Discovery of UV millisecond pulsations and moding in the low mass X-ray binary state of transitional millisecond pulsar J1023+0038
Amruta D. Jaodand, Juan V. Hernández Santisteban, Anne M. Archibald, Jason W. T. Hessels, Slavko Bogdanov, Christian Knigge, Nathalie Degenaar, Adam T. Deller, Simone Scaringi, Alessandro Patruno
DD raft version M arch
1, 2021Typeset using L A TEX twocolumn style in AASTeX63
Discovery of UV millisecond pulsations and moding in the low mass X-ray binary state of transitional millisecond pulsarJ1023 + A mruta D. J aodand , J uan V. H ern ´ andez S antisteban , A nne M. A rchibald , J ason W. T. H essels , S lavko B ogdanov , C hristian K nigge , N athalie D egenaar , A dam T. D eller , S imone S caringi , and A lessandro P atruno ASTRON, Netherlands Institute for Radio Astronomy, Oude Hoogeveensedijk 4, 7991 PD Dwingeloo, The Netherlands Anton Pannekoek Institute for Astronomy, University of Amsterdam, Science Park 904, 1098 XH, Amsterdam, The Netherlands California Institute of Technology, 1200 E Califronia Blvd., Pasadena, CA 91125, USA SUPA School of Physics & Astronomy, University of St Andrews, North Haugh, St Andrews KY16 9SS, UK Columbia Astrophysics Laboratory, Columbia University, 550 West 120th Street, New York, NY 10027, USA School of Physics & Astronomy, University of Southampton, Southampton SO17 1BJ, UK Centre for Astrophysics and Supercomputing, Swinburne University of Technology, John St, Hawthorn, VIC 3122, Australia Department of Physics and Astronomy, Texas Tech University, Lubbock, TX 79409-1051, USA Leiden Observatory, Leiden University, P.O. Box 9513, NL-2300 RA Leiden, The Netherlands Institute of Space Sciences (IEEC-CSIC) Campus UAB, Carrer de Can Magrans, s / n, E-08193 Barcelona, Spain Submitted to ApJABSTRACTPSR J1023 + + Kepler (400 −
800 nm),
Hubble Space Telescope (180 −
280 nm),
XMM-Newton (0 . −
10 keV) and
NuSTAR (3 −
79 keV). We demonstrate that low and high luminosity modesin the UV band are strictly simultaneous with the X-ray modes and change the UV brightness by a factor of ∼
25% on top of a much brighter persistent UV component. We find strong evidence for UV pulsations (pulsefraction of 0 . ± . / X-ray mode changes, but optical modes are inverted compared to the higher frequencies. There appear to betwo broad-band emission components: one from radio to near-infrared / optical that is brighter when the secondcomponent from optical to hard X-rays is dimmer (and vice-versa). We suggest that these components traceswitches between accretion into the neutron star magnetosphere (high-energy high-mode) versus ejection ofmaterial (low-energy high-mode). Lastly, we propose that optical / UV / X-ray pulsations can arise from a shockedaccretion flow channeled by the neutron star’s magnetic field.
Keywords:
Ultraviolet sources–Accretion–Pulsars–Millisecond pulsars–Binary pulsars–Low-mass x-ray binary–Optical pulsars INTRODUCTIONTransitional millisecond pulsars (tMSPs) are a class of neu-tron star binary that has emerged in the last decade with thediscoveries of three systems: PSR J1023 + − − Corresponding author: Amruta [email protected] (also known as M28I; Papitto et al. 2013). These systemsswitch between a radio millisecond pulsar (RMSP) state andanother state, where the system is brighter at most observedwavelengths. In this second state, these systems show an ac-cretion disk, and give the appearance of low-level accretioninto the neutron star magnetosphere. We therefore refer to thisstate as the low-mass X-ray binary (LMXB) state in this paper.However, we caution the reader that the “LMXB state” we a r X i v : . [ a s t r o - ph . H E ] F e b J aodand et al .refer to here is atypical of LMXB systems in general , andIGR J1824 − > erg s − ) state that resembles acanonical LMXB in outburst.Sudden transitions between RMSP and LMXB states oc-cur on a time scale of a few days to weeks, and are accom-panied by drastic changes across the electromagnetic spec-trum. For example, the transition from RMSP to LMXBstate is accompanied by brightening of optical, UV (Papittoet al. 2013; Takata et al. 2014; Patruno et al. 2014), X-rayand γ -ray (Stappers et al. 2013) emissions along with the si-multaneous disappearance of radio pulsations(Patruno et al.2014). What drives these state transitions and the observedphenomena is still debated. Intense multi-wavelength mon-itoring campaigns have been undertaken in the last decadeto understand the rich observational phenomenology in boththe RMSP and LMXB state. PSR J1023 + − − − + d = .
37 kpc; Delleret al. 2012) and prolonged, ongoing LMXB state since.1.1.
X-ray emission
While in the RMSP state, the tMSPs show an orbitally modu-lated X-ray emission with 1 −
10 keV luminosity L X < ergs − (for PSR J1023 + −
15 hr), RMSP binaries with a low-mass( ∼ . . M (cid:12) ), non-degenerate companion (Roberts 2013). Infact, all observed tMSPs so far belong to the redback class ofRMSPs when they are in the radio pulsar state.Upon transition from the RMSP to the LMXB state, theorbitally modulated X-ray emission disappears and is replacedby a much brighter (up to L X ∼ erg s − ) X-ray emission.This X-ray luminosity is still low compared to an LMXB inoutburst, however, and would be considered ‘quiescent’ for acanonical LMXB source (Wijnands et al. 2017). Though on av-erage the X-ray luminosity is stable over multi-year timescales(Archibald et al. 2015; Tendulkar et al. 2014; Jaodand et al. Hence, we also refer to the LMXB state as the “disk state” although the natureand even existence of accretion onto the star in this state remains in doubt.Other papers in the literature have referred to this state as the “sub-luminousaccretion disk state”. (cid:46)
10 s) switches between a steady ‘high’ and ‘low’mode. For PSR J1023 + L X ∼ ergs − and is present ∼
80% of the time, and the low mode is at L X ∼ erg s − and is present for ∼
20% of the time. Thereare also sporadic flares reaching up to L X ∼ erg s − . Co-herent X-ray pulsations at the pulsar’s spin period are observed,but only in the ‘high’ mode (strong upper limits were obtainedin the other modes; Archibald et al. 2015).1.2. γ -ray emission In the RMSP state, the γ -ray emission from tMSPs has beenseen to be pulsed at the spin period of the pulsar (Archibaldet al. 2013; Johnson et al. 2015) – similar to other Fermi -Large Area Telescope (
Fermi -LAT) detected RMSPs. ForPSR J1023 + γ -ray brightness is enhanced by a factorof ∼ Fermi -LAT γ -ray ( ∼ GeV energies) pulsation searches in the LMXB stateare limited by stochastic orbital variations that are not modeledin the rotational ephemeris (Jaodand et al. 2016); this compli-cates the detection of γ -ray pulsations and currently no suchpulsations have been seen in the LMXB state using Fermi -LAT(Jaodand et al., in prep. ). Likewise, γ -ray pulsation searchesusing the Very Energetic Radiation Imaging Telescope ArraySystem ( VERITAS , spanning 50 GeV −
50 TeV; Aliu et al. 2016)have also found no significant γ -ray pulsations above 100 GeV.1.3. UV and Optical emission
Optical emission in the RMSP state is seen to be orbitallymodulated: the optical emission peaks when the companionis at superior conjunction with respect to the pulsar. This isdue to the intra-binary shock and companion surface beingheated by deposition of pulsar wind (de Martino et al. 2013;McConnell et al. 2015). Absorption lines from the companionare also observed and are used to identify the spectral type.In the LMXB state, the optical emission increases and adouble peaked, broad H α emission line shows that an accretiondisk with orbital motion has formed (Wang et al. 2009; Halpernet al. 2013; Patruno et al. 2014). Here, the optical emissioncan be split into two components: i) continuum disk emissionusually modelled with a multi-temperature, geometrically thindisk (Frank et al. 2002), and ii) orbitally modulated opticalemission from the companion’s pulsar-wind-heated surfaceand ellipsoidal modulations (Papitto et al. 2013).Recently, the optical and near-IR emission fromPSR J1023 + ∼ . − ultiwavelength study of PSR J1023 + + ∼ − ) and many di ff erent ionization species,suggestive of an outflow from the system (Hern´andez Santis-teban 2016, Hern´andez Santisteban et al., in prep.). Thesepeculiar UV properties (ionization structure and kinematics)can be explained as being formed in a magnetically drivenoutflow, similar to the one observed in the accreting whitedwarf system AE Aqr (Eracleous & Horne 1996). In addition,the near-UV (and possibly the far-UV) light curves show vari-ability and hints of bi-modality, similar to those present in theX-rays (Archibald et al. 2015; Bogdanov et al. 2015). How-ever, the lack of simultaneous X-ray data prevented us fromreaching any definitive conclusion regarding their correlatednature. 1.4. Radio emission
In the RMSP state, the tMSPs show coherent radio pulsationsat the spin period of the neutron star. In addition, radio eclipsesare seen due to clumpy, intra-binary ionized material (e.g.,Archibald et al. 2013).During the LMXB state, the radio pulsations disappear and,instead, variable, roughly flat-spectrum continuum radio emis-sion, suggestive of a collimated outflow, is observed (Delleret al. 2015). Moreover, using coordinated Very Large Array(VLA) and
Chandra observations, Bogdanov et al. (2018) un-covered an anti-correlation between radio and X-ray brightnessof PSR J1023 + Pulsations at di ff erent wavelengths In the LMXB state, the pulsed radio emission from tM-SPs disappears but Archibald et al. (2015) and Jaodand et al.(2016) have demonstrated the presence of X-ray pulsationsat the neutron star spin period, which are present only in theX-ray high mode. In fast photometry observations, Ambrosino et al. (2017) have found optical pulsations during the highmode in the LMXB state. In addition, Papitto et al. (2019)identify optical pulsations (albeit with reduced flux) duringflares. Using the X-ray pulsations, we recently showed thatPSR J1023 + in prep. ). We suggestedthat the enhanced negative torques on the neutron star couldbe contributed by accretion material being ejected through amechanism like propeller-mode accretion (Papitto & Torres2015), while the pulsar wind remains active. While Archibaldet al. (2015) argued that X-ray pulsations arise from channeledaccretion of the material onto the neutron star surface, Am-brosino et al. (2017) o ff er an alternate explanation in whichsynchrotron emission from the active pulsar wind shocking theaccretion material surrounding it produces the optical photons.Papitto et al. (2019) show that the optical pulsations remainactive during the flares and simultaneous optical and X-ray tim-ing observations, and argue that regions for X-ray and opticalemission should lie within (3 × ( sin ι )) − km or a few kms ofeach other. They also extend the flux power law from X-rayemission to optical to show that the pulsations must arise inthe same region. In this work, we show a slightly di ff erentconclusion. 1.6. Multi-wavelength campaigns
The wealth of observational information from tMSPs in theRMSP and LMXB states enables us to weave together variousmulti-wavelength observations in order to construct a concretepicture of the long-lived, sub-luminous, accretion-dominatedLMXB state in the tMSPs.Previously, these key campaigns have compared joint vari-ability with some combination of X-ray, optical, NIR and / orradio observations such as in de Martino et al. (2013); Takataet al. (2014); Bassa et al. (2014); Bogdanov et al. (2018); Shah-baz et al. (2018); Coti Zelati et al. (2018); Papitto et al. (2019).For instance, Bassa et al. (2014) showed that intense X-rayflares are accompanied by simultaneous UV / optical flaring.Furthermore, the Coti Zelati et al. (2018) campaign with X-rayMulti-Mirror ( XMM-Newton ) and Nuclear Spectroscopic Tele-scope Array (
NuSTAR ) observations showed that the soft andhard X-rays are perfectly correlated with no lags. Moreover,Shahbaz et al. (2015) showed the presence of possible jointX-ray, UV, and optical mode switching. These results wereascribed to the presence of a hot, clumpy accretion flow inthe inner edges of the accretion disk, which can explain thepeculiar top-hat light curves in X-rays and optical. Later, inShahbaz et al. (2018), this argument was extended with jointoptical and near-IR observations to the near-IR componentbelieved to arise from plasmoids in the hot accretion flow andreprocessing of the optical emission. In Papitto et al. (2019)a strictly simultaneous timing and variability campaign is pre-sented for X-ray and optical emission and precise time lags areobserved between the pulsations indicating a similar region oforigin. J aodand et al .These previous campaigns have highlighted the power of asimultaneous multi-wavelength approach.
However, a simulta-neous global campaign spanning high-time-resolution optical,UV, and soft and hard X-ray emission spanning multiple binaryorbits had not yet been performed . Such a campaign is criticalto understand if both optical and UV emission originate fromreprocessing of X-ray emission or changes at outer edges of theaccretion disk. Finding time lags or leads between X-ray andUV emission can inform the mechanisms driving the persistent,low-level accretion regime in tMSPs. A deeper understandingof the optical and UV (if present) pulsed emission similar tothe X-ray pulsations is also called for.Motivated by previous multi-wavelength results and remain-ing open questions, we obtained simultaneous observationswith
XMM-Newton , NuSTAR , Hubble Space Telescope (
HST )and
Kepler targeting PSR J1023 + OBSERVATIONS AND DATA ANALYSIS2.1.
Observation and data reduction
Here we describe the observations and data reduction processfor each observatory used in our campaign. Table 1 summa-rizes these observations.2.2.
XMM-Newton
The
XMM-Newton observations of PSR J1023 + ff ers a time resolution of 29 µ s by foregoing one imagingdimension. The two EPIC MOS cameras (Turner et al. 2001)were employed in “Small Window” mode to minimize thee ff ect of pileup.The X-ray event data were reduced and extracted us-ing the Science Analysis Software (SAS ) package version xmmsas 20180620 1732-17.0.0 . The data were inspectedfor intervals of strong background flaring and none were found.The source photons were extracted from the pn Timing modeobservations from a region of width 6.5 pixels in the imaging(RAWX) direction centered on row 37. This corresponds to anangular size of 27” in the RAWX detector coordinate, whichencircles ∼
87% of the energy from the point source at 1.5keV. The MOS1 / (cid:48)(cid:48) , which encircle ∼
88% of the totalpoint source energy at ∼ The
XMM-Newton
SAS is developed and maintained by the Science Op-erations Centre at the European Space Astronomy Centre and the SurveyScience Centre at the University of Leicester. were 22.7 ks for MOS1, 22.6 ks for MOS2, and 20.7 ks for thepn.To produce the time series X-ray light curve, the data weregrouped in 10 second intervals, and each bin was backgroundsubtracted and exposure corrected using the epiclccor toolin SAS. The resulting light curves from the three detectorswere then stacked to produce a total 0.3-10 keV
XMM-Newton
EPIC light curve. The photon arrival times were barycenteredusing the SAS task barycen assuming the DE405 solar systemephemeris and the pulsar’s astrometric ephemeris presented inDeller et al. (2012).The
XMM-Newton
Optical Monitor (Mason et al. 2001) wasused with the B filter, which has a band pass of 3800 − B filter OMdata were produced using the SAS omfchain pipeline tool usingthe default set of parameters and a 10 s time binning.2.3. HST
NUV observations were taken with the Space TelescopeImaging Spectrograph (STIS) on-board
HST in two visits. Thefirst visit consisted of 3 orbits (orbits a-c hereafter), for a totalexposure time of 8.1 ks, under the DDT program 13630. Thesecond 4 orbits — which were observed in the multiwavelengthcampaign — (orbits 1-4 hereafter), consisted of 8.7 ks totalexposure time, under the DDT program 14934. All observa-tions were taken in TIME-TAG mode with a time resolution of125 µ s with the MAMA detector. We used the G230L gratingand a 52 × R ∼
500 inthe range of 1570 − calstis pipeline within iraf/stsdas .The event files were barycentered using the odelaytime rou-tine. We further processed the event files to assign individualwavelengths with a custom routine and select those eventsbetween 1800 and 2800 Å in order to avoid any emission lines.In addition, we selected events within 10 pixels of the center ofthe slit to minimize contributions from the rest of the detector.We note that while the relative time-stamp accuracy of photonswith STIS within an observation is accurate to the nominalvalue, the absolute timing can di ff er between ∼ − HST helpdesk priv. comm.). Therefore, any relative phasing di ff er-ences between HST and other observatories in timescales lowerthan (cid:46)
NuSTAR
The
Nuclear Spectroscopic Telescope Array ( NuSTAR ; Har-rison et al. 2013) is the first focusing hard X-ray telescope. Itconsists of a pair of co-aligned telescopes with grazing incidentoptics to focus X-rays onto Focal Plane Modules (FPMA andFPMB) that record photons with energies between 3 and 79keV. https: // github.com / Alymantara / stis photons ultiwavelength study of PSR J1023 + Table 1.
Log of observationsFacility Wavelength (nm) Orbit Start BMJD End BMJD Exposure Comments
HST
HST
NuSTAR
XMM-Newton
Kepler . . . . . . N o r m a li ze d c o un tr a t e ( c t s / s ) Optical/UV lightcurves
XMM OMHSTKepler .
65 57917 .
70 57917 .
75 57917 .
80 57917 . . . . . . . N o r m a li ze d c o un tr a t e X-ray lightcurves
NuSTARXMM
Figure 1.
Multi-wavelength light curves of PSR J1023 + Top panel: optical and UV light curves from
Kepler (400 −
800 nm),
XMM-Newton
OM (B filter; 390 −
490 nm) and
HST (180 −
280 nm).
Bottom panel:
X-ray light curves from
XMM-Newton (0 . −
10 keV) and
NuSTAR (3 −
79 keV). The vertical shaded regionsindicate X-ray high (blue), low (grey) and flare (red) modes, as derived from the
XMM-Newton light curve. Unclassified spans are left with a whitebackground; these are either regions with no
XMM-Newton coverage, or where we chose not to classify because of poor overlap with
HST . Following a DDT request, PSR J1023 + NuSTAR for 55 ks during a time stretch starting on 2017June 13 at 12:06 UT and ending on June 14 at 04:01 UT. Weused nuproducts , incorporated in HEASOFT (v. 6.23), toextract data products. A circular region with a radius of 60 (cid:48)(cid:48) was used to extract source events, whereas a region of similardimensions, placed on the same chip away from the source,was used to extract background events. Initial inspection of the FMPA / FMPB X-ray spectrum, com-bined using addascaspec , showed that PSR J1023 + ∼
30 keV.Light curves, with di ff erent time bins, were therefore ex-tracted using an energy range of 3–30 keV. The NuSTAR lightcurves were barycentered using barycorr , using the mostup-to-date clock file at the time of our observations (nuC- J aodand et al .clock20100101v072.fits) and the astrometric ephemeris ofPSR J1023 + KeplerKepler / K2 observed PSR J1023 + + ff ectedby Moire Pattern Drift noise. On account of this extended point-spread-function of PSR J1023 + + + K2 mask is shown in theirFigure 2, as well as the broad-band power spectral density forPSR J1023 + DATA ANALYSIS3.1.
Timestamp verification and mode classification
After having performed the data reduction for the varioustelescopes as outlined above, we obtained a list of valid pho-tons with their arrival times in modified julian date (MJD) ormission elapsed time, and energy associated with each photon.Each of the telescopes assigns photon timestamps in a di ff eringmanner. In order to ensure the uniformity of time conventionswe chose to convert timestamps in all the observation to MJDformat with barycentric dynamical time (TDB) scale.To obtain the light curves presented in Fig.1 we binned the HST and
NuSTAR datasets at 10 s. Here the bin edges, andobservation start and end time limits were dictated by the
XMM-Newton observation. The
Kepler dataset is received in lightcurve form, with bin sizes set by the exposure and readout timelimit to 59 s. Although Fig. 1 and Fig.2 use
Kepler datasetwith 59 s binning, the start and end times for this light curveare also obtained from
XMM-Newton . For all other figures,wherever we have used the
Kepler dataset we re-binned the
XMM-Newton , HST and
NuSTAR datasets according to
Kepler bin edges and while still setting the time limits with
XMM-Newton .We used only the
XMM-Newton light curve to classify timesinto low, high, or flare modes. Fig. 1 shows that the flares areconfined to the end of the
XMM-Newton observation. We there-fore classified all times after MJD 57917.773 as flare-mode,and all times between MJD 57917.765 and MJD 57917.773 asambiguous. For the remaining times (before MJD 57917.765),we used the exposure-corrected
XMM-Newton light curve tocreate a histogram of the count rates in a manner similar to Fig. A1 in Archibald et al. (2015). Using this histogram, we chose acount rate limit of 2.67 s − to classify the observed count ratesinto low and high modes. We group these times into a list oflow, high, and flare intervals. These XMM-Newton based timeclassifications are used as mode markers for
HST , Kepler and
NuSTAR observations. These markers serve as the basis for themodes indicated in various plots in this work: e.g., the varyingbackground colors corresponding to a specific mode in Fig. 1.3.1.1.
Mode stacking
We use the above mode classification to identify transitionedges — to the nearest light curve bin — from low-to-high andhigh-to-low modes. For every transition edge we gather datafrom the ‘before’ and ‘after’ modes, up to a maximum lengthof 300 s in each direction. Next, we stack these data-chunksaround transition edge, averaging all bins that are a commontime before or after a mode transition (Fig. 2).3.2.
Auto and cross correlation analysis
To compute the cross-correlation between two time series S ( t ) and S ( t ) (for autocorrelation we choose S = S ), weuse the auxiliary time series V ( t ) and V ( t ), which are onewhere there is valid flare-mode data at time t in the correspond-ing time series, and zero elsewhere (including past the end ofeach time series). For each delay τ we define the usable datadomain D ( τ ) = { t : V ( t ) V ( t + τ ) (cid:44) } , that is, the values of t where both (shifted) time series are valid. We then computethe correlation function asCorr( τ ) = (cid:80) t ∈ D ( τ ) S ( t ) S ( t + τ ) (cid:113)(cid:80) t ∈ D ( τ ) S ( t ) (cid:113)(cid:80) t ∈ D ( τ ) S ( t ) , where τ is the lag between the two series. This computation hasthe property that correlating a series with a positively scaledcopy of itself gives Corr( τ ) = Pulsation Search
To conduct the UV pulsation search analysis we roughlyfollow the same recipe as outlined in § § § XMM-Newton observations obtained in2017 (Jaodand et al., in prep ). We use this ephemeris and fitfor the time of ascending node ( T asc ) to model the orbit in the XMM-Newton observation. Note that this analysis is only con-ducted on data corresponding to times when PSR J1023 + HST data, we searched a range of three seconds aroundthe
XMM-Newton -derived (true) T asc value to find a T asc valuefor the HST data that compensates for the clock o ff set. Theclock o ff set (as indicated by the dip in Fig. 3) for HST at the ultiwavelength study of
PSR J1023 + − − −
100 0 100 200 300Time (sec)0246810 A v e r ag c o un tr a t e s ( c t s / s ) XMM Low to high
Average − − −
100 0 100 200 300Time (sec)024681012 A v e r ag c o un tr a t e s ( c t s / s ) XMM High to low
Average − − −
100 0 100 200 300Time (sec)405060708090 A v e r ag c o un tr a t e s ( c t s / s ) HST Low to high
Average − − −
100 0 100 200 300Time (sec)4050607080 A v e r ag c o un tr a t e s ( c t s / s ) HST High to low
Average − − −
100 0 100 200 300Time (sec)10001200140016001800 A v e r ag c o un tr a t e s ( c t s / s ) Kepler Low to high
Average − − −
100 0 100 200 300Time (sec)10001200140016001800 A v e r ag c o un tr a t e s ( c t s / s ) Kepler High to low
Average
Figure 2.
Stacked mode plots showing how the count rate changes (on average) during the low-to-high (left column) and high-to-low (rightcolumn) transition for
XMM-Newton (top; 0 . −
10 keV),
HST (middle; 180 −
280 nm) and
Kepler (bottom; 400 −
800 nm). The modal classificationand transition is based on the
XMM-Newton light curve and applied to all simultaneous data from
HST and
Kepler . In each panel, the light greybins show the count rates from individual segments of the light curves; the black bins are a weighted average of these. These are shown withrespect to the time before or after a mode transition, which is also marked with a vertical blue line. Note that when
XMM-Newton and
HST are in alow mode,
Kepler appears to be in a high mode, and vice versa. time of this observation appears to have been − . ff set betweenthe HST and
XMM-Newton data.The ephemerides obtained for
XMM-Newton and
HST withadjusted T asc values (T asc , xmm = − . asc , hst = − . RESULTS We do not expect this o ff set to change significantly between Hubble orbits. The simultaneity between instruments spanning from1.55 eV to 79 keV (800 nm to 0.015 nm) has allowed usto detect correlated changes in the broadband behavior ofPSR J1023 + / high modes, flaring and pulsations. Thoughthe XMM-Newton
OM provided contemporaneous data (Fig. 1),these data are of lower sensitivity and largely redundant withthe wavelength coverage provided by
Kepler ; hence, we do notdiscuss them in detail below. Note that brightness variation inthe optical bands is also a ff ected by the ellipsoidal modulationof the companion star (the sinusoidal modulation visible in the Kepler and
XMM-Newton
OM light curves shown in Fig. 1).However, this is on significantly longer timescales comparedto the few-minute timescale moding / flaring and millisecondpulsations that are the main focus of this present work. J aodand et al . − − − − − − T asc (s) − − − − − l og F PP XMM − − − − − − T asc (s) − . − . − . − . − . − . − . . l og F PP HST
Figure 3. H scores, expressed as false-positive probabilities (FPP), as a function of the time of ascending node T asc compared to a reference orbitalephemeris. Left: XMM-Newton shows a clear optimization at T asc , xmm = − . Right:
The corresponding optimization for
HST where, because the absolute timing is only accurate to within ∼ − XMM-Newton . This value T asc , hst = − . Low and high modes
The well-established switches between X-ray low and highmodes (e.g. Bogdanov et al. 2015) are clearly detected in the
XMM-Newton (0 . −
10 keV) light curve (Fig. 1 and Fig. 4)and appear to extend through to the hard X-ray band probed by
NuSTAR (3 −
79 keV). This has previously been established byTendulkar et al. (2014) and Coti Zelati et al. (2018), and thuswe do not investigate it in more detail here.In the first two
HST orbits, where there is no flaring (Fig. 4),we find clear evidence for mode switching in UV (180 −
280 nm;Fig. 1) occurring synchronously with the X-ray mode switches(Fig. 2). The mode switching is less obvious in
HST data com-pared to the X-ray data (Fig. 5): there is only a ∼
25% averageincrease in count rate from low to the high mode, comparedto the ∼ XMM-Newton . Thenon-moding component in UV is further discussed in §5.The simultaneity with X-ray data has also allowed us toidentify subtle mode switching in the optical to NIR band(400 −
800 nm) provided by
Kepler (Fig. 2). The mode switchesappear to be simultaneous with the X-ray transitions; however,the low time resolution of the
Kepler data complicates thiscomparison and reduces the sensitivity to short lags betweenthe two bands. Most importantly, we find that the
Kepleroptical to NIR signal changes in the opposite direction to the X-ray and UV : when an (X-ray) low mode begins, the optical / NIRincreases by ∼
10% and vice versa (Fig. 2). Because the X-ray data provides the most well-established and most easilyobserved mode switches, we continue to call periods when theX-rays are steady and bright “high modes” and those when theX-rays are steady but faint “low modes”.4.2.
Flares
The observing campaign detected a flaring period lastingfor at least ∼ . + XMM-Newton provides ∼ . Kepler , though
HST and
NuSTAR also clearly show flaring (Fig. 1). The flaring pe-riod is characterized by quasi-periodic flares (Fig. 6), such thatthe flares occur simultaneously in the optical, UV and X-raybands (Fig. 7) and have similar durations of ∼ s on average(Fig. 8) . On the shortest timescales, however, it is clear thatno tight correlation between X-ray and UV flare brightness ispresent (Fig. 5). 4.3. Pulsations
After performing a search in T asc to create a local orbitalephemeris and to correct for the large uncertainty in the HST absolute timing (Fig. 3), we detect pulsations in both the 0 . . The X-raypulsations are detected with 10.3 σ (single-trial) significanceand have a pulse fraction of ∼ σ (single-trial) significance and have a muchlower pulse fraction of 0 . ± . In principle, it may be possible to detect X-ray pulsations in the
NuSTAR data as well, but this is complicated by clock variations as a function of thesatellite’s orbital phase (Tendulkar et al. 2014). ultiwavelength study of
PSR J1023 + .
640 57917 .
645 57917 .
650 57917 .
655 57917 . − . . . . . . . N o r m a li s e d c o un tr a t e HST orbit 1
XMMHST .
705 57917 .
710 57917 .
715 57917 .
720 57917 . − . . . . . . . N o r m a li s e d c o un tr a t e HST orbit 2
XMMHST .
770 57917 .
775 57917 .
780 57917 .
785 57917 . − . . . . . . . N o r m a li s e d c o un tr a t e HST orbit 3
XMMHST .
835 57917 .
840 57917 .
845 57917 .
850 57917 . − . . . . . . . N o r m a li s e d c o un tr a t e HST orbit 4
XMMHST
Figure 4.
Multi-wavelength light curve of PSR J1023 + HST (180 −
280 nm) and
XMM-Newton (0 . −
10 keV). Here wecompare these light curves for each
HST orbit separately. Orbits 1 and 2 show high and low moding behavior, whereas orbits 3 and 4 are dominatedby flaring.
Unfortunately, the absolute alignment between the two bandsis unknown because of the
HST absolute timing accuracy.Given the limited statistics during the low mode, and the highbackground during the flares, it is not possible to constrainwhether the UV pulsations turn on / o ff in sync with the moding— as has been demonstrated for the X-ray pulsations (Archibaldet al. 2015). DISCUSSIONExamining a power budget for the PSR J1023 + . × erg s − Jaodand et al. 2016), the companion’sluminosity ( ∼ . × erg s − , ∼
2% of the neutron-star spin- down; Shahbaz et al. 2019), and the accretion flow. We donot know the amount of mass transferred or its ultimate fate,so estimating the power available from accretion is very di ffi -cult. Nevertheless we can compare these power sources withthe observed luminosities in di ff erent parts of the spectrum.The largest luminosity we observe is in the γ -ray (0 . × erg s − Stappers et al. 2014). In X-rays (0 . ff er-ent luminosities depending on mode: in the high mode 4.2%(3 × erg s − ) and in the low mode 0.7% (5 × erg s − )of the spin-down power (Bogdanov et al. 2015). Finally the op-tical luminosity is about 1.4% (10 erg s − ) of the spin-downpower (Ambrosino et al. 2017). In particular take note of theX-ray emission: matter accreting in the vicinity of the neutronstar can be expected to emit most of its energy in the X-ray, so0 J aodand et al .
100 200 300 400HST (cts/s)010203040 X MM ( c t s / s ) X MM ( c t s / s ) Figure 5.
Comparison of strictly simultaneous
XMM-Newton (0 . −
10 keV) and
HST (180 −
280 nm) count rates. Each point is from an individual10-s bin. The colors represent the low (blue), high (orange) and flare (green) modes, as derived from classifying the
XMM-Newton light curve.
Left: the full count range;
Right: zoom-in on the low and high modes. Though the flares produce a wide range of count rates, the low and high modesshow a more reproducible behavior for both
XMM-Newton and
HST . .
77 57917 .
78 57917 .
79 57917 .
80 57917 .
81 57917 .
82 57917 .
83 57917 . . . . . . . N o r m a li s e d c o un tr a t e ( c t s / s ) Flare region - XMM Kepler
KeplerXMM
Figure 6.
Comparison of X-ray and optical light curves of PSR J1023 + XMM-Newton (0 . −
10 keV)and
Kepler (400 −
800 nm). the fact that the X-ray luminosity of PSR J1023 + some role.During this multi-telescope campaign we observed oneepisode of flaring from PSR J1023 + + XMM-Newton and
XMM-Newton -OM observations in Bassaet al. (2014); Bogdanov et al. (2015); Jaodand et al. (2016) con-firm that every optical flare has a corresponding simultaneous X-ray flare. In contrast, radio flares have been seen to occurwith no corresponding X-ray flare (Bogdanov et al. 2018).Leaving aside the time during which flares occur, we clearlydemonstrate that the X-ray low and high modes have counter-parts in the UV emission. This mode switching in UV happensin the same direction as that in the X-rays: in X-ray high modesthe UV is brighter than in X-ray low modes. We also see modeswitching in the
Kepler data (visible light) but it happens inthe opposite direction: in X-ray high modes the visible light isfainter than in the X-ray low modes.Finally, since we know that X-ray pulsations are only presentduring the high modes, we folded the
HST -UV data dur-ing the high mode by using the established X-ray timingephemeris for PSR J1023 + + HST -UV data set. ultiwavelength study of
PSR J1023 + − − . − . . . . Figure 7.
Cross-correlation of the
XMM-Newton (0 . −
10 keV) and
Kepler (400 −
800 nm) light curves during the flaring mode. Beforecross-correlation, the
XMM-Newton data was binned to the same ∼
59 s resolution as
Kepler . In this discussion, we focus on the broadband mode switch-ing and pulsations from PSR J1023 + Flaring
In PSR J1023 + (cid:38)
10 hrs(Tendulkar et al. 2014). Bogdanov et al. (2015) see occasionalintense optical and UV flares that occur in conjunction withX-ray flares of both types. In contrast, radio flaring behaviouris less directly connected to X-ray flaring, showing a variety ofphenomena such as radio flares during X-ray lows, radio flaresaccompanying X-ray flares and isolated radio flares (Bogdanovet al. 2018). We also note that Kennedy et al. (2018) show thatthe fraction of time spent in optical flaring behaviour seems tovary on a timescale of months.Given this variety, it is not even clear whether there is asingle physical mechanism or emission site that explains allthe flaring behavior. Furthermore, the observing campaignpresented in this work provides only one sampling of a flaringepisode, which may not be representative of flares in general.In this one flaring episode, we do find that the substructure issimilar in shape between the three bands (Fig. 8) and e ff ectivelysimultaneous (Fig. 7).Although previous cross-correlation analysis for flare-onlyregions (Bogdanov et al. 2015) have returned a variety of di ff er-ent lags and leads between X-ray and optical flares, Bogdanovet al. (2015) shows a number of flare light curves in X-ray andB-band (430 nm) where the flares appear to be simultaneous.Shahbaz et al. (2015) seem to have observed the NIR laggingbehind the optical by 300 s in non-flaring times, and Hakala & Kajava (2018) show optical (white light) and NIR (J-band,1200 nm) flare light curves where the flares also appear to be si-multaneous. We point out that Bogdanov et al. (2018) observedsome flares in the radio (8–12 GHz) that had no counterpart inthe X-rays (and others that had). Thus it seems plausible thatsome but not all flares are simultaneous across a broad energyrange.The presence or absence of coherent pulsations might shedlight on the origin of the flare mode, but the observational situ-ation is not simple. Archibald et al. (2015) did not detect X-raypulsations in the flare mode, and in fact were able to show thatthe absolute intensity of the pulsations must decrease duringflares. Papitto et al. (2019) detect optical pulsations duringthe flare mode, but report that the intensity of these pulsationsdecreases by a factor of a few during these modes. We areunable to detect ultraviolet pulsations during the flare modes,but given the increased background and decreased signal, thisis consistent with the idea that the pulsation mechanism isthe same in all three bands. One possibility is that the flaresare produced by a mechanism independent from the high-lowmode switching, that is, when flares are occurring the high-lowmode switching continues undisturbed, but some broad-bandemission process also occurs. The decreased pulsed flux dur-ing flares compared to the high mode might be explained byour inability to remove low modes that occur during flaring,though as low modes only occupy ∼
20% of the time this isnot su ffi cient to explain the size of the pulsed flux decreaseobserved by Papitto et al. (2019).Given that we have only a single example drawn from thediverse population of flares, we cannot usefully address anyexplanation for the multi-wavelength flaring activity. We referthe reader to Bogdanov et al. (2015, 2018); Papitto et al. (2018)for discussions on this topic.5.2. Moding
Unlike the erratic flaring, what we observe during the X-raylow and high mode times is more likely to be representative ofan ongoing low-level accretion phenomenon in tMSPs. Themoding behaviour has displayed remarkably consistent phe-nomenology throughout the LMXB state. The mode switchesare rapid ( ∼ few seconds) and top-hat, steady light curves per-sist in an extremely stable manner on multi-year timescales(Archibald et al. 2015; Bogdanov et al. 2015; Papitto et al.2013; Jaodand et al. 2016). The modes do appear to directlyconnect to pulsations as the X-ray pulsations turn on and o ff with the switches between high and low modes, respectively(Archibald et al. 2015). This suggests that the mode switchesa ff ect matter flow in the immediate vicinity of the neutron star.5.2.1. Broadband spectral behaviour
Previous studies using simultaneous, multi-wavelength ob-servations have tried to investigate if, i) the moding behaviourextends to other frequency bands, and ii) if it is correlated withX-ray moding (e.g., Bogdanov et al. 2015, 2018; Shahbaz et al.2 J aodand et al . − − − − . . . − − − − . . . − − − − . . . − − − − . . . Figure 8.
Auto-correlations for
Kepler (400 −
800 nm),
HST (180 −
280 nm) and
XMM-Newton (0 . −
10 keV) light curves during the flare mode.The bottom panel is an overlay of all three auto-correlations, and demonstrates that the flare durations are similar. ff ort has been invested in dissectingwhat truly is a multiwavelength problem. However, most ofthese studies have been limited to at most a few simultane-ous bands. Given the individual complexities of the system, asingle coherent picture has been elusive. Coupled with thesestudies, our campaign demonstrates that synchronized mod-ing is observable across the electromagnetic spectrum, from ∼
10 GHz radio frequencies up to at least ∼
80 keV hard X-rayenergies.The stability of the high / low mode dichotomy inPSR J1023 + The high-energy power-law which extends from hard X-raysdown to NIR, has now been confirmed by the present study, which shows UV, soft and hard X-rays to mode synchronously.Previous broadband studies had suggested that the X-ray power-law component should extend down to at least the UV bandin order to explain the excess observed in the photometricdata (Baglio et al. 2016; Coti Zelati et al. 2018; Papitto et al.2019). This was further confirmed by the SED modelling ofquasi-simultaneous
HST and VLT / X-Shooter data (Hern´andezSantisteban 2016, Hern´andez Santisteban et al., in prep). Wehave extended this work in Fig. 10, adding an extrapolatedpower-law component as measured in the X-rays (using a pho-ton index
Γ = . ± .
01 and
Γ = . ± .
05 for high- andlow-mode respectively; Bogdanov et al. 2015). It is remarkablethat the HST moding fraction ( ∼ (cid:46)
140 nm, Hern´andez Santisteban ultiwavelength study of
PSR J1023 + Arbitrary Phase (1 cycle)
HSTXMM
Figure 9.
Background-subtracted pulse profiles for
XMM-Newton and
HST observations folded using an extended timing solution basedon Jaodand et al. (2016) and adjusting for the local ∆ T asc measure-ments. An arbitrary vertical o ff set has been added to each profile sothat they do not overlap. Pulsations are detected at 10.3 σ and 3.2 σ significance for XMM-Newton and
HST , respectively. Though theyshow a similar pulse profile structure, we caution that the absolutephasing is unknown because of the inaccuracy in the absolute timestamp of the
HST data. The smooth curves plotted above the his-tograms are obtained directly from photon phases (for more details,see Archibald et al. 2015). γ -ray photons makes it impossible to determineif the moding behaviour extends to the GeV energies (Maat-man et al., in prep ). However, we know that overall emissionextends to GeV ( γ -rays), with an apparent crossover betweenhard X-ray power law and γ -ray power law at ∼ −
10 MeVTendulkar et al. (2014). At the lower energies, however Pa-pitto et al. (2019, see their Fig. 5) surprisingly showed NIR K − band photometry with simultaneous XMM-Newton whereboth bands present correlated behaviour. This shows then thesame spectral component spans from hard X-rays to NIR.However, around the peak of the optical range of the SED(500-600 nm), we observe an anti-correlation between the X-rays. Previous studies already showed evidence of this fact, asseen in the bi-modal distribution of fluxes in r (cid:48) -band (612 nm)is inverted in reference to the X-rays (Shahbaz et al. 2018) and in the overall analysis of the Kepler light curve, wherethey show the same inverted flux distribution (Kennedy et al.2018). We argue that these inverted distributions are in fact In Shahbaz et al. (2018), the modes are referred to as passive- and active-states, which are fainter and brighter respectively in both bands. We suggestthat their passive states correspond to our (X-ray-defined) high modes, andtheir active states correspond to our low modes. the anti-correlated mode switches, that we observe to changesimultaneously (albeit in opposite direction) with the X-raysas presented in this study (see Figure. 2).As we go to wavelengths longer than NIR bands, the high-energy power-law has to turn over in order for the bright flat-spectrum radio emission to show anti-correlated behaviourduring X-ray moding (Bogdanov et al. 2018). Whether this ra-dio component and the optical component are part of the samemechanism or not, it is intriguing that they show a similarbehaviour. In other low-mass X-ray binaries, jet emission canextend and peak in the optical / NIR both in high and low (qui-escent) accretion rates, similar to those in PSR J1023 + ff erent sites, as suggested by Shahbazet al. (2018) where the optical emission arises from reprocess-ing emission and the NIR from synchrotron emission in theoutflow .Finally, we note that measuring moding behaviour in UV,optical, and NIR is somewhat complicated by the presence ofemission from the companion and accretion disk, which aremodulated at the orbital period and heavily dilutes the mode-related signal. Further observations at longer wavelengthssuch as far / mid-infrared and sub-mm might provide additionalevidence to the origin of this component.5.3. Pulsations
Coherent pulsations from PSR J1023 + . −
10 keV Archibaldet al. 2015), optical (320 −
900 nm Ambrosino et al. 2017),and UV (180 −
280 nm). In all three cases, the pulse profileis double-peaked, that is, it is dominated by the first overtone.Ambrosino et al. (2017) argue, based on model-derived spectralestimates, that the optical pulsations must be produced bya di ff erent physical process than the X-ray, but our resultssuggest a di ff erent interpretation: the pulsed fluxes in optical,UV, and X-ray bands all lie on a line (see, Fig.10) with slopeof 1 . ± .
6, which is consistent with the X-ray high-modepower-law index of 1 . ± .
05. It is therefore plausible that thepulsations in these three bands are produced by a single process,and in fact that process also produces the mode switching.The link to mode switching is strengthened by the result ofArchibald et al. (2015) showing that X-ray pulsations disappearin both flare and low modes. We do not have the statisticsto verify such a disappearance for the UV pulsations, andAmbrosino et al. (2017) do not have the mode classificationsto verify this for the optical. Even stronger evidence for abroad-band emission process could be provided by showingphase alignm ent between optical or UV pulsations and X-ray,but due to clock limitations aboard
HST this would requirecareful combination of a ground-based optical observation likethat of Ambrosino et al. (2017) with a quasi-simultaneous
XMM-Newton observation to constrain the pulsar ephemeris.A search for γ -ray pulsations would help explore the limits of4 J aodand et al . Frequency / Hz10 − − − − − − − − − ν F ν / e r g s − c m − VLT+HST spectrumDonorAccretion discShabhaz+18 modingXMM High-modeXMM Low-mode VLA high modeVLA low modeXMM PulseHST modingOptical PulseHST Pulse − − − − Energy / keV 10 Wavelength / nm
Figure 10.
The de-reddened broadband spectral energy distribution of PSR J1023 + the broad-band nature of the pulsation mechanism. However, apossible spectral break in the overall emission near 1 −
10 MeV(Tendulkar et al. 2014) constrains the pulsations to also havea break at some similar energy. It is thus unclear whether toexpect γ -ray pulsations (Jaodand et al., in prep ) or what theirpulsed fraction should be, though sampling this very di ff erentregime would be a valuable physical probe.5.4. Mechanisms
In spite of the broad range of observational data onPSR J1023 + ultiwavelength study of PSR J1023 + ff ected. Thismodel suggests that the enhanced X-rays in this mode comefrom bombardment of the inner edge of the accretion flowby the pulsar wind (which in the low modes would alreadybe divided by the conductive material in the light cylinder).The pulsations would then be produced by the inner regionsof the pulsar wind sweeping around the inner edge of theaccretion flow. This mechanism is di ffi cult in a number of ways.Very few models of non-aligned rotating pulsars launchingwinds exist, and the physics is quite uncertain even beforeincluding the interactions with an accretion flow near to thelight cylinder. In particular it is far from established that theenergy flux — whether leptons rest mass, bulk kinetic energy,thermal, or Poynting — should be strongly focused near the magnetic equator, as needed to produce two hotspots on theinner face of the accretion flow. More, if the inner edge of themoves in or out, the pulses should be advanced or retardedby the light travel time. The extreme stability of the pulsephase was essential to Jaodand et al. (2016) — aside fromthe orbital variations, which have a very di ff erent signature,on long timescales the pulses drift by less than 5 µ s, whichmeans that the inner edge of the disk cannot move by morethan about 1 . + ff ect the spin-down rate of the pulsar. Allthis argues that even in isolation pulsar magnetospheres canhave multiple quasi-stable configurations and can abruptly— on second timescales — switch between them. Such be-haviour was not observed, in spite of long-term monitoring, in PSR J1023 + ffi cult-to-explain behaviour of PSR J1023 + + + CONCLUSIONSWe have established that the mode switching exhibited byPSR J1023 + aodand et al .cylinder and possibly down to the neutron star surface (inflow),and low modes occur when instead matter is ejected in a pos-sibly collimated outflow. This view explains why in the highmode we see the appearance of pulsations and high-energybrightness. This view also explains why in the low mode wesee faint high energy emission, the disappearance of pulsations,and the emergence of flat-spectrum radio emission as wellas increased NIR emission. The flare mode mechanism weleave unexplained. We note that the mode switches happentoo rapidly to be directly controlled by variations in the masstransfer from the companion; turbulent variations in the diskare a more likely direct cause. Precisely what changes whenthe switch from inflow to outflow occurs is not obvious; wediscuss three scenarios.First, a simplistic toy model can explain the mode switchingwith an accretion disk terminated near the co-rotation radiusby magnetospheric pressure. When the disk extends insidethe corotation radius, coupling with the magnetic field canthen lead to channeled accretion onto the neutron-star sur-face, producing a structured inflow. When the ram pressuredrops, the inner edge of the disk recedes outside the corotationradius, and propeller-mode accretion takes over, stopping ac-cretion onto the surface and replacing it with an outflow. Thismodel is simple but the observed X-ray luminosities indicatefar too little material is entering the light cylinder to producethe needed ram pressure to reach corotation in the high mode.The toy model of propeller-mode accretion is also too simplis-tic to work theoretically; magnetohydrodynamic simulations(Romanova et al. 2004) actually predict mixed inflows andoutflows.Another means of switching between inflow and outflow isby having an accretion disk terminate near the light cylinder.Recently, Parfrey & Tchekhovskoy (2017) put forth resultsfrom the first ever fully-relativistic MHD simulation wherethey model magnetized accretion onto millisecond pulsarswith a force-free magnetosphere. They are able to show fourdistinct states depending on the accretion rate. This includesan intermediate state where the accretion disk is usually insidethe light cylinder radius but at other times recedes and leads tosynchrotron radiation from inflowing material being expelledin a classical ‘propeller’ ejection. This model improves uponthe simplistic model by requiring much smaller amounts ofmaterial to reach the neutron-star surface. However, here themodelling is limited to ∼ HST observations suggest that the accretiondisk may actually terminate at ∼ × the light cylinder ra-dius (Hern´andez Santisteban 2016, Hern´andez Santisteban etal., in prep). At this distance, the light crossing time is ∼ + ∼ × the light cylinder radius andthe low modes arise from pulsar-wind-driven outflows.ACKNOWLEDGMENTSWe thank Sergio Campana, James Miller-Jones, Alessan-dro Papitto, Rudy Wijnands, Jakob van den Eijnden andFrancesco Coti Zelati for interesting discussions related toPSR J1023 + / / ERC Starting Grant agreement nr. 337062(“DRAGNET”). JWTH also acknowledges funding from anNWO Vidi fellowship. JVHS and ND are supported by a Vidigrant from NWO, awarded to ND. SB was funded in part byNASA through grant number HST-GO-14934.002-A from theSpace Telescope Science Institute (STScI), which is operatedby the Association of Universities for Reasearch in Astronomy,Inc., under NASA contract NAS 5-26555. The authors aregrateful to Fiona Harrison and the
NuSTAR team for makingthe DDT observations of PSR J1023 + / ESAHubble Space Telescope. This research has made use of dataand software provided by the High Energy Astrophysics Sci-ence Archive Research Center (HEASARC), which is a serviceof the Astrophysics Science Division at NASA / GSFC and theHigh Energy Astrophysics Division of the Smithsonian As-trophysical Observatory. This work has made extensive useof NASA’s Astrophysics Data System Bibliographic Services(ADS) and the arXiv.
Facilities:
HST , Kepler , NuSTAR , XMM-Newton ultiwavelength study of
PSR J1023 + Software:
Astropy (Astropy Collaboration et al. 2018),
Matplotlib (Hunter 2007),
Seaborn (Jones et al. 2001a) and
Scipy (Jones et al. 2001b)REFERENCES
Aliu, E., Archambault, S., Archer, A., et al. 2016, ApJ, 831, 193,doi: 10.3847 / / / / / s41550-017-0266-2Archibald, A. M., Kaspi, V. M., Bogdanov, S., et al. 2010, ApJ, 722,88, doi: 10.1088 / / / / // arxiv.org / abs / / / / / / / aabc4fBacker, D. C. 1970, Nature, 228, 1297, doi: 10.1038 / / / / mnras / stu708Bogdanov, S., Archibald, A. M., Hessels, J. W. T., et al. 2011, ApJ,742, 97, doi: 10.1088 / / / / / / / / / L27Bogdanov, S., Archibald, A. M., Bassa, C., et al. 2015, ApJ, 806, 148,doi: 10.1088 / / / / / / aaaeb9Bucciantini, N. 2011, Astrophysics and Space Science Proceedings,21, 473, doi: 10.1007 / / / / / / / / / / / L25Deller, A. T., Moldon, J., Miller-Jones, J. C. A., et al. 2015, ApJ, 809,13, doi: 10.1088 / / / / // dx.doi.org / / mnras / stx2922Halpern, J. P., Gaidos, E., She ffi eld, A., Price-Whelan, A. M., &Bogdanov, S. 2013, The Astronomer’s Telegram, 5514Harrison, F. A., Craig, W. W., Christensen, F. E., et al. 2013, ApJ,770, 103, doi: 10.1088 / / / / / science.1230960Hermsen, W., Kuiper, L., Basu, R., et al. 2018, MNRAS, 480, 3655,doi: 10.1093 / mnras / sty2075Hern´andez Santisteban, J. V. 2016, PhD thesis, UNIVERSITY OFSOUTHAMPTONHui, C. Y., & Becker, W. 2006, A&A, 448, L13,doi: 10.1051 / / MCSE.2007.55Jaodand, A., Archibald, A. M., Hessels, J. W. T., et al. 2016, ApJ,830, 122, doi: 10.3847 / / / / / / / / // / —. 2001b, SciPy: Open source scientific tools for Python.http: // / Kennedy, M. R., Clark, C. J., Voisin, G., & Breton, R. P. 2018,MNRAS, 477, 1120, doi: 10.1093 / mnras / sty731Lyne, A., Hobbs, G., Kramer, M., Stairs, I., & Stappers, B. 2010,Science, 329, 408, doi: 10.1126 / science.1186683Mason, K. O., Breeveld, A., Much, R., et al. 2001, A&A, 365, L36,doi: 10.1051 / / / / / / / / / / / aabee9Papitto, A., Ambrosino, F., Stella, L., et al. 2019, arXiv e-prints,arXiv:1904.10433. https: // arxiv.org / abs / aodand et al . Parfrey, K., Spitkovsky, A., & Beloborodov, A. M. 2015, ArXive-prints. https: // arxiv.org / abs / / / aa9c85Patruno, A., Archibald, A. M., Hessels, J. W. T., et al. 2014, ApJL,781, L3, doi: 10.1088 / / / / L3Roberts, M. S. E. 2013, in IAU Symposium, Vol. 291, IAUSymposium, ed. J. van Leeuwen, 127–132,doi: 10.1017 / S174392131202337XRomanova, M. M., Ustyugova, G. V., Koldoba, A. V., & Lovelace,R. V. E. 2004, ApJL, 616, L151, doi: 10.1086 / / mnras / sty562Shahbaz, T., Linares, M., Rodr´ıguez-Gil, P., & Casares, J. 2019,MNRAS, 1561, doi: 10.1093 / mnras / stz1652Shahbaz, T., Linares, M., Nevado, S. P., et al. 2015, MNRAS, 453,3461, doi: 10.1093 / mnras / stv1686Stappers, B. W., Archibald, A., Bassa, C., et al. 2013, TheAstronomer’s Telegram, 5513, 1Stappers, B. W., Archibald, A. M., Hessels, J. W. T., et al. 2014, ApJ,790, 39, doi: 10.1088 / / / /
39 Str¨uder, L., Briel, U., Dennerl, K., et al. 2001, A&A, 365, L18,doi: 10.1051 / / / / / / High EnergyAstrophysics Division, Vol. 14, AAS / High Energy AstrophysicsDivision, 122–23Thorstensen, J. R., & Armstrong, E. 2005, AJ, 130, 759,doi: 10.1086 / / // arxiv.org / abs / / / / / //