A Further Drop into Quiescence by the Eclipsing Neutron Star 4U 2129+47
aa r X i v : . [ a s t r o - ph . H E ] O c t A CCEPTED BY A P J 2009, O
CTOBER
Preprint typeset using L A TEX style emulateapj v. 04/20/08
A FURTHER DROP INTO QUIESCENCE BY THE ECLIPSING NEUTRON STAR 4U 2129+47 J INRONG L IN , M ICHAEL
A. N
OWAK , D EEPTO C HAKRABARTY Accepted by ApJ 2009, October
ABSTRACTThe low mass X-ray binary 4U 2129+47 was discovered during a previous X-ray outburst phase and wasclassified as an accretion disk corona source. A 1% delay between two mid-eclipse epochs measured ∼ XMM-Newton observations taken in 2005, providing support to the previoussuggestion that 4U 2129+47 might be in a hierarchical triple system. In this work we present timing andspectral analysis of three recent
XMM-Newton observations of 4U 2129+47, carried out between November2007 and January 2008. We found that absent the two 2005
XMM-Newton observations, all other observationsare consistent with a linear ephemeris with a constant period of 18 857 .
63 s; however, we confirm the time delayreported for the two 2005
XMM-Newton observations. Compared to a
Chandra observation taken in 2000, thesenew observations also confirm the disappearance of the sinusoidal modulation of the lightcurve as reported fromtwo 2005
XMM-Newton observations. We further show that, compared to the
Chandra observation, all of the
XMM-Newton observations have 40% lower 0.5–2 keV absorbed fluxes, and the most recent
XMM-Newton observations have a combined 2–6 keV flux that is nearly 80% lower. Taken as a whole, the timing resultssupport the hypothesis that the system is in a hierarchical triple system (with a third body period of at least 175days). The spectral results raise the question of whether the drop in soft X-ray flux is solely attributable to theloss of the hard X-ray tail (which might be related to the loss of sinusoidal orbital modulation), or is indicativeof further cooling of the quiescent neutron star after cessation of residual, low-level accretion.
Subject headings: accretion, accretion disks – stars:individual(4U 2129+47) – stars:neutron – X-rays:stars INTRODUCTION
4U 2129+47 was discovered to be one of the Accretion DiskCorona (ADC) sources (Forman et al. 1978), which are be-lieved to be near edge-on accreting systems since they haveshown binary orbital modulation via broad, partial V-shape X-ray eclipses (Thorstensen et al. 1979; McClintock et al. 1982,hereafter MC82, White & Holt 1982); the eclipse width was ≈ . ≈
75% of the X-rays were occulted at theeclipse midpoint. The origins of ADCs are not fully under-stood, but they are typically associated with high accretion-rate systems, wherein we are only observing the small frac-tion of the system luminosity that is scattered into our line ofsight. For the prototypical ADC X1822 - .
24 h period (Thorstensen et al. 1979; Ulmer et al. 1980;McClintock et al. 1982; White & Holt 1982). The discoveryof a type-I X-ray burst led to the classification of 4U 2129+47as a neutron star (NS) low mass X-ray binary (LMXB) sys-tem (Garcia & Grindlay 1987), and the companion was sug-gested to be a late K or M spectral type star of ∼ . M J . Thesource distance was estimated to be ∼ ∼ × ergs - . Even ifthe luminosity were 100 times larger, the luminosity wouldhave been somewhat smaller than expected for an ADC. Massachusetts Institute of Technology, Kavli Institute for Astrophysicsand Space Research, Cambridge, MA 02139, USA; [email protected];mnowak,[email protected]
Since 1983, 4U 2129+47 has been in a quiescent state.Optical observations show a flat lightcurve without any ev-idence for orbital modulation between the years 1983 and1987. Additionally, instead of an expected M- or K-typecompanion, the observed spectrum was compatible with alate type F8 IV star (Kaluzny 1988; Chevalier et al. 1989).The refined X-ray source position as determined by
Chan-dra turned out to be coincident with the F star to within0 . ′′ (Nowak, Heinz, & Begelman 2002, hereafter N02). Theprobability of a chance superposition is less than 10 - (seealso Bothwell et al. 2008), therefore, the hypothesis of a fore-ground star is unlikely. A ∼
40 km s - shift in the mean ra-dial velocity was derived from the F star spectrum, provid-ing evidence for a dynamical interaction between the F starand 4U 2129+47 (Cowley & Schmidtke 1990; Bothwell et al.2008). The system was therefore suggested to be a hierar-chical triple system, in which the F star is in a month-longorbit around the binary (Garcia et al. 1989). This hypothesisis tentatively confirmed with two XMM-Newton observationsseparated by 22 days that showed deviations from a simpleorbital ephemeris (Bozzo et al. 2007,hereafter B07). Assum-ing that the F star is part of the system, the source distancewas revised to ∼ . ∼
1% of the averageflux) was a substantial fraction of the Eddington luminosity.The X-ray spectrum of 4U 2129+47, as determined with a
Chandra observation taken in 2000, was consistent with ther-mal emission plus a powerlaw hard tail (N02). The 2–8 keVflux was ≈
40% of the 0.5–2 keV absorbed flux. The contri-bution of the “powerlaw” hard tail to the 0.5–2 keV band was,of course, model dependent, but was consistent with being ≈
20% of the 0.5–2 keV flux, as we discuss below. A sinu-soidal orbital modulation (peak-to-peak amplitude of ± C oun t s / s C oun t s / s F IG . 1.— a) Left: Lightcurve from one of the 2007 XMM-Newton observations, fit by a total eclipse with finite duration ingress and egress. No sinusoidal orbitalmodulation was evident. The 90% confidence level upper limit of the modulation is 10%. b) Right: Lightcurve of six eclipses folded as one. The mid-eclipseepochs are aligned and centered at 2250 s.
The phase resolved spectra showed variation in the hydrogencolumn density. However, in the
XMM-Newton observations(B07), the sinusoidal modulation was absent, while the flux ofthe powerlaw component was constrained to be less than 10%of the 0 . -
10 keV flux (B07).Here we report on recent
XMM-Newton observations of4U 2129+47. We combine analyses of these observations withreanalyses of the previous observations, and show that, absentthe two 2005
XMM-Newton observations (B07), the histori-cal observations are consistent with a linear ephemeris with aconstant period. We outline our data reduction procedure in§2 and present the timing and spectral analysis in §3 and §4respectively. We summarize our conclusions in §5. OBSERVATIONS AND DATA
XMM Newton observed 4U 2129+47 on Nov 29, Dec 20,2007 and on Jan 04, Jan 18, 2008. The total time span foreach of these 4 observations was 43270 sec, resulting in ef-fective exposure times of at least 30 ksec for the EPIC-PN,EPIC-MOS1, and EPIC-MOS2 cameras for the first three ob-servations. The remaining observing time was discarded dueto the background flares. The standard
XMM Newton
Sci-ence Analysis System (SAS 8.0) was used to process the ob-servation data files (ODFs) and to produce calibrated eventlists. We used the EMPROC task for the two EPIC-MOScameras and used the EPPROC task for the EPIC-PN cam-era. We used the high energy (E >
10 keV) lightcurves todetermine the Good Time Intervals, so as to obtain the eventlists that were not affected by background flares. The GoodTime Intervals are slightly different for EPIC-PN and EPIC-MOS cameras; therefore, the overlap Good Time Interval wasused to filter each lightcurve with the EVSELECT keywords“timemin" and “timemax".The low energy (0 . - . . -
12 keV) were extracted within the circles of14 . ′′ radius centered on the source. Larger circles were notused in order to avoid a Digital Sky Survey stellar object (la-beled as S3 - β in N02). We extracted background lightcurvesand spectra within the same CCD as the source region. Thelargest source-free regions near the source, which were withincircles of radii about 116 ′′ , were chosen for background ex-traction. The SAS BACKSCALE task was performed to cal-culate the difference in extraction areas between source andbackground. In order to obtain the mid-eclipse epochs, wetried various bin sizes (75 s, 150 s and 300 s etc.) to obtainthe best compromise between signal to noise ratio in a bin, and time resolution. Finally the lightcurves were extracted withthe bin size of 150 s. The times of all lightcurves were cor-rected to the barycenter of the Solar System by using the SASBARYCEN task. Lightcurves from EPIC-PN and two EPIC-MOS cameras were summed up with the LCMATH task, inorder to get better statistics in the fitting of the eclipse param-eters. Given the low count rate of the EPIC-MOS camerascompared to the EPIC-PN camera, for the spectral analysiswe discuss only the spectrum from the EPIC-PN camera. ORBITAL EPHEMERIS AND ECLIPSE PARAMETERS
The observation made on Jan 18, 2008 was seriously af-fected by background flares, however, 2 eclipses were foundin each of the other 3 observations respectively, resulting in 6eclipses for fitting.The 3 pairs of eclipses were then simultaneously fit withthe same set of eclipse parameters. The ingress duration, theegress duration, and the duration of the eclipse (defined to befrom the beginning of the ingress to the end of the egress)were set to be free parameters and assumed to be the samefor all 6 eclipses, under the hypothesis that these eclipsesshould have the same shape (see §5). The starting time ofthe first eclipse ingress and the count rates within and out ofthe eclipse were fit for each lightcurve individually. The sep-aration between the two connected eclipses within the samelightcurve was set to be 18857.63 s based upon the previ-ous
Chandra results. The second eclipse in each lightcurvetherefore only contributed to the fitting of the eclipse shapes,while only the first eclipse in each lightcurve has been used toestimate the orbital ephemeris. The mid-eclipse epoch wasdefined to be the starting time of the ingress plus half ofthe eclipse full ingress-to-egress duration. The model wasintegrated over each time bin, especially the bins that arecrossed by the ingress and the egress, in order to be com-patible with either the finite ingress/egress durations or the in-finite ingress/egress slopes (for a rectangular eclipse model). χ minimization was performed in order to determine the bestfit to the lightcurve.One of our fitted lightcurves is shown in Fig. 1a. Thebest fits to the mid-eclipse epochs were found to be T (1) = 2454 433 . ± . T (2) = 2454 455 . ± . T (3) = 2454 470 . ± . χ / d.o.f.=782 . /
708 (errors are at 68% confidence level).The best fit result favors a rectangular eclipse model withvery short duration of the ingress and egress durations; upperlimits of 50 s and 30 s were found for the ingress and egresshe Eclipsing Neutron Star 4U 2129+47 3 R e s i du a l ( S ec ) R e s i du a l ( S ec ) F IG . 2.— Each square shows the timing residual (defined as the difference between the observed ephemeris and the predicted ephemeris) at certain numberof orbits from the observations listed in Table. 1. a) Left: The dashed line denotes a linear ephemeris with a period of 18857 .
63, while the solid line denotes aquadratic ephemeris. Absent the 2005
XMM-Newton observations, the ephemeris of the eclipses can be well fitted by a constant period. However, we confirmedthe strong deviations of the 2005
XMM-Newton observations from any linear or quadratic ephemeris. b) Right: Close up of the third body fit with a third bodyperiod of ∼
175 days, showing just the
Chandra and
XMM-Newton data since year 2000. The solid line denotes a sinusoidal ephemeris.TABLE 1M ID - ECLIPSE EPOCH MEASUREMENTS . Observatory
Mid-eclipse Epoch Orbital Period References(JD) (sec)
Einstein
Chandra
XMM-Newton a a XMM-Newton a a XMM-Newton
XMM-Newton
XMM-Newton
OTE . — Numbers in parentheses are the errors (at 1 σ level) on thelast significant digit. a We reprocessed the data for these two observations using a 150 sbinned lightcurves in order to be consistent with our routine. durations respectively. A folded lightcurve of 6 eclipses isshown in Fig. 1b, where the mid-eclipse epochs are alignedand centered. The sinusoidal variation seen in the
Chan-dra observation is absent; the upper limit of the modula-tion amplitude is <
10% at 90% confidence level. The du-ration of the eclipses were derived to be 1565 ±
23 s. Wealso re-analyzed the previous two
XMM-Newton observationsof May 15 and June 6, 2005 with the above methods in or-der to get a direct comparison with our data sets. Thesetwo lightcurves were also extracted with 150 s bins, and therewas one eclipse present in each of them. The best fit mid-eclipse epochs are T ( a ) = 2453 506 . ± . T ( b ) = 2453 528 . ± . χ / d.o.f.=18 . /
21 and87 . /
84, respectively.We considered the above mid-eclipse epochs together withepochs, T n , derived from the previous observations (Table 1)in order to determine a refined orbital solution and to mea-sure any orbital period derivative (which was weakly sug-gested by the analysis of N02). We considered the ephemerisfrom MC82 as reference ( T re f = 2444 403 . ± .
002 JD, P re f = 18 857 . ± .
07 s), and calculated n , the closest in-teger to ( T n - T re f ) / P re f . Our three observations (the firsteclipse in each of the three lightcurves) correspond to n = 45 955 ,
46 055 ,
46 121. The average orbital periods ( P =( T n - T re f ) / n ) inferred from these observations are therefore18857 . ± .
005 s, 18857 . ± .
005 s and 18857 . ± .
004 s. We noticed that, absent the 2005
XMM-Newton ob-servations, the 2007/2008 observations are consistent with alinear ephemeris with a constant period of 18 857 .
63 s. How-ever, we confirmed the strong deviations of the 2005
XMM-Newton observations from any linear or quadratic ephemeris(Fig. 2). We therefore tried to fit the linear timing residualby a third body orbit on top of a steady binary orbital pe-riod. For simplicity, we assume the interaction between a bi-nary system and a third body results in a sinusoidal residual.The ephemeris is not uniquely determined; we found validephemerides for the third body orbit at ∼ ∼
175 days(Fig. 2). SPECTRAL ANALYSIS
The spectra of the 3 observations were accumulated duringthe same time intervals selected for the extraction of the EPIC-PN lightcurves, except that the eclipses were also excluded.Spectral analysis was carried out using ISIS version 1.4.9-55(Houck & Denicola 2000). Since there is no evidence of or-bital modulation in the lightcurves, the spectra we producedwere not phase resolved. Furthermore, preliminary fits in-dicated little or no variability among spectra from individ-ual observations; therefore, we fit all spectra simultaneously.Specifically, we used the ISIS combine_datasets func-tionality to sum the spectra and responses during analysis.The data were binned to have a minimum signal-to-noise of This is essentially equivalent to summing the pulse height analysis (PHA)files and summing the product of the response matrix and effective area files.Whereas this can be accomplished, e.g., using
FTOOLS outside of the analy-sis program (and in fact, “outside of analysis” is the only mode supported by
XSPEC ), performing summing during analysis allows one to examine dataand residuals for each data set individually, choose different binning andnoticing criteria based upon individual spectra, etc. To be explicit, here and throughout we mean that the source counts, ex-cluding the estimated background counts, divided by the estimated error, in-cluding that of the background counts, meets or exceeds the given signal-to-noise threshold in each spectral bin.
Lin, Nowak, & Chakrabarty − − F n ( e r g s c m − s − k e V − ) − c Energy (keV) − − F n ( e r g s c m − s − k e V − ) − c Energy (keV) − − F n ( e r g s c m − s − k e V − ) − c Energy (keV) F IG . 3.— Combined spectra for the Chandra observations of Nowak, Heinz, & Begelman (2002) (left), the
XMM-Newton observations of Bozzo et al. (2007)(middle), and the three
XMM-Newton observations presented in this work (right). Each has been fit with a model consisting of an absorbed blackbody pluspowerlaw; however, the photon index is fixed to Γ = 2 for the XMM-Newton observations. Note that all spectra have been unfolded without reference to theunderlying fitted spectral model . (The unfolded spectra do, however, reference the response matrices of the detectors; see Nowak, et al. 2005 and the Appendixfor further details.) ≈ Γ = 2 when including such a component in thefits. Without the additional powerlaw, the best fit temperatureand column density were 0 . + . - . keV and (0 . ± . × cm - , with a χ = 62 . /
73 degrees of freedom. With theadditional powerlaw, the best fit temperature and column den-sity were 0 . + . - . keV and (0 . ± . × cm - , with a χ = 57 . /
73 degrees of freedom. Both of these sets of pa-rameters are consistent with the thermal component of the fitspresented by N02 for the peak (i.e., least absorbed part) of thesinusoidal modulation of the
Chandra lightcurve. We showthe spectra and fit including the powerlaw in Fig. 3.As indicated by these fits (all of which have reduced χ < H and its associated error bars, while slightly de-creasing the best fit temperature. Our best fit models, withor without a powerlaw, yield an absorbed 0.5–2 keV flux of(7 . ± . × - ergs cm - s - , with no more than 10%,i.e., 0 . × - ergs cm - s - (90% confidence level) beingattributable to a Γ = 2 powerlaw. This 0.5–2 keV flux is 36%lower than the (1 . ± . × - ergs cm - s - found duringthe peaks of the Chandra lightcurve (see below). Note thatabove, and throughout this work, unless stated otherwise wewill quote 68% confidence limits for fluxes, but 90% confi-dence limits for fit parameters.To estimate the 2–6 keV flux, we grouped the spectrum(starting at 2 keV) to have a minimum signal-to-noise of 3and a minimum of 2 channels per grouped energy bin, andthen we fit an unabsorbed powerlaw spectrum between 2–6 keV. (Grouping to higher signal-to-noise left no channelsbetween ≈ . ± × - ergs cm - s - in the 2–6 keV band, which is to be com- pared to the nearly five times larger 2–6 keV flux of (2 . ± . × - ergs cm - s - from the Chandra observations (seebelow).The question then arises as to when the drop in the 0.5–2 keV flux occurred, and whether or not it is solely attributableto the loss of the hard X-ray tail. To further explore these is-sues, we applied absorbed blackbody plus powerlaw fits to thespectra described by B07 (out of eclipse data only, groupedand noticed exactly as for the 2007/2008
XMM-Newton spec-tra described above, with powerlaw index frozen to Γ = 2), andrefit the spectra of N02 (powerlaw index left unfrozen). Weused the same exact data extractions and spectral files fromN02. For the Chandra observations, we follow N02 and fitthe 0.3–2 keV spectra (grouped to signal-to-noise of 4.5 at0.3 keV and above) from the peak of the lightcurve’s sinu-soidal modulation, while we fit the 2–6 keV spectra (groupedto a signal-to-noise of 4 at 2 keV and above) from all of theout of eclipse times. The flux levels quoted above correspondto these new fits of the
Chandra data.The fits to the data from B07 are completely consistent withthe fits to the new data, and yield a blackbody temperature andneutral column of 0 . + . - . keV and 0 . + . - . × cm - , re-spectively, with χ = 32 . /
37 degrees of freedom. The power-law normalization was consistent with 0, and thus excluded apowerlaw from the fits to these data. The absorbed 0.5–2 keVflux was (7 . ± . × - ergs cm - s - , i.e., comparable tothat from the more recent XMM-Newton data. The spectra areshown in Fig. 3.As discussed in N02, and shown in Fig. 3, the
Chan-dra data clearly indicate the presence of a hard tail. Herewe find N H = 0 . + . - . × cm - , kT = 0 . + . - . keV, and Γ = 2 . + . - . , with χ = 10 . /
17 degrees of freedom. Note thatthe larger N H value found here compared to N02 is due tothe inclusion of the 0.3–0.5 keV data, and is partly indica-tive of a systematic dependence of this parameter upon theindex of the fitted powerlaw. For the best-fit powerlaw in-dex, 0 . × - ergs cm - s - of the 0.5–2 keV absorbed fluxis attributable to the powerlaw. This is not enough to ac-count for the drop in flux between the Chandra and
XMM-Newton observations. However, if for the
Chandra spectrawe fix the powerlaw at its 90% confidence level upper limit( Γ = 3 . . × - ergs cm - s - of the 0.5–2 keV ab-sorbed flux is attributable to the powerlaw component (albeithe Eclipsing Neutron Star 4U 2129+47 5with an N h = 0 . × cm - , which is higher than the best fitvalues for the XMM-Newton data). This is more than enoughto account for all of the change in the 0.5–2 keV absorbedflux, and highlights some of the systematic uncertainties in-herent in determining bolometric flux changes (i.e., the needto adequately model the changes in the local column and touse realistic models for the hard X-ray tail). With these cur-rent data, it is difficult to distinguish between resumed coolingof the neutron star thermal component, or mere loss of the ad-ditional (presumed external) hard tail component that here ismodeled with a powerlaw. . . F BBody (0.5−10 keV)/F
Best Fit (0.5−10 keV) F P L ( . − k e V ) / F B e s t F it ( . − k e V ) F IG . 4.— 68%, 90%, and 99% confidence contours for the unabsorbed x -axis) and powerlaw ( y -axis) componentsfrom fits to the 0.3–6 keV spectra from the 2007/2008 XMM-Newton obser-vations (lower left) and
Chandra observation (upper right). The dashed con-fidence contours correspond to fits to the 0.6–6 keV band of the
Chandra spectra. Plus signs indicate the “best fit” values. Note that all flux valuesare shown relative to the total, unabsorbed flux in the 0.5–10 keV band de-termined from the best blackbody plus powerlaw fit to the 2007/2008
XMM-Newton observations.
To highlight some of the issues with determining bolomet-ric flux changes and with attributing any changes to specificmodel components, in Fig. 4 we show error contours forthe unabsorbed
XMM-Newton spectra and the
Chandra spectra. These errors account foruncertainties in the neutral column, but do not address sys-tematic issues with the choice of model itself. We see thatformally the 99% confidence level contours do not overlap ,which would indicate that both the thermal and powerlawcomponents have decreased between the time of the
Chandra and
XMM-Newton observations. However, if we ignore thefirst bin (0.3–0.6 keV) in the
Chandra spectra, the
Chandra contours shift significantly leftward towards lower blackbodyflux. (Additionally, the fitted neutral column also decreases.)There is less of a shift downward in powerlaw flux betweenthese two fits of the
Chandra data . Thus, as stated above, itremains somewhat ambiguous as to what extent the changesbetween the
Chandra and
XMM-Newton data can be attributedsolely to components other than the thermal emission from theneutron star.We have fit also the
XMM-Newton spectra with the neutronstar atmosphere (
NSA ) model of Zavlin, Pavlov, & Shibanov(1996), by fixing the neutron star distance to 6 . M NS (M O. ) R ¥ N S ( k m ) F IG . 5.— 68%, 90%, and 99% confidence contours for the neutron star mass( x -axis) and radius ( y -axis) from neutron star atmosphere ( NSATMOS ) modelfits to the 0.3–6 keV spectra from the 2007/2008
XMM-Newton observations.The neutron star distance was fixed to 6.3 kpc. The plus sign correspondsto the best fit. The jagged contours at the bottom of the figure represent thelimits of the interpolation grids used in the calculations of the NSATMOSmodel. neutron star mass to 1.4 M ⊙ . Results of these fits are pre-sented in Table 2. They are consistent with those of N02 andB07 (although N02 also required the inclusion of a power-law component). Specifically, we find neutron star radii of ≈ × K.If we instead use the
NSATMOS model of Heinke et al. (2006),again fixing the neutron star mass and distance as above, wefind lower neutron star temperatures ( ≈ K) and signifi-cantly larger radii ( >
12 km). This raises the question of thedegree to which one can find a consistent fitted mass and ra-dius among all the datasets discussed here.To explore this question, we have performed a joint fit of the
Chandra and
XMM-Newton spectra. The observations fromthe individual observing epochs (2000, 2005, 2007/2008)were grouped and added as described above, with the addi-tional caveat that we now exclude the
Chandra data below0.6 keV so as to minimize the influence of the soft-end of thepowerlaw component on the fitted neutral column. We includea hard tail, modeled as a Γ = 2 powerlaw, in fits to the Chan-dra spectra, and again use the
NSATMOS model (Heinke et al.2006) to describe the soft spectra from all epochs. For thislatter component we again fix the distance to 6.3 kpc and wefurther constrain the neutron star mass and radius to be thesame for all epochs. The individual epochs, however, are fit-ted with independent neutron star temperatures and neutralcolumns. Results are presented in Table 3.We find that one can find a set of consistent parameters thatare statistically acceptable. There is a modest need for the
Chandra spectra to be described with a slightly larger neutralcolumn than that fit to the 2007/2008
XMM-Newton spectra.The
Chandra spectra also require a slightly higher neutronstar temperature, but here all temperatures fall within eachothers error bars. The best fit mass is 1.66 M ⊙ and the best fitradius is 5.55 km. This is somewhat smaller than the valuespresented in Table 2 when applying the NSATMOS model witha fixed mass of 1.4 M ⊙ , and is more consistent with the valuesfound when using the NSA model.The reasons for this can be elucidated by examining the er- Lin, Nowak, & Chakrabartyror contours for the fitted mass and radius. In Fig. 5 we showthe mass/radius contours obtained from applying an absorbed
NSATMOS model to just the 2007/2008
XMM-Newton data(i.e., our best measured spectrum, with no discernible hardtail). We see that the error contours admit a wide range ofmasses and radii, with the curvature of these contours gen-erally favoring two regimes: small radius (with high tem-perature - not shown in this figure) and large radius (withlower temperture). Formally, the small radius/high temper-ature solution is the statistical minimum; however, it is notstatistically very different than the large radius/low tempera-ture regime. (Comparable contours for the
NSA model lookvery similar, albeit with the contours shifted to lower massvalues.)These different fit parameter regimes add further systematicuncertainty to estimates of the unabsorbed, 0.5-10 keV flux.As shown in Tables 2 and 3, the estimate of this flux rangesfrom 1.5–1.9 × - erg cm - s - for the 2007/2008 XMM-Newton data alone. Furthermore, we see that depending uponwhether we use a phenomenological model (the blackbodyplus powerlaw fits) individually fit to each epoch, or a morephysical model (e.g.,
NSATMOS ) jointly fit to all epochs, weeither find evidence for further neutron star cooling or evi-dence of a consistent neutron star surface temperature. Weagain caution, however, that determining the bolometric fluxin specific model components can be fraught with many sys-tematic uncertainties (e.g., Fig. 4), and we therefore considerany evidence for further neutron star cooling to be ambiguousat best. CONCLUSIONS
We reported on recent
XMM-Newton observations of 4U2129+47 in a quiescent state. Given lower background flar-ing rates and relatively longer durations of these observationscompared to the 2005
XMM-Newton observations reported byBozzo et al. (2007), we were able to make clear the followingthree points:1. The 0.5–2 keV absorbed X-ray flux of the source hasbeen reduced by more than ≈
40% compared to the
Chandra observation. This was even true for the ob-servations discussed by B07.2. The sinusoidal variation seen in the
Chandra obser-vation ( ±
30% peak-to-peak) is absent, or at leastgreatly diminished (90% confidence level upper lim-its of ±
10% modulation). Additionally, the associ-ated neutral column exhibits no orbital variability, andis consistent with the brightest/least absorbed orbitalphases of the
Chandra observation.3. The powerlaw tail seen in the
Chandra observation isabsent. A Γ = 2 powerlaw contributes less than 10% ofthe 0.5–2 keV flux, and is not evident in the 2 - Chandra observation.With this further drop in the X-ray flux of the source, wemay now truly be seeing 4U 2129+47 enter into a quiescentstage solely dominated by neutron star cooling, with little orno contribution from residual weak accretion. The spectrumof the source can be well explained by absorbed blackbodyemission from the surface of a neutron star, with an emissionradius of ∼ . Chandra observation could have been due to a neu-tral hydrogen column being raised by the interaction of theaccretion stream with the outer edge of an accretion disk(Parmar et al. 2000; Heinz & Nowak 2001). The fact that thesinusoidal modulation was absent in the 2007, and likely alsothe 2005
XMM-Newton observations could suggest that thesystem is in a lower accretion state, and might indicate thatthe geometry of the outer disk region has changed. It hasbeen hypothesized that the hard tails of quiescent neutronstars originate from a pulsar wind, only able to turn on in qui-escence, that is interacting with the accretion stream at largeradius (Campana et al. 1998; Campana & Stella 2000). Thevanishing of both the hard tail and the sinusoidal modulationin 4U 2129+47 could suggest that what we had observed in the
Chandra observation was indeed the interaction between thepulsar wind and the disk edge, perhaps as the last remnantsof a disk accreted onto the neutron star. This is consistentwith the fact that for these current
XMM-Newton observationsthe fitted neutral column is ≈ × cm - , which is com-parable to the value measured for the least absorbed/peak ofthe Chandra lightcurve. We note that a dominant nonthermalcomponent has been reported in a “recycled” system: the bi-nary radio millisecond pulsar PSR J0024 - - Chandra observationsto the
XMM-Newton observations, there is no evidence for adecline between the 2005 and 2007/2008 observations. Theinitial drop of the soft X-ray flux is consistent with solely theloss of the hard tail for some parameter regimes of this compo-nent; however, for the best-fit parameters, additional coolingwould have had to occur.We note that the expected luminosity of 4U 2129+47due to neutron star cooling has recently been discussed byHeinke et al. (2009). Even though 4U 2129+47 is among thebrighter of the 23 quiescent sources presented by Heinke et al.(2009), the degree to which it is “too faint”, and thus requires“non-standard” cooling (kaon cooling, pion cooling, etc.; seeHeinke et al. 2009 and references therein) relies on the ratheruncertain average heating history of 4U 2129+47, and theknowledge that the bulk of the observed soft X-ray emissionwas not due to residual accretion. We note that any impliedreduction of the soft X-ray emission of 4U 2129+47, or anycontinued cooling observed in the future, increases the needto invoke “non-standard” cooling mechanisms for this system,and further limits the parameter space for the long term aver-age heating that would allow for “standard” cooling.A question also arises as to whether or not the loss ofthe hard X-ray tail is in any way related to the unusualtiming residuals associated with the 2005
XMM-Newton ob-servations (B07). Strong deviations from a simple linearor quadratic ephemeris have also been noted for the burst-ing neutron star system EXO 0748 -
676 (Wolff et al. 2009),he Eclipsing Neutron Star 4U 2129+47 7with these changes possibly being correlated with the du-ration of the eclipse as viewed by the
Rossi X-ray TimingExplorer (RXTE). (Note that comparable eclipse durationchanges, <
20 s, observed in EXO 0748 -
676 are too smallto have been seen in 4U 2129+47 with our observations.)Wolff et al. (2009) hypothesize that these eclipse duration andephemeris changes are related to magnetic activity within thesecondary. We note, however, that over a comparable fiveyear span, the peak-to-peak residuals of EXO 0748 -
676 (70 s)are three to four times smaller than the peak-to-peak residualsseen in 4U 2129+47 (200–300 s). Furthermore, the trend ofephemeris timing residuals of EXO 0748 -
676 are more per-sistently in one direction (although deviations do occur) thanthose that we show in Fig. 2 (compare to Fig. 4 of Wolff et al.2009).In our work, we confirmed the strong deviations of two2005 XMM-Newton eclipses from any linear or quadraticephemeris, however, this deviation has only been observed once since 1982. Rather than presuming ephemeris residu-als comparable to those of EXO 0748 - REFERENCESBlandford, R. D., & Begelman, M. C. 1999, MNRAS, 303, L1Davis, J. 2001, ApJ, 548, 1010Bogdanov, S., Grindlay, J. E., & van den Berg, M. 2005, ApJ, 630, 1029Bothwell, M.S., Torres, M.A.P., Garcia, M.R., & Charles, P.A. 2008, A&A,485, 773Bozzo, E., Falanga, M., Papitto, A., et al. 2007, A&A, 476, 301 (B07)Campana, S., Stella, L., Mereghetti, S., et al. 1998, ApJ, 499, L65Campana, S., & Stella, L. 2000, ApJ, 541, 849Chevalier, C., Ilovaisky, S. A., Motch, C., et al. 1989, A&A, 217, 108Cottam, J., Sako, M., Kahn, S.M., Paerels, F., & Liedahl, D.A. 2001, ApJ,557, L101.Cowley, A. P., & Schmidtke, P C. 1990, AJ, 99, 678Esin, A. A., McClintock, J. E., & Narayan, R. 1997, ApJ, 489, 865Forman, W., Jones, C., Cominsky, L., et al. 1978, ApJ, 38, 357Garcia, M. R., & Grindlay, J. E. 1987, ApJ, 313, L59Garcia, M. R., Bailyn, C.D., Grindlay J. E., & Molnar, L. A. 1989, ApJ, 341,L75Heinke, C.O., Rybicki, G.B., Narayan, R., & Grindlay, J.E. 2006, ApJ, 644,1090Heinke, C.O., Jonker, P.G., Wijnands, R., Deloye, C.J., & Taam, R.E. 2009,ApJ, 691, 1035 Heinz, S., & Nowak, M. A. 2001, MNRAS, 320, 249Houck, J.C., & Denicola, L.A. 2000, in ASP Conf. Ser. 216: AstronomicalData Analsis Software and Systems IX, Vol. 9, 591Horne, K., Verbunt, F., & Schneider, D. P. 1986, MNRAS, 218, 63Kaluzny, J. 1988, Acta Astron., 38, 207McClintock, J. E., London, R. A., Bond, H. E., & Grauer, A. D. 1982, ApJ,258, 245 (MC82)Nowak, M. A., Heinz, S., & Begelman, M.C. 2002, ApJ, 573, 778 (N02)Nowak, M. A., Wilms, J., Heinz, S., Pooley, G., Pottschmidt, K, & Corbel,S. 2005, ApJ, 626, 1006.Parmar, A. N., Oosterbroek, T., Del Sordo, S., et al. 2000, A&A, 356, 175Thorstensen, J., Charles, P., Bowyer, S., et al. 1979, ApJ, 233, L57White, N.E., & Holt, S.S. 1982, ApJ, 257, 318.Wilms, J., Allen, A., & McCray, R. 2000, ApJ, 542, 914.Wolff, M.T., Ray, P.S., Wood, K.S., & Hertz, P.L. 2009, ApJ, in press.Ulmer, M. P., Shulman, S., Yentis, D., et al. 1980, ApJ, 235, L159Zavlin, V.E., Pavlov, G.G., & Shibanov, Y.A. 1996, A&A, 315, 141. L i n , N o w a k , & C h a k r a b a r t y TABLE 2B
EST - FIT SPECTRAL PARAMETERS
Parameters
XMM-Newton (2007/2008) XMM-Newton (2005) Chandra(no powerlaw) a (no powerlaw) b ( Γ = 2, fixed) (no powerlaw) (no powerlaw) (no powerlaw) a (no powerlaw) b ( Γ free)N H (10 cm - ) 0 . + . - . . ± .
03 0 . ± .
03 0 . ± .
03 0 . + . - . . + . - . . ± .
07 0 . + . - . kT bb (keV) 0 . + . - . . + . - . · · · · · · . + . - . · · · · · · . + . - . R bb (km) c . ± . . + . - . · · · · · · . + . - . · · · · · · . + . - . A PL (10 - γ / keV / cm / sec) d · · · . + . - . · · · · · · · · · · · · · · · . + . - . Γ · · · · · · · · · · · · · · · + . - . log T ef f (log K) · · · · · · . ± .
06 6 . ± . · · · . + . - . . ± . · · · R ∞ NS (km) e · · · · · · . + . - . . + . - . · · · . + . - . . + . - . · · · χ /d.o.f. 62.6/73 57.8/72 56.0/73 55.8/73 32.2/37 33.1/37 32.6/37 10.7/17 F . - (10 - erg cm - s - ) 7 . ± . . ± . . ± . . ± . . ± . . ± . . ± . ± F . -
10 keVf (10 - erg cm - s - ) 8 . ± . . ± . . ± . . ± . . ± . . ± . . ± . ± F . - (10 - erg cm - s - ) 1 . ± . ± .
05) 1 . ± . ± .
08) 1 . ± . ± .
09) 1 . ± . ± .
09) 1 . ± . ± .
2) 1 . ± . ± .
2) 1 . ± . ± .
2) 2 . ± . + . - . ) F . -
10 keVg (10 - erg cm - s - ) 1 . ± . ± .
05) 1 . ± . ± .
08) 1 . ± . ± .
09) 1 . ± . ± .
09) 1 . ± . ± .
2) 1 . ± . ± .
2) 2 . ± . ± .
2) 3 . ± . + . - . ) N OTE . — Parameter errors are 90% confidence for one interesting parameter (i.e., ∆ χ . a NSA model, assuming a fixed neutron star mass of 1.4M ⊙ b NSATMOS model, assuming a fixed neutron star mass of 1.4M ⊙ c Neutron star radius assuming a distance of 6.3 kpc d Powerlaw normalization at 1keV. e Lower limit constrained to > f Absorbed flux. g Unabsorbed flux.
TABLE 3B
EST - FIT
NSATMOS
SPECTRAL PARAMETERS FOR A JOINT FIT OF THE
Chandra
AND
XMM-Newton
SPECTRA
Parameters
Chandra XMM-Newton(2005) (2007/2008)N H (10 cm - ) 0 . ± .
01 0 . ± .
04 0 . + . - . log T ef f (log K) a . + . - . . + . - . . + . - . A PL (10 - γ / keV / cm / sec) 1 . + . - . · · · · · · M NS (M ⊙ ) 1 . + . - . R NS (km) 5 . + . - . F . -
10 keVb (10 - erg cm - s - ) 3 . ± . ± .
3) 2 . ± . ± .
2) 1 . ± . ± . χ / d.o.f. 99.6/128N OTE . — A Γ ≡ Chandra spectra. The neutronstar distance was fixed to 6.3 kpc. Parameter errors are 90% confidence for one interestingparameter (i.e., ∆ χ = 2 . a Upper bound of the log T ef f is constrained to be < . b Unabsorbed flux. he Eclipsing Neutron Star 4U 2129+47 9
APPENDIXFLUX ERROR BARS
It is impossible to uniquely invert the (imperfectly known) X-ray detector response matrices to yield a completely model-independent, yet accurate, estimate of the detected flux for an observed source. This also leads to difficulties in deriving errorbars for any estimate of the flux. Here we use the fact that the “deconvolved flux” is a reasonably close estimate of the “modelflux” to estimate the error bars on this latter quantity as presented in Table 2. Specifically, we define the “deconvolved photonflux” in Pulse Height Analysis (PHA) channel h as: F ( h ) = C ( h ) - B ( h ) T R R ( E , h ) A ( E ) dE , (A1)where C ( h ) - B ( h ) are the background subtracted counts in channel h , T is the observation integration time, R ( E , h ) is the detectorresponse matrix, and A ( E ) is the detector effective area. (See Davis 2001 for a more in depth discussion of the meanings of theseterms.) This is the same deconvolution used to create Fig. 3, and it is independent of assumed model. The deconvolved energyflux in a given band is then determined by multiplying the above photon flux by the midpoint energy of the PHA bin (determinedfrom the EBOUNDS array of the response matrix), and summing over the channels within the given energy band of interest. Theerror is determined from the sum in quadrature of the error from each non-zero bin, which in turn is determined from the countingstatistics of the source and background.This flux estimate can be compared to the model photon flux, which is determined by integrating the best fit model over theenergy band of interest. For all the models considered in this paper, these two estimates agree to within 3% when applied to the absorbed absorbed - σ bound below that ofthe 0.5–2 keV flux. This might be indicative of unmodeled flux in the 3–10 keV band; however, we lack the statistics to describethe flux in this band with any degree of accuracy. From this point of view, the flux estimate with the least amount of systematicuncertainty is the 0.5–2 keV absorbed flux.When determining the statistical errors on the unabsorbed flux, we further scale the above error estimates by the energy-dependent ratio of the unabsorbed to absorbed model flux. The systematic error bars on the unabsorbed fluxes are then determinedindividually for each spectral model. We freeze the fitted value of the neutral column at its ± σ limits, refit the spectra, and thendetermine the fluxes as described above. The deviations of the unabsorbed fluxes at these two absorption limits compared tothe flux obtained when using the best fit parameter values are assigned as the systematic error bars. Note that such systematicerrors are for a given model, and do not reflect the systematic errors on the unabsorbed fluxes obtained from comparing differentassumed spectral models.For the cases of fitting the spectra with a blackbody and a powerlaw we employ one other estimate of the flux error bars forthe individual model components. ISIS allows one to set any given parameter to be an arbitrary function of any other set offit parameters. It therefore was straightforward to recast the spectral fit from dependence upon blackbody normalization andtemperature to dependence upon unabsorbed 0.5–10 keV blackbody flux and temperature (i.e., the blackbody normalization canbe written as a function of those two parameters). Likewise, the powerlaw fit parameters were recast to depend upon unabsorbed0.5–10 keV flux and powerlaw slope. We thus were able to use direct fitting methods to generate the error contours shown inFig. 4.
APPENDIXFLUX ERROR BARS