in an 0.198-day orbit with a ∼ 0.2 M� Roche-lobe-fillingcompanion, and appears to have had an accretion disk in 2001. We observe a hard power-lawspectrum (Γ = 1.26(4)) with a possible thermal component, and orbital variability in X-ray fluxand possibly hardness of the X-rays. We also detect probable pulsations at the pulsar period(single-trial significance ∼ 4.5σ from an 11(2)% modulation), which would make this the firstsystem in which both orbital and rotational X-ray pulsations are detected. We interpret theemission as a combination of X-rays from the pulsar itself and from a shock where materialoverflowing the companion meets the pulsar wind. The similarity of this X-ray emission to thatseen from other millisecond pulsar binary systems, in particular 47 Tuc W (PSR J0024−7204W)and PSR J1740−5340, suggests that they may also undergo disk episodes similar to that seen inJ1023 in 2001.

Subject headings: pulsars: general — pulsars: individual (PSR J1023+0038) — stars: neutron — X-rays:stars

1

1. Introduction

Radio millisecond pulsars are thought to havebeen spun up to their current rapid rotation by along period of accretion in a Low-Mass X-ray Bi-nary (LMXB). This general idea is well supportedby many lines of evidence, but the transition be-tween LMXB and radio millisecond pulsar remainsmysterious. In particular, models of binary evolu-tion often predict that mass transfer should con-tinue indefinitely (Deloye 2008), but ionized ma-terial entering a neutron star’s magnetosphere isexpected to “short out” magnetospheric emission(Stella et al. 1994). How then can we explain theobserved radio millisecond pulsars?

An important system for studying this questionwas recently discovered: FIRST J102347.6+003841(hereafter J1023) is a 0.198-day binary containinga 1.69-ms radio millisecond pulsar, PSR J1023+0038.This system is unique in that it shows evidence forhaving had an accretion disk as recently as 2001,but is now active as a radio pulsar (Archibaldet al. 2009; Wang et al. 2009). This suggeststhat we are observing the end of an LMXB phaseand the birth of a radio millisecond pulsar, dur-ing which the system undergoes transient LMXBphases. This is consistent with a scenario put for-ward by Burderi et al. (2001) in which LMXBs,late in their evolution, may have episodes of masstransfer but spend most of their time in “radioejection” phases, during which the wind from anactive radio MSP sweeps material overflowing thecompanion out of the system.

In its current quiescent state, J1023 has beenextensively observed with optical and radio tele-scopes. The millisecond pulsar’s companion ap-pears as a V ∼ 17 magnitude star with a G-like absorption-line spectrum. The brightness and

1Department of Physics, McGill University, 3600 Uni-versity St, Montreal, QC H3A 2T8, Canada

2Netherlands Institute for Radio Astronomy (AS-TRON), Postbus 2, 7990 AA Dwingeloo, The Netherlands

3Astronomical Institute “Anton Pannekoek,” Universityof Amsterdam, 1098 SJ Amsterdam, The Netherlands

4Department of Physics and Astronomy, University ofBritish Columbia, 6224 Agricultural Road, Vancouver, BCV6T 1Z1

5National Radio Astronomy Observatory, 520 Edge-mont Rd., Charlottesville, VA 22901, USA

6Department of Physics, West Virginia University, Mor-gantown, WV 26505, USA

temperature show a 0.198-day modulation consis-tent with heating due to an irradiation luminosityof roughly 2L� from the primary (Thorstensen& Armstrong 2005). The combination of pulsartiming and optical radial velocity measurementsshows that the mass ratio is 7.1(1), while theo-retical constraints on the pulsar mass imply aninclination angle of ∼ 46◦, a companion mass of∼ 0.2 M� (assuming a pulsar mass of 1.4M�),and, based on Roche lobe size and optical appar-ent magnitude, a distance of ∼ 1.3 kpc (Archibaldet al. 2009). At radio wavelengths, the systemshows variable, frequency-dependent eclipses nearthe orbital phase at which the companion is closestto our line of sight to the pulsar, though the com-panion’s Roche lobe never actually intersects thisline of sight. Irregular short eclipses are also ob-served at all phases (Archibald et al. 2009). Suchvariable radio eclipses have been observed in othermillisecond pulsar systems, both in the so-called“black widow” systems in which the companionis thought to be a very low-mass degenerate starbeing ablated by the pulsar wind (Fruchter et al.1988; Stappers et al. 1996), and in several systemsmore similar to J1023, with companions that ap-pear to be unevolved and Roche-lobe-filling (Freireet al. 2003; D’Amico et al. 2001). Comparison ofJ1023 with these systems, as well as with LMXBsin quiescence, could clarify how accretion comesto an end and radio pulsars become active.

Prior to the discovery that J1023 containeda millisecond pulsar, and unfortunately after ithad returned to quiescence, Homer et al. (2006)observed J1023 with XMM-Newton. They con-cluded that it had a hard power-law spectrum,variability not necessarily correlated with the or-bital phase, and a 0.5–10 keV X-ray luminosityof 2.5 × 1032 erg s−1 (0.065L�) at a distance,based on estimates available at the time, of 2 kpc.This supported the suggestion of Thorstensen &Armstrong (2005) that J1023 had a NS primaryrather than a white dwarf as had previously beenthought.

Not only does the detection of J1023 as a radiomillisecond pulsar confirm that it is a NS binary,radio timing provides an ephemeris predicting therotation of the pulsar. For this reason, we ob-served J1023 for a longer time and with high timeresolution using XMM-Newton. Our observationsconfirm the spectral observations of Homer et al.

2

(2006), show evidence for orbital variability, andshow for the first time probable X-ray pulsationsat the pulsar period.

In section 2 we summarize the observations weused and our initial data reduction. In section3 we describe and give results from our spectralanalysis and our search for orbital and rotationalmodulation in the X-ray data. In section 4 wediscuss the implications for the nature of J1023’sX-ray emission and the relation of J1023 to sim-ilar systems, and in section 5 we summarize andsuggest further avenues of research.

2. Observations

The European Photon Imaging Camera, EPIC,aboard the XMM-Newton satellite, has two MOScameras (Turner et al. 2001) and one PN (Struderet al. 2001). All three cameras may be operated inan imaging mode, but the PN camera can also berun in a fast timing mode, in which the photonsfrom one CCD are accumulated along one dimen-sion, losing two-dimensional imaging, but achiev-ing a time resolution of 29 µs. Calibration uncer-tainties limit the absolute time accuracy of XMM-Newton to approximately 100 µs (Guainazzi et al.2010). XMM-Newton also has a coaxial opticaltelescope, the Optical Monitor, and two reflectiongrating spectrometers, however J1023 is too faintto permit interesting studies with the latter.

We obtained an XMM-Newton observation(OBSID 0560180801) of J1023 of 33 ks, corre-sponding to 1.94 binary orbits, on 2008 Nov 26.We operated the PN camera in fast timing modeand the MOS cameras in “prime full window”(imaging) mode, and we used the thin filter on allthe EPIC cameras. The data were free of soft pro-ton flares, so we were able to use the entire expo-sure time. We also retrieved the data set (OBSID0203050201) used by Homer et al. (2006) from theXMM-Newton archive; it consists of a single 16 ksexposure (0.92 binary orbits) obtained on 2004May 12, with all EPIC instruments using “primefull window” mode and the thin filter. This dataset is also free of soft proton flares, but since allthree EPIC instruments were operated in imagingmode, in this case high-resolution timing data arenot available.

The Optical Monitor was operated in both ob-servations with the B filter and a mode with 10-s

time resolution. Pipeline processing showed opti-cal modulation consistent with that reported byThorstensen & Armstrong (2005), so we did nofurther analysis of the optical data.

We analyzed the X-ray data with the XMM-Newton Science Analysis Software1 version 9.0.02

and xspec version 12.5.0ac3.We reprocessed the MOS and PN data with the

emchain and epchain pipelines, respectively, thenextracted photons meeting the recommended pat-tern, pulse invariant, and flag criteria. In particu-lar, this restricted MOS photons to the 0.2–12 keVenergy range, PN imaging photons to the 0.13–15 keV energy range, and PN timing photons tothe 0.25–15 keV energy range. For all analyses ex-cept the spectral fitting, we restricted the imagingphotons from either type of camera to 0.2–10 keVfor consistency, since this involved discarding onlya handful of photons.

The default pipeline processing showed no ev-idence for extended emission from J1023, so weused the SAS tool eregionanalyse to select ex-traction regions to optimize the ratio of sourcephotons to the square root of source plus back-ground photons. These regions were circles of ra-dius 35′′ (enclosing about 90% and 85% of the1.5 keV flux on the MOS and PN cameras, respec-tively) in imaging observations. For the timing-mode PN observation we adjusted the extractionregion by hand to optimize detection significance(see below), selecting a band 12′′ wide (enclos-ing about 60% of the 1.5 keV flux). We selectedbackground regions from nearby large circles inthe imaging-mode observations, and from a band50′′ wide, to one side of the source, in the timing-mode PN data. All together, this yielded approx-imately 5700 background-subtracted source pho-tons in imaging mode and approximately 2400background-subtracted source photons in timingmode.

We computed photon arrival times using theSAS barycentering tool, barycen, the precise op-tical position 10h23m47.67s +0◦38′41.2′′ (J2000)

1The XMM-Newton SAS is developed and maintained bythe Science Operations Centre at the European Space As-tronomy Centre and the Survey Science Centre at the Uni-versity of Leicester.

2http://xmm.esac.esa.int/sas/3http://heasarc.nasa.gov/docs/xanadu/xspec/

3

given in the NOMAD catalog (Zacharias et al.2004), and the DE405 solar system ephemeris.To produce pulsar rotational phases during the2008 observation, we used the tool tempo4 andthe ephemeris given in Archibald et al. (2009),which is based on radio pulsar timing data brack-eting the X-ray observation epoch. The extremelysensitive phase-coherent timing procedure used byArchibald et al. revealed minuscule variations inthe orbital period of J1023. These are modelledin the published ephemeris by expressing the or-bital period as a quadratic polynomial obtainedby a fit to several months of 2008 timing data.Extrapolating this polynomial back to the 2004XMM-Newton observation results in an unreason-able and poorly constrained orbital period at thetime of ∼ −6 days, a clear indication that theorbital period variations are not actually given bysuch a quadratic polynomial. While it is necessaryto model the orbital period derivatives to obtaingood quality pulsar phase predictions, it is possi-ble to obtain adequate orbital phase predictions byusing a simple model with constant orbital periodof 0.19809620 days and (pulsar) ascending node ofMJD 54801.970652993, based on the model fromArchibald et al. (2009). This model gives adequateorbital phase predictions back to 2004, match-ing the orbital phases observed in Thorstensen& Armstrong (2005) to within 1% of a period aswell as matching the ephemeris given in Archibaldet al. (2009) exactly on 2008 Dec 1. We there-fore used this model to evaluate the orbital phaseof each barycentered photon in both our observa-tions.

3. Results

3.1. Spectral fitting

We used xspec to fit each of several models toall EPIC imaging data sets from both epochs si-multaneously. To verify that the data sets werecompatible, we also fit different absorbed powerlaws to the two observations, obtaining compati-ble spectral indices and normalizations. For theMOS cameras we restricted photons to the recom-mended energy range, namely 0.2–10 keV, and forthe PN camera we used the recommended lowerlimit of 0.13 keV but used the upper limit of

4http://www.atnf.csiro.au/research/pulsar/tempo/

10−6

10−5

10−4

no

rmal

ized

co

un

ts s

−1 k

eV−

1 c

m−

2

10.2 0.5 2 5

1

1.5

2

rati

o

Energy (keV)

Fig. 1.— Absorbed neutron-star atmosphere pluspower-law spectral fit to all EPIC imaging-modephotons from both observations. Each data setis plotted in a different color: black is the 2004MOS1, red the 2004 MOS2, green the 2004 PN,blue the 2008 MOS1, and cyan the 2008 MOS2data set. The vertical axis is normalized by theeffective area, and the expected responses fromthe two additive components are overplotted sep-arately (the harder component is the power law).The lower panel is the ratio of model predicted fluxto observed flux in each group of detector chan-nels.

4

Table 1

Spectral fits to all EPIC image data

power-law neutron star atmosphereplus power law

NH (cm−2) < 5× 1019 < 1× 1020

photon index 1.26(4) 0.99(11)kT (keV) / photon index (Γ) · · · 0.12(2)Thermal emission radius (km)a · · · 0.7+0.5

−0.1

0.5-10 keV unabsorbed flux (erg cm−2 s−1) 4.66(17)× 10−13 4.9(3)× 10−13

0.5-10 keV luminositya (erg s−1) 9.4(4)× 1031 9.9(5)× 1031

Thermal fraction · · · 0.06(2)χ2/degrees of freedom 230/214 208/212Null hypothesis probability 0.21 0.56

aAssuming a distance of 1.3 kpc.

10 keV as there were not enough photons abovethis to provide any constraint. We grouped thephotons to obtain at least 20 counts per spectralbin so as to have approximately Gaussian statis-tics.

We obtained an adequate fit to all imaging datasets with a power-law model including photoelec-tric absorption. We did not obtain adequate fitswith an absorbed single black-body (χ2 = 1203.7with 213 degrees of freedom, for a null hypoth-esis probability of ∼ 10−138) or neutron-star at-mosphere model (χ2 = 1007.1 with 213 degreesof freedom, for a null hypothesis probability of∼ 10−103).

The simple power-law model, with a photon in-dex of 1.26(4), already has a null hypothesis prob-ability of 0.21, which is perfectly satisfactory froma statistical point of view. Adding a neutron-star atmosphere component necessarily improvesthe fit, at the cost of two extra parameters. Al-though the statistics do not call for a thermal com-ponent, we may expect one on physical grounds(for example from the surface of the neutron star).We therefore also consider an absorbed neutron-star atmosphere plus power-law model to fit thesedata, as shown in Figure 1. This model, speci-fied in xspec as phabs(nsa+pow), is based on Za-vlin et al. (1996). During fitting we froze the neu-tron star mass at 1.4M� and the radius at 10 kmfor the purposes of redshift and light bending; the

smaller effective thermal emission radius suggeststhat the emission comes from a small “hot spot”on the surface. In light of the fact that the esti-mated B < 108 G (Archibald et al. 2009), we alsoselected a model in which the magnetic field hasnegligible effects on the atmosphere, i.e. . 109 G.

These two models are summarized in Table 1.Uncertainties given are 90% intervals returnedby xspec’s error command; luminosities are ob-tained by using the cflux model component to es-timate unabsorbed and thermal fluxes in the 0.5–10 keV range and then using an assumed distanceof 1.3 kpc to determine a luminosity. “Thermalfraction” is the fraction of this luminosity due tothe thermal component of the spectrum, if any.

In both cases, fitting for photoelectric absorp-tion in the model gave a very low upper boundon the neutral hydrogen column density, as notedin Homer et al. (2006). Such a low value is tobe expected since the entire Galactic column den-sity in this direction is estimated5 to be only1.9× 1020 cm−2 (based on Kalberla et al. 2005).

3.2. Orbital variability

To test for orbital variability, we selected 0.2–12 keV source photons from the imaging-modedata sets in both our observations and those of

5Using the HEASARC online calculator, http://heasarc.

nasa.gov/cgi-bin/Tools/w3nh/w3nh.pl

5

0.01

0.02

0.03

0.04

cou

nt

rate

(s−1

)

(a)

0.0 0.5 1.0 1.5 2.0Orbital phase

2.4

2.8

Har

dn

ess

rati

o

(b)

Fig. 2.— Panel (a): Average MOS-equivalentcount rate in the 0.2–12 keV energy range asa function of orbital phase. These data are aweighted average of all the imaging-mode data.Dashed and dotted vertical lines indicate the be-ginning and end of the 1.4 GHz radio eclipse, re-spectively. The dashed curve is a best-fit combina-tion of four sinusoids. Panel (b): number of MOSphotons in the 1–12 keV range divided by num-ber of MOS photons in the 0.2–1 keV range, alsoas a function of orbital phase. Both panels showtwo cycles for clarity. The companion’s closest ap-proach to our line of sight to the pulsar happensat orbital phase 0.25; that the radio eclipse is off-center may be due to eclipsing material being farout of the orbital plane, as shown in Figure 5.

0.025

0.050

MJD 53137.4

0.025

0.050

MJD 53137.6

0.025

0.050

MJD 54796.2

0.025

0.050

MJD 54796.4

0.0 0.2 0.4 0.6 0.8 1.0Orbital phase

0.000

0.025

0.050

MJD 54796.6

Cou

nt

rate

(ph

oton

s/s)

Fig. 3.— Average count rate in the 0.2–12 keVenergy range as a function of orbital phase, foreach orbit independently; the starting MJD foreach orbit is indicated. The vertical axis isMOS-equivalent count rate (based on MOS1/2and imaging-mode PN data, scaled to match theMOS1 count rate), and the dashed horizontal lineindicates the average count rate over all orbitalphases and data sets. Where no bar is plotted, nodata are available.

6

Homer et al. (2006) and reduced them to thesolar system barycenter. When combining datafrom multiple instruments, to take into accountthe different orbital coverages and sensitivities, weweighted bin values so that all average count ratesmatch that observed in the MOS1 camera in thesame observation. For the hardness ratio analy-sis, we omitted the PN data so that their differentenergy response and incomplete orbital coveragewould not skew the results.

A weighted histogram of these counts is shownin Fig. 2, along with a best-fit sinusoid based ona finely binned profile. We selected four sinusoidsas this appears to give a good representation ofthe curve and a time resolution similar to the his-togram. A Kuiper test (Paltani 2004) gives a prob-ability of 1.3 × 10−19 (9.0σ) that photons drawnfrom a uniform distribution in orbital phase wouldbe this non-uniform; the reduced χ2 for a constantfit to the histogram is 11.3, with 11 degrees of free-dom. While the evidence for variability is strong,with only ∼ 3 orbits covered by the two observa-tions it is not certain that this variability is linkedto orbital phase, although the fact that the mini-mum in the X-ray light curve occurs near orbitalphase 0.25, when the companion passes closest toour line of sight to the pulsar, suggests a link.To test this we plotted the photon arrival ratesfor each orbital period separately. Figure 3 showsthat the flux during each individual orbit appearsto be lower during phase 0–0.5 than during phase0.5–1, which suggests that the variability is indeedorbital. Note that Homer et al. (2006) detectedvariability, but having only limited orbital cover-age, could not determine whether it was orbital.We also computed a hardness ratio (Fig. 2b), di-viding the number of photons harder than 1 keVby the number of photons softer than 1 keV andcomparing the eclipse versus non-eclipse regions(for this purpose we defined the “eclipse” regionto be phases 0–0.5). The reduced χ2 for a fit ofthese values to a constant hardness ratio is 8.76 for1 degree of freedom, and the probability of such areduced χ2 arising if the hardness ratio were con-stant is 3.1×10−3 (2.7σ). Thus we see marginallysignificant softening in the eclipse region.

3.3. Pulsations at the pulsar period

To test for pulsations, we extracted source andbackground photons from the PN camera using

0.04

0.05

0.06

0.07

0.08

0.09

Cou

nt

rate

(ph

oton

s/s)

(a)

0.0 0.5 1.0 1.5 2.0Pulsar rotational phase

Rad

iofl

ux

den

sity

(arb

.u

nit

s)(b)

Fig. 4.— Panel (a) shows a background-subtracted light curve based on 0.25–2.5 keV PNphotons folded according to the radio ephemeris;the sinusoid (drawn with a dotted line) is a two-component sinusoid fit to the unbinned photonarrival times. Panel (b) shows a 1400 MHz pro-file obtained with Arecibo (Archibald et al. 2009).Both panels show two cycles for clarity. Theuncertainty in relative alignment, due primarilyto the absolute timing uncertainty in the XMM-Newton data, is indicated by the horizontal errorbar. The 20 µs uncertainty relative to the radiodata due to dispersion measure uncertainty is rel-atively unimportant.

7

the energy range 0.25–2.5 keV, selected to givethe most significant detection. We barycenteredthe photon arrival times and used the programtempo6 and the contemporaneous radio ephemerisgiven in Archibald et al. (2009) to assign each pho-ton a rotational phase. We then tested these pho-tons for uniform distribution in phase. The Kuipertest (Paltani 2004) gave a (single-trial) null hy-pothesis probability of 3.7 × 10−6 (4.5σ) and theH test (de Jager et al. 1989) gave a (single-trial)null hypothesis probability of 2.4 × 10−6 (4.6σ),using an optimal number of sinusoids (two). Weconfirmed that no significant pulsations were de-tected with the background photons or with anincorrect ephemeris. We also verified that the pe-riod predicted by the ephemeris is very close to theperiod at which the significance peaks (holding allother ephemeris parameters fixed).

To estimate the degree to which the X-ray fluxis modulated at the pulse period, one could simplytake the lowest bin in the histogram as the back-ground level and compute the fraction of photonsabove it; this yields a pulsed fraction of 0.27(9).However, this method is subject to large uncer-tainties, dependence on binning, and a large sta-tistical upward bias due to the fact that we se-lected the bin in which signal plus noise is lowest,rather than that in which the (unknown) signal islowest. To avoid these problems, we define a root-mean-squared pulsed flux by fitting a model F (x)with two sinusoids:

fRMS =

√∫ 1

0

(F (x)− F )2dx,

where F is the model mean flux. This is in somesense a degree of modulation; if the signal consistsof a constant background plus a variable emissionprocess that drops to zero, this is not directly mea-suring the fraction of emission due to the vari-able process. (There is a conversion factor thatdepends on the exact shape of the pulse profile.)However, this quantity can be computed with sub-stantially less uncertainty and bias. Computation-ally, we estimate the root-mean-squared amplitudein the Fourier domain, based on the amplitudes ofthe two complex Fourier coefficients (since higher-order Fourier coefficients are dominated by noise);since noise always contributes a positive power to

6http://www.atnf.csiro.au/research/pulsar/tempo/

Fourier coefficients, to reduce the bias we sub-tract the expected contribution of noise from thesquared amplitude of each coefficient before tak-ing the square root. We then convert this pulsedflux value to a pulsed fraction by dividing by thetotal background-subtracted flux from J1023.

For the energy range 0.25–2.5 keV we estimatea root-mean-squared pulsed fraction of 0.11(2). Inthe two subbands 0.25–0.6 keV and 0.6–2.5 keV,we find root mean squared pulsed fractions of0.17(5) and 0.14(3), respectively, with profiles thatare broadly similar and in phase. Above 2.5 keVwe find no evidence for pulsations, though a 3σupper limit on the pulsed fraction is only 0.20.

4. Discussion

To summarize our results for J1023, we ob-served an X-ray spectrum dominated by a hardpower-law component, possibly with a small, soft,thermal contribution. The emission appears to bemodulated at the 0.198-day orbital period, withsubstantial dips as the companion passes near theline of sight; these dips are accompanied by a pos-sible softening in the spectrum. We also foundthat the X-ray emission is very likely modulatedat the 1.69-ms rotational period of the radio pul-sar.

4.1. Nature of the X-ray emission

In a review of X-ray emission from millisec-ond pulsars, Zavlin (2007) describes three primarysources of X-ray emission: emission from an in-trabinary shock, emission from the neutron staritself, which is some combination of thermal andmagnetospheric emission, and emission from a pul-sar wind nebula outside the binary system. As allthese mechanisms may be operating in J1023, wenext consider their contributions, if any, to theobserved X-rays from this system.

4.1.1. Emission From an Intrabinary Shock

J1023 likely has a strong pulsar wind, since theoptical data suggest that the companion is be-ing heated by a luminosity of ∼ 2L� from thepulsar (Thorstensen & Armstrong 2005). This ismuch greater than the X-ray luminosity of the sys-tem (∼ 0.03L�), but well below the upper limit(∼ 80L�) on the spin-down luminosity of the pul-sar (Archibald et al. 2009). If material is leaving

8

−1.0 −0.5 0.0x (cm) ×1011

−1.0

−0.5

0.0

y(c

m)

×1011

R�

Fig. 5.— J1023 system geometry at phase 0.25 asseen from within the plane of the sky. The systemorbit is in and out of the page, the dashed lineis our line of sight to the pulsar, and the dottedline is our line of sight to L1, which is eclipsedby the Roche-lobe-filling companion. The pulsaris indicated by a not-to-scale circle, but the com-panion is drawn to scale. The tick mark on theline connecting pulsar and companion marks thecenter of mass of the system. This system geom-etry assumes a pulsar mass of 1.4 M�; a highermass would mean a larger system seen more nearlyface-on, but the range of plausible angles is con-strained to ∼ 53◦–34◦ by the narrow range of plau-sible pulsar masses 1.0–3.0M� (Archibald et al.2009). Note that at this orbital phase the X-raysare near their minimum, and low radio frequen-cies from the pulsar are eclipsed, indicating thepresence of ionized material well out of the orbitalplane.

the companion, either through Roche-lobe over-flow or a stellar wind, we should expect an in-trabinary shock, where this flow meets the pulsarwind. Such a shock could readily produce power-law X-ray emission (Arons & Tavani 1993). If lo-calized, it could easily account for the orbital mod-ulation we observe in the X-ray emission. If theemission were due to Roche-lobe-overflowing ma-terial meeting the pulsar wind we might expect itto be localized at or near L1. Given the systemgeometry described in Archibald et al. (2009) (a46◦ inclination angle, corresponding to a pulsarmass of 1.4M�, and a Roche-lobe-filling compan-ion; see Figure 5), the L1 point itself is eclipsedby the companion for 0.32 of the orbit, centeredon orbital phase 0.25. We would expect spectralsoftening during this period, since the relativelyhard power-law emission is blocked but the, pre-sumably softer, emission from the neutron star isnot. A larger X-ray-emitting region, due either tomaterial streaming away from the L1 point or toa stellar wind shock, would result in a broader,shallower, and potentially asymmetric dip in theX-ray light curve.

The possibility of a wind from the companionis interesting, since it is not clear how Roche-lobeoverflow would provide the ionized material caus-ing the radio eclipses, which occur well above theorbital plane, roughly centered on the companion’sclosest approach to our line of sight. Archibaldet al. (2009) reported dispersion measure (DM)changes of ∼ 0.15 pc cm−3 above a cutoff fre-quency of ∼ 1 GHz at radio eclipse ingress andegress. Assuming that this is the plasma frequencyimplies an electron density of ∼ 1010 cm−3; com-bined with the excess DM measurement, this im-plies a layer of thickness ∼ 4 × 107 cm, rela-tively thin compared to the orbital separation of1.2 × 1011 cm. This thin sheet of material couldin principle arise as the shock front where the pul-sar wind meets a wind from the companion, butit is difficult to explain the companion’s wind be-ing strong enough to support such a shock: giventhe line of sight geometry of the system, the ma-terial causing the radio eclipses must be about asfar from the companion as it is from the pulsar.If we assume that the highly relativistic pulsarwind provides the ∼ 2L� required by the compan-ion heating models of Thorstensen & Armstrong(2005), it seems difficult to explain how the com-

9

panion could have a wind of comparable pressureanywhere along the line of sight. Thus the radioeclipses remain a mystery, and Roche-lobe over-flow seems a more likely explanation for the originof the intrabinary shock.

4.1.2. Emission From the Neutron Star

Some millisecond pulsars, binary or isolated,have X-ray emission modulated at the rotationalperiod. This emission is some combination of mag-netospheric emission, from high-energy particlesin the magnetosphere, and thermal emission, frompolar caps heated by bombardment by high-energyparticles moving along the open field lines. Ob-servationally, Zavlin (2007) draws the distinctionthat magnetospherically dominated emission hashigh pulsed fractions, narrow peaks, and power-law spectra, while polar-cap-dominated emissionhas lower pulsed fractions, broader peaks, andthermal spectra. Zavlin also suggests that pulsarswith spin-down luminosities & 1035 erg s−1 tendto be magnetospherically dominated while thosewith spin-down luminosities . 1034 erg s−1 tendto be thermally dominated, but since few examplesare known, this classification is tentative, and thespin-down luminosity from J1023 is only knownto be < 3 × 1035 erg s−1 (Archibald et al. 2009).In any case, classification of the pulsations we ob-serve in J1023 is difficult.

In systems in which thermal emission from theneutron star provides all the detectable X-rays,detailed modelling indicates that the pulsed frac-tion should normally be . 50% (that is, pulsedemission must be accompanied by roughly equalor greater unpulsed emission) due to light bend-ing and the large size of the polar caps, though forcertain combinations of parameters higher pulsedfractions can occur (Bogdanov et al. 2008). If thepulsations are purely thermal, then our spectralupper limit on the thermal fraction of 0.06(2) isdifficult to reconcile with the fact that we observea fractional modulation of 0.11(2) at the pulsarperiod. Thus it seems likely that at least somemagnetospheric emission is present, as it has anonthermal spectrum and can more readily havehigher pulsed fractions. The poor signal-to-noisein our observations, due to the high background(from the system itself as well as instrumental)and the scarce photons, makes it impractical todetermine the sharpness of the pulse profile or the

hardness of the pulsations. In any case the pul-sations, if real, are a clear sign of X-ray emissionfrom the pulsar itself.

4.1.3. Emission From a Pulsar Wind Nebula

Nebular emission from pulsars arises when theparticle wind driven by the pulsar’s magneto-spheric activity flows out of the pulsar’s immediateneighborhood and meets the surrounding medium;for reviews see Gaensler & Slane (2006) or Kaspiet al. (2006). Pulsars like J1023 that whose windis confined by the ambient interstellar mediumexhibit nebular emission that generally takes acometary form, with an arc-like bow shock pre-ceding the pulsar and a “trail” of ejected materialstreaming back along the pulsar’s track.

The angular size of bow shock emission dependson the pulsar’s spin-down luminosity, its propermotion, and the local interstellar medium den-sity. Kargaltsev & Pavlov (2008) give a formulafor predicting the “stand-off” angle of the X-raybow shock from the pulsar:

θ = 5.4′′ n−1/20.1 µ−1

19 E1/235 D−2

1.3,

where n0.1 is the mean density of the interstellarmedium divided by 0.1 cm−3, µ19 is the propermotion divided by 19 mas yr−1, E35 is the spin-down luminosity divided by 1035 erg s−1, andD1.3 is the distance to J1023 divided by 1.3 kpc.We have supplied best-guess parameters for J1023(Archibald et al. 2009), so that absent further in-formation, we might expect a stand-off distance of∼ 5′′ for the X-ray bow shock. Thus although neb-ular emission is not resolved in our XMM-Newtonimages, higher-resolution images with the Chan-dra X-ray Observatory, or in the Hα or radiobands, might yet resolve a bow shock.

Kargaltsev & Pavlov (2008) find that the ra-tio of neutron-star emission to nebular emissiongenerally lies between ∼ 0.1 and 10, which sug-gests that nebular emission may produce someof the X-rays we observe. Moreover, Kargaltsev& Pavlov find that nebular emission from bowshocks has a power-law spectrum with photon in-dices 1 . Γ . 2 (Kargaltsev & Pavlov 2008), con-sistent with what we observe from J1023, thoughthe latter is somewhat harder than most pulsarwind nebulae. The orbital variability, on the otherhand, argues against that component of the emis-

10

sion arising from an extrabinary pulsar wind neb-ula.

4.2. Comparison to similar systems

In quiescence, most LMXBs have spectra thatare dominated by thermal emission from the neu-tron star. This emission is typically consistentwith a hydrogen atmosphere model with effectiveradius∼ 10 km (Brown et al. 1998). Some LMXBsin quiescence also have a small power-law compo-nent of photon index 1 . Γ . 2 in their spec-trum (Campana et al. 1998), while a few quies-cent LMXBs have spectra completely dominatedby this non-thermal component. For example, thequiescent LMXB SAX J1808.4−3658 has a spec-trum that is a fairly hard power law with no de-tectable thermal emission (Heinke et al. 2009).J1023 resembles this latter category, although if athermal component is present its effective radiusappears to be substantially less than 10 km andthe power-law spectral index is fairly hard. Onthe other hand, J1023’s radio and possible X-raypulsations distinguish it from all known quiescentLMXBs.

SAX J1808.4−3658 is of particular interest be-cause it is the prototype and best-studied exam-ple of a class of LMXBs from which millisecondX-ray pulsations have been detected during ac-tive phases. Many authors have suggested thatSAX J1808.4−3658 should turn on as a radio pul-sar during quiescence (e.g. Wijnands & van derKlis 1998; Chakrabarty & Morgan 1998; Cam-pana et al. 2004), but no radio pulsations havebeen detected in spite of thorough searches (e.g.Burgay et al. 2003; Iacolina et al. 2009). Thenon-detection of radio pulsations could be be-cause SAX J1808.4−3658 is shrouded by previ-ously ejected ionized material, or it could sim-ply be an issue of unfortunate beaming. In anycase, it is natural to compare SAX J1808.4−3658to J1023.

During its active phases, SAX J1808.4−3658reaches a sustained luminosity of∼ 4×1036 erg s−1

(Galloway & Cumming 2006), in sharp contrastto the upper limit of 2 × 1034 erg s−1, availablefor J1023 during its disk phase (Archibald et al.2009). In its quiescent state, SAX J1808.4−3658shows X-ray emission dominated by a hard powerlaw (Γ = 1.74(11)), with a small (. 0.1) frac-tion of thermal emission, and a total X-ray lu-

minosity of 7.9(7) × 1031 erg s−1 (Heinke et al.2009). This is somewhat similar to what we ob-serve in J1023, which has a total luminosity of9.9(5)× 1031 erg s−1, although the spectral indexof 1.26(4) is somewhat harder than that seen fromSAX J1808.4−3658. If J1023 has a thermal com-ponent, its power-law is substantially harder, witha spectral index of 0.99(11). The thermal compo-nent would have an effective radius of 0.7+0.5

−0.1 kmand supply 0.06(2) of the total luminosity. Whilethe spectral index seen in J1023 is harder thanthat seen from SAX J1808.4−3658, it is in linewith the hard spectral indices seen from otherX-ray-detected radio millisecond pulsars (for ex-ample 47 Tuc W, though PSR J1740-530 is softer;see below). Since SAX J1808.4−3658 is knownto cool quickly, even thermal X-ray pulsations inquiescence would be evidence of magnetosphericactivity heating its polar caps. Campana et al.(2002) searched for such X-ray pulsations fromSAX J1808.4−3658 in quiescence, but were unableto conduct meaningful searches due to its faint-ness; SAX J1808.4−3658 is roughly 3.5 kpc away(Galloway & Cumming 2006) compared to theroughly 1.3 kpc we assume for J1023 (Archibaldet al. 2009).

The similarity of the X-ray properties of SAXJ1808.4−3658 in quiescence to those of J1023 sug-gests that SAX J1808.4−3658 may indeed har-bor a radio millisecond pulsar in quiescence; con-versely, the same similarity suggests that J1023may undergo episodes of active accretion.

Among radio pulsars, the two best-studied ex-amples which closely resemble J1023 are 47 Tuc W(Bogdanov et al. 2005) and PSR J1740−5430(D’Amico et al. 2001). Both pulsars exhibit large,variable radio-frequency eclipses, and both arethought to have somewhat massive unevolvedcompanions (& 0.13M� and 0.19–0.8 M�, re-spectively). For comparison, J1023’s companionis thought to be ∼ 0.2 M�, and shows a spectrumcharacteristic of a main-sequence star (Archibaldet al. 2009; Thorstensen & Armstrong 2005). Theradio eclipses are in all three cases evidence formaterial leaving the companion and presumablybeing expelled from the system, and in all threecases the companion appears to fill, and perhapsoverflow, its Roche lobe (Bogdanov et al. 2005;Ferraro et al. 2001; Thorstensen & Armstrong2005).

11

Bogdanov et al. (2005) studied 47 Tuc W andfound an X-ray spectrum consisting of a domi-nant hard power-law (Γ = 1.14(35), contributing∼ 75% of the total flux) plus a thermal compo-nent. They also observed orbital variability, andin particular a substantial X-ray eclipse spanningphases7 0.15–0.45 and centered slightly ahead ofthe optical minimum, at phase 0.2. This is ac-companied by substantial softening of the X-rayspectrum during eclipse. Given these observa-tions, Bogdanov et al. suggest that the power-law X-ray emission originates from a shocked re-gion where material overflowing from the compan-ion of 47 Tuc W meets the pulsar wind and isblown out of the system. Bogdanov et al. (2010)studied Chandra X-ray Observatory observationsof PSR J1740−5430. Although limited by thescarcity of photons, they found a spectrum con-sistent with a power law of index Γ = 1.73(8), orwith a somewhat harder power-law plus a blackbody component. Their estimates of variabil-ity were limited by restricted orbital coverage aswell as by a paucity of photons, but they foundmarginal evidence for orbital modulation, in par-ticular a possible decrease in luminosity during theradio eclipses, possibly accompanied by a soften-ing. Our observations of J1023 are more photon-starved than those of Bogdanov et al. (2005) of47 Tuc W, but we do see evidence for eclipses,and possibly for softening during eclipses. In thecase of J1023, the possibility of overflowing gas issupported by the recent disk phase.

The similarity of the observational propertiesof these three millisecond pulsars suggests thattheir companions are all overflowing their Rochelobes, so that mass flows through the L1 point,where it encounters the pulsar wind and is sweptaway. On the other hand, the similarity also sug-gests the possibility that like J1023, 47 Tuc W orPSR J1740−5430 may undergo episodes in whichthe mass transfer rate from the companion in-creases and an accretion disc forms. Such episodeswould presumably be signalled by optical changes,possibly by extinction of the radio pulsations, andpresumably by modest X-ray brightening. All to-gether, these observations suggest that as a system

7Note that these authors use a different convention for or-bital phase, setting phase 0.5 to the optical minimum, whilewe set phase 0.25 to the optical minimum. All phases givenhere have been converted to our convention.

reaches the end of its life as an LMXB, its accre-tion rate drops and the millisecond pulsar becomesactive. After this point, the wind from the pul-sar generally prevents mass leaving the compan-ion from forming an accretion disk or entering thepulsar magnetosphere. Temporary increases in themass accretion rate may allow the occasional sup-pression of the wind and formation of an accretiondisk, as was observed in J1023 in 2001.

5. Conclusions

J1023 shows X-ray emission probably consistingprimarily of emission from an intrabinary shock,plus a smaller amount of emission from the pul-sar itself and possibly some unresolved nebularemission. The shock emission shows substan-tial variability that appears to be linked to or-bital phase, as well as possible spectral soften-ing during eclipses. Similarities to 47 Tuc Wand PSR J1740−5340 suggest that this may bedue to gas overflowing from the companion meet-ing the pulsar wind in a shock close to the com-panion. The emission from the pulsar itself, ifreal, may either be due to curvature radiation inthe magnetosphere, or thermal radiation from po-lar caps heated by high-energy particles stream-ing downward. Further observations to provideimproved spectra should help clarify the originof both pulsed and unpulsed emission, as wouldfurther high-time-resolution observations to al-low better measurement of pulse profiles, pulsedhardness ratios, and orbital variations of pulsedfraction. X-ray observations with higher spatialresolution might also resolve extended nebularemission around J1023. Better understanding ofthis object promises to clarify the transition fromLMXB to radio millisecond pulsar, and may showthat other known systems are currently in this fas-cinating transition state.

Acknowledgements

The authors thank Mallory S. E. Roberts foruseful discussions. This work was based on ob-servations obtained with XMM-Newton, an ESAscience mission with instruments and contribu-tions directly funded by ESA Member States andNASA. AMA is supported by a Schulich gradu-ate fellowship. VMK acknowledges support fromNSERC, FQRNT, CIFAR, and holds a Canada

12

Research Chair and the Lorne Trottier Chair inAstrophysics and Cosmology. SB is a CIFARJunior Fellow. JWTH is an NWO Veni Fellow.IHS received support from an NSERC DiscoveryGrant. DRL and MAM are supported by a WVEPSCOR Research Challenge Grant.

Facility: XMM-Newton

REFERENCES

Archibald, A. M., Stairs, I. H., Ransom, S. M.,Kaspi, V. M., Kondratiev, V. I., Lorimer,D. R., McLaughlin, M. A., Boyles, J., Hessels,J. W. T., Lynch, R., van Leeuwen, J., Roberts,M. S. E., Jenet, F., Champion, D. J., Rosen,R., Barlow, B. N., Dunlap, B. H., & Remillard,R. A. 2009, Science, 324, 1411

Arons, J. & Tavani, M. 1993, ApJ, 403, 249

Bogdanov, S., Grindlay, J. E., & Rybicki, G. B.2008, ApJ, 689, 407

Bogdanov, S., Grindlay, J. E., & van den Berg, M.2005, ApJ, 630, 1029

Bogdanov, S., van den Berg, M., Heinke, C. O.,Cohn, H. N., Lugger, P. M., & Grindlay, J. E.2010, ApJ, 709, 241

Brown, E. F., Bildsten, L., & Rutledge, R. E. 1998,ApJ, 504, L95+

Burderi, L., Possenti, A., D’Antona, F., Di Salvo,T., Burgay, M., Stella, L., Menna, M. T., Iaria,R., Campana, S., & d’Amico, N. 2001, ApJ,560, L71

Burgay, M., Burderi, L., Possenti, A., D’Amico,N., Manchester, R. N., Lyne, A. G., Camilo,F., & Campana, S. 2003, ApJ, 589, 902

Campana, S., Colpi, M., Mereghetti, S., Stella, L.,& Tavani, M. 1998, Astron. Astropys. Rev., 8,279

Campana, S., D’Avanzo, P., Casares, J., Covino,S., Israel, G., Marconi, G., Hynes, R., Charles,P., & Stella, L. 2004, ApJ, 614, L49

Campana, S., Stella, L., Gastaldello, F.,Mereghetti, S., Colpi, M., Israel, G. L., Bur-deri, L., Di Salvo, T., & Robba, R. N. 2002,ApJ, 575, L15

Chakrabarty, D. & Morgan, E. H. 1998, Nature,394, 346

D’Amico, N., Lyne, A. G., Manchester, R. N., Pos-senti, A., & Camilo, F. 2001, ApJ, 548, L171

D’Amico, N., Possenti, A., Manchester, R. N.,Sarkissian, J., Lyne, A. G., & Camilo, F. 2001,ApJ, 561, L89

de Jager, O. C., Raubenheimer, B. C., &Swanepoel, J. W. H. 1989, A&A, 221, 180

Deloye, C. J. 2008, in American Institute ofPhysics Conference Series, Vol. 983, 40 Yearsof Pulsars: Millisecond Pulsars, Magnetars andMore, ed. C. Bassa, Z. Wang, A. Cumming, &V. M. Kaspi, 501–509

Ferraro, F. R., Possenti, A., D’Amico, N., &Sabbi, E. 2001, ApJ, 561, L93

Freire, P. C., Camilo, F., Kramer, M., Lorimer,D. R., Lyne, A. G., Manchester, R. N., &D’Amico, N. 2003, MNRAS, 340, 1359

Fruchter, A. S., Stinebring, D. R., & Taylor, J. H.1988, Nature, 333, 237

Gaensler, B. M. & Slane, P. O. 2006, ARA&A, 44,17

Galloway, D. K. & Cumming, A. 2006, ApJ, 652,559

Guainazzi, M. et al. 2010, XMM-Newton ScienceOperations Centre, XMM-SOC-CAL-TN-0018

Heinke, C. O., Jonker, P. G., Wijnands, R., De-loye, C. J., & Taam, R. E. 2009, ApJ, 691, 1035

Homer, L., Szkody, P., Chen, B., Henden, A.,Schmidt, G., Anderson, S. F., Silvestri, N. M.,& Brinkmann, J. 2006, AJ, 131, 562

Iacolina, M. N., Burgay, M., Burderi, L., Possenti,A., & di Salvo, T. 2009, A&A, 497, 445

Kalberla, P. M. W., Burton, W. B., Hartmann,D., Arnal, E. M., Bajaja, E., Morras, R., &Poppel, W. G. L. 2005, A&A, 440, 775

Kargaltsev, O. & Pavlov, G. G. 2008, in Ameri-can Institute of Physics Conference Series, Vol.983, 40 Years of Pulsars: Millisecond Pulsars,Magnetars and More, ed. C. Bassa, Z. Wang,A. Cumming, & V. M. Kaspi, 171–185

13

Kaspi, V. M., Roberts, M. S. E., & Harding,A. K. 2006, Isolated neutron stars, ed. Lewin,W. H. G. & van der Klis, M. (Cambridge Uni-versity Press), 279–339

Paltani, S. 2004, A&A, 420, 789

Stappers, B. W., Bailes, M., Lyne, A. G., Manch-ester, R. N., D’Amico, N., Tauris, T. M.,Lorimer, D. R., Johnston, S., & Sandhu, J. S.1996, ApJ, 465, L119

Stella, L., Campana, S., Colpi, M., Mereghetti, S.,& Tavani, M. 1994, ApJ, 423, L47

Struder, L., Briel, U., Dennerl, K., Hartmann,R., Kendziorra, E., Meidinger, N., Pfeffermann,E., Reppin, C., Aschenbach, B., Bornemann,W., Brauninger, H., Burkert, W., Elender,M., Freyberg, M., Haberl, F., Hartner, G.,Heuschmann, F., Hippmann, H., Kastelic, E.,Kemmer, S., Kettenring, G., Kink, W., Krause,N., Muller, S., Oppitz, A., Pietsch, W., Popp,M., Predehl, P., Read, A., Stephan, K. H.,Stotter, D., Trumper, J., Holl, P., Kemmer,J., Soltau, H., Stotter, R., Weber, U., We-ichert, U., von Zanthier, C., Carathanassis, D.,Lutz, G., Richter, R. H., Solc, P., Bottcher, H.,Kuster, M., Staubert, R., Abbey, A., Holland,A., Turner, M., Balasini, M., Bignami, G. F.,La Palombara, N., Villa, G., Buttler, W., Gian-ini, F., Laine, R., Lumb, D., & Dhez, P. 2001,A&A, 365, L18

Thorstensen, J. R. & Armstrong, E. 2005, AJ, 130,759

Turner, M. J. L., Abbey, A., Arnaud, M., Balasini,M., Barbera, M., Belsole, E., Bennie, P. J.,Bernard, J. P., Bignami, G. F., Boer, M., Briel,U., Butler, I., Cara, C., Chabaud, C., Cole,R., Collura, A., Conte, M., Cros, A., Denby,M., Dhez, P., Di Coco, G., Dowson, J., Fer-rando, P., Ghizzardi, S., Gianotti, F., Goodall,C. V., Gretton, L., Griffiths, R. G., Hainaut,O., Hochedez, J. F., Holland, A. D., Jourdain,E., Kendziorra, E., Lagostina, A., Laine, R.,La Palombara, N., Lortholary, M., Lumb, D.,Marty, P., Molendi, S., Pigot, C., Poindron,E., Pounds, K. A., Reeves, J. N., Reppin,C., Rothenflug, R., Salvetat, P., Sauvageot,J. L., Schmitt, D., Sembay, S., Short, A. D. T.,Spragg, J., Stephen, J., Struder, L., Tiengo, A.,

Trifoglio, M., Trumper, J., Vercellone, S., Vi-groux, L., Villa, G., Ward, M. J., Whitehead,S., & Zonca, E. 2001, A&A, 365, L27

Wang, Z., Archibald, A. M., Thorstensen, J. R.,Kaspi, V. M., Lorimer, D. R., Stairs, I., & Ran-som, S. M. 2009, ApJ, 703, 2017

Wijnands, R. & van der Klis, M. 1998, Nature,394, 344

Zacharias, N., Monet, D. G., Levine, S. E., Urban,S. E., Gaume, R., & Wycoff, G. L. 2004, inBulletin of the American Astronomical Society,Vol. 36, Bulletin of the American AstronomicalSociety, 1418

Zavlin, V. E. 2007, Ap&SS, 308, 297

Zavlin, V. E., Pavlov, G. G., & Shibanov, Y. A.1996, A&A, 315, 141

This 2-column preprint was prepared with the AAS LATEXmacros v5.2.

14